Terry A. Slocum, Robert B. McMaster, Fritz C. Kessler, Hugh H. Howard - Thematic Cartography and Geovisualization - Pearson New International Edition-Pearson Education Limited (2013)

Slocum, Terry A., et al.
Thematic Cartography and Geovisualization:

Pearson New International Edition, Pearson Education Limited, 2013. ProQuest Ebook
15:56:40.
Pearson New International Edition
Thematic Cartography
and Geovisualization
Slocum Mcmaster Kessler Howard
Third Edition
15:56:40.
Pearson Education Limited
Edinburgh Gate
Harlow
Essex CM20 2JE
England and Associated Companies throughout the world
Visit us on the World Wide Web at: www.pearsoned.co.uk
© Pearson Education Limited 2014
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without either the
prior written permission of the publisher or a licence permitting restricted copying in the United Kingdom
issued by the Copyright Licensing Agency Ltd, Saffron House, 6–10 Kirby Street, London EC1N 8TS.
All trademarks used herein are the property of their respective owners. The use of any trademark
in this text does not vest in the author or publisher any trademark ownership rights in such
trademarks, nor does the use of such trademarks imply any afﬁliation with or endorsement of this
book by such owners.
ISBN 10: 1-292-04067-X

ISBN 10: 1-269-37450-8
ISBN 13: 978-1-292-04067-7
ISBN 13: 978-1-269-37450-7
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library
Printed in the United States of America
15:56:40.
P E A R S O N C U S T O M L I B R A R Y
Table of Contents
1. Thematic Cartography and Geovisualization

Terry A. Slocum/Robert B. McMaster/Fritz C. Kessler/Hugh H. Howard 1
2. A Historical Perspective on Thematic Cartography
3. Statistical and Graphical Foundation
4. Data Classification
5. Principles of Symbolization
6. Scale and Generalization
7. The Earth and Its Coordinate System
8. Elements of Map Projections
9. Selecting an Appropriate Map Projection
10. Principles of Color
11. Map Elements and Typography
12. Cartographic Design
13. Map Reproduction
I
15:56:53.
14. Choropleth Mapping
15. Isarithmic Mapping
16. Dasymetric Mapping
17. Proportional Symbol and Dot Mapping
18. Multivariate Mapping
19. Cartograms and Flow Maps
20. Visualizing Terrain
21. Map Animation
22. Data Exploration
23. Visualizing Uncertainty
24. Web Mapping
25. Virtual Environments
26. Trends in Research and Development
Color Plates
Index 613
15:56:53.
Thematic Cartography and
Geovisualization
OVERVIEW
This text covers thematic mapping and the associated purpose, (4) design and construct the map, and (5) deter-
expanding area of geographic visualization (or “geovi- mine whether users find the map useful and informative.
sualization”). A thematic map (or statistical map) is Despite some criticism of the appropriateness of such
used to display the spatial pattern of a theme or steps, they are helpful in avoiding the design blunders
attribute. A familiar example is the temperature map that can result from using the most readily available data
shown in daily newspapers; the theme (or attribute) in and software.
this case is the predicted high temperature for the day. Like many disciplines, the field of cartography has
The notion of a thematic map is described in section 1 undergone major technological changes. As recently as
and contrasted with the general-reference map, which the 1970s, most maps were still produced by manual
focuses on geographic location as opposed to spatial pat- and photomechanical methods, whereas today, nearly
tern (e.g., a topographic map might show the location of all maps are produced using computer technology.
rivers). In section 2 the different uses for thematic maps Section 4 considers some of the consequences of this
are described: to provide specific information about technological change, including (1) the ability of virtu-
particular locations, to provide general information ally anyone to create maps using personal computers;
about spatial patterns, and to compare patterns on two (2) new mapping methods, such as animated maps;
or more maps. (3) the ability to explore geographic data in an interac-
An important function of this text is to assist you in tive graphics environment; (4) the ability to link maps,
selecting appropriate techniques for symbolizing spatial text, pictures, video, and sound in multimedia presenta-
data. For example, imagine that you wish to depict the tions; (5) the ability to create realistic representations
amount of forest cleared for agriculture in each country of the environment (virtual environments or virtual
during the preceding year, and have been told that the reality) and the related notion of augmented reality
number of acres of forest cleared by country is available (i.e., enhancing our view of the real world through com-
on the Web. You wonder whether additional data puter-based information); and (6) the ability to access
(e.g., the total acres of land in each country) should be maps and related information via the Web.
collected and how the resulting data should be symbol- In section 5 we consider the origin and definition of
ized. Section 3 presents steps that assist you in tackling geographic visualization. The term “visualization” has
such problems, and ultimately that enable you to com- its roots in scientific visualization, which was devel-
municate the desired information to map readers. These oped outside geography to explore large multivariate
steps are as follows: (1) consider the real-world distribu- data sets, such as those associated with medical imaging,
tion of the phenomenon, (2) determine the purpose for molecular structure, and fluid flows. Borrowing from
making the map, (3) collect data appropriate for the map these ideas, geographers have created the notion of
From Chapter 1 of Thematic Cartography and GeoVisualization, Third Edition. Terry A. Slocum, Robert B. McMaster, Fritz C. Kessler,
Hugh H. Howard. Copyright © 2009 by Pearson Education, Inc. All rights reserved.
1
15:57:04.
Thematic Cartography and Geovisualization
geographic visualization (or geovisualization), which enumeration units (or data-collection units such as
can be defined as a private activity in which unknowns states) are shaded to represent different magnitudes
are revealed in a highly interactive environment. of an attribute (Color Plate 1). A variety of thematic
Communication on traditional printed maps involves maps are possible, including proportional symbol,
the opposite: It is a public activity in which knowns are isarithmic, dot, and flow (see Figure 1). A major pur-
presented in a noninteractive environment. pose of this text is to introduce you to these and other
We want you to be aware of developments in the types of thematic maps, as well as to methods used in
broader realm of geographic information science designing and constructing them.
(GIScience), which can be considered to include cartog- Although cartographers commonly distinguish
raphy and the techniques of geographic information sys- between general-reference and thematic maps, they do
tems (GIS), remote sensing, and quantitative methods. In so largely for the convenience of categorizing maps.
section 6, we consider the increased capability provided The general-reference map also can be viewed as a the-
by GIS and remote sensing, which allow us to create matic map in which multiple attributes are displayed
detailed maps more easily than was possible with man- simultaneously; thus, the general-reference map can be
ual techniques. GIS accomplishes this through its exten- termed a multivariate thematic map. Furthermore,
sive spatial analysis capabilities, and remote sensing although the major emphasis of general-reference
allows us to “sense” the environment, particularly out- maps is on the location of spatial phenomena, they can
side our normal visual capabilities (e.g., detecting previ- also portray the spatial pattern of a particular attribute
sual levels of vegetation stress). The major development (e.g., the pattern of drainage shown on a USGS topo-
in quantitative methods relevant to cartography is that of graphic sheet).
exploratory spatial data analysis (ESDA), which has
close ties with the notion of data exploration that cartog-
raphers utilize. 2 HOW ARE THEMATIC MAPS USED?
While technological advances have had a major
impact on cartography, the discipline has also experi- Thematic maps can be used in three basic ways: to pro-
enced changes in its philosophical outlook. Section 7 vide specific information about particular locations, to
deals with the increasing role that cognition now plays in provide general information about spatial patterns, and
cartography. Traditionally, cartographers approached to compare patterns on two or more maps. As an exam-
mapping with a behaviorist view, in which the human ple of specific information, map A of Color Plate 1 indi-
mind was treated like a black box. Today cartographers cates that between 8.8 and 12.0 percent of the people in
take a cognitive view, in which they hope to learn why Louisiana voted for Ross Perot in the 1992 U.S. presi-
symbols work effectively. Section 8 deals with social and dential election. As another example, Figure 1 indicates
ethical issues in cartography—there we will see that maps that approximately 2 million slaves were transported
often have hidden agendas and meanings, and that our from Africa to Spanish America between 1700 and 1870.
increasing technological capability provides tremendous Obtaining general information requi res an overall analy-
opportunity, but also is fraught with potential problems sis of the map. For example, map B of Color Plate 1 illus-
(e.g., the notion of geoslavery). trates that a low percentage of people voted for Perot in
the southeastern United States, whereas a higher per-
centage voted for him in the central and northwestern
1 WHAT IS A THEMATIC MAP? states; and Figure 1 indicates that the bulk of the slave
trade between 1700 and 1870 occurred outside North
Cartographers commonly distinguish between two America.
types of maps: general-reference and thematic. A pitfall for naive mapmakers is that they often place
General-reference maps are used to emphasize the inordinate emphasis on specific information. Map A of
location of spatial phenomena. For instance, topo- Color Plate 1 is illustrative of this problem. Here one can
graphic maps, such as those produced by the U.S. discriminate the data classes based on strikingly different
Geological Survey (USGS), are general-reference colors and thus determine which class each state belongs
maps. On topographic maps, readers can determine in (as we did for Louisiana), but it is difficult to acquire
the location of streams, roads, houses, and many other general information without carefully examining the leg-
natural and cultural features. Thematic maps (or end. In map B, the reverse is true: Determining class mem-
statistical maps) are used to emphasize the spatial bership is more difficult because the classes are all blue
pattern of one or more geographic attributes shades, but the spatial pattern of voting is readily apparent
(or variables), such as population density, family because there is a logical progression of legend colors.
income, and daily temperature maximums. A common As an illustration of pattern comparison, consider the
thematic map is the choropleth map, in which dot maps of corn and wheat shown in Figure 2. Note that
Slocum, Terry A., et al. Thematic Cartography and Geovisualization:
15:57:04.
NORTH
AMERICA Brit is
hN ASIA
ort
h Am
er
is h Americ
ic
pan a
a
S Oman India
Bri Sahara
tis
hC
ar
F re ib
nc b
Danish hC e Ethiopia
an
Caribbean ari
bb Mogadishu
Du e
tc h a n AFRICA
Ca
SOUTH ribb
e
Angola INDIAN
an Zanzibar
AMERICA OCEAN
Brazil ue
biq
am
PACIFIC
Moz
OCEAN ATLANTIC
OCEAN
Volume of Slave Trade, 1700–1870
Route without
4,000,000 3,000,000 2,000,000 1,000,000 reliable data
FIGURE 1 A flow map: an example of a thematic map. (From BERGMAN, EDWARD F., HUMAN GEOGRAPHY: CULTURE,
CONNECTIONS, AND LANDSCAPES, 1st Edition, © 1995, pg. 167. Reprinted by permission of Pearson Education, Inc., Upper
Saddle River, NJ.)
the patterns on these two maps are quite different. Corn is might mentally correlate the spatial pattern of limestone
concentrated in the traditional Corn Belt region of the in the bedrock with the pattern of limestone houses.
Midwest, whereas wheat is concentrated in the Great A second important issue is that terms other than
Plains, with a less notable focus in the Palouse region of specific and general can be found in the cartographic lit-
eastern Washington. Conventionally, a comparison of pat- erature. We have used specific and general (developed by
terns such as these was limited by their fixed placement on Alan MacEachren 1982b) because they appear frequently
pages of paper atlases, but interactive graphics now allow in the literature. Others have developed a more complex
us to readily compare arbitrarily selected distributions. set of terms. For example, Philip Robertson (1991, 61) dis-
Two further issues are important when considering tinguished among three kinds of information: values at a
the ways in which thematic maps are used. First, one point, local distributions characterized by “gradients
should distinguish between information acquisition and and features,” and the global distribution characterized
memory for mapped information.* Thus far, we have by “trends and structure”; Robertson argued that these
focused on information acquisition, or acquiring infor- levels corresponded closely with Jacques Bertin’s (1981)
mation while the map is being used. We can also elementary, intermediate, and superior levels.†
consider memory for map information and how that
memory is integrated with other spatial information 3 BASIC STEPS FOR COMMUNICATING
(obtained through either maps or fieldwork). For exam- MAP INFORMATION
ple, a cultural geographer might note that houses in a
particular area are built predominantly of limestone.
In this section, we consider basic steps involved in com-
Recalling a geologic map of bedrock, the geographer
municating map information to others. For instance,
imagine that you wish to create a map for a term paper
* Technically, memory for mapped information would be equivalent
to what psychologists term long-term memory, but for simplicity, the † Additional terminology can be found in Olson (1976a) and Board
word memory is normally used. (1984).
3
15:57:04.
Corn for Grain or Seed, 1992
1 Dot: 10,000 Acres

United States Total
69,339,869
Wheat for Grain, 1992
1 Dot: 10,000 Acres

United States Total
59,089,470
FIGURE 2 An illustration of pattern comparison, one of the fundamental ways in which thematic maps are used. (From
U.S. Bureau of the Census 1995.)
15:57:04.
wish to map the distribution of total population in the

United States from the 1990 decennial census.
Step 1. Consider what the real-world distribution of
Step 1 the phenomenon might look like. One way to implement
this step is to ask yourself, “What would the distribution
Consider what the real-world
distribution of the phenomenon of the phenomenon look like if I were to view it while
might look like. traveling across the landscape?” In the case of our pop-
ulation example, you might know (based on your travels
or knowledge as a geographer) that a large percentage
of people are concentrated in major cities and that such
cities are much more densely populated than rural areas.
Step 2 Often, however, it is unrealistic to presume a single
Determine the purpose of the map “objective” real world. In the case of population,
and its intended audience. “correct” population values are unknown for several
reasons. The U.S. Bureau of the Census is never able to
make an exact count of population; after every census,
some city officials dispute the census figures for their
city. Also, census figures do not necessarily count the
Step 3 homeless or illegal aliens (the latter would account for
Collect data appropriate for the a significant percentage of the population in some
map’s purpose. states, such as California). Another problem is that
population obviously varies locationally during the day
and throughout the year—that is, people commute to
work, travel to the beach on weekends, and take vaca-
tions far away from home.* In spite of these problems, it
Step 4 is useful to think of a “real-world” distribution. Such an
approach forces you to think about the distribution at
Design and construct the map.
its most detailed level, and then to decide what degree
of generalization meets the purpose of the map.
Step 2. Determine the purpose of the map and its
intended audience. One purpose would be to attempt to
match the real world as closely as possible (within the
Step 5 constraints of the map scale used). In the case of popula-
tion, you might want to distinguish clearly between urban
Determine whether users find the and rural population. Another purpose might be to map
map useful and informative.
the distribution at a particular geographic level (say, the
county level); such views are often sought by government
officials for political reasons. From your viewpoint as a
mapmaker, it is important for you to realize that map-
ping at a particular geographic level can introduce error
FIGURE 3 Basic steps for communicating map information into the resulting map because each enumeration unit is
to others. represented by a single value, and thus the variation
within units cannot be portrayed. This error might be
unimportant if the focus is on how one enumeration unit
compares to another, but it can be a serious problem
or that you are working for a local newspaper and need
if readers infer more from the map than was intended;
to create a map for a major news story. Traditionally, the
for example, readers might erroneously assume that the
basic steps for communicating map information were
population density is uniform across a county on a choro-
taught within the framework of map communication
pleth map. A key point is that mapmakers often display
models (e.g., Dent 1996, 12–14; Robinson et al. 1984,
data at the level of a convenient political unit (e.g.,
15–16). Although such models have received criticism
county, state, or nation) because data are available for
(e.g., MacEachren 1995, 3–11), their use often leads to
that level, rather than consider the purpose of the map.
better designed maps. The map communication model
that we use is shown in Figure 3 as a set of five idealized * For a statistical approach for handling mobile populations, see li
steps. Let’s examine these steps by assuming that you (1998).
5
15:57:04.
The nature of your intended audience may also play Now, consider two maps that could result from efforts
an important role in designing the map. For instance, if to create a population map of the United States: a
you are utilizing thematic maps in a grade school combined proportional symbol–dot map (Color Plate
textbook, you might utilize symbols that are particu- 2) for the real-world view, and a choropleth map for
larly suggestive of the phenomenon being mapped the county-level view (map A in Figure 4). The pro-
(e.g., pizzas to represent the amount of pizza con- portional symbol–dot map is particularly illustrative
sumed in census tracts throughout a city). In many of how one can attempt to match the mapped spatial
instances, however, it is difficult to anticipate exactly distribution to the real world. Note that the overall
who will read the map (such as when designing maps population is split into urban and rural categories, and
for a daily newspaper). that urban population is further subdivided into
Step 3. Collect data appropriate for the map’s purpose. “urbanized areas” and “places outside urban areas.”
In general, spatial data can be collected from primary Step 5. Determine whether users find the map useful
sources (e.g., field studies) or secondary sources (e.g., and informative. Possibly the most important point to
Census data). For something close to the real-world keep in mind is that you are designing the map for oth-
view of population, you would likely consult the U.S. ers, not for yourself. For example, you might find a par-
Census of Population for information on urban and ticular color scheme pleasing, but you should ask your-
rural population; additionally, you would collect ancil- self whether others also will find it attractive. Ideally,
lary data that could assist you in locating the population you should answer such questions by getting feedback
data within rural areas. For a county-level view of pop- about the map from potential users. Admittedly, time
ulation, the Census figures for individual counties and monetary constraints make this task difficult, but it
would suffice. is necessary to undertake because you could discover
Step 4. Design and construct the map. Designing and not only whether a particular mapping technique
constructing the map involves not only selecting an works, but also the nature of information that users
appropriate symbology (e.g., using a dot map rather acquire from the map. Moreover, if you plan to employ
than a choropleth map), but also selecting and position- the map to illustrate a particular concept (as for a class
ing the various map elements (e.g., title, legend, and lecture), then you will want to know whether users
source) so that the resulting map is both informative acquired this concept when using the map.
and visually pleasing. This step is a complex one that If your analysis reveals that the map is not useful and
involves assessing the following questions: informative, then you will need to redesign the map.
This possibility is shown as a dashed line in Figure 3. It
1. How will the map be used? Will it be used to
is also conceivable that you might have to return to an
portray general or specific information?
earlier step, but it is more likely that you will have to
2. What is the spatial dimension of the data? For
modify some design aspect, such as the color scheme.
instance, are the data available at points, do they
Unfortunately, naive mapmakers are unlikely to
extend along lines, or are they areal in nature?
follow the five steps we have outlined. Instead, their
3. At what level are the data measured—nominal,
decisions are frequently based on readily available
ordinal, interval, or ratio?
data and mapping software. As an example, imagine
4. Is data standardization necessary? If the data are
that for a term paper, a student wished to map the
raw totals, do they need to be adjusted?
distribution of population that we have been dis-
5. How many attributes are to be mapped?
cussing. Rather than considering steps 1 and 2 of the
6. Is there a temporal component to the data?
model, the student might simply use state totals,
7. Are there any technical limitations? For example,
either because fewer numbers would have to be
a journal might not be willing to reproduce maps
entered into the computer or because the data are
in color.
readily available (as on a Web site). Furthermore, in
8. What are the characteristics of the intended audi-
step 4 the student might choose a choropleth map
ence? Is the map intended for the general public or
(map B of Figure 4) because software for creating
for professional geographers? What limitations
choropleth maps is readily available. Presuming that
might particular members of the audience have
the student had collected data in raw-total form, the
(e.g., do any have color vision impairments)?
choropleth map would be a poor choice because it
9. What are the time and monetary constraints? For
requires standardized data. Cartographers would
example, creating a high-quality dot map will cost
argue that a proportional symbol map (map C of
more than creating a choropleth map, regardless of
Figure 4) would be a better choice if raw totals were
the technical capabilities available.
to be mappe d.
15:57:04.
Population Density, 1990
Persons per Square Mile
0 to 11 12 to 28 29 to 51 52 to 119 120 to 63,231
B C
Population, 1990 Population, 1990
Millions of People Millions of People
0.5 15.1 29.8

0.5 to 1.8 2.4 to 4.4 4.8 to 7.7 9.3 to 18.0 29.8
FIGURE 4 Potential maps of the population distribution in the United States in 1990: (A) a standardized choropleth map
at the county level; (B) an unstandardized choropleth map at the state level; and (C) a state-level proportional symbol map.
(Data source: U.S. Bureau of the Census 1994.)
7
15:57:04.
4 CONSEQUENCES OF TECHNOLOGICAL CHANGE part of the United States, whereas map B suggests a pat-
IN CARTOGRAPHY tern with an eastern dominance. Map B portrays the
more accurate distribution because it is based on
Ever since the 1960s, the field of cartography has been standardized data (i.e., the number of acres of forested
undergoing major technological change, evolving from a land in relation to the area of each state); in contrast,
discipline based on pen and ink to one based on com- map A is based on raw totals (i.e., the number of acres
puter technology. In this section we consider several of forested land). Choropleth maps of raw totals tend to
consequences of this technological change. One is that portray large areas as having high values of a mapped
map production is no longer the sole province of trained attribute; in the case of map A, readers might incor-
cartographers, as virtually anyone with access to a per- rectly interpret the dark shades as indicating a high pro-
sonal computer can create maps. Although this provides portion of forested land.
more people with the opportunity to make maps (this is An alternative to cartographic instruction is the
sometimes referred to as the democratization of cartog- development of expert systems in which a computer
raphy; Morrison (1997)), there is no guarantee that the automatically makes decisions about symbolization by
resulting maps will be well designed and accurate. The using a knowledge base provided by experienced car-
maps shown in Color Plate 1 illustrate a good example tographers. Although prototypical expert systems have
of design problems that can arise. Map A uses an illogi- been developed (Buttenfield and Mark 1991 provide an
cal set of unordered hues, whereas map B uses logically overview; see Zhan and Buttenfield 1995, and Forrest
ordered shades from the same hue. Although map A 1999a, for more recent work), commonly used carto-
allows users to discriminate easily between individual graphic and GIS software, to date, has not implemented
states, it does not readily permit perception of the over- expert systems. In part, this is because cartographers do
all spatial pattern, which is one of the major reasons for not agree on the rules for symbolization (Wang and
creating a thematic map. Ormeling 1996).
Another error commonly committed by naive map- A second consequence of technological change is the
makers is illustrated in Figure 5. Map A suggests that ability to produce maps that would have been difficult
forested land is more likely to be found in the western or impossible to create using manual methods. An early
example was the unclassed map, introduced by Waldo
A Forested Land, 1992 B Forested Land, 1992

(unstandardized) (standardized)
Millions of Acres Percentage of Total Land Area
0.4 to 5.0 7.5 to 14.0 15.3 to 19.3 19.6 to 28.0 37.3 1.0 to 19.7 23.9 to 36.8 40.3 to 52.5 55.7 to 66.4 73.7 to 84.0
FIGURE 5 A comparison of the effect of data standardization. Map A is based on raw totals (i.e., the number of acres of forested
land), whereas map B is based on standardized data (i.e., the number of acres of forested land relative to the area of each state).
Map A is misleading because states with large areas tend to have more forest. (Data source: Powell et al. 1992.)
15:57:04.
A Foreign-Owned Agricultural Land, 1992 B Foreign-Owned Agricultural Land, 1992

(classed) (unclassed)
Percentage of
Privately Owned Land
Percentage of Privately Owned Land
0 to 0.4 0.6 to 1.2 1.5 to 3.1 5.2 14.1 0 3.5 7.1 10.6 14.1
FIGURE 6 A comparison of (A) classed and (B) unclassed maps. On the classed map, states are grouped into classes of similar
value, with a gray tone assigned to each class, whereas on the unclassed map, gray tones are selected proportional to the data value
associated with each state. (Data source: DeBraal 1992.)
Tobler (1973).* Figure 6 compares a traditional classed ally created one “best” map for users. In contrast, inter-
map (A) with its unclassed counterpart (B). On the active graphics now permit users to examine spatial
classed map, data are grouped into classes of similar data dynamically and thus develop several different
value, with a progressively darker gray tone assigned to representations of the data—a process termed data
each class, whereas on the unclassed map, gray tones exploration. The software package MapTime exempli-
are assigned proportionally to each data value. Some fies the nature of data exploration. MapTime permits
people have promoted unclassed maps, arguing that users to explore data using three approaches: anima-
they more accurately reflect the real-world distribution, tion, small multiples (in which individual maps are
whereas others have promoted classed maps on the shown for each time period), and change maps (in
grounds that they are easier to interpret. which an individual map depicts the difference between
Animated maps (or maps characterized by continu- two points in time). The examination of population val-
ous change while the map is viewed) are particularly ues for 196 U.S. cities from 1790 to 1990 (a data set dis-
representative of the capability of modern computer tributed with MapTime) illustrates the different per-
technology. Although the notion of animated mapping spectives provided by these approaches. For example,
has been around at least since the 1930s (Thrower 1959; an animation reveals major growth in northeastern
1961), only recently have cartographers begun to recog- cities over most of the period, with an apparent drop
nize its full potential (Campbell and Egbert 1990). for some of the largest cities from about 1950 to 1990.
Probably the most common forms of animation are those In contrast, a change map, showing the percent of pop-
representing changing cloud cover, precipitation, and ulation gains or losses between 1950 and 1990, reveals a
fronts on daily television weather reports. Animations of distinctive pattern of population decrease throughout
spatial data are now also found in popular computerized the Northeast (Color Plate 3). One of the keys to data
encyclopedias such as Grolier’s (DiBiase 1994). exploration is that displays such as these can be created
A third consequence of technological change is that in a matter of seconds.
it alters our fundamental way of using maps. With the A fourth consequence of technological change is that
communication model approach, cartographers gener- it enables mapmakers to link maps, text, pictures, video,
and sound in multimedia presentations. For example, in
* The earliest choropleth maps were actually unclassed, but they the Grolier encyclopedia already mentioned, anima-
rapidly gave way to classed maps (see Robinson 1982, 199). tions include sound clips to assist in understanding.
9
15:57:04.
These animations are integrated, of course, with the rest Web can serve as a source of data, maps, software,
of the encyclopedia, which includes text, pictures, and electronic atlases, tools for developing software, and
videos. In a more sophisticated form of multimedia teaching materials (Table 1). In the course of reading
known as hypermedia, the various forms of media can this text, you will find references to numerous Web sites.
be linked transparently in ways not anticipated by sys- In addition to providing a wealth of maps and related
tem designers (Buttenfield and Weber 1994). information, the Web also has changed the way in which
Closely associated with animation, data exploration, maps are produced and used. Prior to the existence of the
and multimedia is the electronic atlas. Initially, elec- Web, cartography could be characterized as supply-
tronic atlases emulated the appearance of paper atlases driven—that is, cartographers created maps and mapping
(e.g., the Electronic Atlas of Arkansas developed by software that they and other experts thought would be
Richard Smith in 1987). More recently, electronic useful. With the advent of the Web, cartography can be
atlases have incorporated animation, data exploration, considered demand-driven, as the nature of maps and
and multimedia capability. We consider some online mapping software is driven by what users want (e.g., a
electronic atlases (such as The National Atlas of the Web site containing maps that receives few “hits” proba-
United States; http://www.nationalatlas.gov/ ) within the bly is not particularly useful or desirable).* As part of this
larger realm of Web mapping. demand-driven character, users might want to design their
A fifth consequence of technological change is own maps. User-designed maps were possible prior to the
the capability to create realistic representations of the Web with mapping software, but now there are millions of
Earth’s natural and built environments. To illustrate, Web users, whose cartographic abilities obviously vary. As
Color Plate 4 is a frame taken from a fly-through we have already suggested, this raises the possibility of
animation of a portion of Lawrence, Kansas, that was millions of poorly designed maps. Poorly designed maps
created by geography students at the University of are problematic even if only the individual making the
Kansas. Such realistic fly-throughs provide users a sense map will use that map, because the individual might derive
of being immersed in a 3-D landscape. When users are incorrect information as a function of poor design.
able to navigate through and interact with a realistic 3-D A distinct advantage of the Web is that it provides the
environment, we term this a virtual environment or general public with access to spatial information that they
virtual reality. The notions of visual realism and virtual normally might not have access to. For example, imagine
environments certainly challenge our traditional that officials are planning to build a bypass around the city
thoughts about what cartography is. Normally, we have that you live in, and that the city has hired a consulting
thought of maps as consisting of abstract symbols; now firm to model the anticipated impact on the city. If the
we need to extend our notion of maps to include realism. consulting firm makes its predictions available via the
Closely aligned with virtual environments is the Web, you should be able to examine them from the
notion of augmented reality, in which computer-based comfort of your home; you might even be able to run their
information is used to enhance our view of the real models under different scenarios—for example, you might
world. For instance, imagine that you are a physical wonder what would happen if the bypass were located
geographer studying vegetation changes in a particular north as opposed to south of the city. The ability of the
region, and you wish to examine the current vegetation general public to work with such spatial information is
in the field and compare it with past vegetation patterns. commonly termed public participation GIS (PPGIS).†
Traditionally, you would accomplish this by taking maps One aspect of the Web (and associated Internet tech-
into the field and comparing them with current vegeta- nology) that has received considerable interest is map-
tion patterns. With augmented reality, however, you can ping the technology and its associated information
don a wearable computer (and associated specialized spaces. For instance, Color Plate 5 is an example of map-
viewing hardware) and actually see historic vegetation ping the technology itself—in this case, early Internet
patterns overlaid on the present-day landscape. traffic. The red balls, and the white lines connecting
A sixth consequence of technological change is the them, represent the backbone of NSFNET in 1991
ability to access maps and related information via the (NSFNET formed the basis for the present-day
World Wide Web (WWW, or simply “the Web”), which
is part of the larger Internet. The Web has, of course,
changed our daily lives as evidenced by the common * The terms supply-driven and demand-driven were taken from the
listing of Web addresses in newspaper and television work of Kraak and Brown (2001).
advertisements. From the standpoint of cartography, the † For an overview of PPGIS, see Sieber (2006).
15:57:04.
TABLE 1 Internet uses for cartography

Function Example Address (URL)
Locational data Longitude and latitude http://www.census.gov/cgi-bin/gazetteer
of major U.S. cities
Attribute data Population estimates http://www.census.gov/popest/counties/
for counties
Maps
Static 3-D representation http://www.rpi.edu/locker/69/000469/dx/harbor.www/maxus.dieldrin.cr.gif
of pollution in
New York Harbor
Animated Animation of urban http://www.ncgia.ucsb.edu/projects/gig/v2/About/abApps.htm
sprawl in the San
Francisco Bay area
Software for Software for creating http://factfinder.census.gov/jsp/saff/SAFFInfo.jsp?_pageId=thematicmaps&_
creating thematic maps submenuId=maps_0
static maps
Software for CommonGIS http://www.ais.fhg.de/SPADE/products/index_eng.html
exploring
data
Electronic The Atlas of Canada http://atlas.nrcan.gc.ca/site/index.html
atlases
Tools for GeoVISTA Studio http://geovistastudio.sourceforge.net/
developing
software
Teaching The Geographer’s Craft http://www.colorado.edu/geography/gcraft/contents.html
materials
for students
Teaching Cartography: The http://www.colorado.edu/geography/virtdept/contents.html
materials for Virtual Geography
instructors Department
Internet). The colored vertical lines depict inbound data represent similar commonly occurring news topics. Note
to each of the nodes of the backbone, with low and high that this is not a map in the conventional sense because
amounts of data depicted by purple and white, respec- we are mapping abstract space. Numerous examples of
tively. Figure 7 is an example of mapping the informa- mapping the Internet and its associated information
tion space of the Internet. Here, more than 900 online spaces can be found in Martin Dodge and Rob Kitchin’s
news reports of the Kosovo crisis of the late 1990s have Atlas of Cyberspace (2001a) and on Martin Dodge’s
been “mapped” using the isarithmic method; dark peaks Web site (http://www.cybergeography.org/atlas/).
FIGURE 7 An example of mapping

the information space of the Internet.
The relationship and frequency of
topics associated with more than 900
online news reports of the Kosovo
crisis are depicted using the isarith-
mic method. (Originally produced by
Newsmaps, Cartia, Inc.; the software
used to create this figure is now
known as ThemeView and is pro-
duced by the Pacific Northwest
National Laboratory (http://infoviz.
pnl.gov/technologies.html)).
11
15:57:04.
An excellent recent example of the capability pro-

vided by the Web is Google Earth (http://earth.google.
com/index.html ) and the companion product Google
n
Maps (http://maps.google.com/ ). By draping high-reso-
tio
ica
lution, remotely sensed imagery on a digital elevation
un
model and permitting users to zoom in or fly over the
Public
mm
landscape, Google Earth provides breathtaking, realis-
Co
tic views of the Earth’s landscape, all of which are free
in the basic version.
n
tio
5 GEOVISUALIZATION
liza
Pr
ua
Kn ese
Vi s
Private
Outside geography, the term visualization has its ori- ow nti
ns ng
gins in a special issue of Computer Graphics authored R
U ev
by Bruce McCormick and his colleagues (1987). To nk e
no alin
McCormick and his colleagues, the objective of w g High Low
ns
scientific visualization was “to leverage existing scien- Human–map Interaction
tific methods by providing . . . insight through visual
methods” (p. 3). Today, scientific visualization gener-
ally involves using sophisticated workstations to FIGURE 8 A graphical representation of how maps are used.
explore large multivariate data sets (Color Plate 6). Note that visualization is contrasted with communication
Work in scientific visualization extends far beyond along three dimensions: private versus public, revealing
the realm of spatial data, which geographers deal unknowns versus presenting knowns, and the degree of
with, to include topics such as medical imaging and human–map interaction. (After MacEachren 1994b, 7.)
visualization of molecular structure and fluid flows.
Peter and Mary Keller (1993) provided numerous
through computer displays or other media—to make spa-
examples of the use of scientific visualization. The tial contexts and problems visible, so as to engage the most
most recent developments in scientific visualization powerful human information-processing abilities, those
can be found in the proceedings of the Institute of associated with vision.
Electrical and Electronics Engineers (IEEE)
Visualization conference, which has been held every Using this definition, geographic visualization could be
year since 1990. applied to the visual analysis of a paper map created by
The notion of information visualization is closely pen-and-ink methods or to the visual analysis of a map
related to scientific visualization. Information visualiza- created on an interactive graphic display.
tion focuses on the visual representation and analysis of The second, and narrower, definition is based on
nonnumerical abstract information. The “map” of MacEachren’s (1994b, 6) cartography-cubed repre-
online news reports of the Kosovo crisis shown in sentation of how maps are used, which is shown in
Figure 7 is an example of information visualization. In Figure 8.* In this graphic, visualization is contrasted
Chapter 1 of their book Readings in Information with communication along three dimensions: private
Visualization: Using Vision to Think, Stuart Card and versus public, revealing unknowns versus presenting
his colleagues (1999) provide numerous examples of knowns, and the degree of human–map interaction.
information visualization. Based on this diagram, MacEachren argued that
Although those outside geography were the first to geographic visualization is a private activity in which
popularize visualization, the idea has existed in cartog- unknowns are revealed in a highly interactive envi-
raphy at least since the 1950s (MacEachren and Taylor ronment, whereas communication is the opposite: a
1994, 2). As a result, cartographers have struggled to public activity in which knowns are presented in a
define the term. Thus far, two basic definitions have noninteractive environment.
emerged. The first is a broad one that encompasses both As an example of geographic visualization,
paper and computer-displayed maps. According to Alan MacEachren employed Joseph Ferreira and Lyna
MacEachren and his colleagues (1992, 101):
Geographic visualization [can be defined] as the use of * MacEachren used the term (Cartography) 3 ; we use the word cubed
concrete visual representations—whether on paper or to avoid confusion with the superscript 3.
15:57:04.
FIGURE 9 A small-scale version (1:21,400,000) of the USGS map Landforms of the Conterminous United States: A Digital
Shaded-Relief Portrayal, developed by Thelin and Pike (1991). (Courtesy of Gail P. Thelin.)
Wiggin’s (1990) “density dial,” in which class break The notions of map communication models and data
points on choropleth maps were manipulated to iden- exploration introduced previously can be associated
tify and enhance spatial patterns. The idea is that with communication and visualization, respectively, in
users of the density dial could discover previously MacEachren’s cartography-cubed representation. Thus,
unknown spatial patterns. In contrast, MacEachren when using the five steps of the communication model
argued that communication is exemplified by the “you presented in section 3, the intention generally is that the
are here” maps used to locate oneself in a shopping map is being made for the general public, there will be
mall. MacEachren stressed that certain map uses do low human–map interaction, and the focus is on pre-
not fit neatly into either category (thus the need for senting knowns. In data exploration, the emphasis is on
the cartography-cubed representation). For example, revealing unknowns via high human–map interaction in
Gail Thelin and Richard Pike’s (1991) dramatic a private setting; in this sense the word “exploration”
shaded relief map shown in Figure 9 fits into the com- could easily be substituted for “visualization” in
munication realm because it is available to a wide MacEachren’s graphic.
readership (making it “public”) and because the user On an informal basis, visualization has been used to
cannot interact with the paper version of it. The describe any recently developed, novel method for
Thelin and Pike map, however, fits visualization in the displaying data. Thus, cartographers have placed ani-
sense that it can reveal unknowns; for instance, many mation and virtual environments under the rubric of
readers would be unfamiliar with the flat-iron-shaped “visualization.” Additionally, novel methods that
plateau in the north central part of the map—the might result in static maps (e.g., Daniel Dorling’s car-
Coteau des Prairies. The net result is that the Thelin tograms) are also placed under the heading of “visual-
and Pike map falls in the upper right corner of ization.” More recently, MacEachren has simplified
MacEachren’s diagram. the term geographic visualization to geovisualization
13
15:57:04.
(e.g., MacEachren et al. 1999c). We will follow techniques the ability to create more effective carto-
MacEachren’s lead and use the term geovisualization graphic products.
throughout the remainder of the book. GISs are computer-based systems that are used to
One limitation of the term geovisualization is its analyze spatial problems. For example, we might use a
implication that our interpretation of spatial patterns is GIS to determine optimal bus routes in a school dis-
solely a function of our sense of vision. Cartographers trict, or to predict the likely location of a criminal’s res-
have begun to explore how our other senses (e.g., idence based on a series of apparently related crimes.
sound, touch, and even smell) might be used to inter- Traditionally, most digital thematic maps were created
pret spatial patterns. Thus, it would seem desirable to using so-called thematic mapping software (e.g.,
use a term that is more inclusive. One term that has MapViewer).* Today, GIS software typically has consid-
been proposed (by Fraser Taylor) is cybercartography, erable thematic mapping capability, in addition to its
which, he argued, includes the following elements: inherent spatial analysis capabilities. These spatial
• Is multisensory, using vision, hearing, touch, and analysis capabilities also enable GIS to handle more
eventually smell and taste; sophisticated mapping problems than traditional the-
• Uses multimedia formats and new telecommunica- matic mapping software. Dot maps (such as those
tions technologies such as the World Wide Web; shown in Figure 2) provide a good illustration of the
• Is highly interactive and engages the user in new capability of GIS software. Ideally, dots on a dot map
ways; should be located as accurately as possible to reflect the
• Is applied to a wide range of topics of interest to soci- underlying phenomenon. Thus, in the case of the wheat
ety, not only location finding and the physical envi- map, we would want to place dots where wheat is most
ronment; likely to be grown (say, on level terrain with fertile soil)
• Is not a stand-alone product like the traditional map and not where it cannot be grown (in water bodies).
but is part of an information/analytical package; Traditional thematic mapping software did not accom-
• Is compiled by teams of individuals from different plish this, as dots would normally be based solely on the
disciplines; basis of enumeration units (e.g., counties). In contrast,
• Involves new research partnerships among acade- GIS software enables a large number of factors (or
mia, government, civil society, and the private sector. layers, in GIS terminology) to be accounted for.
(Taylor 2003, 407) The basic purpose of remote sensing is to record
information about the Earth’s surface from a distance
A number of these elements overlap with the notion of
(e.g., via satellites and aircraft). For instance, we might
geovisualization (e.g., “Is highly interactive and
use remote sensing to determine temporal changes
engages the user in new ways”), while others suggest
resulting from forest fires (e.g., acreage burned, effects
a broader view (e.g., “Involves new research part-
of erosion, and regrowth) or the health of crops. The
nerships . . .”). As an example of the latter, Taylor and
importance of remote sensing to cartography can be
his colleagues have been developing the Cyber-
illustrated by again presuming that we wish to map the
cartographic Atlas of Antarctica (Pulsifer et al. 2007),
distribution of wheat across the United States. Rather
which includes a broad range of partnerships (Taylor
than use a GIS approach, in which layers of related
2003, 414). Although the term cybercartography has
information are considered, we could use remote sens-
not been widely adopted by cartographers, it seems
ing to directly determine the precise location of wheat
clear that ultimately we will need a term that is more
fields. For instance, Stephen Egbert and his colleagues
inclusive than geovisualization.
(1995) found that wheat fields could be identified with
an accuracy as high as 99 percent by using remotely
6 RELATED TECHNIQUES sensed imagery for three time periods.
Quantitative methods are used in the statistical
The picture of cartography today would not be com- analysis of spatial data. For instance, we might develop
plete without including the broader realm of geo- an equation that relates deaths due to drunk driving to
graphic information science (GIScience), which can various attributes that we think might explain the spa-
be considered to include cartography and the related tial variation in the death rate, such as the severity of
techniques of geographic information systems (GIS), laws that penalize drunk driving, the extent to which
remote sensing, and quantitative methods. In this the laws are enforced, the percentage of the population
section, we consider not only how cartographers can
use these techniques, but also how knowledge of *MapViewer is still marketed as a thematic mapping tool, but now
cartography can provide those working with these has some GIS analytic capability.
15:57:04.
that are members of various religious groups, the traffic would be desirable because users could associate cer-
density, and many other attributes. The major develop- tain colors with the raw image and a different set of col-
ment in quantitative methods relevant to cartography is ors with the changes.
that of exploratory spatial data analysis (ESDA), which
refers to data exploration techniques that accompany a
statistical analysis of the data. As an example, we might 7 COGNITIVE ISSUES IN CARTOGRAPHY
explore a map of predicted deaths due to drunk driving
to see how the pattern is affected by various attributes Understanding the role that cognition plays in cartogra-
that we include in the model. phy requires contrasting cognition with perception.
Thus far, we have considered how cartographers can Perception deals with our initial reaction to map symbols
benefit from knowledge of other geographic tech- (e.g., that a symbol is there, that one symbol is larger or
niques. We argue that those working with these other smaller than another, or that symbols have different
techniques can also benefit from a knowledge of colors). In contrast, cognition deals not only with percep-
cartography. For example, imagine that you are working tion but also with our thought processes, prior experi-
in the GIS department of a city and wish to examine the ences, and memory.* For example, contour lines on a
distribution of auto thefts, and so you create a choro- topographic map can be interpreted without looking at a
pleth map depicting the number of auto thefts for each legend because of one’s past experience with such maps.
census tract in the city. Based on the high incidence of Alternatively, one might correlate the pattern of soils on
auto thefts in a contiguous set of census tracts, you a particular map with the distribution of vegetation seen
recommend to the police department that they focus on a previous map.
their patrols in those areas. Unfortunately, your solu- The principles of cognition are important to cartog-
tion might be inappropriate, because you failed to raphers because they can explain why certain map sym-
consider the population (and possibly the number of bols work (i.e., communicate information effectively).
cars owned) in each census tract. Instead of mapping Traditionally, cartographers were not so concerned with
the raw number of auto thefts, you probably would want why symbols worked but rather with determining which
to adjust for the population (or number of cars owned) symbols worked best. This was known as a behaviorist
in each tract. This is another example of the data view, in which the human mind was treated like a black
standardization problem that we mentioned earlier. box. Today, cartographers are more likely to utilize a
Effective use of GIS requires an understanding of basic cognitive view, in which they hope to determine why
cartographic principles such as these. symbols work effectively. A cognitive view provides a
As another example of how those working in other theoretical basis for map symbol processing that not
areas can benefit from cartography, imagine that you only assists in explaining why particular symbols work,
are a remote sensor working on the GreenReport but also provides a basis for evaluating other map sym-
(http://koufax.kgs.ku.edu/kars/kars_map.cfm), which bols, even those not yet developed.
uses remotely sensed images to depict the health of To illustrate the difference between the behaviorist
crops and natural vegetation throughout the United and cognitive views, consider an experiment in which a
States. Color Plate 7A illustrates a basic greenness map cartographer wishes to compare the effectiveness of
that you might use to represent current vegetation con- two color schemes for numerical data: five hues from a
ditions. One can argue that this is a logical color scheme yellow-to-red progression and five hues from the elec-
because dark green represents “High Biomass” whereas tromagnetic spectrum (red, orange, yellow, green, blue).
dark brown indicates “Low Biomass.” Color Plate 7B Let us presume that the results of such an experiment
depicts a change map that might be used to compare reveal that the yellow–red progression works best. The
current conditions to those two weeks earlier. Note that traditional behaviorist would report these results but
in this case, the “Little or no Change” category is simi- probably would not provide any indication as to why
lar to some of the categories on the basic greenness one sequence worked better than another. In contrast,
map, which could confuse a map reader. An improved the cognitivist would consider how color is processed
symbology would be to use a gray tone for the “Little or by the eye–brain system, possibly theorizing that the
no Change” category, as shown in Color Plate 7C. An yellow–red progression works best because of oppo-
alternative symbology would be to use completely dif- nent process theory. Effectiveness of the spectral hues
ferent colors to represent the changes in greenness; for might be emphasized on the grounds that different hues
example, Color Plate 7D uses shades of blue to repre-
sent increased greenness and orange and red shades to *For a more in-depth discussion of perception and cognition, see
represent decreased greenness. The latter symbology Goldstein (2007) and Matlin (2006), respectively.
15
15:57:04.
will appear to be at slightly different distances from the After the information has been rehearsed in the
eye and thus will form a logical progression (e.g., red short-term visual store, it is ultimately passed on to
will appear nearer than, say, blue. Spectral hues might long-term visual memory (Figure 10). Note that arrows
also be considered effective because of their common are shown going in both directions between short-term
use on maps, and the likelihood that readers have expe- visual store and long-term visual memory. When some-
rience in using them. thing is initially memorized, information must be
An important concept of cognition is the three moved from the short-term store to long-term mem-
types of memory: iconic memory, short-term visual ory; when something is retrieved from memory, the
store, and long-term visual memory (Peterson 1995). opposite is the case. As an example of the latter, imag-
Iconic memory deals with the initial perception of an ine that you are shown a map of Texas and asked to
object (in our case, a map or portion thereof) by the indicate which state it is. To make your decision, you
retina of the eye. Calling this “memory” is somewhat must retrieve the image of Texas from long-term mem-
of a misnomer, because it exists for only about one- ory and compare it with the image in the short-term
half second and because we have no control over it. store to make your decision. We have covered here
Visual information initially recorded in iconic memory only some of the basic concepts of cognitive psychol-
is passed on to the short-term visual store. Only ogy that are necessary for understanding this text. For
selected information is passed on at this stage; for a broader overview, see Peterson (1995, Chapter 2) and
example, the boundary of Texas shown in Figure 10 MacEachren (1995).
will likely be simplified to some extent in moving from
iconic to short-term visual store. Keeping information
8 SOCIAL AND ETHICAL ISSUES IN
in short-term visual store requires constant attention
CARTOGRAPHY
(or activation). This is accomplished by rehearsal of
the items being memorized (e.g., staring at the map of
Texas and telling yourself to remember its shape). Although maps, and especially interactive digital maps,
have tremendous potential for visualizing spatial data,
we should also consider various social and ethical
issues associated with their use. The notion of social
and ethical issues in cartography was first developed in
the context of postmodernism. Those who subscribe to
a postmodernist view believe that problems can best
Iconic Memory Short-Term be approached from multiple perspectives or view-
(Visual Register) Visual-Spatial
Scratch Pad
Visual Store points; for example, in a study of an urban neighbor-
hood, a postmodernist would want to acquire not only
the perspective of those in positions of political and
economic power, but also the perspectives of as wide a
sample of inhabitants as possible: men and women,
Long-Term children, the elderly, each of the social classes, ethnic
Visual Memory and racial groups, or any others who might shed light
Eye on the dynamics of the neighborhood (Cloke et al.
1991, 170–201).
An important notion of postmodernism is that a text
Stimulus (or map) is not an objective form of knowledge, but
rather has numerous hidden agendas or meanings.
Deconstruction enables us to uncover these hidden
agendas and meanings. Brian Harley (1989, 3) stated:
FIGURE 10 Three forms of memory used in cartography. A
perceived map is initially stored in iconic memory (the retina Deconstruction urges us to read between the lines of the
of the eye). The image is then passed to the short-term visual map . . . to discover the silences and contradictions that
store in the brain (where it is rehearsed). Finally, information challenge the apparent honesty of the image. We begin to
is stored for later use in long-term visual memory in the brain. learn that cartographic facts are only facts within a specific
(From PETERSON, MICHAEL P., INTERACTIVE AND cultural perspective.
ANIMATED CARTOGRAPHY, 1st Edition, © 1995, pg. 27.
Reprinted by permission of Pearson Education, Inc., Upper As an example of deconstruction, consider Mark
Saddle River, NJ.) Bockenhauer’s (1994) examination of the official state
15:57:04.
highway maps of Wisconsin over a seven-decade The notions of postmodernism and map decon-
period. Although the major purpose of highway maps struction led to an examination of the social and ethi-
is presumably to assist a motorist in getting from one cal implications of the broader field of GIScience (Sui
place to another, Bockenhauer (17) argued that “three 2004); today, an examination of such issues is known as
dominant ‘cultures’ [could] be identified as factors critical GIS (Harvey et al. 2005). Early on, those tak-
influencing the map product: a transportation/ ing a critical view were concerned that, because of its
modernization culture, a culture of promotion, and a quantitative and empirical nature, GIS failed to prop-
subtle, beneath-the-surface culture of dominion.” As erly consider the everyday world that individuals live
an example of the transportation/modernization cul- in, that the cost and complexity of GIS software pre-
ture, Bockenhauer recounted the removal of surfaced vented access by the full range of the population, and,
county roads from recent editions of the map, and thus consequently, that GIS was largely under the control
the greater emphasis on “getting us from here to there of those in positions of political and economic power.
by encouraging travelers onto the freeways” (p. 21). To understand the nature of such concerns, consider
An illustration of highway maps as promotional the role that GIS might play in tracking the where-
devices was their use by government officials; for abouts of individuals. The Digital Angel Corporation
example, in the 1989–1990 edition, Governor Tommy (http://www.digitalangelcorp.com/ ) markets a wrist-
Thompson showcased himself next to a replica of a band that can be affixed to an individual and locked or
Duesenberg automobile. Finally, an example of the unlocked remotely, enabling another party to track the
culture of dominion was the portrayal of women on movement of that individual in real time via the global
the maps: “Among the most common and prominent positioning system (GPS) and GIS. Although such
images appearing on the . . . maps . . . are those of technology is arguably useful for tracking children and
women in swim suits and fishermen. Nearly all of the the elderly, its availability raises interesting ethical
photos of people enjoying Wisconsin fishing . . . are of questions. For instance, what if a child does not wish to
white men. . . . [The] women seem to be part of the be tracked (not surprising for a teenager)? And should
package of ‘pleasure’ offered to white men in spouses be able to track one another? Jerry Dobson
Wisconsin” (p. 24). Although some would disagree and Peter Fisher (2003) are particularly wary of such
with Bockenhauer’s interpretations of these maps, it is technology, noting that if a transponder were added to
clear that maps can convey information other than the wristband, it would be possible to administer a
their supposed primary purpose. form of punishment to the individual. They term the
Although we do not focus on postmodernism and net result geoslavery, suggesting that the results would
map deconstruction in this text, both mapmakers and be far worse than George Orwell’s 1984, as one “mas-
map users need to recognize their importance. ter” could potentially monitor and enslave thousands
Mapmakers must realize that maps can communicate of people.
unintended messages, and that the data they have cho- Although those involved with critical GIS are
sen to include on a map or the method of symbolizing concerned about the potential misuse of GIS, they also
that data might be a function of the culture of which note that effective visualizations can be produced if
they are a part. Conversely, map users must recognize GIS is used with care. For instance, in a paper dealing
that a single map might depict only one representation with feminist visualization, Mei-Po Kwan (2002)
of a spatial phenomenon (e.g., a map of percent forest describes how she has mapped the spacetime paths of
cover is only one representation of vegetation). a sample of African-American women in Portland,
Note that there is some overlap between the notions Oregon. Kwan argues that “not only do the homes and
of data exploration and postmodernism because both workplaces of these women concentrate in a small
promote the notion of multiple representations of data. area . . . but their activities’ locations are much more
In the context of data exploration, “multiple represen- spatially restricted when compared to those of all
tations” refers to the various methods of symbolizing other gender/ethnic groups.” (p. 654). As another
the data (e.g., using MapTime to display population example, Jeremy Crampton (2004) has criticized
data as both an animation and a change map). The post- the traditional choropleth map because “it produces a
modernist would likely support this approach because view of human life as crammed into pre-given political
it concurs with the notion that there is no single, “cor- units.” Utilizing the work of Holloway and his
rect” way of visualizing data. Additionally, however, the colleagues (1999), Crampton illustrates how a consid-
postmodernist would be interested in the multiple eration of ancillary information such as land use/land
meanings and potentially hidden agendas found in a cover can be used to create a much more detailed
particular thematic map. dasymetric map.
17
15:57:04.
FURTHER READING
Board, C. (1984) “Higher-order map-using tasks: Geographical Kraak, M.-J., and Brown, A., eds. (2001) Web Cartography:
lessons in danger of being forgotten.” Cartographica 21, no. Developments and Prospects. London: Taylor & Francis.
1:85–97. Provides a broad overview of issues related to the Web.
Discusses issues relevant to the kinds of information that can be MacEachren, A. M. (1994b) “Visualization in modern cartog-
acquired from maps. raphy: Setting the agenda.” In Visualization in Modern
Buttenfield, B. P., and Mark, D. M. (1991) “Expert systems in Cartography, ed. by A. M. MacEachren and D. R. F. Taylor, pp.
cartographic design.” In Geographic Information Systems: The 1–12. Oxford: Pergamon.
Microcomputer and Modern Cartography, ed. by D. R. F. Discusses definitions for visualization.
Taylor, pp. 129–150. Oxford: Pergamon.
MacEachren, A. M. (1995) How Maps Work: Representation,
Although a bit dated, this work provides a good summary of the Visualization, and Design. New York: Guilford.
potential for using expert systems in map design; for more recent
An advanced treatment of cognitive issues in cartography.
work, see Zhan and Buttenfield (1995) and Forrest (1999a).
MacEachren, A. M., and Kraak, M.-J. (1997) “Exploratory car-
Campbell, C. S., and Egbert, S. L. (1990) “Animated cartogra-
tographic visualization: Advancing the agenda.” Computers &
phy: Thirty years of scratching the surface.” Cartographica 27,
Geosciences 23, no. 4:335–343.
no. 2:24–46.
Provides an alternative view of MacEachren’s cartography-cubed
An overview of early work in animated cartography, along with
model.
some suggestions for the potential of animation.
McCormick, B. H., DeFanti, T. A., and Brown, M. D. (1987)
Cartographica 40, no. 4, 2005 (entire issue).
“Visualization in scientific computing.” Computer Graphics
This special issue focuses on critical GIS; also see Cartographica 21, no. 6 (entire issue).
39, no. 1, 2004 (entire issue).
A classic early work on visualization.
DiBiase, D. (1990) “Visualization in the earth sciences.” Earth
Monmonier, M. S. (1985) Technological Transition in
and Mineral Sciences 59, no. 2:13–18.
Cartography. Madison, WI: University of Wisconsin Press.
Discusses visualization and its role in geographic research.
A text dealing with technological change in cartography.
Dodge, M., and Kitchin, R. (2001a) Atlas of Cyberspace. Monmonier, M. (2002) Spying with Maps: Surveillance
Harlow, England: Addison-Wesley. Technologies and the Future of Privacy. Chicago: University of
Provides numerous examples of mapping the Internet and its Chicago Press.
associated information spaces.
Presents myriad ways in which modern GPS and GIS technologies
Environment and Planning B: Planning and Design, 28, no. 6, can be used for surveillance and discusses related privacy issues;
2001 (entire issue). also see Monmonier (2003).
This special issue deals with public participation in GIS. Also see Peterson, M. P. (1997) “Cartography and the Internet:
vol. 25, no. 2, 1998 of Cartography and Geographic Information Introduction and research agenda.” Cartographic Perspectives
Science. no. 26:3–12.
Hallisey, E. J. (2005) “Cartographic visualization: An assessment An overview of the Internet and its role in cartography. Other
and epistemological review.” The Professional Geographer 57, articles in this issue also deal with the Internet. The complete issue
no. 3:350–364. can be found at http://maps.unomaha.edu:16080/NACIS/cp26/. For
more recent work, see Peterson (2003).
Considers epistemological issues in geovisualization (what
Hallisey terms “cartographic visualization”). Rundstrom, R. A., ed. (1993) “Introducing cultural and social
cartography.” Cartographica 30, no. 1 (entire issue).
Keller, P. R., and Keller, M. M. (1993) Visual Cues: Practical
Data Visualization. Los Alamitos, CA: IEEE Computer A set of articles dealing with postmodern issues in cartography.
Society Press. Taylor, D. R. F., ed. (2005) Cybercartography: Theory and
Provides examples of visualization from a wide variety of Practice. Amsterdam:Elsevier.
disciplines. An edited volume dealing entirely with the theory and practice of
. cybercartography.
Kolác̆ny, A. (1969) “Cartographic information: A fundamental
concept and term in modern cartography.” The Cartographic Wood, D., and Fels, J. (1992) The Power of Maps. New York:
Journal 6, no. 1:47–49. Guilford.
Presents a cartographic communication model that MacEachren An extensive essay on postmodern issues in cartography. Also see
(1995) claimed had “the greatest initial impact on cartography” (p. 4). Wood and Fels (1986).
15:57:04.
GLOSSARY
animated maps: maps characterized by continuous change, dot map: a map in which small symbols of uniform size
such as the daily weather maps shown on television newscasts (typically, solid circles) are used to emphasize the spatial pat-
that depict changes in cloud cover. tern of a phenomenon (e.g., one dot might represent 1,000
attribute: a theme or variable you might wish to display on a head of cattle).
map (e.g., the predicted high temperature for each major U.S. electronic atlas: a collection of maps (and databases) avail-
city). able in a digital environment; sophisticated electronic atlases
augmented reality: a combination of a real-world experi- enable users to take advantage of the digital environment
ence with a virtual representation in which the real world is through Internet access, data exploration, map animation, and
dominant, such as overlaying the boundaries of “old-growth” multimedia.
forests on your view of a mountainous environment; a subset enumeration unit: a data-collection unit such as a county or
of a mixed environment. state.
change map: a map representing the difference between two expert system: a software application that incorporates rules
points in time. derived from experts to make decisions and solve problems.
choropleth map: a map in which enumeration units (or data- exploratory spatial data analysis (ESDA): data exploration
collection units) are shaded with an intensity proportional to techniques that accompany a statistical analysis of spatial
the data values associated with those units (e.g., census tracts data.
shaded with gray tones that have an intensity proportional to general information: the type of information that one
population density). acquires when examining spatial patterns on thematic maps.
classed map: a map in which data are grouped into classes of general-reference map: a type of map used to emphasize the
similar value, and the same symbol is assigned to all members location of spatial phenomena (e.g., a U.S. Geological Survey
of each class (e.g., a data set with 100 different data values (USGS) topographic map).
might be depicted using only five shades of gray).
geographic information systems (GIS): automated systems
cognition: the mental activities associated with map reading for the capture, storage, retrieval, analysis, and display of
and interpretation, including the initial perception of the map, spatial data.
thought processes, prior experiences, and memory.
geographic visualization (or geovisualization): a private
communication: the transfer of known spatial information activity in which previously unknown spatial information is
to the public; typically, this is done via paper maps. revealed in a highly interactive computer graphics environ-
compare patterns: when two or more maps are analyzed ment.
visually, readers are expected to compare the patterns found geoslavery: the use of GPS and GIS technologies to monitor
on the maps. and control the movement of people.
critical GIS: a consideration of the social implications of head-mounted display (HMD): helmet-like device that
using GIS. shields a person from the real-world view and provides images
cybercartography: a term used to describe the character of of a virtual environment to each eye.
modern cartography; some of its characteristics include the hypermedia: a sophisticated form of multimedia in which
capability to interpret spatial patterns using senses other than various forms of media can be linked in ways not anticipated
vision (e.g., hearing and touch); the use of multimedia formats by system designers.
and new communication technologies; and new research part-
iconic memory: refers to the initial perception of an object
nerships among academia, government, and the private sec-
(in our case, a map or portion thereof) by the retina of the
tor.
eye; contrast with short-term visual store and long-term visual
data exploration: examining data in a variety of ways to memory.
develop different perspectives of the data; for example, we
information acquisition: the process of acquiring spatial
might view a choropleth map in both its unclassed and its
information while a map is being used; contrast with memory
classed forms. Data exploration is sometimes equated with
for mapped information.
geovisualization.
information visualization: involves the visualization and
deconstruction: the process of analyzing a map (or text) to
analysis of nonnumerical abstract information such as the top-
uncover its hidden agendas or meanings.
ics that are discussed on the front page of a newspaper over a
democratization of cartography: the notion that modern monthlong period.
technology (e.g., the Web) provides virtually anyone the
capability to design maps. long-term visual memory: an area of memory that does not
require constant attention for retainment; contrast with iconic
digital elevation model (DEM): topographic data com- memory and short-term visual store.
monly available as an equally spaced gridded network.
19
15:57:04.
map communication model: a diagram or set of steps sum- specific information: the information associated with partic-
marizing how a cartographer imparts spatial information to a ular locations on a thematic map.
group of map users. standardized data: a term used when raw totals are adjusted
memory for mapped information: remembering spatial to account for the area over which the data are collected; for
information that was previously seen in mapped form; con- example, we might divide the acres of tobacco harvested in
trast with information acquisition. each county by the areas of those counties.
multimedia: the combined use of maps, graphics, text, statistical map: see thematic map.
pictures, video, and sound. thematic map: a map used to emphasize the spatial distribu-
multimedia systems: interactive mapping systems that tion (or pattern) of one or more geographic attributes (e.g., a
emphasize multimedia aspects. map of predicted snowfall amounts for the coming winter in
paper atlas: a bound collection of maps; modern paper the United States).
atlases also include text, photographs, tables, and graphs that, thematic mapping software: software that is expressly
when combined, tell a story about a place or region. intended for the creation of thematic maps; MapViewer is a
raw totals: raw numbers that have not been standardized to traditional example.
account for the area over which the data were collected (e.g., unclassed map: a map in which data are not grouped into
we might map the raw number of acres of tobacco harvested classes of similar value, and thus each data value can theoret-
in each county, disregarding the size of the counties). ically be represented by a different symbol (e.g., a data set
scientific visualization: the use of visual methods to explore with 100 different data values might be depicted using 100 dif-
either spatial or nonspatial data; the process is particularly ferent shades of gray on a choropleth map).
useful for large multivariate data sets. virtual environment (VE): a three-dimensional computer-
short-term visual store: an area of memory that requires based simulation of a real or imagined environment that users
constant attention (or activation) to be retained; contrast with are able to navigate through and interact with.
iconic memory and long-term visual memory. virtual reality (VR): the use of computer-based systems to
small multiple: many small maps are displayed to show the create lifelike representations of the real world.
change in an attribute over time or to compare many attrib-
utes for the same time period.
REFERENCES
Bertin, J. (1981) Graphics and Graphic Information- Crampton, J. W. (2004) “GIS and geographic governance:
Processing. Berlin: Walter de Gruyter. Reconstructing the choropleth map.” Cartographica 39,
Board, C. (1984) “Higher order map-using tasks: no. 1:41–53.
Geographical lessons in danger of being forgotten.” DeBraal, J. P. (1992) “Foreign ownership of U.S. agricultural
Cartographica 21, no. 1:85–97. land through December 31, 1992.” United States
Bockenhauer, M. H. (1994) “Culture of the Wisconsin Department of Agriculture, Economic Research Service,
Official State Highway Map.” Cartographic Perspectives, Statistical Bulletin 853.
no. 18:17–27. Dent, B. D. (1996) Cartography: Thematic Map Design (4th
Buttenfield, B. P., and Mark, D. M. (1991) “Expert systems in ed.). Dubuque, IA: William C. Brown.
cartographic design.” In Geographic Information Systems: DiBiase, D. (1994) “Designing animated maps for a multime-
The Microcomputer and Modern Cartography, ed. by D. R. dia encyclopedia.” Cartographic Perspectives, no. 19:3–7,
F. Taylor, pp. 129–150. Oxford, England: Pergamon. 19.
Buttenfield, B. P., and Weber, C. R. (1994) “Proactive Dobson, J. E., and Fisher, P. F. (2003) “Geoslavery.” IEEE
graphics for exploratory visualization of biogeographical Technology and Society Magazine 22, no. 1:47–52.
data.” Cartographic Perspectives, no. 19:8–18. Dobson, J. E., Bright, E. A., Coleman, P. R., and Bhaduri, B.
Campbell, C. S., and Egbert, S. L. (1990) “Animated cartogra- L. (2003) “LandScan: A global population database for
phy: Thirty years of scratching the surface.” Cartographica estimating populations at risk.” In Remotely Sensed Cities,
27, no. 2:24–46. ed. by V. Mesev, pp. 267–279. London: Taylor & Francis.
Card, S. K., Mackinlay, J. D., and Shneiderman, B. (eds.). Dodge, M., and Kitchin, R. (2001a) Atlas of Cyberspace.
(1999) Readings in Information Visualization: Using Vision Harlow, England: Addison-Wesley.
to Think. San Francisco, CA: Morgan Kaufmann. Egbert, S. L., Price, K. P., Nellis, M. D., and Lee, R.-Y. (1995)
Cloke, P., Philo, C., and Sadler, D. (1991) Approaching “Developing a land cover modelling protocol for the high
Human Geography: An Introduction to Contemporary plains using multi-seasonal thematic mapper imagery.”
Theoretical Debates. London: Paul Chapman. 1995 ACSM/ASPRS Annual Convention & Exposition,
Technical Papers, Volume 3, Charlotte, NC, pp. 836–845.
15:57:04.
Ferreira, J. J., and Wiggins, L. L. (1990) “The density dial: A Olson, J. M. (1976a) “A coordinated approach to map com-
visualization tool for thematic mapping.” Geo Info Systems munication improvement.” The American Cartographer 3,
1, no. 0:69–71. no. 2:151–159.
Forrest, D. (1999a) “Developing rules for map design: A Orwell, G. (1984) 1984. London: Secker & Warburg.
functional specification for a cartographic-design expert Peterson, M. P. (1995) Interactive and Animated Cartography.
system.” Cartographica 36, no. 3:31–52. Englewood Cliffs, NJ: Prentice Hall.
Goldstein, E. B. (2007) Sensation and Perception (7th ed.). Powell, D. S., Faulkner, J. L., Darr, D. R., Zhu, Z., and
Belmont, CA: Thomson Wadsworth. MacCleery, D. W. (1992) “Forest resources of the United
Harley, J. B. (1989) “Deconstructing the map.” Cartographica States, 1992.” United States Department of Agriculture,
26, no. 2:1–20. Forest Service, Rocky Mountain Forest and Range
Experiment Station, General Technical Report RM-234
Harvey, F., Kwan, M.-P., and Pavlovskaya, M. (2005)
(Revised).
“Introduction: Critical GIS.” Cartographica 40, no. 4:1–4.
Pulsifer, P. L., Caquard, S., and Taylor, D. R. F. (2007)
Holloway, S., Schumacher, J., and Redmond, R. L. (1999)
“Toward a new generation of community atlases—The
“People and place: Dasymetric mapping using ARC/INFO.”
Cybercartograhic Atlas of Antarctica.” In Multimedia
In GIS Solutions in Natural Resource Management:
Cartography, ed. by W. Cartwright, M. P. Peterson, and G.
Balancing the Technical-Political Equation, ed. by S. Morain,
Gartner, pp. 195–216. Berlin: Springer-Verlag.
pp. 283–291. Sante Fe: NM: OnWord Press.
Robertson, P. K. (1991) “A methodology for choosing data
Kraak, M.-J., and Brown, A. (eds.). (2001) Web Cartography:
representations.” IEEE Computer Graphics &
Developments and Prospects. London: Taylor & Francis.
Applications 11, no. 3:56–67.
Kwan, M.-P. (2002) “Feminist visualization: Re-envisioning GIS
Sieber, R. (2006) “Public participation geographic informa-
as a method in feminist geographic research.” Annals of the
tion systems: A literature review and framework.” Annals
Association of American Geographers 92, no. 4:645–661.
of the Association of American Geographers 96, no.
MacEachren, A. M. (1982b) “The role of complexity and 3:491–507.
symbolization method in thematic map effectiveness.”
Sui, D. Z. (2004) “GIS, cartography, and the ‘Third Culture’:
Annals of the Association of American Geographers 72, no.
Geographic imaginations in the computer age.” The
4:495–513.
Professional Geographer 56, no. 1: 62–72.
MacEachren, A. M. (1994b) “Visualization in modern cartog- Taylor, D. R. F. (2003) “The concept of cybercartography.” In
raphy: Setting the agenda.” In Visualization in Modern Maps and the Internet, ed. by M. P. Peterson, pp. 405–420.
Cartography, ed. by A. M. MacEachren and D. R. F. Taylor, Amsterdam: Elsevier Science Ltd.
pp. 1–12. Oxford, England: Pergamon.
Thelin, G. P., and Pike, R. J. (1991). Landforms of the conter-
MacEachren, A. M. (1995) How Maps Work: Representation, minous United States—A digital shaded-relief portrayal,
Visualization, and Design. New York: Guilford. Map I-2206. Washington, DC: U.S. Geological Survey.
MacEachren, A. M., and Taylor, D. R. F. (1994) Visualization (Scale 1:3,500,000.)
in Modern Cartography. Oxford, England: Pergamon. Thrower, N. J. W. (1959) “Animated cartography.” The
MacEachren, A. M., Buttenfield, B. P., Campbell, J. B., Professional Geographer 11, no. 6:9–12.
DiBiase, D. W., and Monmonier, M. (1992) “Visualization.” Thrower, N. J. W. (1961) “Animated cartography in the United
In Geography’s Inner Worlds: Pervasive Themes in States.” International Yearbook of Cartography, pp. 20–29.
Contemporary American Geography, ed. by R. F. Abler, M.
G. Marcus, and J. M. Olson, pp. 99–137. New Brunswick, Tobler, W. R. (1973) “Choropleth maps without class inter-
NJ: Rutgers University Press. vals?” Geographical Analysis 5, no. 3:262–265.
MacEachren, A. M., Wachowicz, M., Edsall, R., and Haug, D. Treinish, L. A. (1992) “Climatology of global stratospheric
(1999c) “Constructing knowledge from multivariate spa- ozone.” Videotape, IBM Corporation.
tiotemporal data: Integrating geographical visualization U.S. Bureau of the Census. (1994) County and City Data
with knowledge discovery in database methods.” Book: 1994. Washington, DC: U.S. Government Printing
International Journal of Geographical Information Science Office.
13, no. 4:311–334. U.S. Bureau of the Census. (1995) Agricultural Atlas of the
Matlin, M. W. (2006) Cognition (6th ed.). Hoboken, NJ: John United States. Washington, DC: U.S. Government Printing
Wiley. Office.
McCormick, B. H., DeFanti, T. A., and Brown, M. D. (1987) Wang, Z., and Ormeling, F. (1996) “The representation of
“Visualization in scientific computing.” Computer quantitative and ordinal information.” The Cartographic
Graphics 21, no. 6. Journal 33, no. 2:87–91.
Morrison, J. L. (1997) “Topographic mapping for the twenty- Zhan, F. B., and Buttenfield, B. P. (1995) “Object-oriented
first century.” In Framework for the World, ed. by D. knowledge-based symbol selection for visualizing statisti-
Rhind, pp. 14–27. Cambridge: Geoinformation cal information.” International Journal of Geographical
International. Information Systems 9, no. 3:293–315.
21
15:57:04.
This page intentionally left blank
15:57:04.
A Historical Perspective
on Thematic Cartography*
OVERVIEW
This chapter covers several aspects related to the history of Frenchmen Charles Joseph Minard and D’Angeville, were
cartography and, more specifically, thematic cartography. prominent early thematic cartographers.
Section 1 provides a brief historical overview of the broad In section 3, we discuss the developments in the history
field of cartography.We will see that the earliest known map of U.S. academic cartography by considering four major
dates back 4,500 years and was produced on a clay tablet, time periods: the incipient, the post-war era, the growth of
obviously a different medium from the computer screens secondary programs, and the integration of GIScience
we use today! Although the Egyptians and the Chinese (Table 1). The incipient period, spanning from the early
produced some of our earliest maps, many of our modern part of the twentieth century to the 1940s, represents what
cartographic ideas date to the time of the Greeks. For example, might be called nodal activity, in which academic cartogra-
the Greek Claudius Ptolemy discussed basic principles of phy was centered at only two or three institutions and
cartography including the construction of projections. As in involved individuals who were not necessarily educated in
many disciplines, cartography flourished during the cartography. An example was John Paul Goode at the
Renaissance (e.g., the first globe was created during that University of Chicago. The post-war era, from the 1940s to
period). In the eighteenth and nineteenth centuries, the first the 1960s, saw the building of core programs that incorpo-
national surveys of countries were completed and methods rated multiple faculty, strong graduate programs, and PhD
for base mapping were developed. students who ventured off to create their own programs.
Section 2 provides an overview of the historical devel- Three core programs stand out—those of the University of
opment of thematic cartography. Although much of the Wisconsin, the University of Kansas, and the University of
development of thematic cartography came about in the Washington. The third period, from the 1960s to the 1980s,
nineteenth and twentieth centuries, we’ll see that significant involved the growth of secondary programs, such as those
advances took place years earlier—for example, Halley at UCLA, Michigan, and Syracuse. It was in this period
developed statistical maps of trade winds and monsoons. that academic cartography emerged as a true subdiscipline
With the compilation of national-level statistics, the field of of geography. Since the 1980s, cartography has become
social cartography, which involves mapping human phe- more integrated with GIScience, though recently there has
nomena, particularly various aspects of population, came been a resurgence in the focus on cartography, geovisual-
into being. As a result, the world saw the development of ization, and representation.
many thematic mapping techniques, including the dot, In section 4, we consider the notion of paradigms in car-
choropleth, isarithmic, and even the dasymetric method in tography, focusing on the paradigm of analytical
the 1800s. The Englishman Edmund Halley, and the cartography, which considers the mathematical concepts
and methods underlying cartography and includes topics
* Portions of this chapter were taken from McMaster, R., and
such as cartographic data models, coordinate transforma-
McMaster, S. (2002) “A history of twentieth-century American acade-
mic cartography.” Cartography and Geographic Information Science tions and map projections, interpolation, and analytical
29, no. 3: 305–321. Reprinted with permission from the American generalization. Waldo Tobler is widely recognized as the
Congress on Surveying and Mapping and Mark Monmonier. founder of analytical cartography.
23
15:58:09.
A Historical Perspective on Thematic Cartography
TABLE 1 The four major periods of U.S. academic Equator and the divisions of the Earth into hot, temper-
cartography ate, and frigid zones; and an accurate measurement of
Incipient period (early 1900s to mid-1940s) the size of the Earth. It was Eratosthenes of Cyrene
J. Paul Goode (276–195 BCE), the head of the Library of Alexandria,
Erwin Raisz who, using some basic knowledge of earth-sun geometry,
Guy-Harold Smith calculated the circumference of the Earth to be 250,000
Richard Edes Harrison stadia, or 40,500 kilometers (25,170 miles), just 425 kilo-
Post-war era of building core programs (mid-1940s to meters (264 miles) from the true value of 40,075 kilome-
mid-1960s) ters (24,906 miles). As it turned out, it was a series of
University of Wisconsin compensating errors that had led to this incorrect esti-
University of Kansas mation. It was also during this period that Claudius
University of Washington Ptolemy of Alexandria (90–168 CE) produced his semi-
Growth of secondary programs (late 1960s to late 1980s) nal Geographia volumes; the eighth volume discussed
UCLA basic principles of cartography including the construc-
University of Michigan tion of projections. Accompanying these volumes were
University of South Carolina numerous maps that located many of the known places
Syracuse University on Earth (Raisz 1948, 10).
Curricula integrated with geographic information Important cartographic works from the Roman period
systems/science (1990s) included The Peutinger Table—really, a scroll 21 feet long
University of California, Santa Barbara and 1 foot high—that depicted the worldwide road net-
SUNY Buffalo work of that time. It was, for all practical purposes, a type
University of South Carolina of premodern road map, or more specifically a TripTik®
(http://en.wikipedia.org/wiki/TripTik). Although cartog-
raphy slipped backward during the Middle Ages, with the
1 A BRIEF HISTORY OF CARTOGRAPHY period’s significant focus on ecclesiastical knowledge in
its maps, such as the common T-in-O map (Obis
Cartography has a long and distinguished history that Terrarum), this period did produce the portolan chart,
dates back thousands of years. Although this chapter will which appears to have been the basis for precise water
focus on some of the major developments in thematic car- navigation. It was also during the Middle Ages that Edrisi,
tography, you should be aware of the rich literature on the an Arabic cartographer, produced a world map that was
broader subject of cartography, including R. V. Tooley’s based on Ptolemy’s representations. However, it wasn’t
Maps and Map-Makers (1987), Lloyd Brown’s The Story until the Renaissance that cartography began to flourish
of Maps (1979), Norman Thrower’s Maps and Civilization once again, with the Age of Discoveries; developments in
(2008), Leo Bagrow and R. A. Skelton’s History of printing and engraving that led to the widespread dissem-
Cartography (1964), and P. D. A. Harvey’s The History of ination of cartographic knowledge; the creation of the
Topographical Maps (1980). You can also refer to the first globe (which still exists today); and the prolific cre-
remarkable series on the history of cartography produced ation of atlases by the Dutch. Specific examples of carto-
by the History of Cartography Project at the University of graphic accomplishments from this period are the Martin
Wisconsin (http://www.geography.wisc.edu/histcart/). Behaim globe of 1492, the printing press, and the Dutch
The earliest known map, dating back 4,500 years, was atlases of Mercator, Ortelius, Hondius, and Blaeu.
created in ancient Babylon. This map, produced on a The eighteenth and nineteenth centuries also saw the
clay tablet, depicts a river valley (most probably the completion of the first national surveys of countries and
Euphrates), mountains, and a larger body of water the development of methods for base mapping. Key to
(Raisz 1948, 5). But many cultures began mapping in accurate base mapping is the method of triangulation,
ancient times, including the Egyptians, who focused on whereby a series of measured baselines and angles are
land surveys for taxation purposes, and the Chinese, calculated. In 1744, César Francois Cassini completed
who had already established basic principles of car- one of the very first baseline projects in France—the
tography in the first few centuries CE. Some of these first triangulation, which became the foundation for the
principles included rectilinear divisions, orientation, complete survey of that country. The survey itself, also
accurate indications of distances, indications of lower designed by Cassini, was completed in the late 1700s
and higher altitudes, and attention to the right and left at a scale of 1:86,400 and was composed of 182 sheets
angles of bends in roads (Raisz 1948, 6). depicting the cultural and natural features of France.
Many of our modern cartographic ideas date to the Another of the early national surveys was the Ordnance
Greeks, who in the 4th century introduced the ideas of Survey of Great Britain, published in 1801 at a scale
the spherical shape of the Earth; the concept of the of 1:63,360.
15:58:09.
2 HISTORY OF THEMATIC CARTOGRAPHY growth, and some simply on the burgeoning interest in
descriptive geography. For a considerable period in the two
centuries the concern focused on such topics as popula-
The history of thematic mapping is more recent, dating tion, fertility, mortality, and the comparative characteristics
largely to the nineteenth and twentieth centuries, of nations.
although we will see that there were developments
before the nineteenth century. Many of the original One can find numerous examples of creative the-
developments occurred in Europe, driven by the acqui- matic cartography during the seventeenth century, but
sition of new forms of statistical data. As described by perhaps the most interesting work is that of Edmund
Robinson (1982, 32): Halley, a seminal member of the British scientific com-
munity (Thrower 1969; Robinson 1982). Halley’s most
The development of a concern for science and for a uni-
significant discovery, of course, was the comet named
versal system of measure from the mid-seventeenth to the
mid-nineteenth century in Europe was accompanied by a
after him, but he also created a series of ingenious sta-
parallel growth of interest and competence in statistics and tistical maps depicting the trade winds and monsoons
statistical method, especially in physical science. In the (1686), the path of the eclipse of the sun (1715), and,
realm of social affairs some of this was based on theoreti- perhaps the most famous, his 1701 chart of compass
cal aspects, for example probability theory, some on prob- variations in the Atlantic Ocean that uses isarithmic
lems resulting from growing populations and industrial lines (isogones) (Figure 1). The 1715 eclipse map, which
FIGURE 1 Halley’s 1701 chart of

compass variations in the Atlantic
Ocean using the isarithmic technique
(from Robinson, Arthur H. 1982.
Early Thematic Mapping in the History
of Cartography. University of Chicago
Press; originally acquired from the U.S.
Library of Congress).
25
15:58:09.
depicts the duration and timing of the eclipse social cartography involves mapping all types of human
as it passes across England, was widely distributed phenomena (e.g., language, religion, ethnicity, eco-
throughout England to help the population understand nomic activity, and transportation), various aspects of
the path of this phenomenon. population were the focus in nineteenth-century
Much of the early work in thematic cartography Europe. Some of the seminal works included Frère de
focused on the mapping of natural phenomena and was Montizon’s map of the population of France, dated
published in major atlases. Some of this early work con- 1830 (using the dot-mapping technique; see Figure 2);
centrated on the portrayal of relief such as those the 1848 world map of population density and food
in Heinrich Berghaus’s Physikalischer Atlas (1842), J. L. habits in Berghaus’s Physikalischer Atlas (using the
Dupain-Triel’s 1791 map of France, and Alexander choropleth technique); D’Angeville’s 1836 map of the
Johnston’s 1852 map of “The Mountains, Table Lands, number of persons per square myriamètre (using the
Plains, & Valleys of Europe” (Robinson 1982). In this choropleth technique; see Figure 3); Harness’s 1837
same period, maps of the biological world began to map of the number of persons per square mile in
appear. Some depicted agricultural distributions, such Ireland (using the dasymetric technique); and
as Schouw’s 1823 map of cereals (rye, wheat, maize) and Petermann’s 1848 map of cholera in the British Isles (a
the 1839 world map of cultivated plants in Berghaus’s map based on continuous tones; see Figure 4)
Physikalischer Atlas. One of the most interesting maps (Robinson 1982). Some of the more interesting works of
is that of the world distribution of plants that depicts this era were produced by Charles Joseph Minard, who
Schouw’s 25 phytogeographic regions. Robinson (1982, was trained as an engineer but then became interested in
103–105) pointed out that many of the cartographic geography. As documented in Robinson’s Early
techniques used for the mapping of natural vegetation Thematic Mapping, Minard produced maps of the
and cultivated plants—including using colors, patterns, amount of butcher’s meat sent to Paris from each depart-
boundaries of extents, and isolines—were straightfor- ment (1858), the extent of the markets for foreign oil and
ward, while the acquisition of data was more difficult. coke (1961), the volume of French wines exported by sea
(1865), and the movement of travelers on the principal
railways of Europe (1865). Minard was a prolific and
2.1 The Rise of Social Cartography creative cartographer who used a variety of innovative
cartographic methods including segmented graduated
With the acquisition of national-level statistics, the circles, colored regions, and graduated flow maps.
field of social cartography rapidly evolved. Although During the rise of social cartography, cartographers and
FIGURE 2 Frère de Montizon’s

1830 dot map of the population of
France; each dot represents 10,000
people (from Robinson, Arthur H.
1982. Early Thematic Mapping in the
History of Cartography. University of
Chicago Press; original acquired from
Bibliothèque Nationale, Paris).
15:58:09.
FIGURE 3 D’Angeville’s 1836

choropleth map of the number of
persons per square myriamètre (100
square kilometers); in contrast to the
convention followed today, the light-
est tone represented the highest pop-
ulation density, and the darkest tone
represented the lowest population
density (from Robinson, Arthur H.
1982. Early Thematic Mapping in the
History of Cartography. University
of Chicago Press; original acquired
from Bibliothèque Nationale, Paris).
statisticians also began to map health data such as of this chapter will focus on some of the major activ-
diseases and sanitation, the most famous being ity in the United States, where academic thematic car-
Dr. John Snow’s map depicting the spread of cholera. tography grew out of geography departments. But
It is interesting to study these (mostly European) early academic thematic cartography also existed in other
examples of thematic maps, as they utilized symbology places, particularly Europe, where, for instance, many
that we still use today. key ideas in terrain representation and label place-
ment (Imhof 1975), map symbolization (Bertin 1981,
1983), cartographic design (Keates 1973), and gener-
3 HISTORY OF U.S. ACADEMIC CARTOGRAPHY alization (Eckert 1921) were formulated. It was
also the European cartographers who initiated the
Thematic cartography morphed into a true academic development of the International Cartographic
discipline during the twentieth century. A rich history Association (ICA) in the late 1950s, with the first
of ideas and a complex relationship with other International Cartographic Conference held in 1961
disciplines, most often geography, exist. The remainder in Paris.
27
15:58:09.
FIGURE 4 Petermann’s 1848 con-

tinuous-tone map of cholera in the
British Isles; the original map was
tinted in red, with darker shades rep-
resenting a higher mortality ratio
(from Robinson, Arthur H. 1982.
Early Thematic Mapping in the
History of Cartography. University of
Chicago Press; original acquired
from the British Library).
3.1 Period 1: The Incipient Period
John Paul Goode

Although basic training in cartography started as early
as 1900 in the United States, “It could be argued that the
first genuine American academic cartographer was John
Paul Goode at the University of Chicago” (McMaster
and Thrower 1987, 346). Goode (Figure 5) was one of
the most professionally active geographers of his time.
During the early part of the twentieth century, Goode
became a charter member of the Association of
American Geographers (AAG), served as co-editor of
the Journal of Geography from 1901 through 1904,
helped organize the Geographic Society of Chicago, and
was appointed by President Taft to assist in leading a
U.S. tour for a distinguished group of Japanese
financiers. However, Goode is best known for the devel-
opment of Goode’s Homolosine Projection, first
presented at the AAG’s meetings in 1923, and the devel-
opment of Goode’s World Atlas, also published in 1923
and now in its 21st edition. As an example of his teach-
ing, in 1924 Goode taught a course at the University of
Chicago entitled “A Course in Graphics,” which
included four major themes: The Graph, The Picture,
Preparation of Illustrative Material for Reproduction, FIGURE 5 J. Paul Goode, who was arguably the first
and Processes of Reproduction. It is clear that Goode genuine American academic cartographer. (After McMaster
used this course as a prerequisite to the more traditional and Thrower 1991. First published in Cartography and
cartography class that he taught. Geographic Information Systems 18(3), p. 151. Reprinted
Goode’s influence was extended through his stu- with permission from the American Congress on Surveying
dents at Chicago. Most of these students did not devote and Mapping.)
15:58:09.
themselves to cartography specifically, but some were

able to influence the course of the field through their
positions in the private sector, government, and acade-
mia. Two of Goode’s students at the University of
Chicago, Henry Leppard (University of Chicago,
Washington, and UCLA) and Edward Espenshade
(Northwestern University), devoted their careers to
cartographic education, and continued Goode’s work
with both base map development and the many gener-
ations of the Goode’s World Atlas, published by Rand
McNally.
Erwin Raisz
Erwin Raisz (Figure 6) was the leading American acade-
mic cartographer between the time of Goode’s death and
the emergence of major graduate programs at Wisconsin,
Kansas, and Washington. Born in Hungary in 1893, Raisz
immigrated to the United States after World War I. While
working on his PhD in geology at Columbia, Raisz studied
with Douglas Johnson, a William Morris Davis–trained
geomorphologist at Harvard who had strong interests in
the construction of block diagrams and the representation
of landscapes.* As an instructor at Columbia, Raisz offered
the first cartography course there.
In 1931, Raisz published “The Physiographic Method FIGURE 6 Erwin Raisz, who was a leading American acade-
of Representing Scenery on Maps” in the Geographical mic cartographer between the time of Goode’s death and the
Review, where he expressed his own individualized emergence of major graduate programs in cartography. (After
approach to landscape representation. Based on this McMaster and Thrower 1991. First published in Cartography
work and on the recommendation of his mentor, and Geographic Information Systems 18(3), p. 153. Reprinted
Douglas Johnson, Davis himself offered Raisz the posi- with permission from the American Congress on Surveying
tion of lecturer in cartography at the Institute of and Mapping.)
Geographical Exploration at Harvard. During the 1930s,
Raisz continued to publish and work on his physio-
graphic techniques. In 1938, he published the first edi-
tion of the influential book General Cartography, which Although the geography department at Harvard
was to remain the only general English textbook on car- folded in 1947 (Smith 1987a), Raisz remained in
tography for 15 years. The book was part of the geogra- the Boston area, teaching at Clark University until
phy series published by McGraw-Hill, which at the same 1961, and finished his career at the University of
time was publishing such classics as Finch and Florida, where he published his second textbook,
Trewartha’s Physical Elements of Geography, Platt’s Principles of Cartography, in 1962. What is curious
Latin America, and Whitbeck and Finch’s Economic about Raisz is that he never held a regular academic
Geography. In addition to the predictable emphasis on appointment. Thus, he was unable to produce a gener-
“representation,” the contents of General Cartography ation of students who would perpetuate his brand of
were what one would expect to find at this time period: cartography. Raisz is best known for his production of
projections, lettering, composition, and drafting. A sig- “landform” maps of various parts of the world. His
nificant part of Raisz’s book also covered the history of “Landform Outline Map of the United States” (1954),
cartography, with sections on manuscript maps, the perhaps one of the best examples of cartography in
renaissance of maps, the reformation of cartography, the twentieth century, has become a standard refer-
and American cartography. What is most interesting is ence in U.S. geography classes (see section 20.2.3 for a
the relatively small amount of text that was devoted to discussion of this map). Robinson and Sale (1969)
“statistical mapping” as we know it today. asserted that landform maps or physiographic dia-
grams, such as those created by William Morris Davis,
A. K. Lobeck, Guy-Harold Smith, and Raisz, were “one
*One of Davis’s students, A. K. Lobeck, carried on the tradition of the more distinctive contributions of American
of using block diagrams to represent landscapes. cartography” (p. 187).
29
15:58:09.
Guy-Harold Smith Richard Edes Harrison

Cartography was taught as early as 1925 at Ohio Richard Edes Harrison (Figure 8), who was born in
State University—a class in Map Construction and Baltimore in 1901, was the son of Ross Granville
Interpretation was offered by Fred Carlson—making Harrison, one of the most distinguished biologists of his
this one of the oldest cartography courses in the coun- time. Although Harrison graduated with a degree in
try. In 1927, Guy-Harold Smith (Figure 7), a recent architecture from Yale University in 1930, his interests
PhD recipient from the University of Wisconsin soon turned to scientific illustration, and in the years
(A. K. Lobeck was his advisor), took over the cartog- after completing his degree, he eventually was attracted
raphy program, and taught for nearly 40 years until to cartography. He drew his first map for Time maga-
his retirement in 1965. Although chair of the depart- zine in 1932. This initial exposure to mapping piqued his
ment for 29 years, Smith was also a prolific thematic curiosity, and he soon became a freelance cartographer
cartographer, producing his famous “Relative Relief for Time and Fortune magazines. In the late 1940s,
Map of Ohio” and “Population Map of Ohio” using Harrison flew to Syracuse University once a week to
graduated spheres. A talented teacher, his best-known teach a course in cartography (George Jenks was one of
student was Arthur Robinson, who created the influ- his students). Although not formally an educator,
ential program in cartography at the University of Harrison nonetheless influenced the discipline of
Wisconsin, which we describe later in this chapter. cartography through his specific technique and intrinsic
cartographic abilities. He can also be considered one of
the first “popular” cartographers because of his work in
media mapping.
3.2 Period 2: The Post-War Era and the Building

of Core Academic Programs
A seminal event in the evolution of American acade-

mic cartography—the first meeting of the Committee
FIGURE 7 Guy-Harold Smith, who was an early thematic

cartographer who produced a famous map of graduated
spheres. Arthur Robinson was one of Smith’s students. (After
McMaster and Thrower 1991. First published in Cartography
and Geographic Information Systems 18(3), p. 153. Reprinted FIGURE 8 Richard Edes Harrison, who was one of the first
with permission from the American Congress on Surveying “popular” cartographers because of his work in media map-
and Mapping.) ping; George Jenks was one of his students.
15:58:09.
on Cartography—was organized by Erwin Raisz on completing a stint in the Geography Division of the
April 6, 1950, at Clark University during the national Office of Strategic Services (OSS) in World War II. In
meetings of the AAG. Based on the results of this his faculty position at Wisconsin, Robinson was respon-
important conference, it was clear at the time that car- sible for establishing a cartography and map use
tography was positioning itself in relation to geogra- instructional program that, at the outset, included two
phy and other disciplines even though it was still basic cartography courses (i.e., introductory and inter-
rather descriptive and atheoretical. Fortunately, a mediate cartography) as well as an aerial photo inter-
series of academic cartographers with strong interests pretation course. Later, other courses were added
in more conceptual and theoretical issues emerged including Seminar in Cartography, Cartographic
during the 1950s and led the development of basic Production, and Use and Evaluation of Maps. These
research programs. Although only a sporadic set of were followed by another series of courses in Map
institutions had offered courses in cartography before Projections and Coordinate Systems, Problems in
World War II, the demands of the war accelerated the Cartography, Computer Cartography, History of Map
development of cartographic curricula. Over the fol- Making, and Cartographic Design. In the late 1960s, the
lowing decade (1950–1960), three major programs cartography staff increased when Randall Sale became
would emerge at the Universities of Wisconsin, associate director of the University of Wisconsin
Kansas, and Washington. Cartographic Laboratory.
In a 1979 paper on World War II’s influence on car-
University of Wisconsin tography, Robinson wrote about these early years:
Although Arthur Robinson (Figure 9) traced the teach-
ing of cartography at the University of Wisconsin, In the development of cartography in American academic
geography, probably the most notable event prior to World
Madison, back to 1937 (1991, 156), the key beginning
War II was the publication of Erwin Raisz’s General
occurred when Robinson was hired in 1945 after Cartography, in 1938. By the mid-1930s the majority of
graduate students in geography (probably few if any
undergraduates) took one course in cartography. Mine, at
Wisconsin, came before Raisz’s book appeared and our
“textbooks” were Deetz and Adams’ Elements of Map
Projections and Loebeck’s Block Diagrams. Besides con-
structing map projections and making crude dot map, iso-
pleth, and pie chart maps, we were taught how to tint glass
lantern slides with Japanese water colors. When I trans-
ferred to Ohio State University for doctoral work, I was
not allowed to take the cartography course because I
already had one! Raisz’s pioneering book provided a small
beginning for an academic program in cartography, but
nothing really got started before World War II began.
(1979, 97)
Arthur Robinson established himself as the unofficial

“Dean” of American academic cartographers, and built
the cartography program at the University of Wisconsin
into the very best in the United States during the 1970s
and early 1980s. His seminal volume The Look of Maps,
based on his doctoral dissertation at Ohio State
University, was arguably the seed for three decades of
cartographic research. Robinson established the first
American journal in cartography, The American
Cartographer, in 1974; wrote six volumes of Elements of
Cartography; and was president of the International
Cartographic Association. Robinson also influenced
several generations of students who themselves
ventured off and established graduate programs in car-
FIGURE 9 Arthur Robinson, who was responsible for devel- tography. Robinson and Sale guided the cartography
oping the cartography program at the University of Wisconsin. program at Madison until 1968, when Joel Morrison, a
31
15:58:09.
Robinson PhD, and later Philip Muehrcke joined them. had collected during this grant year into an Annals of
Both brought strong mathematical expertise to the pro- the Association of American Geographers paper enti-
gram. Thus, in the mid-1970s, when many geography tled “An Improved Curriculum for Cartographic
departments were struggling to maintain a cartography Training at the College and University Level,” and
program with a single faculty member, Wisconsin had adopted it for the cartography program at Kansas.
four individuals, separate BS and MS degrees in cartog- The grant enabled Jenks (1953a) to identify a series
raphy, and the very best cartography laboratory (within of key problems for cartographers, including: (1) mass-
a geography department). It was a cartographic tour production techniques had to be improved; (2) new
de force. The cartography program at Wisconsin has inks, papers, and other materials were needed; and (3)
produced more than 100 students with Master’s degrees additional personnel had to be trained. Jenks pointed
and 20 students with doctoral degrees over its 60-year out that, at the time, several factors impeded carto-
history. graphic training: inexperienced instructors, poorly
equipped cartographic facilities and map libraries, lim-
University of Kansas ited research and limited access to research, and too
The cartography program at the University of Kansas much emphasis on theory. It is interesting that Jenks,
was started and nurtured for more than 35 years by who spent much of his research career building carto-
George Jenks (Figure 10). Jenks received his PhD in graphic theory in design, symbolization, and classifica-
agricultural geography from Syracuse University and tion, would make such an argument. But he wrote, “too
also studied with Richard Edes Harrison, the cartogra- little time and effort has been spent on the practical
pher for Time and Fortune magazines, at Syracuse. A application of theory. Theorizing on art does not make
significant event in Jenks’s career, and for the program an artist, knowledge of medical theory does not make a
itself, was his receiving a grant from the Fund for the qualified doctor, and talking about maps (and listening
Advancement of Science that allowed him to visit all to lectures on cartography) does not mean that the stu-
major mapmaking establishments of the federal dent can execute a map” (Jenks 1953a, 319).
government as well as a number of quasi-public labora- The grant also led Jenks to propose the following
tories in 1951–52. He incorporated the information he five-course core sequence in cartography:
Course 1. Elementary training in projections, grids,
scales, lettering, symbolization, and simple
map drafting.
Course 2. The use, availability, and evaluation of maps.
Course 3. Planning, compiling, and constructing small-
scale maps, primarily subject-matter maps.
Course 4. Planning, compiling, and constructing large-
scale maps, primarily topographic maps.
Course 5. Nontechnical training in the preparation of
simple manuscript maps for persons wishing
the minimum in the manipulative aspects of
cartography.
The importance of Jenks’s landmark study cannot be
overemphasized. Cartography had emerged from World
War II as a true discipline, in part due to the great
demand for war-effort maps and mapping. Both those
who had been practicing before the war, such as Arthur
Robinson and Erwin Raisz, and those who emerged
after, such as George Jenks and John Sherman, realized
that comprehensive cartographic curricula could be
maintained within geography departments. Jenks’s
study, in parallel with the previously described efforts
by Erwin Raisz and the AAG, provided the intellectual
infrastructure for those attempting to build cartography
programs at universities.
FIGURE 10 George Jenks, who was responsible for develop- Other significant influences on Jenks’s early career
ing the cartography program at the University of Kansas. were his relationship with John Sherman of the
15:58:09.
University of Washington and their establishment in the

1960s of the National Science Foundation–funded
Summer Institutes for College Teachers. The institutes,
nine weeks each in length, were designed to educate
college professors in the modern techniques of cartog-
raphy. Jenks wrote, “We were surprised to find that a
number of professors had been assigned arbitrarily by
their deans or chairman to teach mapmaking that fall. HANS
Moreover, several were going to have to teach without

a laboratory, equipment, or supplies. These activities
greatly enhanced my teaching and were the basis for
numerous changes in our curriculum” (1963, 163).
Despite the presence of faculty at Kansas who had
interests closely related to cartography—in particular,
statistics and remote sensing—at the end of the 1960s,
Jenks was still the only cartographer on the staff.
Robert Aangeenbrug, who had strong interests in com-
puter cartography and urban cartography and was the FIGURE 11 John Sherman discussing lunar modeling with a
director of two of the International Symposium on group of students. Sherman was responsible for developing
Computer-Assisted Cartography (Auto-Carto) confer- the cartography program at the University of Washington.
ences, joined the Kansas faculty in the 1960s. Thomas
Smith, who had arrived in the department as its second
When Donald Hudson joined the staff at the
hire in 1947, established coursework during the 1970s
University of Washington in 1951, he implemented a
and 1980s in the history of cartography.
new program for the Geography Department that con-
The Kansas program grew rapidly in the 1970s. As
centrated on Anglo-America, the Far East, economic
explained by Jenks, “George McCleary joined the staff,
geography, and cartography. He asked Sherman to lead
and with his help we renovated and broadened the
the cartography concentration and also invited Henry
offerings in cartography. More emphasis was placed on
Leppard, who had studied under J. Paul Goode at the
map design and map production, and new courses at the
University of Chicago, to assist. As explained by
freshman and sophomore level were added” (1991,
Sherman:
164). During this period, Jenks initiated research pro-
jects on 3-D maps, eye-movement research, thematic
By 1953 six cartography courses were in place, including
map communication, and geostatistics. By the end of
Maps and Map Reading, Introductory Cartography,
the 1970s, Jenks had turned his attention to carto- Intermediate Cartography, Techniques in the Social Sciences,
graphic line generalization, and supervised 10 PhDs, Map Reproduction, and Map Intelligence. In 1954, Leppard
15 MA candidates, and four postdoctoral cartographers. left for UCLA and in 1958 Willis Heath, having completed
Many of these individuals accepted academic appoint- his Ph.D. in the department, joined Sherman in carrying on
ments, continued the “Jenks school,” and made signifi- the cartography program. (1991, 169)
cant contributions to the field of cartography through
their teaching and research. Jenks continued to teach One noteworthy event in the early history of the
and be engaged in research until his retirement at program was Heath and Sherman’s participation in
Kansas in 1986. the Rand McNally-sponsored Second International
Cartographic Conference at Northwestern University,
University of Washington held in June 1958. According to Sherman (1991, 169), a
Although the first course at the University of group of some 50 international cartographers were able
Washington that was formally identified as a “cartogra- to discuss “the graphic philosophy, functional analysis,
phy” course had been offered by William Pierson in the and technological developments that were then influ-
Geography Department during the 1937–38 academic encing the field.” Based on discussions at the conference,
year, John Sherman (Figure 11) is primarily considered changes and additions were made to the cartography
to have developed the cartography program there. program at Washington. Another event influencing
Sherman received his MA from Clark University in 1944 Sherman, and thus the cartography program at
and his PhD from Washington in 1947. Interestingly, Washington, was the Summer Institute for College
unlike Jenks and Robinson, who had both received Teachers in Cartography that Sherman and Jenks led
formal training in cartography, Sherman never had any (see our earlier discussion). Later, in 1968, Sherman
coursework in cartography. developed a proposal to establish a National Institute
33
15:58:09.
of Cartography, as requested by the National Academy earlier days when students would pursue a general
of Sciences/National Research Council (NAS/NRC) graduate program in cartography, individual graduate
Committee on Geography. A panel of prestigious programs were tailored to students’ particular research
cartographers, including Arch Gerlach, Norman Thrower, specialty, such as cognitive or analytical cartography.
Richard Dahlberg, Waldo Tobler, George McCleary, This period also witnessed rapid growth in academic
George Jenks, and Arthur Robinson, assisted Sherman. cartography in terms of faculty hired, students trained,
Unfortunately, the proposed institute was never journals started, and professional societies strength-
created. ened. It was in this period that cartography emerged as
The program Sherman and Heath put together a true academic subdiscipline, nurtured within acade-
strongly emphasized both design and production. mic geography departments that had strong research
Personal correspondence with Carlos Hagen (a grad- programs and well-established graduate education pro-
uate student at Washington during this period) sup- grams. The pinnacle of academic cartography in the
ports this. Hagen wrote, “One thing that particularly United States was in the mid-1980s, when cartography
impressed me at that time was the importance that had reached its maximum growth but the effect of the
John Sherman and Bill Heath gave to a sort of sacred emerging discipline of GIScience had not yet been felt.
trilogy, ‘Drafting-Printing-Reproduction.’ In the Latin
American and European traditions, these production 3.4 Period 4: Integration with GIScience
techniques are certainly not considered part of the
academic environment. They are very respected and The intellectual landscape of cartography has changed
much appreciated, but generally you will find them not significantly over the past 10 to 15 years, in large part
in academia, but in the realm of a very professional and due to the rapid growth of GIS and GIScience. Fifteen
dedicated tradition of craftsmanship” (personal com- years ago, the prognosis for a PhD in cartography
munication 1987). Whereas Sherman’s main research being able to acquire an academic position was excel-
interests were in communication, map design, and lent, whereas today’s job market seeks out the geo-
tactile mapping, many of his doctoral students pursued graphic information scientist. One can certainly still
dissertation topics related to analytical and computer study cartography at most major institutions, but the
cartography. number of cartographic courses has decreased as the
number of GIS-related courses has increased.
3.3 Period 3: Growth of Secondary Programs Additionally, the field of geographic visualization
increasingly used by many departments instead of
During the 1970s and 1980s, a series of what might be cartography, has caused a further erosion of the pro-
called secondary programs, many established by fessional base of cartography. However, one hope for
PhDs from the Universities of Wisconsin, Kansas, and the discipline is that as GISs become almost ubiqui-
Washington, were created in the United States. tous in our society, some researchers are realizing that
Although not exhaustive, one can point to programs at a deeper knowledge of maps, cartography, and map
UCLA with Norman J. W. Thrower (a Wisconsin PhD), symbolization and design is still crucial and necessary.
Michigan with Waldo Tobler (a Washington PhD), Kraak and Ormeling (1996), in their textbook
South Carolina with Ted Steinke and Patricia Gilmartin Cartography: Visualization of Spatial Data, made the
(Kansas PhDs), SUNY Buffalo with Kurt Brassel (a following point with respect to the relationship
Zurich PhD) and Duane Marble (a Washington PhD), between GIS and cartography:
Michigan State with Richard Groop (a Kansas PhD)
and Judy Olson (a Wisconsin PhD), Northern Illinois Many of the concepts and functions of GIS were first con-
University with Richard Dahlberg (a Wisconsin PhD), ceived by cartographers. This is not only valid for the GIS
Oregon State University with A. Jon Kimerling (a output module, but for many of the processing actions (e.g.,
Wisconsin PhD), Syracuse with Mark Monmonier transformations, analyses) and input functions (e.g., digitiz-
ing, scanning) of a GIS as well. There are conflicting views
(a Pennsylvania State PhD), Penn State University with
regarding the relations between cartography and GIS, viz.
Alan MacEachren (a Kansas PhD), and Ohio State with
whether GIS is a technical-analytical subset of cartography,
Harold Moellering (a Michigan PhD). Key activities in or whether cartography is just a data visualization subset of
these departments included Tobler’s development of GIS. For the purpose of this book, also written for GIS ana-
analytical cartography, Thrower’s expertise in the his- lysts who have to learn to use the cartographic method, car-
tory of cartography and remote sensing, Moellering’s tography will be regarded as an essential support for nearly
animated cartography and emphasis on a numerical car- all aspects of handling geographical information. (p. 16)
tography, Monmonier’s statistical mapping, and Olson’s
work in cognitive research. Each of the institutions There have been major changes in the way cartogra-
developed its own area of expertise in which, unlike the phy is taught in American universities. Some of the
15:58:09.
most significant changes include: (1) closer integration

with GIS education; (2) the nearly complete transition
to digital methods; (3) less emphasis on procedural
programming (e.g., Fortran and Pascal) and greater
emphasis on object-oriented, user interface, and
Windows programming; (4) greater emphasis on the
dynamic aspects of cartography, including animation
and multimedia; and (5) greater use of the Internet for
acquiring spatial data (e.g., census data) and for online
mapping applications.
4 THE PARADIGMS OF AMERICAN CARTOGRAPHY
As academically oriented graduate programs emerged

in the post–World War II period, basic research in car-
tography accelerated. Although we could document
many research paradigms, some of the more substantial
efforts were made in communication models, a theory
of symbolization and design, experimental cartography,
analytical cartography, and the recent series of debates
on social and ethical issues in cartography.
4.1 Analytical Cartography
If any one paradigm within cartography has an FIGURE 2.12 Waldo Tobler, who was the developer of the
“intellectual leader,” it is analytical cartography. Waldo notion of analytical cartography.
Tobler (Figure 12) originated (in the 1960s) and
nurtured (in the 1970s and 1980s) the idea of mathe-
matical, transformational, and analytical approaches to only remain flexible if he has command of a theoretical
structure as well as specific implementation. The spirit of
cartography. Tobler laid out the agenda for analytical
Analytical Cartography is to try to capture this theory,
cartography in his seminal 1976 paper “Analytical in anticipation of the many technological innovations
Cartography,” published in American Cartographer. which can be expected in the future; wrist watch
This paper, and Tobler’s ideas, have had a profound latitude/longitude indicators, for example, and pocket
effect on American academic cartography. calculators with maps displayed by colored light emitting
What exactly is “analytical cartography”? Kimerling, diodes, do not seem impossible. In a university environ-
in his 1989 Geography in America review of cartogra- ment one should not spend too much time in describing
phy, described it as “the mathematical concepts and how things are done today. (1976, 29)
methods underlying cartography, and their application
in map production and the solution of geographic prob- Tobler finished his PhD in 1961 at the University of
lems” (p. 697). Analytical cartography includes the top- Washi ngton under John Sherman, completing a doc-
ics of cartographic data models, digital cartographic toral dissertation entitled “Map Transformations of
data-collection methods and standards, coordinate Geographic Space.” While at Washington, Tobler was
transformations and map projections, geographic data influenced not only by some of the faculty’s (William
interpolation, analytical generalization, and numerical Garrison, for instance) strong emphasis on quantifica-
map analysis and interpretation. Tobler’s original tion, but also by the large number of graduate students
syllabus described a series of topics steeped in theory who were interested in mathematical geography, includ-
and mathematics, and his goal for the course was futur- ing Duane Marble, Arthur Getis, Brian Berry, and John
istic (as well as prescient): Nystuen, among others. In the early 1960s, the
Department of Geography at Washington was ground
What is easy, convenient, or difficult depends on the tech- zero for the quantitative revolution in geography. Many
nology, circumstances, and problem. The teaching of car- of its students had enrolled in J. Ross MacKay’s
tography must reflect this dynamism, and the student can Statistical Cartography course, which MacKay taught in
35
15:58:09.
the late 1950s. In a personal interview, Tobler (2001) had been developed at the Harvard Laboratory for
also discussed the influence of Carlos Hagen, a gradu- Spatial Analysis and Computer Graphics, including the
ate student at Washington who had arrived from Chile concept of topological data structures, were directly
in the late 1950s hoping to pursue graduate work in transferred to modern GISs, such as the Environmental
mathematical cartography. Tobler himself actually had System Research Institute’s ARC/Info product
little training in formal mathematics, but was self-taught (Chrisman 1998a).
and was intrigued by Hagen’s work in projections.
After completing his dissertation at Washington,
Tobler joined the faculty at the University of Michigan, SUMMARY
where his graduate student colleague from Washington,
John Nystuen, had also moved. It was at Michigan that We have seen that many of our modern cartographic
Tobler honed his ideas on analytical cartography, in part ideas date to the time of the Greeks, who introduced
assisted by a relatively obscure event in American geog- the ideas of the spherical shape of the Earth; the
raphy: the meetings of the Michigan IntraUniversity concept of the Equator and the divisions of the Earth
Community of Mathematical Geographers (MIC- into hot, temperate, and frigid zones; and an accurate
MOG). Many of the topics presented at these Detroit- measurement of the size of the Earth. Although cartog-
based meetings were strongly cartographic in nature, raphy slipped backward during the Middle Ages, it
including Gould’s “Mental Maps,” Perkal’s “Epsilon flourished again during the Renaissance with the devel-
Filtering,” and Tobler’s own work on generalization. opments of printing and engraving, which led to the
Tobler’s work, which significantly influenced the disci- widespread dissemination of cartographic knowledge,
plines of both cartography and geography, led to his the creation of the first globe, and the prolific creation
election into the prestigious National Academy of of atlases by the Dutch. The eighteenth and nineteenth
Sciences, making him the only cartographer to hold that centuries saw the completion of the first national
honor. surveys of countries and the development of scientific
What emerged from the concept of “analytical methods for base mapping.
cartography” was a cadre of individuals working on Compared to the history of the broader discipline of
problems that can be identified as analytical, computa- cartography, the history of thematic cartography is rela-
tional-digital, and mathematical in nature. Some were tively recent, with major developments not taking place
Tobler’s own PhD students, such as Stephen Guptill until the nineteenth and twentieth centuries. In the early
(USGS), Harold Moellering (Ohio State University), 1800s, with the compilation of national-level statistics
and Phil Muehrcke (Universities of Washington and and the evolution of the field of social cartography, the
Wisconsin). Others, immersed in the paradigm without world saw the development of several basic thematic
necessarily having obtained a formal education in it, mapping techniques, including the dot, choropleth, and
were Mark Monmonier, the author of the first textbook dasymetric methods. Charles Joseph Minard was one of
on computer cartography; Carl Youngmann (a Jenks- the pioneers of thematic.
educated cartographer at Kansas who joined Sherman The history of U.S. academic cartography can be
at Washington); and Jean-Claude Muller (another Jenks divided into four periods: the incipient, the post-war
student who worked at the University of Georgia, the era, the growth of secondary programs, and the inte-
University of Alberta, the International Training Center gration with GIScience. In the incipient period (from
in the Netherlands, and the University of Bochum in the early part of the twentieth century to the 1940s),
Germany). Additionally, a large group of individuals academic cartography was centered at only two or
who had been educated in the late 1970s through the three institutions and involved individuals who were
early 1980s considered themselves computer or analyti- not necessarily educated in cartography; key figures
cal cartographers, including Terry Slocum (PhD, of this era included John Paul Goode, Erwin Raisz,
University of Kansas), Keith Clarke (PhD, University Guy-Harold Smith, and Richard Edes Harrison. The
of Michigan), Nicholas Chrisman (PhD, Bristol), post-war era (1940s to 1960s) saw the creation of
Timothy Nyerges (PhD, Ohio State University), Marc three core programs at the University of Wisconsin
Armstrong (PhD, University of Illinois), Barbara (Arthur Robinson), the University of Kansas
Buttenfield (PhD, University of Washington), and (George Jenks), and the University of Washington
Robert McMaster (PhD, University of Kansas). (John Sherman). Secondary programs developed
A strong argument can be made that the principles during the third period (1960s to 1980s) included
of numerical, analytical, and digital cartography became those at UCLA, Michigan, and Syracuse. It was in
the “core” of modern GISs. For instance, many of the this period that cartography emerged as a true
basic ideas in analytical and computer cartography that subdiscipline, nurtured within academic geography
15:58:09.
departments that had strong research programs and We can document several research paradigms that
well-established graduate education programs. The came about as graduate programs evolved after World
integration of cartography with GIScience has been War II, including communication models, a theory of
occurring since the 1980s and has raised questions symbolization and design, experimental cartography,
about the direction that cartography will take in the analytical cartography, and the recent series of debates
future. While it can be argued that cartographic on social and ethical issues in cartography. In this
principles are critical to GIS (Kraak and Ormeling chapter, we considered the analytical cartography
1996), the number of courses in cartography has paradigm, which involves the mathematical concepts
decreased as the field of geographic information and methods underlying cartography and includes
science/systems—which, many assume, “contains” topics such as cartographic data models, coordinate
cartography/representation—has begun to dominate transformations and map projections, interpolation,
most university curricula. and analytical generalization.
FURTHER READING
Bagrow, L. Revised and enlarged by R. A. Skelton. (1964) Harvey, P. D. A. (1980) The History of Topographical Maps:
History of Cartography. Cambridge, MA: Harvard University Symbols, Pictures, and Surveys. London: Thames and Hudson.
Press. A complete history of the topographic map.
One of the definitive volumes on the history of cartography. Keates, John. (1973) Cartographic Design and Production.
Brown, L. A. (1979) The Story of Maps. New York: Dover London: Longman.
Publications. A classic book on cartographic design and production.
An excellent general survey of the history of cartography. Montello, D. R. (2002) “Cognitive map-design research in the
Cartographic Perspectives no. 51, 2005 (entire issue). twentieth century: Theoretical and empirical approaches.”
Cartography and Geographic Information Science 29, no. 3:283–304.
This special issue celebrates the lives and careers of Arthur
Robinson and David Woodward. Reviews the development of cartographic research related to
cognition and map design.
Cartography and Geographic Information Systems 18, no. 3,
1991 (entire issue). Raisz, E. (1948) General Cartography, 2nd ed. New York:
McGraw-Hill.
Papers in this issue discuss the history and development of acade-
mic cartography in the United States. A classic early textbook in cartography.
Cartography and Geographic Information Science 29, no. 3, Robinson, A. H. (1982) Early Thematic Mapping in the
2002 (entire issue). History of Cartography. Chicago: University of Chicago Press.
Reviews the development of individual thematic mapping
This special issue concentrates on the history of cartography in the
techniques.
twentieth century.
Thrower, N. J. W. (2008) Maps and Civilization: Cartography in
Chrisman, N. (1998a) “Academic Origins of GIS.” In The
Culture and Society, 3rd ed. Chicago: University of Chicago Press.
History of Geographic Information Systems: Perspectives from
the Pioneers. Ed. by T. W. Foresman, pp. 33–43. Upper Saddle A general survey of the major maps and mapmakers, with a focus
River, NJ: Prentice Hall. on their social impacts.
A brief overview of academic activity and the history of GIS. Tobler, W. R. (1976) “Analytical cartography.” The American
Cartographer 3, no. 1:21–31.
Fabrikant, S. I. (2003) “Commentary on ‘A history of twentieth-
century American academic cartography.’” Cartography and Tobler’s classic piece on analytical cartography.
Geographic Information Science 30, no. 1:81–84. Tooley, R. V. (1987) Maps and Map-Makers. London: B. T.
Provides a perspective on developments in thematic cartography
Batsford Ltd.
in Germany, Austria, and Switzerland. For related discussion, see A good general overview of the history of cartography, organized
McMaster and McMaster (2003) and Thrower (2003). by region.
GLOSSARY
analytical cartography: a theoretical branch of cartography

developed by Waldo Tobler that deals with analytical topics
such as cartographic data models, coordinate transformations
and map projections, interpolation, and generalization.
37
15:58:09.
REFERENCES
Bagrow, L., Revised and enlarged by R. A. Skelton. (1964) Kimerling, A. J. (1989) “Cartography.” In Geography in
History of Cartography. Cambridge, MA: Harvard America, ed. by G. S. Gaile and C. J. Willmott, pp. 686–717.
University Press. Columbus, OH: Merrill.
Bertin, J. (1981) Graphics and Graphic Information- Kraak, M.-J., and Ormeling, F. J. (1996) Cartography:
Processing. Berlin: Walter de Gruyter. Visualization of Spatial Data. Essex, England: Addison
Bertin, J. (1983) Semiology of Graphics: Diagrams, Networks, Wesley Longman.
Maps. Madison: University of Wisconsin Press. McMaster, R. B., and Thrower, N. J. W. (1991) “The early
Brown, L. A. (1979) The Story of Maps. New York: Dover years of American academic cartography: 1920–45.”
Publications. Cartography and Geographic Information Systems 18, no.
3:151–155.
Chrisman, N. (1998a) “Academic Origins of GIS.” In The
History of Geographic Information Systems: Perspectives McMaster, R., and McMaster, S. (2002) “A history of twenti-
from the Pioneers, ed. by T. W. Foresman, pp. 33–43. Upper eth-century American academic cartography.” Cartography
Saddle River, NJ: Prentice Hall. and Geographic Information Science 29, no. 3:305–321.
Eckert, M. (1921) Die Kartenwissenschaft. 2 volumes. Berlin: Raisz, E. (1948) General Cartography (2nd ed.). New York:
Walter de Gruyter & Co. McGraw Hill.
Harvey, P. D. A. (1980) The History of Topographical Maps: Sherman, J. C. (1991) “The development of cartography at
Symbols, Pictures, and Surveys. London: Thames and the University of Washington.” Cartography and
Hudson. Geographic Information Systems 18, no. 3:169–170.
Imhof, E. (1975) “Positioning names on maps.” The American Smith, N. (1987a) “Academic war over the field of geogra-
Cartographer 2, no. 2:128–144. phy: The elimination of geography at Harvard, 1947–1951.”
Annals of the Association of American Geographers 77, no.
Jenks, G. F. (1953a) “An improved curriculum for carto-
2:155–172.
graphic training at the college and university level.”
Annals of the Association of American Geographers 43, no. Thrower, N. J. W. (1969) “Edmund Halley as a Thematic Geo-
4:317–331. Cartographer.” Annals of the Association of American
Geographers 59:652–676.
Keates, J. (1973) Cartographic Design and Production.
London: Longman. Tooley, R. V. (1987) Maps and Map-Makers. London: B.T.
Batsford Ltd..
15:58:09.
Statistical and Graphical
Foundation
39
16:02:07.
Statistical and Graphical
Foundation
OVERVIEW
One purpose of this chapter is to provide a statistical plots), but other material will likely be new (e.g.,
and graphical foundation for the remainder of the text. hexagon bin plots, the scatterplot matrix, the reduced
For example, in this chapter we will define basic statisti- major axis, scatterplot brushing, and parallel coordi-
cal measures such as the mean and standard deviation. nate plots).
A second purpose of this chapter is to introduce a range One limitation of the numerical summaries covered
of techniques (tables, graphs, and numerical sum- in section 3 is that spatial location is not an integral
maries) that can be used along with maps to analyze part of the summaries. In contrast, section 4 deals with
spatial data. numerical summaries in which spatial location is an
Sections 1 and 2 briefly deal with the principles of pop- integral part—these are often referred to as
ulation versus sample and descriptive versus inferential geostatistical methods. Some geostatistical methods
statistics. A population is the total set of elements or things covered in section 4 analyze just spatial location; for
that could potentially be studied, whereas a sample is the example, the formula for computing the centroid, or
portion of the population that is actually examined. balancing point for a geographic region, uses just the
Descriptive statistics are used to summarize the character x and y coordinates bounding a region. Other methods
of a sample or population, and inferential statistics are consider both spatial location and the values of an
used to make a guess (or inference) about a population attribute; for example, spatial autocorrelation mea-
based on a sample. The focus of this chapter is on descrip- sures the likelihood that similar attribute values occur
tive statistics because these are most useful in mapping. near one another.
Section 3 covers a broad range of methods for ana- To illustrate many concepts in this chapter, we will ana-
lyzing data using tables, graphs, and numerical sum- lyze the relationship between murder rate (number of
maries. The section is split into three major parts: ana- murders per 100,000 people) and the following attributes
lyzing the distribution of individual attributes (3.1), for 50 U.S. cities with a population of 100,000 or more in
analyzing the relationship between two (or more) 1990: (1) percentage of families whose income was below
attributes (3.2), and exploratory data analysis (3.3). the poverty level; (2) percentage of those 25 years and
Readers with coursework in statistics will find some of older who were at least high school graduates; (3) the drug
this material a review (e.g., histograms, measures of arrest rate (number of arrests per 100,000 people); (4)
central tendency, correlation, and stem-and-leaf population density (number of people per square mile);
McMaster, Robert B., et al. Thematic Cartography and Geovisualization:
16:02:07.
Statistical and Graphical Foundation
and (5) total population (in thousands).* The raw data are population and split into 10 classes using Jenks’s (1977)
shown in Table 1 (ordered on the basis of murder rate). optimal method. A proportional number of cities were
One problem with these data is that an analysis at the sampled from each of the 10 classes to obtain a broad
city level might be inappropriate because it fails to range of city sizes that would be representative of those
account for the variation within a city; it might instead be found in the United States.
desirable to look at finer geographic units, such as census
tracts, or at individual murder cases. We chose the city 2 DESCRIPTIVE VERSUS INFERENTIAL STATISTICS
level for analysis, however, because it is easier to relate to
individual cities than to individual census tracts. Later in
the chapter we will consider the effect of aggregation of Statistical methods can be split into two types: descriptive
enumeration units on measures of numerical correlation. and inferential. Descriptive statistics describe the charac-
ter of a sample or population. For example, to assess the
current president’s job performance, you might ask a sam-
1 POPULATION AND SAMPLE ple of 500 people, “Is the President doing an acceptable
job?” The percentage responding yes, say 52 percent,
In statistics, a population is defined as the total set of would be an example of a descriptive statistic. Inferential
elements or things that one could study, and a sample is statistics are used to make an inference about a popula-
the portion of the population that is actually examined (in tion from a sample. For example, based on the 52 percent
this text, the number of elements in each will be repre- figure just given, you might infer that 52 percent of the
sented by N and n, respectively). Generally, scientists col- entire population thinks the President is doing an accept-
lect samples because they don’t have the time or money able job.You would be surprised, of course, if the 52 percent
to examine the entire population. For example, a geomor- figure truly applied to the population because the figure
phologist studying the effect of wave behavior on beach is based on a sample. To correct for this problem, in infer-
development would collect data at a series of points along ential statistics it is necessary to compute a margin of
a shoreline, rather than examine the entire shoreline. error (e.g., plus or minus 3 percent) around the sampled
The data in Table 1 have characteristics of either a pop- value; we often find such errors reported in media cover-
ulation or sample, depending on one’s perspective of the age of polling.
data. Consider first a perspective from the standpoint of
attributes. Murder rate, drug arrest rate, population, and 3 METHODS FOR ANALYZING SPATIAL DATA,
population density are all based on the entire population IGNORING LOCATION
of each city (as defined by the census); for example, in the
case of the murder rate, all murders occurring within a city
This section considers methods for analyzing data, ignor-
are considered relative to the entire population of that
ing the spatial location of the data.The section is split into
city. The other attributes, percentage of families below the
three parts: (1) analyzing the distribution of individual
poverty level and percentage of high school graduates, are
attributes, (2) analyzing the relationship between two or
based on sampling approximately one of every six hous-
more attributes, and (3) exploratory data analysis.
ing units (U.S. Bureau of the Census 1994, A-2).
From the perspective of observations, the 50 cities
shown in Table 1 were sampled from 200 cities. Sampling 3.1 Analyzing the Distribution of Individual
was done in two stages. The first stage involved eliminat- Attributes
ing cities with political boundaries that extended beyond
the limits of where most people live within those cities Tables
(using the “extended city” definition provided by the Raw Table. The simplest form of tabular display is
U.S. Census Bureau; see http://www.census.gov/popula- the raw table in which the data for an attribute of interest
tion/www/documentation/twps0027.html). This was done are listed from lowest to highest value, as for the murder
because one of the attributes being analyzed, population rate data in Table 1. Tabular displays are useful for pro-
density, was a function of city area. In the second stage, viding specific information about observations (e.g.,
the remaining cities were ordered on the basis of total Buffalo, New York, had a murder rate of 11.3 per 100,000
people in 1990), but they can provide additional informa-
* The murder and drug arrest data were obtained from the Sourcebook tion if they are examined carefully. Note, for example, that
of Criminal Justice Statistics 1991 (Flanagan and Maguire 1992). The the sorted values provide the minimum and maximum of
remaining data were taken from the 1994 City and County Data Book the data (0 for Irvine, California, and 77.8 for Washington,
(U.S. Bureau of the Census 1994). All data were for either 1989 or 1990. DC). With some mental arithmetic, the range can be
41
16:02:07.
TABLE 1 Sample data for 50 U.S. cities (sorted on murder rate)

Families below High School Drug Population Total Population
City Murder Rate* Poverty Level (%) Graduates (%) Arrest Rate † Density ‡ (in Thousands)
Irvine, CA 0.0 2.6 95.1 780 2607 110

Cedar Rapids, IA 0.9 6.6 84.5 110 2034 109
Overland Park, KS 0.9 1.9 94.1 255 2007 112
Livonia, MI 1.0 1.7 84.7 665 2823 101
Lincoln, NE 1.6 6.5 88.3 294 3033 192
Madison, WI 1.6 6.6 90.6 57 3311 191
Glendale, CA 1.7 12.3 77.2 452 5882 180
Allentown, PA 1.9 9.3 69.4 1078 5934 105
Tempe, AZ 2.1 7.0 89.9 295 3590 142
Boise City, ID 2.4 6.3 88.6 512 2726 126
Lakewood, CO 2.4 5.2 88.2 216 3100 126
Mesa, AZ 3.1 6.9 84.8 223 2653 288
Pasadena, TX 3.4 11.1 69.8 370 2727 119
San Jose, CA 4.5 6.5 77.2 1289 4568 782
Waterbury, CT 4.6 9.9 66.8 1326 3815 109
Springfield, MO 5.0 11.6 77.0 446 2068 140
Chula Vista, CA 5.2 8.6 75.7 808 4661 135
St. Paul, MN 6.6 12.4 81.1 260 5157 272
Arlington, VA 7.0 4.3 87.5 758 6605 171
Alexandria, VA 7.2 4.7 86.9 834 7281 111
Portland, OR 7.6 9.7 82.9 1001 3508 437
Des Moines, IA 8.3 9.5 81.0 118 2567 193
Lansing, MI 8.7 16.5 78.3 780 3755 127
Pittsburgh, PA 9.5 16.6 72.4 723 6649 370
Yonkers, NY 9.6 9.0 73.6 917 10403 188
Riverside, CA 9.7 8.4 77.8 1703 2916 227
Elizabeth, NJ 10.0 13.7 58.5 929 8929 110
Berkeley, CA 10.7 9.4 90.3 1569 9783 103
Buffalo, NY 11.3 21.7 67.3 580 8080 328
Raleigh, NC 11.5 7.7 86.6 634 2360 208
Sacramento, CA 11.7 13.8 76.9 1555 3836 369
Tacoma, WA 14.1 12.5 79.3 673 3677 177
Knoxville, TN 15.2 15.3 70.8 328 2137 165
Beaumont, TX 16.7 16.6 75.2 693 1427 114
Winston-Salem, NC 16.8 11.6 77.0 1343 2018 143
Montgomery, AL 18.2 14.4 75.7 131 1386 187
Waco, TX 21.2 19.7 68.4 400 1367 104
Jackson, MS 22.3 18.0 75.0 693 1804 197
Savannah, GA 23.9 18.5 70.1 707 2198 138
Norfolk, VA 24.1 15.1 72.7 624 4859 261
Los Angeles, CA 28.2 14.9 67.0 1391 7426 3485
Chicago, IL 30.6 18.3 66.0 1157 12251 2784
New York, NY 30.7 16.3 68.3 1255 23701 7323
Houston, TX 34.8 17.2 70.5 555 3020 1631
Newark, NJ 40.7 22.8 51.2 1751 11554 275
Baltimore, MD 41.4 17.8 60.7 2063 9108 736
Gary, IN 55.6 26.4 64.8 261 2322 117
Detroit, MI 56.6 29.0 62.1 1052 7410 1028
Atlanta, GA 58.6 24.6 69.9 2330 2990 394
Washington, DC 77.8 13.3 73.1 1738 9883 607
* Murders per 100,000 people.
† Arrests per 100,000 people.
‡ Number of people per square mile.
calculated by simply subtracting the minimum from the murder rate, duplicate values (e.g., 0.9 for Cedar Rapids,
maximum (77.8 - 0 = 77.8 in this case). Iowa, and Overland Park, Kansas) are unimportant, as they
Raw tables can also reveal any duplicate values and out- are a function of the number of significant digits reported.
liers in the data that might have special significance. For For some attributes, however, duplicates can be quite
16:02:07.
meaningful. For example, an examination of the distribu- adding the class interval to the lowest value in the data.
tion of Major League Baseball player salaries would reveal The result is the right-hand side of the Limits column in
numerous duplicates because several players on each team Table 2. Note that the highest calculated limit does not
typically earn exactly the same amount, the minimum match the highest data value (77.8) exactly because the
required salary for all of Major League Baseball. class interval is not a simple fraction; a more precise
Outliers—values that are quite unusual or atypical—can class interval would be 5.186666667, but even with this
also be seen (Barnett and Lewis 1994). For murder rate, no interval, the match would not be exact (the highest
value is considerably different from the rest (although value would be 77.80000001).
Washington, DC, is clearly larger), but for the total popula- Step 4. Determine the lower limit of each class. Lower
tion attribute, note that several cities are quite large, with limits for each class are specified so that they are just
New York more than twice the value of any other city in above the highest value in a lower-valued class. This is
the sample. done so that any observation will fall in only one class.
Although raw tables are useful for providing specific For example, the lowest value in the second class is
information, they are not very good at providing an 5.188, which is 0.001 larger than 5.187, the highest value
overview of how data are distributed along the number in class 1.
line. For example, in studying Table 1, note that roughly Step 5. Tally the number of observations falling in each
half the cities have murder rates below 10.0, but it is dif- class. These numbers are shown as Number of
ficult to develop a feel for what the overall murder rate Observations in Class in Table 2. Also shown in the table
distribution really looks like. Grouped-frequency tables are the percent and cumulative percent in each class.
are more useful for this purpose. In comparison to the raw table, the grouped-frequency
table provides a somewhat better overview of the
Grouped-Frequency Table. To construct grouped- data. For example, note that 80 percent of the cities fall in
frequency tables, we divide the data range into equal the five lowest murder rate classes. Both raw and
intervals and then tally the number of observations that grouped-frequency tables, however, do not take full
fall in each interval. Although grouped-frequency tables advantage of our visual processing powers; for this we
generally are constructed using software (e.g., SPSS and turn to graphs.
SAS), it is useful to consider the actual steps involved so
that you have a clear understanding of what the resulting Graphs
table reveals. Point and Dispersion Graphs. In a point graph or
one-dimensional scatterplot, each data value is repre-
Step 1. Decide how many groups (or classes) you sented by a small point symbol plotted along the num-
want to use. When grouped-frequency tables were cre- ber line (Cleveland 1994, 133; in Figure 1A an open
ated manually, this was an important step because of circle is used). For the murder rate attribute, the point
the time and thought needed to construct the table. graph shows that the data are concentrated at the lower
With the ready availability of software, it is reasonable
end of the distribution (based on the graph, under a
to construct tables for various numbers of classes. For
this discussion, presume that 15 classes will be chosen. TABLE 2 Grouped-frequency table for the murder
Step 2. Determine the width of each class (the class rate data
interval). The class interval is computed by dividing the
Number of
range of the data by the number of classes. The follow- Observations Percent Cumulative
ing would be the computation for the murder rate data Class Limits in Class in Class Percent
using 15 classes:
1 0.000–5.187 16 32 32
2 5.188–10.374 11 22 54
Range High - Low 3 10.375–15.561 6 12 66
= 4 15.562–20.748 3 6 72
Number of classes Number of classes
5 20.749–25.935 4 8 80
77.8 - 0.0 6 25.936–31.122 3 6 86
= = 5.187
15 7 31.123–36.309 1 2 88
8 36.310–41.496 2 4 92
9 41.497–46.683 0 0 92
Note that we use more decimal places for the class 10 46.684–51.870 0 0 92
interval than appear in the actual data to avoid round- 11 51.871–57.057 2 4 96
ing errors that could cause the last class not to match 12 57.058–62.244 1 2 98
the highest value in the data. 13 62.245–67.431 0 0 98
Step 3. Determine the upper limit of each class. The 14 67.432–72.618 0 0 98
15 72.619–77.805 1 2 100
upper limit for each class is computed by repeatedly
43
16:02:07.
0 20 40 60 80
Murder Rate
B
Median Mean
9.7 16.0
0 20 40 60 80
Murder Rate
FIGURE 1 Point (A) and dispersion (B) graphs for the murder rate data presented in Table 1. Note the locations of the median
and mean in (B).
rate of about 12). One obvious problem with a point in the tails of the distribution (Figure 3). Curves repre-
graph is that individual symbols might overlap, thus senting a normal distribution have been overlaid on the
making the distribution difficult to interpret (note the histograms shown in Figure 2. Distributions lacking the
“smearing” in the left-hand portion of Figure 1A); du- symmetry of the normal distribution are termed skewed.
plicate values, in particular, cannot be detected at all Positively skewed distributions have the tallest bars con-
by this method. centrated on the left-hand side (as for total population,
An alternative to the point graph is the dispersion murder rate, and population density), whereas negatively
graph (Hammond and McCullagh 1978), in which data skewed distributions have the tallest bars on the right-
are grouped into classes, the number of values falling in hand side. (There is no distinctive example of a nega-
each class are tallied, and dots are stacked at each class tively skewed distribution in Figure 2, although per-
position (Figure 1B). Intervals for classes are defined in a cent with high school education has a slight negative
fashion identical to the grouped-frequency table, except skew).
that a large number of classes are used. For example, 99 Because many inferential tests require a normal dis-
classes were used for Figure 1B. The dispersion graph for tribution, raw data are often transformed to make them
the murder rate data portrays a distribution similar to the approximately normal. Transformation involves applying
point graph, except that potential confusion in the over- the same mathematical operation to each data value of
lapping areas is eliminated. an attribute; for example, we might compute the log10 of
each murder rate value, or alternatively, we could com-
Histogram. A histogram is constructed in a manner pute the square root of each murder rate value.* log10
analogous to a dispersion graph, except that fewer classes and square root transformations are commonly used to
are generally used and bars of varying height are used to convert a positively skewed distribution to a normal dis-
represent the number of values in each class. Because the tribution; to illustrate, Figure 4 portrays such transforma-
histogram is more commonly used than either the point or tions for the murder and drug arrest data, respectively.
dispersion graph, all of the attributes shown in Table 1 are
graphed using the histogram in Figure 2. Looking first at the Numerical Summaries
murder rate histogram, note that the up-and-down nature Although tables and graphs are useful for analyzing data,
of the dispersion graph has been smoothed out; there is they are prone to differing subjective interpretations, and
clearly a peak in the graph on the left, with a decreasing the limited space of formal printed publications often
height in bars as one moves to the right. limits their use. As an alternative, statisticians frequently
Histograms are often compared with a hypothetical use numerical summaries, which typically are split into
normal (or bell-shaped) distribution. For a normal dis-
tribution, most of the observations fall near the mean (in * The log10 for a number is the power that we would raise 10 to get
the middle of the distribution), with fewer observations that number. For example, log10100 = 2 because 102 = 100.
16:02:07.
20 8
16
6
Frequency
Frequency
12
4
8
2
4
0 0
2.6 13.0 23.4 33.8 44.2 54.6 65.0 75.4 2.0 6.0 10.0 14.0 18.0 22.0 26.0 30.0
Murder Rate Percent of Families below Poverty Level
10 10
8 8
Frequency
Frequency
6 6
4 4
2 2
0 0
51.5 57.7 63.8 69.9 76.1 82.2 88.3 94.5 80 400 720 1040 1360 1680 2000 2320
Percent with High School Education Drug Arrest Rate
20 50
16 40
Frequency
Frequency
12 30
8 20
4 10
0 0
1800 5000 8200 11400 14600 17800 21000 24200 355 1375 2395 3415 4435 5455 6475 7495
Population Density Total Population
FIGURE 2 Histograms for the data presented in Table 1.
two broad categories: measures of central tendency and value, and is thus generally useful for only nominal data,
measures of dispersion. such as on a land use/land cover map. The median is the
middle value in an ordered set of data or, alternatively,
Measures of Central Tendency. Measures of central the 50th percentile, because 50 percent of the data are
tendency are used to indicate a value around which the below it. For the murder rate data, the median is 9.7. Note
data are most likely concentrated.Three measures of cen- its location in Figure 1B.The mean is often referred to as
tral tendency are commonly recognized: mode, median, the “average” of the data and is calculated by summing
and mean. The mode is the most frequently occurring all values and dividing by the number of values. Separate
45
16:02:07.
where the Xi are individual data values. Because the

data in Table 1 are a sample from 200 cities, the sample
formula is appropriate in this case. The mean for the
murder rate data is 16.0. Note its location in Figure 1B.
Frequency
One problem with the mean is that either a skew or

outliers in the data affect it, whereas the median is
resistant to these characteristics. We can see this in
Figure 1B, in which the median falls where most of the
data are concentrated, whereas the mean is pulled to the
-2 -1 0 1 2 z - values
right by the positive skew.
68.3%
Measures of Dispersion. Measures of dispersion
provide an indication of how data are spread along the
number line. The simplest measure is the range, which
95.4% was defined earlier as the maximum minus the mini-
mum. Obviously, the range is of limited usefulness
FIGURE 3 An example of a normal curve. Histograms will
because it is based on only two values, the maximum
approximate this shape if the data are normal. For a perfectly and the minimum of the data.
normal data set, approximately 68 percent and 95 percent of More useful measures of dispersion are the
the observations will fall within 1 and 2 standard deviations, interquartile range and standard deviation, which
respectively, of the mean. should be used with the median and mean, respectively.
The interquartile range is the absolute difference
formulas are used to distinguish mean values for the sam- between the 75th and 25th percentiles, or where the
ple and population, as follows:* middle 50 percent of the data lie. For the murder rate
n
data, the result is |22.700 - 3.325| = 19.4. An important
characteristic of the interquartile range is that it, like
a Xi
i=1 the median, is unaffected by outliers in the data. For
Sample: X =
n example, if the highest murder rate were replaced by a
N value of 150, the interquartile range would still be 19.4.
a Xi In a fashion similar to the mean, separate formulas
Population: m =
i=1 normally are provided for the sample and population
N standard deviation:
n
2
*g is the symbol for summation, indicating that all data values should a (Xi - X )
be summed. Readers unfamiliar with summation notation should con- i=1
Sample: s =
sult an introductory statistics book such as Burt and Barber 1996, 68–70. Q n - 1
8 10
8
6
Frequency
Frequency
6
4
2
2
0 0
0.1 0.2 0.3 0.5 0.6 0.7 0.8 1.0 1.1 1.2 1.4 1.5 1.6 1.7 1.9 7.7 10.5 13.3 16.2 19.0 21.8 24.7 27.5 30.3 33.2 36.0 38.8 41.7 44.5 47.3
Log of Murder Rate Square Root of Drug Arrest Rate
FIGURE 4 Histograms of two transformed attributes: log of murder rate and square root of drug arrest rate (compare with cor-
responding histograms in Figure 2).
16:02:07.
TABLE 3 Tabular relationship between murder rate and percent of families below
poverty level for cities in Table 1
Percent Families below Poverty Level
Murder Rate 1.70–7.16 7.17–12.62 12.63–18.08 18.09–23.54 23.55–29.00
62.25–77.80 1
46.69–62.24 3
31.13–46.68 2 1
15.57–31.12 1 6 3
0.00–15.56 13 14 5 1
N percent of families below the poverty level. In general,

2
a (Xi - m) the attributes appear related (low poverty values are asso-
Population: s = i=1
ciated with low murder rates, and high poverty values are
Q N associated with high murder rates), but it is difficult to
summarize the relationship. To simplify the process, we
In comparing these formulas, note that they differ
can class both attributes using the grouped-frequency
principally in the denominator: a value of 1 is subtracted in
method and create a matrix of the result (Table 3). The
the sample case, but not in the population case. Subtracting
same general relation between the attributes is still appar-
a value of 1 is necessary because a sample estimate using
ent, but it is more easily seen; also, the matrix reveals that
just n tends to underestimate the population value (Burt
Washington, DC (in the highest murder rate class), does
and Barber 1996, 64). Using the sample formula, the
not fit the general trend of the data, which extends from
standard deviation for the murder rate data is 17.5. In
the lower left to the upper right of the table.
contrast to the interquartile range, outliers in the data do
affect the standard deviation. To illustrate, replacing the
highest murder rate by a value of 150 results in a standard Graphs
deviation of 24.4, an increase of nearly 40 percent. A scatterplot is used to examine the relationship of
attributes against one another in two-dimensional
3.2 Analyzing the Relationship between Two space. To illustrate, Figure 5 portrays scatterplots of
or More Attributes murder rate and log of murder rate against percent of
families below poverty level (also shown are best-fit
Tables regression lines, which we will consider shortly). On
In the previous section, we saw that considerable infor- scatterplots, dependent and independent attributes
mation could be derived from raw tables when examining normally are plotted on the y and x axes, respectively.
an individual attribute. When trying to relate two attrib- Because it seems reasonable that the murder rate might
utes, however, this task becomes difficult. To convince depend, in part, on poverty, murder rate has been plot-
yourself of this, try using Table 1 to relate murder rate to ted on the y axis.
A B
80 2.0
Log of Murder Rate
60 1.5
Murder Rate
40 1.0
20 0.5
0 0
0 10 20 30 0 10 20 30
Percent of Families Below Poverty Level Percent of Families Below Poverty Level
FIGURE 5 Scatterplots of (A) murder rate against percent of families below poverty level and (B) log of murder rate against the same
attribute. Also shown are best-fit regression lines.
47
16:02:07.
In examining Figure 5A, note the similarity of the dis-

Log of
tribution of points to the pattern of cells in Table 3. Also Murder
note that after transforming murder rate (Figure 5B), Rate
Washington, DC, is not quite so different from the rest of
Percent
the data; thus, data transformations affect not only the below
values for individual attributes but also the relationship Poverty
Level
between attributes.
Percent
When the number of observations is large, scatter- with
plots can become difficult to interpret, just as the smear- High
School
ing of dots made the point graph of murder rate hard to Education
interpret. One solution to this problem is the hexagon
Log of
bin plot, which is shown in the right-hand portion of Total
Figure 6. Such a plot is created by laying a grid of hexa- Population
gons onto a scatterplot (see the left-hand portion of

Square
Figure 6), and then filling each hexagon grid cell with a Root of
solid hexagon of size that is proportional to the number Drug
Arrest
of dots falling in each cell.*
To examine the relationship among three attributes Log of
simultaneously, the basic scatterplot can be extended to Population
Density
a three-dimensional scatterplot by specifying x, y, and z
axes. Many statistical packages have options for creat-
ing such plots, and even permit rotating them interac- FIGURE 7 A scatterplot matrix for the data presented in
tively. Our experience is that these plots are difficult to Table 1. Also shown are best-fit regression lines.
interpret; moreover, they cannot be extended to handle
more than three attributes. above the diagonal of attribute names and once below it.
One method that can handle three or more attributes This approach enables each attribute to be shown as
is the scatterplot matrix (Figure 7). This graph might at either an independent or dependent attribute in relation
first appear rather complex, but the principles underlying to other attributes. By scanning a row, you can see what
it are actually quite simple. Note first that the relation happens when an attribute is considered dependent, and,
between any two attributes is displayed twice, once by scanning a column, you can see what happens when it
is considered independent. If our focus were on log of
murder rate as a potential dependent attribute, we would
want to examine the first row, where we see that the
strongest relationship appears to be between log of mur-
der rate and percent below poverty level.
A second approach for displaying three or more attrib-
utes simultaneously is the parallel coordinate plot. In this
graph, attributes are depicted by a set of parallel axes, and
observations are depicted as a series of connected line seg-
ments passing through the axes. To illustrate, Figure 8 por-
trays a parallel coordinate plot for a subset of 10 of the 50
cities shown in Table 1. The six attributes in Table 1 are
represented by six parallel vertical lines, whereas the 10
FIGURE 6 Hexagon bin plot: an alternative to the scatter- observations are represented by horizontal line segments.
plot. The size of the hexagons shown on the right is propor- Figure 8A shows all 10 observations plotted in black, a
tional to the number of dots falling within the hexagons on conventional approach for black-and-white presentations;
the left. The x and y axes represent sulfate and nitrate deposi- Figure 8B shows Washington, DC, highlighted. At the top
tion values for sites in the eastern United States. (After Carr of the figures are correlation coefficients, or r values, which
et al. 1992. First published in Cartography and Geographic are described in detail in the “Numerical Summaries” sec-
Information Systems 19(4), p. 229. Reprinted with permission tion later.
from the American Congress on Surveying and Mapping.) In interpreting the parallel coordinate plot, it is
important to recognize the characteristic shape of plots
* For other approaches to handling a large number of observations, associated with particular values of r. For instance,
see Carr 1991 and Cleveland 1994. Figure 9 illustrates plots for r = 1 (parallel line
16:02:07.
A
r = 0.81 r = -0.77 r = -0.38 r = 0.17 r = 0.69
Maximum
Value
Minimum
Value
Percent below Log of Percent with Log of Square Root of Log of

Poverty Level Murder Rate High School Education Total Population Drug Arrest Rate Population
Density
B
r = 0.81 r = -0.77 r = -0.38 r = 0.17 r = 0.69
Maximum
Value
Minimum
Value
Percent below Log of Percent with Log of Square Root of Log of

Poverty Level Murder Rate High School Education Total Population Drug Arrest Rate Population
Density
FIGURE 8 (A) A parallel coordinate plot for 10 of the 50 cities shown in Table 1; (B) the same plot, with Washington, DC,
highlighted.
49
16:02:07.
r=1 r = -1 r=0
1 2 3 4
FIGURE 9 Characteristic parallel coordinate plots for particular values of r.
segments), r = - 1 (line segments intersecting one is used to summarize the nature and strength of the
another in the middle of the graph), and r = 0 (line relationship, and bivariate regression provides an equa-
segments crossing one another at different angles). In tion for a best-fit line passing through the data when
Figure 8, the highly positively correlated percent below shown in a scatterplot.
poverty level and log of murder rate (r = .81 for the 10
cities) have relatively parallel line segments Bivariate Correlation. One value commonly computed
(Washington, DC, is a notable exception). In contrast, in bivariate correlation is the correlation coefficient, or r:
line segments for the highly negatively correlated log
of murder rate and percent with high school education n
(r = - .77 for the 10 cities) tend to cross one another
a (Xi - X)(Yi - Y )
near the middle of the graph. Obviously, the appear- i=1
r = n n
ance of the parallel coordinate plot will change, 2
a (Xi - X ) (Yi - Y )2
depending on the order in which the axes are plotted. A i=1 A ia
=1
Ideally, a successful interpretation of the plot requires
an interactive program that can easily change the order
of the attributes and highlight selected observations. Table 4 shows r values for the transformed sample
data. (Note that the table is symmetric about a diago-
nal extending from upper left to lower right; this
Numerical Summaries occurs because the correlation between attributes A
Bivariate correlation-regression is the most widely and B is identical to that between attributes B and A.)
used approach for summarizing the relationship Extreme values for r range from - 1 to + 1. A posi-
between two numeric attributes. Bivariate correlation tive r value indicates a positive relationship in which
16:02:07.
TABLE 4 Matrix of correlation coefficients for the transformed city data

Percent Percent
Log of below with Log of Square Root Log of
Murder Poverty High School Total of Drug Population
Rate Level Education Population Arrest Rate Density
Log of 1.00 .82 -.70 .52 .49 .27

murder rate
Percent below .82 1.00 -.79 .38 .29 .17
poverty level
Percent with -.70 -.79 1.00 -.39 -.44 -.36
high school
education
Log of total .52 .38 -.39 1.00 .39 .54
population
Square root of .49 .29 -.44 .39 1.00 .52
drug arrest rate
Log of population
density .27 .17 -.36 .54 .52 1.00
increasing values on one attribute are associated with distances shown in Figure 10. The values of a and b that
increasing values on another attribute; for example, minimize the differences are:
the r value for log of murder rate and percent below
poverty level is .82. Conversely, a negative r value a = Y - bX
indicates a negative relationship in which increasing n n n
values on one attribute are associated with decreasing n a Xi Yi - a Xi a Yi
values on another attribute; for example, log of mur- i=1 i=1 i=1
b = n n 2
n a X i2 - a a X i b
der rate and percent with high school education have
a correlation of -.70. Note that as r values approach i=1 i=1
-1 or + 1 , the points cluster more tightly about the
best-fit line. (Compare Table 4 and Figure 7, and note
that the graph of log of murder rate and percent
below poverty level is more tightly clustered than the Y
one for log of murder rate and square root of drug
arrest.)
The value of r is primarily used to indicate the
direction of a relationship (whether it is positive or
negative). To properly compute the strength of a rela-
tionship, one must compute the coefficient of deter-
mination, or r2 , which measures the proportion of (Yi - Yi)
variation in one attribute explained, or accounted for,
by another attribute. For example, r2 between log of (Xi - Xi) (Xi - Xi)
murder rate and percent below poverty level is .67,
indicating that 67 percent of the variation in murder (Yi - Yi)
rate can be accounted for by a linear function of the
poverty attribute.
Bivariate Regression. In general, the equation for

any straight line is Yi = a + bXi, where Xi, Yi is a point
on the line, a is the y intercept, and b is the slope. In
X
regression analysis, Xi and Yi are raw values for the in-
dependent and dependent attributes, respectively, and
the line of best fit is defined as YN i = a + bXi, where YN i FIGURE 10 Possible approaches for determining best-fit
is a predicted value for the dependent attribute.The best- lines for a set of data include (1) minimizing vertical distances,
fit line is found by minimizing the differences between (2) minimizing horizontal distances, and (3) minimizing the
the actual and predicted values (Yi - YN i), the vertical area of the resulting triangles.
51
16:02:07.
where X and Y are the mean of the X and Y values, (Fotheringham 1998). A solution to the aggregation
respectively. The best-fit lines shown in Figures 5 and 7 problem is to examine data at the individual level (look-
were derived using this approach. ing at individual murders and collecting data regarding
the people involved). An example is The Project on
Major-Axis Approach. When one does not wish to spec- Human Development in Chicago Neighborhoods, which
ify a dependent attribute (as when relating murder rate and is analyzing data from a variety of Chicago neighbor-
race), it makes sense to minimize the vertical and horizontal hoods; here the emphasis is on individual interviews
distances shown in Figure 10 simultaneously, effectively min- and extensive fieldwork (see http://www.icpsr.umich.edu/
imizing the area of the triangles; this is known as the PHDCN/ ).
reduced major-axis approach (Davis 2002, 214–218).The A third consideration is that specialized regression
equations for this approach turn out to be simpler than techniques have been developed to handle the fact
those for standard regression. The slope is just the ratio of that geographical data tend to be spatially autocorre-
the standard deviations of the attributes, or sY/sX, and the lated, meaning that like values tend to be located near
intercept is still Y = bX . one another (e.g., a high-income census tract will tend
to occur near another high-income census tract). For a
Multiple Regression and Other Multivariate Tech- discussion of these specialized techniques, see Griffith
niques. When the concept of bivariate correlation- (1993) and Rogerson (2006). SpaceStat (http://www.
regression is extended into the multivariate realm, it is terraseer.com/products.php) and GeoDa (http://www.
termed multiple regression. In multiple regression, csiss.org/clearinghouse/GeoDa/ ) are software pack-
there is still a single dependent attribute, but multiple ages that have been developed explicitly for handling
independent attributes are possible. In a fashion sim- such techniques.
ilar to the preceding, it is possible to perform correla- Finally, we need to recognize that regression tech-
tion and regression analyses that summarize the niques can be applied both globally and locally. By glob-
relationship between the dependent and independent ally, we mean that a regression equation is applied
attributes. For example, we might attempt to explain uniformly throughout a geographic region. In contrast,
the murder rate as a function of percent below poverty local regression techniques consider the notion that a
level, percent with high school education, and the drug single model might not be appropriate for an entire
arrest rate. A discussion of multiple regression and region, as local variation might necessitate different
other multivariate techniques, such as principal com- models within subregions. As with modifications due to
ponents analysis, is beyond the scope of this text. You spatial autocorrelation, a full discussion of this issue is
should consult statistical textbooks such as Davis beyond the scope of this text. For an overview of the
(2002) and Rogerson (2006) for related information. local–global issue applied to regression and other statis-
tical methods, see Fotheringham (1997); a detailed dis-
Considerations in Using Correlation-Regression. There
cussion of the related technique of geographically-
are several things that you should consider carefully when
weighted regression (GWR) can be found in
using bivariate correlation-regression and multiple regres-
Fotheringham et al. (2002).
sion. One is that high correlations do not necessarily imply
a causal relationship.The high correlation between murder
rate and poverty that we found could be a result of chance 3.3 Exploratory Data Analysis
(although it is unlikely that two random data sets could
result in a correlation of this magnitude); or some other One of the most important advances in statistical analysis
attribute or set of attributes could be influencing both of in the last 30 years was John Tukey’s (1977) development
these attributes. of exploratory data analysis (EDA). Rather than trying to
A second consideration is that the magnitude of r can make one “best” map, interactive graphics systems should
be affected by the level at which data have been aggre- provide multiple representations of a spatial data set. In
gated, which is known as the modifiable areal unit prob- much the same way, Tukey proposed that rather than try-
lem (Clark and Hosking 1986; Barrett 1994). Generally, ing to fit statistical data to standard forms (normal,
coarser levels of aggregation (e.g., analyzing at the city Poisson, binomial), data should be explored, much as a
level rather than at the census tract level) will lead to detective investigates a crime. In the process of exploring
higher r values because “aggregation reduces the between data, the purpose should not be to confirm what one
unit variation in an attribute, making the attribute seem already suspects, but rather to develop new questions or
more homogeneous” (Clark and Hosking, 1986, 405). A hypotheses.
related issue is that the magnitude of r might be a func- One technique representative of Tukey’s approach is
tion of the size and arrangement of enumeration units the stem-and-leaf plot, which is depicted in Figure 11
when the number of enumeration units remains constant using the murder rate data. To construct a stem-and-leaf
16:02:07.
is that it is possible to determine approximate values

0 0111222222233 for each observation. For the murder rate data, this
* might not be particularly useful, but for some data
0 . 5555777889
sets it can be. For example, Burt and Barber (1996,
1 * 000011224
542–544) describe a stem-and-leaf plot that portrays
1 . 5778 when houses were constructed in a neighborhood.
2 * 1244 The stem-and-leaf plot reveals that seven houses
2 . 8 were built during the post–World War II period
3 11 (1945–1947), but that no houses were built from 1948
* to 1950.
3 . 5
Another technique representative of Tukey’s work
4 * 11
is the box plot (Figure 12). Here, a rectangular box
4 . represents the interquartile range (the 75th minus the
5 * 25th percentile), and the middle line within the box
5 . 679 represents the median, or 50th percentile. The position
6 of the median, relative to the 75th and 25th per-
*
6 . centiles, is an indicator of whether the distribution is
symmetric or skewed; for the murder rate data, the
7 *
. position indicates a positive skew (a tail toward higher
7 8
values on the number line). The horizontal lines out-
side the rectangular box represent the maximum and
minimum values in the data. The relative positions of
FIGURE 11 Stem-and-leaf plot of the
these lines also permit an examination of the symme-
murder rate data presented in Table 1.
try of the distribution, and their positions relative to
the box indicate how extreme the maximum and mini-
mum values are. In this case, their relative positions
again suggest a positive skew.†
plot, one first separates the digits of the data values into
three classes: sorting digits, display digits, and digits not
displayed because of rounding.* For the murder rate
data, we chose the 10s place as a sorting digit, the 1s
place as the display digit, and did not display the 10ths 80 Maximum
place because we rounded to the nearest whole percent.
Sorting digits are placed to the left of the vertical line
shown in Figure 11 and are known as stems, whereas dis-
play digits are placed to the right and are known as 60
leaves. For Figure 11, Tukey’s conventional system
Murder Rate
of asterisks and dots was used to split the 10s place

into two parts (leaf values of 0 to 4 are plotted on one
row and leaf values of 5 to 9 on the next row). For exam- 40
ple, Norfolk’s murder rate of 24.1 appears as the fourth
leaf (“4”) in the fifth row (“2*”), and Los Angeles’s
rate of 28.2 appears as the only leaf (“8”) in the sixth
75th Percentile
row (“2–”). 20
If you mentally rotate the stem-and-leaf plot so
that the stems are on the bottom, and then compare Median
it to those graphical methods discussed previously for
25th Percentile
individual attributes (Figures 1 and 2), you will note 0 Minimum
a great deal of similarity to the histogram. Both
methods portray a peak on the left side of the graph
with a distinct positive skew. Although the graphs are FIGURE 12 Box plot of the murder rate data
similar, the major advantage of the stem-and-leaf plot presented in Table 1.
* Tukey (1977) indicated that digits beyond the display digit could † Numerous variations of the box plot have been developed; for an
either be used for rounding or simply ignored. overview and suggestions for an alternative using color, see Carr (1994).
53
16:02:07.
Log of
60 Maximum Murder
Rate
Percent
below
Poverty
Murder Rate
40 Level
Percent
with
Maximum High
75th Percentile School
Education
20
Median
Log of
25th Percentile 75th Percentile Total
Median Population
Minimum 25th Percentile
0 Minimum Square
Root of
Drug
Deep South California Arrest
Log of
FIGURE 13 Box plots comparing murder rate data for cities Population
Density
in the Deep South and California. California cities appear to
have a distinctly smaller range, a smaller maximum, and an
interquartile range that is smaller and associated with lower FIGURE 14 Scatterplot brushing. As the rectangular box (or
murder rates. brush) is moved, dots (cities) within it are highlighted in all
scatterplots.
Box plots are most frequently used to compare two or

more distributions having the same units of measurement. Mark Monmonier (1989) developed the notion of
For example, Figure 13 compares murder rate data for integrating scatterplot brushing with maps. Monmonier
cities falling in two distinct regions (the “Deep South” and indicated that as the brush is moved within the scatter-
“California,” as defined by Birdsall and Florin 1992). Note plot, mapped areas corresponding to dots within the
that California cities have a distinctly smaller range, a brush should also be highlighted (Figure 15). He also
smaller maximum, and an interquartile range that is suggested the notion of a geographic brush: as areas of a
smaller and shifted toward lower murder rates. map are brushed, corresponding dots in the scatterplot
It is important to realize that many of the methods matrix should be highlighted.
covered thus far can also be used in an exploratory In this section we have focused on the notion of EDA
manner. For example, in a fashion similar to the stem- as developed by statisticians, and suggested how you, as
and-leaf plot, a dispersion graph can uncover nuances in a cartographer, might make use of these ideas. You
the data not revealed by a histogram. The scatterplot should realize that statisticians within geography have
matrix can also be used in an exploratory fashion, espe- also borrowed ideas from EDA, developing an area they
cially if the number of attributes is large. Remember that call exploratory spatial data analysis (ESDA). Anselin
the key to exploratory analysis is to reveal hidden char- (1998) provides a good overview of ESDA, and the
acteristics; the broadest possible range of approaches January 2006 issue of Geographical Analysis describes a
should be considered for achieving this. range of software developed for ESDA.
A technique that fits especially well under the
heading of EDA is scatterplot brushing, which is
illustrated in Figure 14 for the data for the 50 U.S. 4 NUMERICAL SUMMARIES IN WHICH
cities. Note that the scatterplot for log of murder LOCATION IS AN INTEGRAL COMPONENT
rate–percent below poverty level contains a gray
rectangular box, or “brush,” and that all dots (cities) This section considers numerical summaries in which
within this box are highlighted by a solid fill; these spatial location is an integral component. Some of these
same cities are also highlighted within other scatter- methods analyze just spatial location (formulas for the
plots in the matrix. Brushing involves using an inter- centroid and various shape indices), whereas others
active graphics display to move the box; as the box consider both spatial location and the values of an
moves, observations falling within it are highlighted attribute (e.g., spatial autocorrelation).
within all scatterplots.
16:02:07.
FIGURE 15 How scatterplot brushing can

be integrated with a map. As the brush (the
rectangular box) is moved in the scatterplot,
dots within it are highlighted within the scat-
terplot and on the map. (Monmonier 1989.
Reprinted by permission from Geographical
Analysis, Vol. 21, No. 1, Jan. 1989. Copyright
1989 by Ohio State University Press. All
rights reserved.)
4.1 Analysis of Spatial Location n
a (Xi+1Yi - Xi Yi+1) (Yi + Yi+1)

i=1
Centroid Yr = n
The centroid is defined as 3 a (Xi+1Yi - Xi Yi+1)
i=1
a (Xi Yi+1 - Xi+1Yi)(Xi + Xi+1) where the Xi Yi are a sequence of points defining the
i=1
Xr = n
boundary of a region, assuming that the n + first point is
3 a (Xi Yi+1 - Xi+1Yi) identical to the first point (Bachi 1999, 106). These formu-
i=1 las appear rather complicated, but the concept is really
55
16:02:07.
quite simple. Imagine cutting the outline of the 48 con-

tiguous states out of a thin metal sheet and then attempt-
ing to balance the sheet on the eraser end of a pencil. The
point at which the sheet balances (which happens to be
located near Lebanon, Kansas) would be its centroid (or
center of gravity). It should be recognized, however, that
the centroid derived by the preceding formula might not
necessarily fall within a region if that region is highly con-
voluted, such as Florida. As a result, it might be more
appropriate to consider other measures for computing the
center of region, such as the center of a rectangle sur-
rounding a region—the so-called bounding rectangle
(Carstensen 1987).
FIGURE 17 Two regions having different shapes but similar CI
Indexes for Measuring Shape values.
Geographers have developed a wide variety of indexes
for measuring the shape of geographic regions. One of
the simplest is the compaction index (CI) (Hammond One application of shape indices is redistricting,
and McCullagh 1978, 69–70), which is defined as the which involves the combination of voting precincts to
ratio of the area of a shape to the area of a circum- form legislative or congressional districts. This process
scribing circle (a circle that just touches the bounds must be done at each decennial census to account for
of the shape). Values for CI range from 0 to 1, with 0 migration and natural increases and decreases in popu-
representing the least compact shape (e.g., a narrow lation. You might be familiar with the related term
rectangular box) and 1 representing the most compact gerrymandering, in which districts are purposely struc-
shape (e.g., a circle). As an example, consider the tured for partisan benefit. Shape indexes can be used as
shapes of Tennessee and Arkansas relative to a circle objective measures of the degree of gerrymandering
(Figure 16); based on the CI, Arkansas is clearly more (Morrill 1981; Monmonier 2001).
compact than Tennessee. One limitation of the CI is
that it does not differentiate well between some shapes. 4.2 Analyzing an Attribute in Association
For example, Figure 17 shows two regions with shapes with Spatial Location
that differ, but that have similar CI values. This sort of
problem led geographers to develop other methods This section considers various methods for analyzing
for analyzing shape, many of which are more complex spatial data in which an attribute is linked with its spa-
than the CI; examples include the Boyce-Clark index tial location. Topics covered include (1) weighting
(Unwin 1981) and Moellering and Rayner’s (1982) attribute data to account for sizes of enumeration
harmonic-analysis method. units, (2) central tendency and dispersion for point
data, and (3) spatial autocorrelation and map com-
plexity measures.
Weighting Attribute Data to Account for Sizes

of Enumeration Units
Arthur Robinson and his colleagues (1995) argued that
when data are associated with enumeration units, basic
summary measures such as the mean and standard devi-
ation should be weighted to account for the differing
sizes of enumeration units. The formulas for a popula-
tion are as follows:
n n
2
a ai Xi a ai Xi
i=1 i=1
mw = n sw = n - m2w
a ai a a
ai
i=1 i=1
FIGURE 16 Computing the compaction index (CI) for two
states. CI is computed as a ratio of the area of the region to a
circle enclosing that region; the CI value for Arkansas clearly where Xi is a value for an attribute for the ith enumer-
would be greater than that for Tennessee. ation unit, and ai is the area of the ith enumeration unit.
16:02:07.
These weighted measures are appropriate if the being that larger enumeration units should impact r
intention is to impart to the reader the visual impact more. The formula is as follows (Robinson 1956):
that large enumeration units have on a distribution. You
n n
should realize, however, that different means and stan-
dard deviations will arise, depending on how enumera- a ai a aiXiYi
i=1 i=1
tion units are defined. To illustrate, consider the single n n
enumeration unit and its potential four subregions - a aiXi a aiYi
i=1 i=1
shown in Figure 18. Presume that within the single enu- rw = n
a aiXi b
n n 2
meration unit, 39 out of 1,300 people, or 3 percent, are
A a ai a aiX i - ia
2
college graduates; but that in three of the four subre- i=1 =1
i=1
gions, 13 out of 100 people, or 13 percent, are college n
* A a ai a aiYi 2 - a a aiYi b
n n 2
graduates; and in the fourth subregion, there are no
i=1 i=1 i=1
college graduates. The average of the four subregions is
9.75 percent, a value that is quite different from the
where ai is the area of the ith enumeration unit. Edwin
3 percent for the aggregated data.
Thomas and David Anderson (1965) showed that such a
Another weighting procedure relevant to enumera-
weighting procedure is inappropriate from a statistical
tion units would be to modify the formula for r to
perspective because unweighted correlation coefficients
account for the sizes of the enumeration units—the logic
resulting from different arrangements of enumeration
unit boundaries are not significantly different from
one another. Although inappropriate from a statistical
perspective, such a measure might be used as an indicator
of the visual correlation between two maps.
Central Tendency and Dispersion for Point Data

Earlier in this chapter we considered measures of central
tendency and dispersion that ignored spatial location. For
example, in computing the mean for murder rate, we
ignored the location of individual cities. We now consider
1300 People analogous measures of central tendency and dispersion
for explicitly spatial data. Imagine a set of x and y coordi-
39 College nates defining the location of small towns in a portion of
Graduates Minnesota.The central tendency for such a set of points is
termed the mean center, and is defined as
n n
a Xi a Yi
i=1 i=1
Xc = Yc =
n n
Note that these formulas simply apply the nonspatial mean

100 People 100 People formula twice, once for the x axis and once for the y axis.
As with the nonspatial mean, there is some danger that
13 College 13 College
Graduates Graduates the mean center will not fall where the data are concen-
trated. This problem can be handled using a spatial version
of the median (Hammond and McCullagh 1978, 48–53).
A measure of dispersion associated with the mean
100 People 1000 People center is standard distance, SD, which is computed as*
13 College 0 College n
Graduates Graduates 2
a dic
i=1
SD = = 2s2X + s2Y
Q N
FIGURE 18 As explained in the text, the areally * The formulas given are for a population. Griffith and Amrhein
weighted mean is not necessarily a logical mea- (1991, 123) indicate that for standard distance, there is little difference
sure to use. between sample and population formulas.
57
16:02:07.
n n n n
a a wij (Xi - X )(X j - X ) n a a w ij

where dic is the distance from the ith point to the
mean center. The formula on the left indicates that i=1 j=1 i=1 j=1
SD is a measure of spread about the mean center, and MC = n
the one on the right is valuable when using a calcula- 2
a (Xi - X ) /n
tor, which generally does not include functions for i=1
spatial statistics.
For some applications, it can be argued that mea- where wij = 1 if enumeration units i and j are adjacent
sures of central tendency and dispersion for points (or contiguous) and 0 otherwise.* Computations for the
should be weighted to reflect the magnitude of an MC for a hypothetical region consisting of nine enu-
attribute at the points. For example, imagine that you meration units are shown in Figure 19. Within each enu-
have population values for point locations (e.g., the meration unit, the upper left number is an identifier
centers of counties) and wish to find the mean center for that unit, and the middle number is the value for a
and standard distance for the population. The formu- hypothetical attribute. Note that the formula involves
las for the weighted mean center and weighted standard multiplying a weight for two enumeration units (w ij)
distance are: times the product of the difference between the
attribute values for the enumeration units and the mean
n n of the data (X i - X )(X j - X ). Computations are
a wiXi a wi Yi shown only for adjacent enumeration units, because wij
i=1 i=1
X cw = n Ycw = n
will be 0 for nonadjacent units and thus the product
i j (X i - X )(X j - X ) also will be 0.
w
aw
i a wi
i=1 i=1 The formula for MC resembles the formula for the
n correlation coefficient, r, discussed previously. For
2
a wi dic both equations, the denominator contains a measure
i=1
SDw = n
of the variation in the attribute about the mean, and
the numerator contains a measure of how adjacent
a a
wi
i=1 enumeration units covary; compare the equation
shown here with that for r in section 2. Thus, it is not
Using formulas similar to these, the U.S. Bureau of the surprising that MC also ranges from - 1 to + 1 . A
Census found that the population mean center moved value close to + 1 indicates that similar values are
from near Baltimore, Maryland, in 1790 to about 3 miles likely to occur near one another, whereas a value
east of Edgar Springs, Missouri, in 2000 (U.S. Department close to -1 indicates that unlike values are apt to
of Commerce 2001). occur near one another. Finally, a value near 0 is
indicative of no autocorrelation, or a situation in
Spatial Autocorrelation and Measuring Spatial Pattern which values of the attribute are randomly distrib-
uted. “Moderate” values of positive spatial autocorre-
Although maps allow us to visually assess spatial lation are most frequently observed in the real world
pattern, they have two important limitations: their (Griffith 1993, 2).
interpretation varies from person to person, and To illustrate how the MC might be used, consider
there is the possibility that a perceived pattern is mapping the rates for respiratory cancer for white
actually the result of chance factors, and thus not males in Louisiana counties (Figure 20). A visual
meaningful. For these reasons, it makes sense to assessment of this map suggests that there is a strong
compute a numerical measure of spatial pattern, positive spatial autocorrelation, with high cancer rates
which can be accomplished using spatial autocorre- occurring both in the southern part of the state and
lation. In this section we will consider how spatial along an east–west strip in the northern part of the
autocorrelation can be applied to choropleth maps. state. The MC for this pattern is .14, with a probability
For a discussion of measures appropriate for exam- of less than 3 percent that this value would occur by
ining patterns on other types of maps, see Unwin chance (Odland 1988). The value of .14 indicates that
(1981) and Davis (2002). the pattern is not quite as strongly positively autocor-
Spatial autocorrelation is the tendency for like related as a visual examination of the map suggests, but
things to occur near one another in geographic space. the associated probability indicates that the pattern is
For example, expensive homes likely will be located significant and worthy of further exploration.
near other expensive homes, and soil cores with high
clay content likely will be found near other soil cores
with high clay content. A common measure of spatial * The weights can be modified to account for differing sizes of enu-
autocorrelation is the Moran coefficient (MC), which is meration units (Odland 1988, 29–31). We have used weights of 0 and
defined as 1 for computational simplicity.
16:02:07.
Respiratory Cancer
White Males, 1970 - 1979
Deaths per 100,000

34 to 46
46 to 60
60 to 74
74 to 85.6
85.6 to 99
FIGURE 20 A map of rates of respiratory cancer for white

males in Louisiana counties. The Moran coefficient (a measure
of spatial autocorrelation) for this map is .14, with a probabil-
ity of less than 3 percent that this value occurs by chance. Thus,
the pattern is significant and worthy of further exploration.
(Courtesy of John Odland.)
An important recent development in statistical geog-

raphy is a consideration of spatial autocorrelation in the
context of inferential statistics. We touched on this idea
earlier in the chapter with respect to regression, but it
should be recognized that virtually all traditional infer-
ential statistics can be modified to account for spatial
autocorrelation. For further discussion of this issue, see
Griffith (1993) and Rogerson (2006).
MacEachren’s Measures of Map Complexity

The notion of spatial autocorrelation and associated
measures such as the MC were developed by statisti-
cal geographers. Cartographers have also been inter-
ested in measures for describing spatial pattern. As an
example, we’ll consider some map complexity mea-
sures developed by Alan MacEachren (1982a).
MacEachren defined map complexity as “the degree
to which the combination of map elements results in a
pattern that appears to be intricate or involved” (32).
He argued that when a map is used for presentation,
very complex maps might hinder the communication
of information.
MacEachren’s measures are based on Muller’s
(1974) application of graph theory to choropleth maps.
To compute measures of complexity, a base map is
treated as a set of faces (enumeration units), edges
FIGURE 19 Computation of the Moran coefficient (MC) for (the boundaries of the units), and vertices (where the
spatial autocorrelation. edges intersect; see Figure 21A). When mapping a
59
16:02:07.
A
Edges
Face
Vertex
1
1
1 3
1 CF = 5/15 = 0.33
CV = 8/28 = 0.29
2 3 CE = 19/42 = 0.45
3
3
2 2 2
2
2
2
1
1 3
2 CF = 12/15 = 0.80
CV = 22/28 = 0.79
3 3 CE = 27/42 = 0.64
2 1
1 2
1
3 3
1
FIGURE 21 Computation of MacEachren’s (1982a) face, vertex, and edge complexity measures. (A) Faces, edges, and
vertices for a county-level map of Arizona; (B) three-class map for which low complexity values result; (C) three-class
map for which high complexity values result.
16:02:07.
distribution, edges and vertices between like values (e.g., the mean and standard deviation) are a solution to
are omitted (Figures 21B and 21C). Complexity mea- these problems. A weakness of numerical summaries,
sures are computed by dividing the number of faces, however, is that they hide the detailed character of the
edges, and vertices on the mapped distribution by the data; for example, an outlier might be missed if only a
corresponding number on the base map:* numerical summary is used. Because of these advantages
and disadvantages, data exploration software should
Observed number of faces
CF = include a broad range of methods for analyzing spatial
Number of original faces data.
Observed number of vertices When examining two or more attributes at once, tab-
CV =
Number of original vertices ular displays are of limited use. Much more suitable are
Number of original edges between categories graphical displays, such as the scatterplot and scatterplot
CE =
Number of original edges matrix. The numerical method of bivariate correlation-
regression can also be useful if one wishes to summarize
the relationship between two attributes using a single
In the example illustrated in Figure 21, map C has the
number or equation. As with univariate data, care should
higher complexity measures, and so would presumably
be taken in using bivariate numerical summaries, as they
be more difficult for a user to interpret.
hide the detailed character of the relationship between
attributes. An important problem with bivariate correla-
SUMMARY tion is that it is a function of the level of aggregation (i.e.,
a different correlation coefficient might arise at the cen-
In this chapter, we have examined a variety of techniques sus tract level as opposed to at the block-group level), an
(tables, graphs, and numerical summaries) that can be issue known as the modifiable areal unit problem.
used along with maps to analyze spatial data. These tech- Given the emphasis on data exploration in this
niques each have their advantages and disadvantages. In text, special emphasis in this chapter was placed on
the univariate realm, raw tables are useful for providing exploratory data analysis—that data should be
specific information (e.g., the murder rate for Atlanta, explored, much as a detective investigates a crime. We
Georgia, in 1990 was 58.6 per 100,000 people), but they looked at three common methods for exploratory data
fail to provide an overview of a data set (e.g., that the analysis: the stem-and-leaf plot, the box plot, and
murder rate for U.S. cities in 1990 had a distinct positive scatterplot brushing. We saw that the notion of scatter-
skew). Grouped-frequency tables do provide an overview, plot brushing could be extended to a geographic
but are not as effective as graphical methods (e.g., the brush—as areas on a map are highlighted, correspond-
dispersion graph and histogram). Potential weaknesses of ing dots in a scatterplot matrix are highlighted. This is
graphical methods are the subjectivity of interpretation the sort of capability we would expect to find in data
and the physical space they require. Numerical summaries exploration software.
The latter section of the chapter considered several
numerical summaries for which spatial location is an
* MacEachren used the “number of original edges between cate- integral part; here we examined the centroid, compaction
gories” for the numerator of the edge measure to attempt to account index, mean center (and associated standard distance),
for the size of faces. spatial autocorrelation, and map complexity measures.
FURTHER READING
Anselin, L. (1998) “Exploratory spatial data analysis in a Barrett, R. E. (1994) Using the 1990 U.S. Census for Research.
geocomputational environment.” In Geocomputation: A Thousand Oaks, CA: Sage.
Primer, ed. by P. A. Longley, S. M. Brooks, R. McDonnell, Summarizes data available from the U.S. Bureau of the Census
and B. MacMillan, pp. 77–94. Chichester, England: Wiley. and discusses problems associated with using such data.
Summarizes the notion of exploratory spatial data analysis Burt, J. E., and Barber, G. M. (1996) Elementary Statistics for
(ESDA); for information on recently developed ESDA software, Geographers. 2nd ed. New York: Guilford.
see the January 2006 issue of Geographical Analysis.
An introductory statistics book for geographers.
Bachi, R. (1999) New Methods of Geostatistical Analysis and
Graphical Representation: Distributions of Populations over Cleveland,W. S. (1994) The Elements of Graphing Data. Rev. ed.
Territories. New York: Kluwer Academic/Plenum. Summit, NJ: Hobart Press.
Covers a broad range of graphical methods for summarizing data.
Presents geostatistical methods for summarizing the distribution
The text has more of a statistical emphasis than the Kosslyn text
of population over geographic regions.
listed below. Also see Cleveland (1993).
61
16:02:07.
Davis, J. C. (2002) Statistics and Data Analysis in Geology. A cartographer’s view on redistricting; Chapter 5 considers two
3rd ed. New York: Wiley. numerical measures that can be utilized to gauge the compactness
of congressional districts.
Although intended primarily for geologists, this text covers a
broad range of statistical methods, many of which are of interest Odland, J. (1988) Spatial Autocorrelation. Newbury Park, CA:
to cartographers. Sage.
Dykes, J. A. (1994) “Visualizing spatial association in area- A primer on spatial autocorrelation.
value data.” In Innovations in GIS, ed. by M. F. Worboys, Rogerson, P. A. (2006) Statistical Methods for Geography: A
pp. 149–159. Bristol, PA: Taylor & Francis. Student’s Guide. London: Sage.
Considers problems of using a mathematical measure of spatial auto- A statistics book for geographers that covers both introductory
correlation to represent the perceived degree of autocorrelation. and advanced concepts.
Fotheringham, A. S. (1999) “Trends in quantitative methods Tufte, E. R. (1983) The Visual Display of Quantitative
III: Stressing the visual.” Progress in Human Geography 23, Information. Cheshire, CT: Graphics Press.
no. 4:597–606.
Covers a broad range of techniques (both graphs and maps) for
Presents some directions that researchers might take in develop- representing numerical data.
ing methods for displaying multivariate data that would be useful
Tukey, J. W. (1977) Exploratory Data Analysis. Reading, MA:
to geographers.
Addison-Wesley.
Jacoby, W. G. (1997) Statistical Graphics for Univariate and
The classic reference on exploratory data analysis.
Bivariate Data. Thousand Oaks, CA: Sage.
Unwin, D. (1981) Introductory Spatial Analysis. London:
An overview of graphical methods for summarizing both univari-
ate and bivariate data. This short work (98 pages) is a useful com- Methuen.
plement to Cleveland’s book mentioned above. For graphical Covers a broad range of methods for analyzing spatial data.
methods for summarizing multivariate data, see Jacoby (1998).
Wegman, E. J. (1990) “Hyperdimensional data analysis using
Kosslyn, S. M. (1994) Elements of Graph Design. New York: parallel coordinates.” Journal of the American Statistical
W. H. Freeman. Association 85, no. 411:664–675.
A how-to text on constructing graphs; the text is liberally illus- Explains the mathematics underlying parallel coordinate plots and
trated with do and don’t examples. how to interpret them. For a more detailed discussion of the math-
Monmonier, M. (2001) Bushmanders & Bullwinkles. Chicago: ematics, see Inselberg (1985).
The University of Chicago Press.
GLOSSARY
bivariate correlation: a numerical method for summarizing a variety of ways, much as a detective investigates a crime; this
the relationship between two interval or ratio-level attributes. is in contrast to fitting data to standard forms, such as the
bivariate regression: the process of fitting a line to two inter- normal distribution.
val or ratio-level attributes; the dependent attribute is plotted on geographic brush: as areas of a map are selected (e.g., using
the y axis, and the independent attribute is plotted on the x axis. a mouse), corresponding dots are highlighted within a scatter-
bounding rectangle: a rectangle that just touches but plot matrix.
completely encloses an arbitrary shape. geostatistical methods: numerical summaries for which
box plot: a method of exploratory data analysis that illus- spatial location is an integral part (e.g., centroid and spatial
trates the position of various numerical summaries along the autocorrelation measures).
number line (e.g., minimum, maximum, median, and lower gerrymandering: the purposeful distortion of legislative or
and upper quartiles). congressional districts for partisan benefit.
centroid: the “balancing point” for a geographic region. grouped-frequency table: constructed by dividing the data
compaction index (CI): a measure of shape involving the ratio range into equal intervals and tallying the number of observa-
of the area of the shape to the area of a circumscribing circle. tions falling within each interval.
correlation coefficient: a numerical expression of the rela- hexagon bin plot: a method of data display in which a set of
tionship between two interval or ratio-level attributes. hexagons is placed over a conventional scatterplot, and the
hexagons are filled as a function of the number of dots falling
descriptive statistics: used to describe the character of a
within them.
sample or population, such as computing the mean square
footage of a sample of 50 stone houses. histogram: a type of graph in which data are grouped into
classes, and bars are used to depict the number of values
dispersion graph: a graph in which data are grouped into
falling in each class.
classes, and the number of values falling in each class is repre-
sented by stacked dots along the number line. inferential statistics: used to make an inference (or guess)
about a population based on a sample.
exploratory data analysis (EDA): a method for analyzing
statistical data in which the data are examined graphically in interquartile range: the absolute difference between the
75th and 25th percentiles of the data.
16:02:07.
map complexity: a term used to indicate whether the set of range: the maximum minus the minimum of the data.
elements composing a map pattern appears simple or intri- raw table: a form of tabular display in which the actual data
cate (i.e., complex); for example, does the pattern of gray are listed from lowest to highest value.
tones on a choropleth map appear simple or intricate?
redistricting: the process of assigning voting precincts to leg-
mean: the average of a data set; computed by adding all islative or congressional districts to equalize voter representa-
values and dividing by the number of values. tion.
mean center: a measure of central tendency for point data; reduced major-axis approach: the process of fitting a line to
computed by independently averaging the x and y coordinate two attributes when we do not wish to specify a dependent
values of all points. attribute.
measures of central tendency: used to specify a value around sample: the portion of the population that is actually exam-
which data are concentrated (e.g., the mean and median). ined, such as sampling only 50 stone houses out of the 5,000
measures of dispersion: used to specify the dispersion or existing within a city.
variability in a data set (e.g., the interquartile range and stan- scatterplot: a diagram in which dots are used to plot the
dard deviation). scores of two attributes on a set of x and y axes.
median: the middle value in an ordered set of data. scatterplot brushing: subsets of data within a scatterplot
mode: the most frequently occurring value in a data set. matrix are focused on by moving a rectangular “brush”
modifiable areal unit problem: the notion that the magni- around the matrix; dots falling within the brush are high-
tude of the correlation coefficient (r) can be affected by the lighted in all scatterplots.
level at which data have been aggregated. scatterplot matrix: a matrix of scatterplots is used to exam-
multiple regression: a statistical method for summarizing ine the relationships among multiple attributes.
the relationship between a dependent attribute and a series of skewed distribution: when data are placed along the number
independent attributes. line, the distribution appears asymmetrical; most common is a
normal distribution: a term used to describe the bell-shaped positive skew, in which the bulk of data are concentrated
curve formed when data are distributed along the number toward the left and there may be outliers to the right.
line. spatial autocorrelation: the tendency for like things to occur
one-dimensional scatterplot: see point graph. near one another in geographic space (technically, this is
termed positive spatial autocorrelation).
outliers: unusual or atypical observations.
standard deviation: the square root of the average of the
parallel coordinate plot (PCP): a method of graphical dis-
squared deviations of each data value about the mean.
play in which attributes are depicted by a set of parallel axes
and observations are depicted as a series of connected line standard distance (SD): a measure of dispersion for point
segments passing through the axes. data; essentially, a measure of spread about the mean center
of the distribution.
point graph: a type of graph in which each data value is rep-
resented by a small point symbol plotted along the number stem-and-leaf plot: a method for exploratory data analysis
line. in which individual data values are broken into “stem” and
“leaf” portions; the result resembles a histogram but provides
population: the total set of elements or things that one could
greater detail for individual data values.
study, such as all stone houses within a city.
three-dimensional scatterplot: a diagram that uses small
principal components analysis: a method for succinctly
point symbols in which the scores for observations on three
summarizing the interrelationships occurring within a set of
attributes are plotted on a set of x, y, and z axes; viewing all
attributes.
points in the plot requires interactive graphics.
REFERENCES
Anselin, L. (1998) “Exploratory spatial data analysis in a Birdsall, S. S., and Florin, J. W. (1992) Regional Landscapes of
geocomputational environment.” In Geocomputation: A the United States and Canada. New York: Wiley.
Primer, ed. by P. A. Longley, S. M. Brooks, R. McDonnell, Burt, J. E., and Barber, G. M. (1996) Elementary Statistics for
and B. MacMillan, pp. 77–94. Chichester, England: Wiley. Geographers (2nd ed.). New York: Guilford.
Bachi, R. (1999) New Methods of Geostatistical Analysis and Carr, D. B. (1991) “Looking at large data sets using binned
Graphical Representation: Distributions of Populations data plots.” In Computing and Graphics in Statistics, ed. by
over Territories. New York: Kluwer Academic/Plenum. A. Buja and P. A. Tukey, pp. 7–39. New York: Springer-
Barnett, V., and Lewis, T. (1994) Outliers in Statistical Data Verlag.
(3rd ed.). Chichester, England: Wiley. Carr, D. B. (1994) “A colorful variation on box plots.”
Barrett, R. E. (1994) Using the 1990 U.S. Census for Statistical Computing & Statistical Graphics Newsletter 5,
Research. Thousand Oaks, CA: Sage. no. 3:19–23.
63
16:02:07.
Carr, D. B., Olsen, A. R., and White, D. (1992) “Hexagon Jenks, G. F. (1977) Optimal data classification for choropleth
mosaic maps for display of univariate and bivariate maps. Occasional Paper No. 2. Lawrence, KS: Department
geographical data.” Cartography and Geographic of Geography, University of Kansas.
Information Systems 19, no. 4:228–236, 271. MacEachren, A. M. (1982a) “Map complexity: Comparison
Carstensen, L. W. J. (1987) “A comparison of simple mathe- and measurement.” The American Cartographer 9, no.
matical approaches to the placement of spot symbols.” 1:31–46.
Cartographica 24, no. 3:46–63. McMaster, R. B. (1987) “The geometric properties of numeri-
Clark, W. A. V., and Hosking, P. L. (1986) Statistical Methods cal generalization.” Geographical Analysis 19, no.
for Geographers. New York: Wiley. 4:103–116.
Cleveland, W. S. (1993) Visualizing Data. Summit, NJ: Hobart Moellering, H., and Rayner, J. N. (1982) “The dual axis
Press. fourier shape analysis of closed cartographic forms.” The
Cleveland, W. S. (1994) The Elements of Graphing Data (rev. Cartographic Journal 19, no. 1:53–59.
ed.). Summit, NJ: Hobart Press. Monmonier, M. (1989) “Geographic brushing: Enhancing
Davis, J. C. (2002) Statistics and Data Analysis in Geology exploratory analysis of the scatterplot matrix.”
(3rd ed.). New York: Wiley. Geographical Analysis 21, no. 1:81–84.
Eyton, J. R. (1984a) “Complementary-color two-variable Monmonier, M. (2001) Bushmanders & Bullwinkles.
maps.” Annals of the Association of American Geographers Chicago: University of Chicago Press.
74, no. 3:477–490. Morrill, R. L. (1981) Political Redistricting and Geographic
Flanagan, T. J., and Maguire, K. (1992) Sourcebook of Theory. Washington, DC: Association of American
Criminal Justice Statistics 1991. Washington, DC: U.S. Geographers.
Department of Justice, Bureau of Justice Statistics. Muller, J. C. (1974) “Mathematical and statistical compar-
Fotheringham, A. S. (1997) “Trends in quantitative methods isons in choropleth mapping.” Unpublished PhD disserta-
I: Stressing the local.” Progress in Human Geography 21, tion, University of Kansas, Lawrence, KS.
no. 1:88–96. Odland, J. (1988) Spatial Autocorrelation. Newbury Park,
Fotheringham, A. S. (1998) “Trends in quantitative methods CA: Sage.
II: Stressing the computational.” Progress in Human Rogerson, P. A. (2006) Statistical Methods for Geography: A
Geography 22, no. 2:283–292. Student’s Guide. London: Sage.
Fotheringham, A. S., Brunsdon, C., and Charlton, M. (2002) Thomas, E. N., and Anderson, D. L. (1965) “Additional
Geograhically Weighted Regression: The Analysis of comments on weighting values in correlation analysis of
Spatially Varying Relationships. West Sussex, England: areal data.” Annals of the Association of American
John Wiley. Geographers 55, no. 3:492–505.
Griffith, D. A. (1993) Spatial Regression Analysis on the PC: Tukey, J. W. (1977) Exploratory Data Analysis. Reading, MA:
Spatial Statistics Using SAS. Washington, DC: Association Addison-Wesley.
of American Geographers. Unwin, D. (1981) Introductory Spatial Analysis. London:
Griffith, D. A., and Amrhein, C. G. (1991) Statistical Analysis Methuen.
for Geographers. Englewood Cliffs, NJ: Prentice Hall. U.S. Bureau of the Census. (1994) County and City Data
Hammond, R., and McCullagh, P. (1978) Quantitative Book: 1994. Washington, DC: U.S. Government Printing
Techniques in Geography: An Introduction (2nd ed.). Office.
Oxford, England: Clarendon. U.S. Department of Commerce. (2001) “Centers of population
Inselberg, A. (1985) “The plane with parallel coordinates.” computation for 1950, 1960, 1970, 1980, 1990 and 2000.”
The Visual Computer 1:69–91. http://www.census.gov/geo/www/cenpop/calculate2k.pdf.
Jacoby, W. G. (1998) Statistical Graphics for Visualizing
Multivariate Data. Thousand Oaks, CA: Sage.
16:02:07.
Data Classification
OVERVIEW
Data classification involves combining raw data into the method can assist in selecting an appropriate number
classes or groups, with each class represented by a of classes.
unique symbol. For instance, data for a choropleth map Many cartographers have promoted the optimal
might be grouped into five classes, with each class method of classification because it does the best job of
depicted by a different shade of gray. The result of data considering how data are distributed along the number
classification is a classed map; in contrast, if each raw line (by placing similar data values in the same class and
data value is depicted by a unique symbol, an unclassed dissimilar data values in different classes). The optimal
map results. method, however, does not score well on some of the other
Although data classification is relevant to a wide vari- criteria; for example, the legend is difficult to understand
ety of mapping techniques, it is commonly discussed in and the method is unacceptable for ordinal data. As a
association with choropleth maps. Traditionally, cartog- result, it is important that you learn about the advantages
raphers argued for classed choropleth maps on two and disadvantages of other methods of classification.
grounds: readers’ inability to discriminate among many One limitation of the classification methods described
differing areal symbols, and the difficulty of creating in section 1 is that they do not consider the spatial context
unclassed maps using traditional photomechanical proof the data. If the purpose of classification is to simplify
cedures. Today the latter constraint has been eliminated, the appearance of a map, it can be argued that spatial
as computer hardware is capable of producing unclassed context also should be considered. In light of this, section
maps; as a result, some cartographers now question 2 covers two approaches for simplifying choropleth
whether data classification is a necessity. maps that do consider spatial context. The first uses the
In this chapter, we consider two issues relevant to the optimal classification method, but incorporates a spatial
classification of a single attribute: methods of classifica- constraint by requiring that values falling in the same
tion and how spatial context can be used to simplify the class be contiguous, whereas the second employs no clas-
appearance of choropleth maps. Section 1 considers six sification approach, but rather smooths the raw data by
common methods of data classification: equal intervals, changing values for enumeration units as a function of
quantiles, mean–standard deviation, maximum breaks, values of neighboring units.
natural breaks, and optimal. A common problem you Taking a broader view of the classification issue, Marc
will face is determining which of these methods you Armstrong and his colleagues (2003) have argued that we
should use. Criteria that can assist in selecting a classifi- should consider multiple criteria when classifying data,
cation method include: (1) whether the method considers including minimizing the difference between data values
how data are distributed along the number line, (2) ease in classes, minimizing the difference between classes for
of understanding the method, (3) ease of computation, contiguous enumeration units, maximizing the overall
(4) ease of understanding the legend, (5) whether the spatial autocorrelation in the classed map, and equalizing
method is acceptable for ordinal data, and (6) whether the area in each class. The first of these is a nonspatial
65
16:05:32.
Data Classification
constraint, while the latter three are spatial constraints. We in 1990 (U.S. Bureau of the Census 1994). Raw data
discuss Armstrong et al.’s approach in section 3. (sorted from low to high data values) and a dispersion
graph for this data set are given in Table 1 and
Figure 1, respectively. Note that the dispersion graph
1 COMMON METHODS OF DATA has a positive skew, with a distinctive outlier (i.e.,
CLASSIFICATION Dade County, which includes the city of Miami).
Before attempting to classify data, it is useful to
In this section, we consider six common methods of data consider whether the data have either a natural or a
classification: equal intervals, quantiles, mean–standard meaningful dividing point that can be used to partition
deviation, maximum breaks, natural breaks, and opti- the data (the data are bipolar. For example, a data set
mal. These methods range from those that do not con- of “percent population change” has a natural dividing
sider how data are distributed along the number line point of zero, which can be used to create two classes:
(e.g., equal intervals) to those that do (e.g., optimal). We values at or above zero and values below zero. Once
point out advantages and disadvantages of each data have been split in such a fashion, it might be
method, and develop a general set of criteria by which appropriate to apply one of the methods discussed in
the methods can be evaluated. For simplicity, we assume this section to each subset of data. Even if there is no
that the intent is to visualize an attribute for a single natural dividing point in the data, it might be desirable
point in time, as opposed to visualizing how an attribute to create a meaningful dividing point prior to classify-
changes over time. ing. For example, we might compute a mean percent-
For illustrative purposes, we will map the percentage value for the foreign-born data and then split the
age of foreign-born residents for counties in Florida data into values above and below the mean.
TABLE 1 Percentage of foreign-born population in Florida counties in 1990

Observation County % Foreign-Born Observation County % Foreign-Born
1 Madison 0.6 35 Leon 3.7
2 Calhoun 0.8 36 Lafayette 4.1
3 Dixie 0.8 37 Okaloosa 4.3
4 Baker 0.8 38 Glades 4.5
5 Taylor 0.9 39 Highlands 4.6
6 Jefferson 1.0 40 Citrus 4.9
7 Liberty 1.0 41 Lee 5.2
8 Bradford 1.0 42 Brevard 5.3
9 Wakulla 1.2 43 Manatee 5.4
10 Gadsden 1.2 44 DeSoto 5.5
11 Gilchrist 1.3 45 Hernando 5.5
12 Gulf 1.3 46 Volusia 5.8
13 Holmes 1.4 47 Pasco 5.9
14 Suwannee 1.5 48 Alachua 5.9
15 Hamilton 1.6 49 Sarasota 6.0
16 Walton 1.6 50 Indian River 6.1
17 Nassau 1.6 51 Okeechobee 6.3
18 Columbia 1.7 52 Seminole 6.3
19 Franklin 1.8 53 St. Lucie 6.3
20 Sumter 1.9 54 Charlotte 6.3
21 Jackson 2.2 55 Hardee 6.3
22 Putnam 2.2 56 Martin 6.8
23 Sana Rosa 2.2 57 Osceola 7.1
24 Levy 2.3 58 Pinellas 7.1
25 Union 2.4 59 Orange 7.5
26 Washington 2.4 60 Hillsborough 7.6
27 Escambia 2.7 61 Flagler 8.3
28 Clay 3.1 62 Monroe 10.1
29 Bay 3.4 63 Collier 10.5
30 Duval 3.5 64 Palm Beach 12.2
31 Lake 3.5 65 Hendry 14.6
32 St. Johns 3.6 66 Broward 15.8
33 Polk 3.6 67 Dade 45.1
34 Marion 3.6
16:05:32.
Data Classification
Equal Intervals (A)
5 10 15 20 25 30 35 40 45
Class 1 Class 2 Class 3 Class 4 Class 5
Quantiles (B)
5 10 15 20 25 30 35 40 45
Class 1 Class 3 Class 5
Class 2 Class 4
Mean–Standard Deviation (C)
5 10 15 20 25 30 35 40 45
Class 3 Class 4 Class 5
Maximum Breaks (D)
5 10 15 20 25 30 35 40 45
Class 1 Class 2 Class 4 Class 5
Class 3
Natural Breaks (E)
5 10 15 20 25 30 35 40 45
Class 2
Optimal (median) (F)
5 10 15 20 25 30 35 40 45
Class 2
FIGURE 1 Dispersion graphs of the foreign-born data shown in Table 1 along with class breaks for various methods of data
classification.
67
16:05:32.
Data Classification
An important consideration in any method of TABLE 2 Class limit computations for equal-intervals
classification is selecting an appropriate number of classification
classes. To compare methods of classification in this sec- Class Calculated Limits Legend Limits
tion, we assume five classes, which are easy to discrimi-
Florida Foreign-Born Data (Five Classes)
nate on the gray-tone maps shown here.
Before selecting a classification method, it is essential 1 0.6 to 9.5 0.6 to 9.5
to determine precisely how you wish to portray the data 2 9.6 to 18.4 9.6 to 18.4
on the map. The precision you select will be a function of 3 18.5 to 27.3 18.5 to 27.3
4 27.4 to 36.2 27.4 to 36.2
the initial data you have available, your impression of
5 36.3 to 45.1 36.3 to 45.1
the quality of the data, and how easily you think readers
can interpret the numeric values you provide. In our Florida Foreign-Born Data (Four Classes)
case, we felt that the data were of sufficient quality to 1 0.6 to 11.725 0.6 to 11.7
report to the nearest tenth of a percent and that readers 2 11.726 to 22.850 11.8 to 22.9
of this text would be comfortable with these values. The 3 22.851 to 33.975 23.0 to 34.0
argument could certainly be made, however, that we 4 33.976 to 45.100 34.1 to 45.1
should have rounded to the nearest whole percent.
reflect the rounded raw data on which the classification
1.1 EQUAL INTERVALS is based. Because we began with values rounded to the
nearest tenth of a percent, we also should report class
In the equal intervals (or equal steps) method of classifi- limits to the nearest tenth of a percent. This step does not
cation, each class occupies an equal interval along the affect a five-class map for the foreign-born data because
number line. As a result, this method is identical to creat- the class limits are already reported to the nearest tenth.
ing a grouped-frequency table, except that cartographers (Note that the Calculated Limits and Legend Limits in
commonly distinguish between the calculated class limits Table 2 match.) To illustrate how this step might have an
and the limits actually used for mapping. The steps for impact, consider computing class limits for a four-class
computation are as follows: map for this same data set. The class interval would be
computed as follows:
Step 1. Determine the class interval, or width that each
class occupies along the number line. This is computed Range High - Low
by dividing the range of the data by the number of =
Number of classes Number of classes
classes. The result is as follows for the foreign-born data:
45.1 - 0.6
= = 11.125
Range High - Low 4
=
Number of classes Number of classes Using this class interval, we derive the calculated class
45.1 - 0.6 limits shown at the bottom of Table 2. Assuming that
= = 8.9 we wished to report class limits to the nearest 0.1, we
5
would round the calculated limits to create the legend
Step 2. Determine the upper limit of each class. The limits shown at the bottom of Table 2.
upper limit for each class is computed by repeatedly Step 5. Determine which observations fall in each
adding the class interval to the lowest value in the data. class. This involves simply comparing the raw values
(For the first class, adding the class interval 8.9 to 0.6 with the legend limits from step 4. Figure 1A presents
yields a value of 9.5.) The result is the right-hand set of the resulting classes in graphic form, and a map of the
numbers in the Calculated Limits column at the top of classified data appears in Figure 2A.
Table 2. An advantage of equal intervals is that these steps
Step 3. Determine the lower limit of each class. Lower can be completed using a calculator, or even pencil and
limits for each class are specified so that they are just paper. As a result, this method was often favored before
above the highest value in a lower-valued class (the mapping software became available. A second advan-
lower limit of class 2 is 9.6, which is 0.1 more than the tage is that the resulting equal intervals will, in some
upper limit of class 1). An alternative approach is to let cases, be easy for map users to interpret. For example, if
the lowest value of each class equal the highest value in you were making a five-class map of “percent urban
a lower-valued class, but then it may be unclear into population,” and the data ranged from 0 to 100, the
which class a data value falls if it matches the lowest resulting classes would be convenient rounded multiples
value in a class. (0–20, 21–40, 41–60, 61–80, and 81–100). For the foreign-
Step 4. Specify the class limits actually shown in the leg- born data, however, note that the mapped limits do not
end. The class limits actually shown in the legend should readily reveal what the class interval might be.
16:05:32.
Methods of Data Classification
Foreign Born in Florida, 1990
A B
Equal Intervals Quantiles
Percentage Percentage
Foreign Born Foreign Born
0.6 to 9.5 0.6 to 1.4
9.6 to 18.4 1.5 to 2.7
18.5 to 27.3 3.1 to 4.9
27.4 to 36.2 5.2 to 6.3
36.3 to 45.1 6.8 to 45.1
C D
Mean–Standard Deviation Maximum Breaks
< -6.9 0.6 to 8.3
-6.9 to -1.0 10.1 to 10.5
-0.9 to 10.9 12.2
11.0 to 16.8 14.6 to 15.8
16.9 to 45.1 45.1
E F
Natural Breaks Optimal (median)
0.6 to 2.7 0.6 to 2.4
3.1 to 4.9 2.7 to 4.9
5.2 to 8.3 5.2 to 8.3
10.0 to 15.8 10.1 to 15.8
45.1 45.1
FIGURE 2 Choropleth maps illustrating various methods of data classification for the foreign-born data shown in Table 1. (A)
Equal intervals, (B) quantiles, (C) mean–standard deviation, (D) maximum breaks, (E) natural breaks, and (F) optimal, based on
medians.
69
16:05:32.
Data Classification
data until the desired number of members in a class is

obtained. Identifying class membership for the foreign-
born data is somewhat complicated because the number
of observations desired in each class, 13.4, is not an integer
0.6 9.5 18.4 27.3 36.2 45.1 value. In such a situation, one should attempt to place
approximately the same number of observations in each
class. Thus, we placed 13 observations in the first class and
FIGURE 3 A legend for the equal-intervals method that 14 observations in the second class so that the total for the
takes advantage of the continuous nature of class limits. first two classes was 27 (approximately 13.4 * 2).
Because identical data values should not be placed in
different classes, ties can complicate the quantiles
A third advantage of equal intervals is that the method. This can be seen for the foreign-born data if a
legend limits contain no missing values (or gaps): For class break is attempted between the 54th (13.4 * 4) and
the foreign-born data, the difference between the upper 55th numeric positions. Defining a break here doesn’t
value in a class and the lower value in the next class is make sense because the data value for both positions is
0.1, which is the precision of the data. Gaps, which we 6.3. Rather, it is more logical to choose a break between
will see occur in other classification methods (e.g., the 55th and 56th positions.
quantiles), might cause a reader to wonder why some of Two approaches are possible for defining legend
the data are missing from the legend. A related advan- limits for the quantiles method. One is to specify the
tage for equal intervals is that the legend limits can be lowest and highest values of members in a class; for the
simplified so that only the lowest and highest values in foreign-born data, the 14 members of the second class
the data and the upper limit of each class are shown range from 1.5 to 2.7, and so these limits would be
(Figure 3). This approach should permit faster map shown in the legend. The other approach is to compute
interpretation, but it might also create confusion con- a class boundary as an average of the highest value in a
cerning the bounds of each class (e.g., the reader might class and the lowest value of the next class; using this
wonder whether 9.5 falls in the first or second class). approach for the foreign-born data, the upper limit for
The major disadvantage of equal intervals is that the the second class would be (2.7 + 3.1)/2, or 2.9. We pre-
class limits fail to consider how data are distributed fer the former approach because it more accurately
along the number line. For example, if you inspect the reflects the range of data values falling in a class. An
dispersion graph for the foreign-born data (Figure 1), advantage of the latter approach, however, is that only
you will note that classes 3 and 4 seem to be wasted, as the lowest and highest values in the data and the upper
these classes have no observations within them. On the limit of each class would need to be shown, as was done
plus side, however, note that unusual data values (such for the equal-intervals method in Figure 3.
as the Dade County outlier) will appear in their own As with equal intervals, an advantage of quantiles is
unique class. Thus, although two classes are not used for that class limits can be computed manually. A second
the foreign-born data, the mapped distribution does advantage is that because an equal number of obser-
correctly show Dade County as being distinctly differ- vations fall in each class, the percentage of observa-
ent from the rest of the data. tions in each class will also be the same. Depending on
the number of classes, this might simplify our discus-
1.2 Quantiles sion of the mapped data. For instance, on a five-class
map we can refer to the upper or lower 20 percent of
In the quantiles method of classification, data are rank- the data, whereas on a six-class map we would have
ordered and equal numbers of observations are placed to refer to the upper or lower 16.7 percent of the
in each class. Different names for this method are used, data, which is not a convenient round number. A
depending on the number of classes; for example, four- related advantage is that the 50th percentile or
and five-class quantiles maps are referred to as quartiles median (a measure of central tendency in the data)
and quintiles, respectively. To compute the number of will be logically associated with the classes. For an odd
observations in a class, the total number of observations number of classes, the median will fall in the center of
is divided by the number of classes. For the foreign-born the middle class, whereas for an even number of
data, we have: classes, the median will fall between the two middle
Total observations classes.
Number in class = = 67/5 = 13.4 A third advantage of quantiles is that because class
Number of classes
assignment is based on rank order, quantiles are useful
To determine which observations should be placed in each for ordinal-level data. For example, if the 50 states of
class, one simply progresses through the rank-ordered the United States were ranked on “quality of life,” the
16:05:32.
Data Classification
resulting ranks could be split into five equal groups: two different classes, and the limits of the lowest and
no numeric information would be necessary to create highest classes reflect the lowest and highest values in
the classification. A fourth advantage is that if enu- the data.
meration units are approximately the same size, each A major disadvantage of the mean–standard devia-
class will have approximately the same map area. tion method is that it works well only with data that are
The quantiles method shares the major disadvantage normally distributed. This is particularly evident with
of equal intervals: It fails to consider how the data are the foreign-born data, in which the two lowest classes
distributed along the number line. For example, note contain solely negative values and therefore have no
that for the foreign-born data, the outlier for Dade members (Figure 2C). One solution to this problem is
County is included in the same class with values of con- to transform the data, but this is inappropriate if the
siderably lower magnitude (Figure 1B). Thus, we have intention is to examine the raw data. Another disadvan-
eliminated the problem of empty classes, but distinctly tage is that the mean–standard deviation method
unlike data values have been placed in the same class. requires an understanding of some basic statistical con-
Another disadvantage of quantiles is that gaps result cepts; a message on the map or in the text indicating
(for the foreign-born data, the highest value in class 2 is that “classes were developed based on the mean and
2.7 and the lowest value in class 3 is 3.1, resulting in a standard deviation” would not be meaningful if one had
gap of 0.4; Figure 2B). Gaps are problematic because no statistical training.
the reader might wonder why they occur. However, they A distinct advantage of the mean–standard devia-
do permit the legend to reflect the range of data actu- tion method, however, is that if the data are normally
ally occurring in each class. distributed (or near normal), the mean serves as a
useful dividing point, enabling a contrast of values
1.3 Mean–Standard Deviation above and below it. This is most effectively accom-
plished if an even number of classes is used; for exam-
The mean–standard deviation method is one of several ple, a six-class map could consist of the positive
classification techniques that do consider how data are q to x
classes ( x q + s, xq + s to x
q + 2s, and 7 x q + 2s)
distributed along the number line. In this method, classes and the negative classes ( x q to x
q - s, qx - s to x
q - 2s,
are formed by repeatedly adding or subtracting the stan- and 6 x q - 2s). (For the five-class map shown in
dard deviation from the mean of the data, as shown in Figure 2C, the two middle classes were combined.)
Table 3.* As with the equal-intervals method, both cal- Another advantage is that the legend contains no gaps
culated and legend limits can be computed. Calculated that might confuse the reader.
limits are computed using the mean and standard devia-
tion values listed in column 2 of Table 3 (Normal 1.4 Maximum Breaks
Distribution Limits). To create legend limits, calculated
limits are adjusted so that identical values cannot fall in The mean–standard deviation method considers how
data are distributed along the number line in a holistic
sense—by trying to fit a normal distribution to the data.
An alternative approach is to consider individual data
values and group those that are similar (or, alterna-
TABLE 3 Class limit computations for mean–standard
deviation classification tively, avoid grouping those that are dissimilar). The
maximum breaks method is a simplistic means for
Normal accomplishing this.† In this method, raw data are
Distribution Calculated Legend
Class Limits Limits Limits
ordered from low to high, the differences between adja-
cent values are computed, and the largest of these
Florida Foreign-Born Data: xq = 4.96, s = 5.94 differences serve as class breaks.
1 6 xq - 2s 6 - 6.9 6 - 6.9 The foreign-born data provide a good illustration of
2 xq - 2s to qx - 1s -6.9 to - 1.0 - 6.9 to - 1.0 how the maximum breaks method groups similar values
3 xq - 1s to xq + 1s - 1.0 to 10.9 - 0.9 to 10.9 (and avoids grouping dissimilar values): In Figure 1, note
4 xq + 1s to xq + 2s 10.9 to 16.8 11.0 to 16.8 how the six highest data values are broken into four
5 7 xq + 2s 7 16.8 16.9 to 45.1 classes. In addition to considering how data are distrib-
uted along the number line, the maximum breaks method
is easy to compute, simply involving subtracting adjacent
* We use mean and standard deviation formulas appropriate for
a sample; because of the relatively large number of observations
in each case, similar results would be obtained with population † We have borrowed the term “maximum breaks” from the teaching
formulas. of George Jenks.
71
16:05:32.
Data Classification
values. A disadvantage of maximum breaks is that by TABLE 4 Computing the sum of absolute deviations
paying attention only to the largest breaks, the method about class medians (ADCM)
seems to miss natural clusters of data along the number Raw Data: 11, 12, 13, 14, 31, 32, 33, 99, 100
line. For example, for the foreign-born data, the maximum
breaks method ignores the peaks and dips in class 1 that Quantiles Classification Optimal Classification
seem to reflect natural clustering there. Class Values Error Class Values Error
1 11, 12, 13 2 1 11, 12, 13, 14 4

1.5 Natural Breaks
2 14, 31, 32 18 2 31, 32, 33 2
3 33, 99, 100 67 3 99, 100 1
The natural breaks method is one solution to the failure
of maximum breaks to consider natural groupings of ADCM = 87 ADCM = 7
data. In natural breaks, graphs (e.g., the dispersion graph
or histogram) are examined visually to determine logi-
cal breaks (or, alternatively, clusters) in the data. Stated method places similar values in the same class (the first
another way, the purpose of natural breaks is to mini- class consists of 11, 12, 13, and 14, whereas the second
mize differences between data values in the same class class consists of 31, 32, and 33).
and to maximize differences between classes. Later One measure of classification error commonly used
we will see that this is also the objective of the optimal in the optimal method is the sum of absolute devia-
method, but with the optimal method, the classification tions about class medians (ADCM). Computing this
is done using a numerical measure of classification measure involves calculating the median of each class,
error, whereas with natural breaks, the classification is which is the sum of absolute deviations of class
subjective. members about each class median, and then adding
To illustrate the computation of natural breaks, the resulting sums of absolute deviations. For exam-
consider how we might divide the foreign-born data ple, for quantiles, the median in the first class is 12
into five classes (you should examine the dispersion (remember, the median is simply the middle value in
graph associated with the natural breaks method in an ordered set), and the sum of absolute deviations
Figure 1E). The highest value in the data (45.1) for the class is |11 - 12| + |12 - 12| + |13 - 12| = 2.
appears to be quite different from the rest of the data, The sum of absolute deviations for all classes is
so we will place it in a class by itself. Our next deci- 2 + 18 + 67 = 87. In contrast, ADCM for the optimal
sion is how to handle the five values ranging from 10.1 method is 4 + 2 + 1 = 7, which is obviously a smaller
to 15.8. We could divide these into two separate value and thus is indicative of a better classification.
classes (10.1–12.2 and 14.6–15.8), but then we would The data for this hypothetical example were selected
have to divide the remaining data into two classes. It so that the results would be clear-cut. In the real world,
seems easier to divide the remaining data into three the desired minimum-error classification is normally not
classes, roughly corresponding to the three peaks in obvious, so researchers have developed computer-based
the dispersion graph, and so we group the values rang- algorithms for determining possible solutions. Here we
ing from 10.1 to 15.8 in one class. We can see in this consider two algorithms: the Jenks–Caspall and the
example that an obvious problem with natural breaks Fisher–Jenks.
is that decisions on class limits are subjective, and
therefore can vary among mapmakers. The Jenks–Caspall Algorithm
The Jenks–Caspall algorithm, developed by George Jenks
1.6 Optimal and Fred Caspall (1971), is an empirical solution to the
problem of determining optimal classes. It minimizes the
The optimal classification method is a solution to the sum of absolute deviations about class means (as opposed
limitations noted for maximum and natural breaks. The to medians). The algorithm begins with an arbitrary set of
optimal method places similar data values in the same classes (say, the quantiles classes shown in Table 4), calcu-
class by minimizing an objective measure of classifica- lates a total map error analogous to ADCM (but involv-
tion error. To illustrate, consider how a small hypothet- ing the mean), and attempts to reduce this error by mov-
ical data set of nine values would be classified by the ing observations between adjacent classes. Observations
quantiles and optimal methods (Table 4). The quantiles are moved using what Jenks and Caspall termed reitera-
method assigns the same number of observations to tive and forced cycling. In reiterative cycling, movements
each class (three, in this case), and thus places similar are accomplished by determining how close an observa-
values in different classes (e.g., 14 appears with 31 and tion is to the mean of another class; for example, for the
32 in class 2, even though 14 is more similar to 11, 12, quantiles data in Table 4, the value 14 is closer to the
and 13, the members of class 1). In contrast, the optimal mean of class 1 (12) than 13 is to the mean of class 2
16:05:32.
Data Classification
(25.7), so 14 would be moved to the first class. Movements map. They computed a measure of boundary accuracy
based on the relation of observations to class means are by dividing the sum of the n actual cliffs used on the
repeated until no further reductions in total map error classed map by the sum of the n largest cliffs occurring
can be made. in the raw data.
In forced cycling, individual observations are moved Of these three kinds of error, cartographers (including
into adjacent classes, regardless of the relation between Jenks himself) have focused on tabular error, primarily
the mean value of the class and the moved observation. because of its simplicity. We have presented the three
After a movement, a test is made to determine whether different kinds of error, however, to provide a broader
any reduction in total map error has occurred. If error perspective on the classification problem. In section 3, we
has been reduced, the new classification is considered will return to the notion of the boundary error when we
an improvement, and the movement process continues consider the possibility of using multiple criteria to deter-
in the same direction. Forcing is done in both direc- mine class intervals.
tions (from low to high classes and from high to low
classes). At the conclusion of forcing, the reiterative The Fisher–Jenks Algorithm
procedure described earlier is repeated to see whether In contrast to the empirical approach used by Jenks and
any further reductions in error are possible. Although Caspall, the Fisher–Jenks algorithm has a mathematical
this approach does not guarantee an optimal solution, foundation that guarantees an optimal solution. Walter
Jenks and Caspall (1971, 236) indicated that they were Fisher (1958) was responsible for developing the mathe-
“unable to generate, either purposefully or by accident, matical foundation, and George Jenks (1977) introduced
a better . . . representation in any set of data.” the idea to cartographers. Cartographers generally have
In addition to developing an automated algorithm chosen to recognize only Jenks for this contribution, so
for determining optimal classes, Jenks and Caspall the reader might find the algorithm referred to as
(1971, 225) introduced three criteria for selecting a “Jenks’s optimal method.”
“best” classification. They introduced these criteria by To understand the Fisher–Jenks algorithm, it is
posing three questions: worthwhile to consider how an optimal solution might
be computed using brute force. Imagine that you
1. Which map provides the reader with the most wanted to develop an optimal two-class map of the
accurate intensity values for specific places? data 1, 3, 7, 11, and 22. With such a small data set, it is
2. Which map creates the most accurate overview? easy to list all possible two-class solutions and compute
3. Which map contains boundaries that occur along associated error measures (Table 5). If the process is so
major breaks in the statistical surface? simple for a small data set, it would seem that for large
data sets, a computer could be used to determine an
Corresponding to these questions, Jenks and Caspall optimal solution by simply considering all possibilities.
discussed three kinds of error: tabular, overview, and Unfortunately, for large data sets, the number of possi-
boundary. ble solutions becomes prohibitively large; for example,
To understand these forms of error, it is helpful to Jenks and Caspall (1971, 232) calculated that for the
consider a 3-D prism map, which has no error due to 102 counties of Illinois, there would be over 1 billion
classification because each county is raised to a height possible seven-class maps.
proportional to the data. If the data were classed, error Rather than consider all solutions, the Fisher–Jenks
would arise as a result of counties in the same class algorithm takes advantage of the mathematical foun-
being raised to the same height. The difference in dation provided by Fisher, which states that any opti-
height between corresponding prisms would constitute mal partition is simply the sum of optimal partitions of
tabular error, and overview error would be the differ- subsets of the data. We illustrate this concept by con-
ence in volume between corresponding prisms. Tabular sidering some initial steps for handling the data 1, 3, 7,
error is equivalent to the error measure described pre- 11, and 22 (Table 6). For computational simplicity, we
viously (minimizing the sum of absolute deviations use the median (and associated sum of absolute devia-
about the median, or mean, of each class); thus, mini- tions); another version of the algorithm uses the mean
mizing tabular error produces an optimal classification. (and associated sum of squared deviations about the
In contrast, overview error is weighted to account for mean).
the size of enumeration units. Step 1 involves computing the sum of absolute devia-
Jenks and Caspall used the term boundary error to tions about the class median for any ordered subset of
describe the error occurring along the boundary the raw data, ignoring how these subsets might fit into a
between two enumeration units on a classed map. They particular classification. For example, the sum of absolute
argued that the highest cliffs appearing on the unclassed deviations for the first through third observations (the
prism map should ideally appear on the classed prism subset 1, 3, and 7) is |1 - 3| + |3 - 3| + |7 - 3| = 6.
73
16:05:32.
Data Classification
TABLE 5 Computing ADCM for all potential two-class TABLE 6 Initial steps in the Fisher–Jenks algorithm for
maps optimal data classification
Raw Data: 1, 3, 7, 11, 22 Raw Data: 1, 3, 7, 11, 22
Solution 1 Solution 2 Step 1. Compute the sum of absolute deviations about the class
median for all ordered subsets of the data.
Class Values Error Class Values Error
The following matrix shows the sum of absolute deviations
1 1 0 1 1, 3 2 about the median for the i th through the j th observation;
2 3, 7, 11, 22 23 2 7, 11, 22 15 for example, if i = 1 and j = 3, then the sum is
ADCM = 23 ADCM = 17 |1 - 3| + |3 - 3| + |7 - 3| = 6. (Note that this result appears
in the first row and the third column of the matrix.) The resulting
Solution 3 Solution 4 sums of absolute deviations are commonly termed the diameter (D)
and are represented by D(i, j) ; for this example D (1,3) = 6.
Class Values Error Class Values Error
j th observation
1 1, 3, 7 6 1 1, 3, 7, 11 14
1 2 3 4 5
2 11, 22 11 2 22 0
1 0 2 6 14 29
ADCM = 17 ADCM = 14
i th 2 0 4 8 23
Solution 4 is optimal because it has the smallest total error
observation 3 0 4 15
(or ADCM).
4 0 11
5 0
This result appears in row 1, column 3, of the matrix Step 2. Compute all optimal two partitions.
shown in step 1 of Table 6. The resulting sum of absolute
(a) The results for the optimal two-class map of the complete
deviations is termed the diameter (D), and is represented
data set are as follows:
by D(i, j), where i and j identify the observations in step
1 | 3 7 11 22
1 (Hartigan 1975, Chapter 6); thus, D(i, j) for this exam-
D(1,1) + D(2,5) = 0 + 23 = 23
ple would be D(1,3) = 6. 1 3 | 7 11 22
In step 2, the optimal solution for a two-class map D(1,2) + D(3,5) = 2 + 15 = 17
of the complete data set is computed, along with opti- 1 3 7 | 11 22
mal two-class solutions for subsets of the data. D(1,3) + D(4,5) = 6 + 11 = 17
1 3 7 11 | 22
Together, these are termed the optimal two partitions.
D(1,4) + D(5,5) = 14 + 0 = 14
Calculations for the optimal two-class map for the
complete data set are shown in part (a) of step 2 (b) The following are some results for optimal two-class partitions
(Table 6), and some of the results for subsets of the of subsets of the data:
data are shown in part (b) of step 2. Although the sub- 1 | 3 7 11
D(1,1) + D(2,4) = 0 + 8 = 8
set calculations are not used in determining the opti-
1 3 | 7 11
mal two-class map, they are used to determine optimal D(1,2) + D(3,4) = 2 + 4 = 6
classifications for maps with a greater number of 1 3 7 | 11
classes, such as a three-class map, for which some cal- D(1,3) + D(4,4) = 6 + 0 = 6
culations are shown in step 3. 1|3 7
D(1,1) + D(2,3) = 0 + 4 = 4
1 3|7
Advantages and Disadvantages of Optimal D(1,2) + D(3,3) = 2 + 0 = 2
Classification Ideally, all optimal two partitions would be computed!
The obvious advantage of the optimal method is that it Step 3. Compute all optimal three-partitions (let Opt-2 represent
considers, in detail, how data are distributed along the an optimal two partition.)
number line. It is the “best” choice for classification (a) The results for the optimal three-class map of the complete
when the intention is to place like values in the same data set would be calculated. Some of the calculations follow.
class (and unlike values in different classes) based on The question marks represent optimal two-class partitions not
computed above.
the position of values along the number line.
Another advantage is that the optimal method can 1 | 3 7 11 22
D(1,1) + Opt-2 = 0 + ?
assist in determining the appropriate number of classes. 1 3 | 7 11 22
When the median is used as the measure of central ten- D(1,2) + Opt-2 = 2 + ?
dency, this is accomplished by computing the goodness 1 3 7 | 11 22
of absolute deviation fit (GADF), which is defined as: Opt-2 + D(4,5) = 2 + 11 = 13
1 3 7 11 | 22
ADCM Opt-2 + D(5,5) = 6 + 0 = 6
GADF = 1 -
ADAM (b) Calculate all other optimal three-class partitions.
16:05:32.
Data Classification
where ADCM is the sum of absolute deviations about the foreign-born data, the curve appears to flatten out at
class medians for a particular number of classes, and about five classes. A flattening at this point indicates that
ADAM is the sum of absolute deviations about the a larger number of classes would not contribute sub-
median for the entire data set. An analogous measure can stantially to a reduction in the classification error. (One
be computed when the mean is used as the measure of should also bear in mind that a map with more classes
central tendency (and the error in a class is the sum of would be more difficult to use, as a greater number of
squared deviations about the mean), and is known as the areal symbols would have to be differentiated.)
goodness of variance fit (GVF; Robinson et al. 1984, 363). Another approach is to determine the number of
GADF ranges from 0 to 1, with 0 representing the low- classes for which the GADF first exceeds a certain
est accuracy (a one-class map) and 1 representing the value, say, 0.80 (i.e., the accuracy is 80 percent). For the
highest accuracy. If there are no ties in the data, then a foreign-born data, this approach yields a six-class map
GADF value of 1 will result only when each observation (Figure 4). Note, however, that if a more stringent value
is a separate class (an n-class map will be required, where were used, say, 0.90, a nine-class map would be required.
n is the number of classes). Ties can, however, consider- Admittedly, both of these approaches are subjective,
ably reduce the number of classes needed to achieve a but they are an improvement over choosing an arbi-
GADF of 1. For example, in the case of the foreign-born trary number of classes.
data, only 46 classes are needed (there are 67 data values). In addition to helping you determine an appropri-
It is important to note that an n-class map is equivalent ate number of classes for the optimal method, the
to an “unclassed map.” This might be confusing because GADF technique could also be used to compute the
the term unclassed suggests no classes, and there are actu- accuracy of other classification methods, and thus
ally n classes on an n-class map. “Unclassed” is commonly determine whether those methods might be appropri-
used to indicate that no classing or grouping has been ate. For example, if you were to compute similar
applied to the data; for instance, it is not necessary to run GADF values for optimal and quantiles, you might
the optimal program to create an unclassed map. choose the quantiles method because it would be eas-
GADF calculations can assist in selecting an appro- ier for the user to understand how the class limits were
priate number of classes in two ways. One approach is to created. Alternatively, GADF values might be com-
construct a graph of the number of classes against puted for various numbers of classes for quantiles to
GADF values (Figure 4) and look for a point at which a assist in determining an appropriate number of classes
curve fit to the data begins to flatten out. In the case of for that method.
Disadvantages of the optimal method include the dif-
ficulty of understanding the concept and the appearance
of gaps in the legend. Another traditional disadvantage
was that software packages generally did not include the
1.0 optimal method as an option, but this has changed (e.g.,
the popular ArcGIS package includes an optimal option,
which it terms “Natural Breaks (Jenks)”).
0.8
1.7 Criteria for Selecting a Classification Method
0.6 In discussing common methods of classification, we

GADF
have pointed out numerous criteria that might be

used to judge these methods’ usefulness. Figure 5
0.4 summarizes these criteria and rates each classification
method as very good, good, or poor on each measure
(and acceptable or unacceptable in the case of
0.2 “acceptable for ordinal data”). One problem with any
rating system is that it is a function of the computer
environment the mapmaker has available and the
0 knowledge of the map user. For Figure 5, we assume
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
that computer software is available for creating all of
Number of Classes the classification methods we have considered and
that the map user is a college-level student with a
FIGURE 4 Graph of the number of classes plotted against basic foundation in introductory statistics.
GADF values. The curve appears to flatten out at about five Note that “ease of understanding legend” is a func-
classes, indicating that a greater number of classes would not tion of whether or not there are gaps in the legend:
substantially reduce the error on the map. Remember that gaps between class limits can make the
75
16:05:32.
Data Classification
FIGURE 5 Criteria for selecting a method of classification.
legend difficult to understand. The equal-intervals An analysis of Figure 5 reveals that there is no single
method receives a very good rating on this criterion best method of classification. Although the optimal
because not only are there no gaps, but the rounded approach is often touted as the best method, it is best
intervals can be very easy to understand (e.g., 0–25, only in terms of grouping like values together (as a
26–50, etc.).You might wish to avoid the problem of gaps function of their position along a number line) and in
by creating continuous legends for all classification selecting an appropriate number of classes. Clearly,
methods (as in Figure 3). Remember, though, that this there are several other criteria for which it is not the
approach will not indicate the actual range of values best. Ultimately, you must consider the purpose of the
falling in a class (the latter is dealt with in the criterion map and the knowledge of the intended audience
“legend values match range of data in a class”). before selecting a classification method.
16:05:32.
Data Classification
A good illustration of the role of map purpose and between 6 and 9 and between 13 and 16. The GADF
intended audience is the effort of Cynthia Brewer and for this map is 0.76, and the complexity using the CF
Trudy Suchan to create an atlas of the 2000 U.S. Census face measure is 10/16 or 0.63.
data (Brewer 2001). Rather than use one of the stan- A spatial constraint can be applied by starting with
dard classification methods presented in this chapter, the optimal solution and then allowing data values to
Brewer and Suchan used meaningful breaks (e.g., a per- shift between classes so that the map pattern is simpli-
centage figure for the entire United States), rounded fied. One potential solution is shown in the right-hand
breaks, and breaks that were identical across a set of map, where the values 9 and 16 have been shifted from
maps. Brewer (2001, 225) argued that the resulting map the second and third classes to the first and second
set is much more useful than a “map-by-map optimiza- classes, respectively. The result is a slightly lower GADF
tion approach.” (0.70 as opposed to 0.76), but a considerably simpler
complexity (0.38 as opposed to 0.63).
2 USING SPATIAL CONTEXT TO SIMPLIFY Franky Declercq (1995) developed an algorithm
CHOROPLETH MAPS that will find the least complex map for a particular
number of classes, assuming that one is willing to shift
a certain percentage of observations between classes.
One limitation of the classification methods described
Obviously, one problem with this approach is that there
in the preceding section is that they do not consider the
is a trade-off between complexity and accuracy: As
spatial context of the data. If the purpose of classifica-
complexity decreases, accuracy will also decrease.
tion is to simplify the appearance of the map, it can be
Consequently, there is some subjectivity in determin-
argued that spatial context also should be considered.
ing what magnitude of GADF is acceptable for a
This section considers two approaches that do consider
particular application.
spatial context: one that begins with the optimal classi-
fication approach but incorporates a spatial constraint,
2.2 Spatial Context Only
and one that works solely with spatial context.
Another approach for simplifying map pattern is to
2.1 Optimal Classification with a Spatial Constraint adjust the values of enumeration units as a function of
values of surrounding units, an idea developed by Waldo
To illustrate how a spatial constraint can be combined Tobler (1973) and promoted by Adrian Herzog (1989).
with optimal classification, consider Figure 6. The top This notion is based on the concept of random error: If
portion of the figure lists raw data for 16 hypothetical we repeatedly measure a value for an enumeration unit,
enumeration units, and the bottom portion portrays we will likely get a different value each time (think of
two maps of these data. The map on the left is an opti- trying to calculate the population for a census tract for
mal classification; note that breaks for this map occur the 2000 Census—numerous variables could affect your
result). This suggests that some change in the data we
have collected is permissible (Clark and Hosking 1986,
Raw Data 14). Moreover, the spatially autocorrelated nature of
geographic data provides a mechanism for making
1, 2, 3, 4, 5, 6, 9, 10, 11, 12, 13, 16, 17, 18, 19, 20 adjustments: Nearby units can assist in determining
appropriate values for a particular enumeration unit
because similar values are likely to be located near one
Optimal Map another.
Optimal Map (with a spatial constraint)
Although this approach could be used with a wide
1 9 5 6 1 9 5 6 variety of data, it is easiest to implement for propor-
tion data (e.g., the proportion of adults who smoke cig-
20 3 16 12 20 3 16 12 arettes) because the statistical theory for such data is
well known. To illustrate, consider the hypothetical
19 4 13 11 19 4 13 11
portion of a map shown in Figure 7, where it is
18 10 17 2 18 10 17 2 assumed we wish to change the value for the central
enumeration unit as a function of the surrounding
GADF = 0.76 GADF = 0.70 units. The simplest formula for determining the value
CF = 10/16 = 0.63 CF = 6/16 = 0.38 of the central unit would be to average all four values,
but such a formula would not consider two factors:
(1) that we have presumably taken some care in col-
FIGURE 6 Applying a spatial constraint to the optimal clas- lecting the data for the central unit (and thus would
sification approach. like to place greater weight on it), and (2) that those
77
16:05:32.
Data Classification
outside of a 95 percent confidence interval around its

original value (see Burt and Barber 1996, 272–274, for
a discussion of confidence intervals for proportions).
Second, surrounding values differing dramatically from
the central value should not be used in the calculations.
0.22
For proportion data, Herzog suggested a difference of
0.26 proportions test (see Burt and Barber 1996, 322–324),
with proportions differing significantly from the cen-
tral unit not being used. Third, the process should be
0.15 implemented in an iterative fashion, meaning that the
entire map should be simplified several times; for
example, this process could continue until all enumera-
tion units reached the bounds of their confidence inter-
vals or until the largest change made in a unit was less
0.19 than a small tolerance. Finally, the map resulting from
the iterative process should be further smoothed by
combining areas with nearly equal values.
Figure 8 illustrates the net effect of using Herzog’s
approach for the percentage of students in each com-
mune of Switzerland who study at the University of
FIGURE 7 A hypothetical set of enumeration units for which a
Zurich: The top map is an original (unsimplified)
simplified value is desired for the central enumeration unit.
unclassed map, and the bottom map is the simplified
result. Although the bottom map is still unclassed, note
that it is much simpler to interpret than the top map.
units having a longer common boundary with the cen- It should be noted that the methods we have exam-
tral unit should have greater impact. Thus, a more ined for simplifying map appearance are relatively new
appropriate formula is and have not yet been incorporated into commercial
n software. Assuming that such methods do become read-
Ve = Wc Vc + Ws a a Vi b
Li
ily available, it will be interesting to see to what extent
i=1 LT
they are used, and what impact their use has on the use
where Ve = the estimate value for the central unit of more traditional classification methods that do not
Vc = the original value for the central unit consider spatial context.
Vi = the original value for the ith
surrounding unit
Wc = the weight for the central unit 3 USING MULTIPLE CRITERIA TO DETERMINE CLASS
Ws = the weight for the surrounding units INTERVALS
Li = the length of the boundary between
the ith unit and the central unit In section 2.1, we showed how we might combine the
LT = the total length of the central optimal classification approach with a spatial con-
unit boundary straint. We utilized two criteria to accomplish this: (1)
n = the number of surrounding units minimizing the difference between data values in
classes (i.e., minimizing the tabular error, a nonspatial
If we assume Wc and Ws have values of 0.67 and 0.33, constraint), and (2) simplifying the map pattern by
respectively, then for Figure 7, the formula would be defining class limits such that contiguous enumeration
calculated as follows: units fall in the same class (a spatial constraint). Marc
Armstrong and his colleagues (2003) have argued that
Ve = .67(.15) + .33 a (.26) +
1
6 we should consider a broader set of criteria when clas-
sifying data, including minimizing the tabular error,
(.22) + (.19) b = .17
2 3
6 6 minimizing the difference between classes for contigu-
ous enumeration units (i.e., minimizing the boundary
Herzog indicated that a more complete algorithm error), maximizing the overall spatial autocorrelation
should involve several other considerations. First, some in the resulting classed map, and equalizing the area in
limit should be placed on how much change is permit- each class (which is often useful when we wish to com-
ted to the central unit. For proportion data, Herzog pare choropleth maps). Note that the latter three of
suggested that the central value should not be moved these are all spatial constraints.
16:05:32.
Data Classification
Percentage of Students Studying

at the University of Zurich
University of Zurich
Percentage of Students Studying

at the University of Zurich
University of Zurich
FIGURE 8 A comparison of an original unclassed map (top) and a simplified map (bottom) resulting from applying
Herzog’s method for simplifying choropleth maps. (An adaptation of Figure 1 from p. 214 of Accuracy of Spatial Databases,
M. Goodchild and S. Gopal (eds.), 1989; courtesy of Taylor & Francis.)
79
16:05:32.
Data Classification
Armstrong et al. considered these multiple criteria by criterion we choose to focus on. If you compare Figures
using the technique of genetic algorithms (Michalewicz 2 and 9, you will note that the “Minimize Tabular Error”
1998), which “do not guarantee a global optimum solu- map matches exactly the “Optimal (median)” map. This
tion, [but] normally yield very good solutions to large should not be surprising because, as we indicated in
problems” (Armstrong et al., 602). Since a full under- section 1.6, minimizing the tabular error is equivalent to
standing of genetic algorithms is beyond the scope of the producing an optimal classification. Also note that the
present text, we will focus on the results Armstrong et al. “Equalize Area in Each Class” map is similar in appear-
obtained when utilizing genetic algorithms. Figure 9 illus- ance to the “Quantiles” map. This also should not be sur-
trates the four maps for the Florida foreign-born data prising, either, because we indicated that if enumeration
that result if we consider only one of the four criteria at units are approximately the same size, each class on a
a time. As with Figure 2, we see that the appearance of quantiles map will have approximately the same map
the distribution obviously differs as a function of which area. In contrast to these similarities, note that the
Armstrong et al.’s Multiple Criteria for Data Classification

A B
Minimize Tabular Error Minimize Boundary Error
0.6 to 2.4 0.6 to 3.6
2.7 to 4.9 3.7 to 7.6
5.2 to 8.3 8.3 to 12.2
10.1 to 15.8 14.6 to 15.8
45.1 45.1
C D
Maximize Spatial Autocorrelation Equalize Area in Each Class
0.6 to 0.9 0.6 to 1.6
1.0 to 3.6 1.7 to 3.5
3.7 to 10.1 3.6 to 5.4
10.5 to 12.2 5.5 to 7.5
14.6 to 45.1 7.6 to 45.1
FIGURE 9 Maps that are ideal from the standpoint of the data classification criteria considered by Armstrong et al. (2003): (A)
minimizes the tabular error, (B) minimizes the boundary error, (C) maximizes the spatial autocorrelation, and (D) equalizes the
area in each class.
16:05:32.
Data Classification
“Maximize Spatial Autocorrelation” and “Minimize are labeled EVF and GEA. EVF stands for “equality of
Boundary Error” maps provide a different view of the variance fit” and is a measure of tabular error (in this
data than the standard methods of classification shown case, minimizing the sum of squared deviations about
in Figure 2. class means); low EVF values are desirable. The G in
Rather than focusing on only one criterion, the pro- GEA stands for the “Gini coefficient” and EA stands for
gram ChoroWare (Xiao and Armstrong 2006) can be uti- “equal areas.” The Gini coefficient is an index of dissim-
lized to trade off one criterion for another. For instance, ilarity (Hammond and McCullagh 1978; Smith 1977) that
Figure 10 shows two maps that represent trade-offs measures the difference between the areas in each class
between minimizing tabular error and equalizing the for a given map with a theoretical best map in which the
area in each class. Also shown in this figure is a scatter- area in each class is identical. Low values on GEA are
plot of potential solutions that arise when considering also desirable because a low value on the Gini coefficient
these two criteria. Note that the axes on the scatterplot tends to equalize the area in each class. Also note that
Trade-Offs Using Armstrong et al.’s Multiple Criteria

A B
Trade-Off #1 (TO1) Trade-Off #2 (TO2)
0.6 to 2.3 0.6 to 1.7
2.4 to 4.3 1.8 to 3.7
4.5 to 6.8 4.1 to 6.0
7.1 to 15.8 6.1 to 10.5
45.1 12.2 to 45.1
Best
EVF
TO1
Best
GEA
TO2
FIGURE 10 Trade-offs using Armstrong et al.’s (2003) multiple criteria approach: (A) the EVF (the tabular error) is close to the pos-
sible minimum and the GEA is also reasonably low (classes are close to being equal in area); (B) the GEA is very low and EVF is
moderate; (C) a scatterplot of potential values of EVF and GEA determined using the genetic algorithm approach.
81
16:05:32.
Data Classification
the Best EVF and Best GEA values (corresponding to One limitation of traditional classification methods
maps A and D, respectively, in Figure 9) fall to the (e.g., equal intervals and optimal) is that they fail to con-
extreme left and bottom of the scatterplot, reflecting the sider the spatial context of the data. If the purpose of
fact that these values are minimized. In contrast, the two classification is to simplify the appearance of a map, it
trade-off maps (TO1 and TO2) are not minimized on can be argued that spatial context should also be consid-
either criterion, but represent a balance between the cri- ered. In this chapter, we initially looked at two
teria. For instance, TO1 has a value that is very close to approaches that consider spatial context when simplify-
the minimum EVF value, but is also a reasonably low ing a choropleth map. The first used the optimal classifi-
GEA value. In contrast, TO2 has a very low GEA value cation method but incorporated a spatial constraint by
and a moderate EVF value. Selection of one of these two requiring that values falling in the same class be contigu-
solutions would be a function of the relative weight the ous, whereas the second employed no classification
user wished to specify for the two criteria. approach, but rather smoothed the raw data by changing
values for enumeration units as a function of values of
SUMMARY neighboring units. We then examined an approach based
on genetic algorithms that allowed us to consider a broad
range of criteria when selecting class intervals, including
In this chapter, we initially examined several common
the tabular error, the boundary error, the spatial auto-
methods of data classification that combine raw data
correlation of the data, and a desire to equalize the area
values into groups (or classes). Methods that do con-
in each class. We saw that optimizing one criterion (say,
sider how data are distributed along the number line
tabular error) would not necessarily optimize another
(e.g., natural breaks and optimal) are desirable
(say, boundary error), but that the program ChoroWare
because they place similar data values in the same
could be used to trade off one criterion for another.
class (and dissimilar data values in different classes).
In the simplest view, data classification involves
Methods that do not consider the distribution of data
grouping observations that have similar scores on a sin-
along the number line (e.g., equal intervals and
gle numeric attribute such as median income in
quantiles) might, however, also be desirable because
California counties. It is also possible, however, to con-
they satisfy other criteria. For example, equal intervals
sider classifying observations on the basis of multiple
is desirable because it is easy to understand, easy to
attributes; for example, we might ask which counties
compute, and has an easily understood legend (at least
have similar incomes and similar voting behavior. Such
when rounded percentage data are used, such as 0–100
a process is termed cluster analysis.
percent urban population).
FURTHER READING
Armstrong, M. P., Xiao, N., and Bennett, D. A. (2003) “Using Cromley, R. G. (1996) “A comparison of optimal classification
genetic algorithms to create multicriteria class intervals for strategies for choroplethic displays of spatially aggregated
choropleth maps.” Annals of the Association of American data.” International Journal of Geographical Information
Geographers 93, no. 3:595–623. Systems 10, no. 4:405–424.
Uses a variety of data sets to illustrate the notion of considering Illustrates a variety of criteria for optimally classifying data by
multiple criteria in data classification. treating classification as an integer programming problem; also
Brewer, C. (2001) “Reflections on mapping Census 2000.” see Cromley and Mrozinski (1997; 1999).
Cartography and Geographic Information Science 28, no. Evans, I. S. (1977) “The selection of class intervals.”
4:213–235. Transactions, Institute of British Geographers (New Series) 2,
Describes how meaningful breaks, rounded breaks, and breaks
no. 1:98–124.
that were identical across a set of maps were used to design an A classic article on methods of data classification.
atlas of 2000 U.S. Census data.
Herzog, A. (1989) “Modeling reliability on statistical surfaces
Coulson, M. R. C. (1987) “In the matter of class intervals for by polygon filtering.” In Accuracy of Spatial Databases, ed. by
choropleth maps: With particular reference to the work of M. Goodchild and S. Gopal, pp. 209–218. London: Taylor &
George F. Jenks.” Cartographica 24, no. 2:16–39. Francis.
Summarizes data classification methods, with an emphasis on A detailed discussion of Herzog’s method for simplifying patterns
Jenks’s work. on choropleth maps.
16:05:32.
Data Classification
Jenks, G. F., and Caspall, F. C. (1971) “Error on choroplethic Monmonier, M. S. (1982) “Flat laxity, optimization, and
maps: Definition, measurement, reduction.” Annals, Associ- rounding in the selection of class intervals,” Cartographica
ation of American Geographers 61, no. 2:217–244. 19, no. 1:16–27.
The classic article on optimal data classification. Argues that round-number breaks should be used with optimal
data classification to create a more readable map.
Lindberg, M. B. (1990) “Fisher: A Turbo Pascal unit for optimal
partitions.” Computers & Geosciences 16, no. 5:717–732.
Describes a computer implementation of the Fisher–Jenks optimal
method of data classification.
GLOSSARY
boundary error: if classed data are conceived as a prism maximum breaks: a method of data classification in which
map, boundary error describes how close the resulting cliffs the largest differences between ordered observations are used
come to matching the cliffs on an unclassed prism map of the to define classes.
data. mean–standard deviation: a method of data classification in
class interval: in creating an equal-interval map, the width which the mean and standard deviation of the data are used
that each class occupies along the number line. to define classes.
classed map: a map in which data are grouped into classes of natural breaks: a method of data classification in which a
similar value, and the same symbol is assigned to all members graphical plot of the data (e.g., a histogram) is examined to
of each class (e.g., a data set with 100 different data values determine natural groupings of data.
might be depicted using only five shades of gray). optimal: a method of classification in which like values are
cluster analysis: a mathematical method for grouping obser- placed in the same class by minimizing an objective measure
vations (say, counties) based on their scores on a set of attrib- of classification error (e.g., by minimizing the sum of absolute
utes. deviations about class medians).
equal-intervals: a method of data classification in which overview error: if classed data are conceived of as a prism
each class occupies an equal portion of the number line. map, overview error is the difference in volume between this
Fisher–Jenks algorithm: a method for classifying data in map and an unclassed prism map of the data.
which an optimal classification is guaranteed by essentially quantiles: a method of data classification in which an equal
considering all possible classifications of the data. number of observations is placed in each class.
goodness of absolute deviation fit (GADF): a measure of random error: the notion that if we repeatedly measure a
the accuracy of a classed choropleth map when the median is value for an enumeration unit, we will likely get a different
used as a measure of central tendency. value each time.
goodness of variance fit (GVF): a measure of the accuracy tabular error: if classed data are conceived as a prism map,
of a classed choropleth map when the mean is used as a mea- tabular error is the difference in height between prisms on
sure of central tendency. this map and those on an unclassed prism map of the data.
Jenks–Caspall algorithm: an optimal data classification that unclassed map: a map in which data are not grouped into
is achieved by moving observations between classes through classes of similar value, and thus each data value can theoret-
trial-and-error processes known as reiterative and forced ically be represented by a different symbol (e.g., a data set
cycling. with 100 different data values might be depicted using 100 dif-
ferent shades of gray on a choropleth map).
REFERENCES
Armstrong, M. P., Xiao, N., and Bennett, D. A. (2003) “Using Cromley, R. G., and Mrozinski, R. D. (1997) “An evaluation
genetic algorithms to create multicriteria class intervals for of classification schemes based on the statistical versus the
choropleth maps.” Annals of the Association of American spatial structure properties of geographic distributions in
Geographers 93, no. 3:595–623. choropleth mapping.” 1997 ACSM/ASPRS Annual
Brewer, C. A. (2001) “Reflections on mapping Census 2000.” Convention & Exposition, Technical Papers, Volume 5
Cartography and Geographic Information Science 28, no. (Auto-Carto 13), Seattle, WA, pp. 76–85.
4:213–235. Cromley, R. G., and Mrozinski, R. D. (1999) “The classifica-
Burt, J. E., and Barber, G. M. (1996) Elementary Statistics for tion of ordinal data for choropleth mapping.” The
Geographers (2nd ed.). New York: Guilford. Cartographic Journal 36, no. 2:101–109.
Clark, W. A. V., and Hosking, P. L. (1986) Statistical Methods Declercq, F. A. N. (1995) “Choropleth map accuracy and the
for Geographers. New York: Wiley. number of class intervals.” Proceedings of the 17th
83
16:05:32.
Data Classification
International Cartographic Conference, Volume 1, Jenks, G. F., and Caspall, F. C. (1971) “Error on choroplethic
Barcelona, Spain, pp. 918–922. maps: Definition, measurement, reduction.” Annals of the
Fisher, W. D. (1958) “On grouping for maximum homogene- Association of American Geographers 61, no. 2:217–244.
ity.” Journal of the American Statistical Association 53, Michalewicz, Z. (1998) Genetic Algorithms + Data Structures
December:789–798. = Evolution Programs (3rd ed.). Berlin: Springer-Verlag.
Hammond, R., and McCullagh, P. (1978) Quantitative Smith, D. M. (1977) Human Geography: A Welfare Approach.
Techniques in Geography: An Introduction (2nd ed.). New York: St. Martin’s Press.
Oxford, England: Clarendon. Tobler, W. R. (1973) “Choropleth maps without class inter-
Hartigan, J. A. (1975) Clustering Algorithms. New York: vals?” Geographical Analysis 5, no. 3:262–265.
Wiley. U.S. Bureau of the Census. (1994) County and City Data
Herzog, A. (1989) “Modeling reliability on statistical surfaces Book: 1994. Washington, DC: U.S. Government Printing
by polygon filtering.” In Accuracy of Spatial Databases, ed. Office.
by M. Goodchild and S. Gopal, pp. 209–218. London: Xiao, N., and Armstrong, M. P. (2006) “ChoroWare: A
Taylor & Francis. software toolkit for choropleth map classification.”
Jenks, G. F. (1977) Optimal data classification for choropleth Geographical Analysis 38, no. 1:102–120.
maps. Occasional Paper No. 2. Lawrence, KS: Department
of Geography, University of Kansas.
16:05:32.
Principles of Symbolization
85
16:18:26.
OVERVIEW
The purpose of this chapter is to cover basic principles of the soil type of a region as an entisol, as opposed to a
symbolizing geographic phenomena. An overarching mollisol; such a categorization of soils would be termed
goal is to assist you in selecting among four common the- a nominal level of measurement. We consider four basic
matic mapping techniques: choropleth, proportional levels of measurement: nominal, ordinal, interval, and
symbol, isopleth, and dot. For example, imagine that you ratio. The latter two levels are commonly combined into
wish to map the spatial pattern of income in Washington, numerical data.
DC, and that you have collected data on the annual The term visual variables is commonly used to
income of all families in each census tract of the city. You describe the various perceived differences in map sym-
might wonder which of these four techniques would be bols that are used to represent spatial data. For example,
appropriate. Determining the appropriate technique will the visual variable spacing involves varying the distance
require that you first consider (1) the nature of geographic between evenly spaced marks (e.g., horizontal lines).
phenomena, (2) the various levels at which we can mea- Section 3 covers a host of visual variables, including
sure geographic phenomena, and (3) the types of symbols spacing, size, perspective height, orientation, shape,
that can be used to represent spatial data. arrangement, hue, lightness, and saturation.
Section 1 discusses the nature of geographic phenom- Section 4 introduces four common thematic mapping
ena. One way to think about geographic phenomena is to techniques (choropleth, proportional symbol, isopleth,
consider a phenomenon’s extent or spatial dimension— and dot) and considers how a mapmaker selects among
whether a phenomenon can be conceived of as points, them. We will see that the selection is a function of both
lines, areas, or volumes. For example, water well sites in a the nature of the underlying phenomenon and the pur-
rural area constitute a point phenomenon, whereas a city pose for making the map. Section 4 also introduces the
boundary is representative of a linear phenomenon. notion of data standardization to account for the area
Another approach is to contrast discrete and continuous over which data are collected; here we will consider the
phenomena. Discrete phenomena occur at isolated point most direct form of standardization, which involves
locations, whereas continuous phenomena occur every- dividing raw totals by the areas of enumeration units
where. For example, water towers in a city would be dis- (e.g., dividing acres of wheat for each county by the area
crete, but the distribution of solar insolation during the of each county).
month of January is continuous. Discrete and continuous Section 5 considers the issue of selecting an appropri-
phenomena can also be classified as smooth or abrupt. ate visual variable for choropleth mapping, which has
For instance, rainfall and sales tax rates for states are both traditionally been the most common thematic mapping
continuous in nature, but the former is smooth, whereas the method. Selecting an appropriate visual variable requires
latter is abrupt (varying at state boundaries). creating a logical match between the level of measure-
Section 2 considers levels of measurement, which ment of the data and the visual variable (e.g., if data are
refers to the various ways of measuring a phenomenon numerical, the visual variable should appear to reflect
when a data set is created. For instance, we might specify the numerical character of the data).
16:18:26.
1 NATURE OF GEOGRAPHIC PHENOMENA In contrast, true 3-D phenomena are multivalued

because each x and y location can have multiple values
1.1 Spatial Dimension associated with it. With true 3-D phenomena, any point
on the surface is specified by four values: an x coordi-
One way to think about geographic phenomena is to nate, a y coordinate, a z coordinate (which is the height
consider their extent or spatial dimension. For our pur- above, or depth below, sea level), and the value of the
poses, we consider five types of phenomena with phenomenon. Consider mapping the concentration of
respect to spatial dimension: point, linear, areal, 21/2-D carbon dioxide (CO2) in the atmosphere. At any point
and true 3-D. in the atmosphere, it is possible to define longitude, lat-
Point phenomena are assumed to have no spatial itude, height above sea level, and an associated level of
extent and are thus termed “zero-dimensional.” CO2. Color Plate 1 illustrates another true 3-D phenom-
Examples include weather station recording devices, oil enon: geologic material underneath the earth’s surface.
wells, and locations of nesting sites for eagles. Locations Although ideally we should distinguish true 3-D phe-
for point phenomena can be specified in either two- or nomena from 21/2-D phenomena, in this text we some-
three-dimensional space; for example, places of reli- times refer to a map of either as a 3-D map.
gious worship are defined by x and y coordinate pairs It is important to realize that map scale plays a major
(longitude and latitude), whereas nesting sites for role in determining how we handle the spatial dimension
eagles are defined by x, y, and z coordinates (the z coor- of a phenomenon. For example, on a small-scale map
dinate would be the height above the earth’s surface). (e.g., a page-sized map of France), places of religious
Linear phenomena are one-dimensional in spatial worship occur at points, but on a large-scale map (e.g., a
extent, having length, but essentially no width. map of a local neighborhood), individual buildings would
Examples include a boundary between countries and be apparent, and thus the focus might be on the area cov-
the path of a stunt plane during an air show. Locations ered by the place of worship. Similarly, a river could be
of linear phenomena are defined as an unclosed series considered a linear phenomenon on a small-scale map,
of x and y coordinates (in two-dimensional space), or an but on a large-scale map, the emphasis could be on the
unclosed series of x, y, and z coordinates (in three- area covered by the river.
dimensional space).
Areal phenomena are two-dimensional in spatial 1.2 Models of Geographic Phenomena
extent, having both length and width. An example
would be a lake (assuming that we focus on its two- The notion of spatial dimension is just one way of think-
dimensional surface extent). Data associated with polit- ing about how geographic phenomena are arranged in
ical units (e.g., counties) can also fit this framework, the real world. Another approach is to consider the
because the location of each county can be specified as arrangement of geographic phenomena along discrete–
an enclosed region. In two-dimensional space, areal continuous and abrupt–smooth continua. In this section
phenomena are defined by a series of x and y coordi- we define the terms associated with these continua and
nates that completely enclose a region (computer sys- show how they provide a useful set of models of geo-
tems generally require that the first coordinate pair graphic phenomena, a notion developed by Alan
equal the last). MacEachren and David DiBiase (1991).
When we move into the realm of volumetric phe- The terms “discrete” and “continuous” are often
nomena, it is convenient to consider two types: 21/2-D used in statistics courses to describe different types
and true 3-D. The first of these, 2 1/2 -D phenomena, can of data along a number line; here we consider their
be thought of as a surface, in which geographic location use by cartographers in a spatial context. Discrete
is defined by x and y coordinate pairs and the value of phenomena are presumed to occur at distinct locations
the phenomenon is the height above a zero point (or (with space in between). Individual people living in a
depth below a zero point). Probably the easiest exam- city would be an example of a discrete phenomenon; for
ple to understand is elevation above sea level, because an instant in time, a location can be specified for each
we can actually see the surface in the real world; here, person, with space between individuals. Continuous
height above a zero point is the elevation of the land phenomena occur throughout a geographic region of
surface above sea level. A more abstract example would interest. The examples presented previously for 21/2-D
be precipitation falling over a region during the course phenomena would also be considered continuous phe-
of a year; in this case the height of the surface would be nomena. For instance, when considering elevation,
the total amount of precipitation for the year. every longitude and latitude position has a value above
Another way of thinking about 21/2-D surfaces is that or below sea level.
they are single-valued in the sense that each x and y Discrete and continuous phenomena can also be
coordinate location has a single value associated with it. described as either abrupt or smooth. Abrupt phenomena
87
16:18:26.
change suddenly, whereas smooth phenomena change the abruptness–smoothness continuum. In some cases,
in a gradual fashion. This concept is most easily under- average income would exhibit the abrupt changes of
stood for continuous phenomena. The number of elec- percent sales tax (as at the boundary between urban
toral votes for each state in the United States would neighborhoods), while in others it would exhibit a more
be considered an abrupt continuous phenomenon gradual change (as one moves up a hill toward a region of
because although each enumeration unit (a state) has a more attractive views, average income should increase).
value, there are abrupt changes at the boundaries In contrast to the bottom row, the top row of
between states. In contrast, the distribution of total pre- Figure 1A represents a range of discrete phenomena.
cipitation over the course of a year for a humid region The number of employees located at a county courthouse
would be a smooth continuous phenomenon because we is clearly an abrupt discrete phenomenon, as there can
would not expect such a distribution to exhibit abrupt be only one value for a county and it occurs at an
discontinuities. isolated location. In contrast, the number of employed
Figure 1A provides a graphic portrayal of a variety of people (based on where they live, as opposed to where
models of geographic phenomena that result when they work) is a smooth discrete phenomenon, because it
we combine the discrete–continuous and abrupt–smooth gradually changes over geographic space. The number of
continua. We’ll discuss the models in detail because government employees (again, based on where they
considering the nature of geographic phenomena is live) falls somewhere between these phenomena; it
extremely important in selecting an appropriate method might exhibit an abrupt characteristic in the sense that
of symbolization. First, consider the continuous phenom- government employees might live near government
ena shown in the bottom row of Figure 1A. Percent sales offices, but it will not exhibit the extreme abruptness of
tax is an obvious abrupt continuous phenomenon, as it the courthouse example.
changes suddenly at the boundary between enumeration The middle row of Figure 1A represents phenomena
units (e.g., one state’s sales tax is different from another that can be classified as not clearly continuous or dis-
state’s). In contrast, average farm size is an example of a crete, and that also span the abruptness–smoothness
smooth, continuous phenomenon because we would continuum. This row is probably most easily understood
expect it to vary in a relatively gradual fashion (as the cli- by considering the influenza case first. Because
mate becomes drier, we would expect the average farm influenza is an infectious disease, it should exhibit a
size to increase). Average income falls somewhere smooth character. Although individual influenza cases
between percent sales tax and average farm size on could be represented at discrete locations, it makes
A B
Abrupt Smooth Abrupt Smooth
Discrete
Discrete
# of employees at # of government # of employed

county courthouse employees people
Proportional Symbol Dot
# of telephone # of influenza
Continuous
subscribers # of AIDS cases cases

Continuous
percent sales tax average income average farm size

Choropleth Isopleth
FIGURE 1 (A) Models of geographic phenomena arranged along discrete–continuous and abrupt–smooth continua. (B) A set of
symbolization methods appropriate for these models. (After MacEachren 1992, 16; courtesy of North American Cartographic
Information Society and Alan MacEachren.)
16:18:26.
5 Principles of Symbolization 79
sense to suggest some degree of continuity if we wish to that have been collected and the phenomenon that is
stress the potential of infection. At the other end of the being mapped. Which type of map is used will be a func-
row is the number of subscribers to a particular tele- tion of both the nature of the underlying phenomenon
phone company. Competition between telephone com- and the purpose of the map. We consider this issue in
panies could lead to a distribution that changes abruptly greater depth in section 4.
but that exhibits continuity between the lines of abrupt
change. Finally, the number of people with AIDS is in
the middle of the diagram. AIDS occupies a more 2 LEVELS OF MEASUREMENT
abrupt position than influenza because of its mode of
transmission (sexual intercourse, sharing of needles, and When a geographic phenomenon is measured to create
blood transfusions). a data set, we commonly speak of the level of measure-
Figure 1B is a set of symbolization methods that ment associated with the resulting data. Conventionally,
MacEachren and DiBiase argued would be appropriate four levels of measurement are recognized—nominal,
for the models shown in 1A. Note that we have labeled ordinal, interval, and ratio—with each subsequent level
the four corner maps (proportional symbol, dot, choro- including all characteristics of the preceding levels.
pleth, and isopleth), which will be discussed in depth in The nominal level of measurement involves grouping
section 4. (or categorization), but no ordering. The classic exam-
ple is religion, in which individuals might be identified
1.3 Phenomena versus Data as Catholic, Protestant, Jewish, or Other; here, each
religious group is different, but one is not more or less
When mapping geographic phenomena, it is important religious in value than another. Another example would
to distinguish between the actual phenomenon and the be classes on a land use/land cover map; for example,
data collected to represent that phenomenon. For grassland, forest, urban, water, and cropland differ from
example, imagine that we wish to map the percentage one another, but one class is not more or less in value
of forest cover in South Carolina. If we try to visualize than another.
the phenomenon, we can conceive of it as smooth and The second level of measurement, ordinal, involves
continuous in some portions of the state where the per- categorization plus an ordering (or ranking) of the data.
centage gradually increases or decreases. In other areas, For example, a geologist asked to specify the likelihood
we can conceive of relatively abrupt changes where the of finding oil at each of 50 well sites might be unwilling
percentage shifts very rapidly (when, say, an urban area to provide numerical data, but would feel comfortable
is bounded by a hilly, forested region). specifying a low, moderate, or high potential at each site.
One form of data that we might use to represent per- Here three categories (low, moderate, and high) are
centage of forest cover would be individual values for provided, with a distinct ordering among them. Another
counties, which can be found in the state statistical example of ordinal data would be rankings resulting
abstract for South Carolina (South Carolina State Budget from a map comparison experiment. Imagine that you
and Control Board 1994, 45). We might consider mapping constructed dot maps for 10 different phenomena and
these data directly by creating the prism map shown in asked people to compare these maps with another dot
Figure 2B. Note that in this case, there are abrupt changes map (say, of population) and to rank the maps from
at the boundaries of each county. Such a map might be “most like” to “least like” the population map. The 10
appropriate if we wished to provide a typical value for maps ranked by each person would constitute a distinct
each county, but it obviously hides the variation within ordering, and thus represent ordinal data.
counties and misleads the reader into thinking that An interval level of measurement involves an
changes take place only at county boundaries. ordering of the data plus an explicit indication of the
A potentially better approach would be the smooth, numerical difference between two categories. Classic
continuous map (fishnet map) shown in Figure 2A; this examples are the Fahrenheit and Celsius tempera-
map indicates that the percentage of forest cover does ture scales. Consider temperatures of 20°F and 40°F
not coincide with county boundaries, but rather changes recorded in Fairbanks, Alaska, and Chattanooga,
in a gradual fashion. A still better map would be one that Tennessee, respectively. These two values are ordered,
shows some of the abrupt changes that are likely to and they reveal the precise numerical difference
occur. Creating such a map would require detailed infor- between the two cities. One characteristic of interval
mation about the location of forest within the state, as scales is the arbitrary nature of the zero point. In the
might be available from a remotely sensed image. Our case of the Celsius scale, 0 is the freezing point for
purpose at this point in the text is not to create the most pure water, whereas on the Fahrenheit scale, 0 is the
representative map of the phenomenon, but to stress that lowest temperature obtained by mixing salt and ice. A
the mapmaker must carefully distinguish between data result of an arbitrary zero point is that ratios of two
89
16:18:26.
80
60
40
C
B
45º
FIGURE 2 Approaches for mapping a data set of percentage of forest cover by county for the state of South Carolina: (A) The
data are treated as coming from a smooth, continuous phenomenon; (B) the data are treated as an areal phenomenon. (C) Map C
illustrates that maps A and B have been rotated 45° from a traditional north-oriented map. For A, values outside the state are
extrapolated, and thus must be treated with caution. (Data Source: South Carolina State Budget and Control Board 1994.)
interval values cannot be interpreted correctly; for A ratio level of measurement has all the characteris-
example, 40°F is numerically twice the value of 20°F, tics of the interval level, plus a nonarbitrary zero point.
but it is not twice as warm (in terms of the kinetic Continuing with the temperature example, the Kelvin
energy of the molecules). An example of an interval scale is ratio in nature because at 0°K all molecular
scale familiar to academics is SAT scores, which range motion ceases; thus, a temperature of 40°K is twice as
from a minimum of 200 to a maximum of 800. Note warm as 20°K (in terms of the kinetic energy of the
that it is not possible to say that an individual scoring molecules). Ratio data sets are more common than inter-
800 on an SAT exam did four times better than an val data sets. For example, a perusal of the maps shown in
individual scoring 200; all that can be said is that the this text will reveal that most are based on ratio-level
individual scored 600 points better. A geographical data. Because many symbol forms can be used with both
example of interval-level data is elevation, where interval and ratio scales, these two levels of measurement
the establishment of mean sea level represents an are often grouped together and referred to as numerical
arbitrary zero point. data. The basic scales that we have discussed can also be
16:18:26.
divided into qualitative (nominal data) and quantitative data and can be used intuitively to divide the data into
(ordinal, interval, and ratio data) scales. two parts. An example would be a value of 0 for percent-
The four levels of measurement we have considered age of population change, which would divide the data
are the only ones usually covered in geographic text- into positive and negative percent changes. A meaningful
books. Nicholas Chrisman (1998; 2002, 25–33) argues that dividing point does not occur inherently in the data, but
these are insufficient for working with geographic data, can logically divide the data into two parts. An example
and so has proposed several extensions. One is to create would be the mean of the data, which enables differenti-
a separate level of measurement for data sets that are ating values above and below the mean. Balanced data
constrained to a fixed set of numbers, such as probability are characterized by two phenomena that coexist in a
(the range is 0 to 1) or percentages (the range is 0 to 100). complementary fashion. An example is the percentages
Chrisman terms this an absolute level of measurement, as of English and French spoken in Canadian provinces—a
no transformations are possible that could retain the high percentage of English-speaking people implies a low
meaning of measurement. We instead use the term percentage of French-speaking people (the two are in
constrained ratio recommended by Forrest (1999b). “balance” with one another). Unipolar data have no
A second extension proposed by Chrisman is the natural dividing points and do not involve two comple-
cyclical level of measurement, which is appropriate for mentary phenomena. Per capita income associated with
phenomena that have a cyclical character. For instance, countries of Africa or states of the United States would
angular measurements have a cycle of 360°, with angle x be an example of unipolar data.
as far from 0° as 360° -x. This notion is not dealt with in
the linear unbounded number line associated with ratio
measurements. As another example, the seasons are cycli- 3 VISUAL VARIABLES
cal and can be specified with different starting points—we
could specify spring–summer–fall–winter–spring or The term visual variables is commonly used to describe
fall–winter–spring–summer–fall. A third extension is the the various perceived differences in map symbols that are
notion of counts, in which individual objects, such as used to represent geographic phenomena. The notion of
people, are counted.Although counts have a nonarbitrary visual variables was developed by the French cartogra-
zero point, Chrisman argues that they cannot be rescaled pher Jacques Bertin (1983) and subsequently modified by
as readily as ratio scales because we cannot conceive of a others, including McCleary (1983), Morrison (1984),
fraction of an object (e.g., half of a person). DiBiase et al. (1991), and MacEachren (1994a). Our
A final extension is the notion of fuzzy categories.* approach is similar to that of MacEachren, but differs pri-
Normally, we think of individual items as falling wholly marily in our inclusion of 21/2-D and true 3-D phenom-
within a particular nominal category; for instance, an ena and the use of the perspective-height visual variable.
individual is either a Protestant or not (in terms of church In this chapter, we consider only visual variables for
membership). In practice, category memberships are static maps. We will find that abstract sound variables
often fuzzy, as it might not be entirely clear whether an can be especially useful for visually impaired map users.
item is within a particular category. A good illustration of The visual variables for static maps are illustrated in
fuzzy categories is some remote sensing classification Figures 3 (for quantitative phenomena) and 4 (for qualita-
procedures, which produce a probability that each pixel tive phenomena) and Color Plates 2 and 3. Note that the
falls in a particular land use (e.g., a pixel might have an visual variables appear in the rows, and that the columns
85 percent probability of being wheat and a 15 percent represent the dimensions of spatial phenomena discussed
probability of being corn). Increasingly, race and ethnic- in the preceding section. In discussing the visual variables,
ity can be considered “fuzzy” in that individuals identify we sometimes need to distinguish between the overall
themselves as more than one race. In fact, the 2000 U.S. symbol and the marks making up the symbol. For exam-
Census, for the first time, allowed for multiple ethnicities. ple, note that the spacing visual variable shown for point
Another extension to the basic levels of measurement phenomena consists of circular symbols, and that each cir-
is the three kinds of numerical data proposed by J. Ronald cle is composed of parallel horizontal marks.
Eastman (1986): bipolar, balanced, and unipolar. Bipolar
data are characterized by either natural or meaningful 3.1 Visual Variables for Quantitative Phenomena
dividing points. A natural dividing point is inherent to the
Let’s first consider the visual variables for quantitative
phenomena—here the visual variables should reflect
* Chrisman terms this extension “graded membership in categories.” either an ordinal, interval, or ratio level of measurement.
91
16:18:26.
Visual Variables
for Quantitative Phenomena
Point Linear Areal 2½-D True 3-D
Spacing
Size
Perspective None
Height Possible
Red
Yellow Red
Color ra
O
Red
Ye
ow ng
Orange
ll
Orange
(Hue) Orange
e
Re
Yellow Yellow
Yellow
d
Red Orange Yellow
Dark Green
Dark
Light Green Green
Li
re Dark
G
gh
Gr en Green
Green
t
Green een Green

Da Green
Green
Light Light
rk
Green Green
Green
Gr n
Dark Green Green Light Green

Color
(Lightness)
Vivid Red
Vivid
Muted Red Red
M
Color Re Vivid
V ivid
ut
d Red
ed
Red R ed Red
(Saturation) Viv Red
Muted
Muted
id R
Red Red
Vivid Red Red
e
Muted Red
d
FIGURE 3 Visual variables for quantitative phenomena. For visual variables for color maps, also see Color Plate
2.
16:18:26.
Visual Variables
for Qualitative Phenomena
None
Orientation Recommended
None
Shape Recommended
None
Arrangement Recommended
Blue Green
Gr
Re
Color Green
Green
ee
d None
n
Red Red
(Hue) Recommended
Bl
Blue Blue
ue
Green Red
FIGURE 4 Visual variables for qualitative phenomena. For visual variables for color maps, also see Color
Plate 3.
Spacing the size of the entire areal unit could also be changed, as
The spacing visual variable involves changes in the dis- is done on cartograms.
tance between the marks making up the symbol—a
smaller distance between marks suggests a higher data Perspective Height
value (Figure 3). Cartographers traditionally have used Perspective height refers to a perspective 3-D view of a
the term texture to describe these changes (e.g., Castner phenomenon (Figure 3). It is interesting to consider
and Robinson 1969), but we use the term spacing some of the potential applications of this visual vari-
because texture has varied usages in the literature. able. In the case of point phenomena, oil production at
well locations might be represented by raised cylinders
Size above each well, with the cylinder height proportional
Cartographers have used size as a visual variable for to well production. For linear phenomena, total traffic
quantitative phenomena in two different ways. One has flow between two cities over some time period could be
been to change the size of the entire symbol, as is shown represented by a fencelike structure above each road-
for the point and linear phenomena (Figure 3). Another way, with the height of the “fence” proportional to
is to change the size of individual marks making up the traffic flow. For areal phenomena, each enumeration
symbol, as for the areal, 21/2-D and true 3-D phenomena. unit is raised to a height proportional to the data, just as
This inconsistency might be a bit confusing, but the term we did for the forest cover data in South Carolina in
size seems to reflect quantitatively the visual differences Figure 2; note that a similar approach is used for 21/2-D
that arise in each case. Note that for areal phenomena, phenomena. Perspective height cannot be used for true
93
16:18:26.
3-D phenomena because three dimensions are needed recommend using it for 21/2-D phenomena, which
to locate the phenomenon being mapped. are inherently numerical. Note that in Figure 4, the
shape visual variable is handled in a fashion similar to
Color Hue, Lightness, and Saturation orientation.
The visual variables hue, lightness, and saturation are
commonly recognized as basic components of color.* Arrangement
Hue is the dominant wavelength of light making up a Understanding the arrangement visual variable re-
color. In everyday life, hue is the parameter of color quires a careful examination of Figure 4. For areal and
most often used; for example, you might note that one true 3-D phenomena, note that arrangement refers to
person has on a red shirt and another a blue shirt. For how marks making up the symbol are distributed; marks
quantitative phenomena, we would want to select hues for some areas are part of a square arrangement, whereas
that suggest a quantitative difference; for instance, yel- marks for other areas appear to be randomly placed.
low, orange, and red would be appropriate hues because For linear phenomena, arrangement refers to splitting
orange is seen as a mixture of yellow and red. Note that lines into a series of dots and dashes, as might be found on
these hues are shown in words in Figure 3 and as sym- a map of political boundaries. Finally, for point phenom-
bols in Color Plate 2. ena, arrangement refers to changing the position of the
Lightness (or value) refers to how dark or light a white marker within the black symbol. Regardless of how
particular color is, while holding hue constant; for these differing arrangements are created, they seem to
instance, differing shades of a green hue might be suggest qualitative (nominal) differences rather than
used (Figure 3 and Color Plate 2). Lightness also can quantitative ones.
be shown as shades of gray (in the absence of what we
commonly would call color), as in Figure 3. Color Hue
Saturation (or chroma) can be thought of as a mixture We noted above that color is comprised of hue,
of gray and a pure hue. It is the intensity of a color; lightness, and saturation. Only color hue is appropri-
for instance, we might speak of different intensities of ate for qualitative phenomena, as lightness and
colorful shirts. In Figure 3 we have used the terms saturation differences suggest quantitative differences.
“muted red,” “red,” and “vivid red” to suggest the Furthermore, we must select color hues that imply
differing amounts of red found in each shade. Color qualitative differences. For instance, red, green, and
Plate 2 is especially useful in understanding saturation, blue would be appropriate choices for qualitative
as we can actually see the symbols that result as we mix phenomena because these colors are not associated
gray and red. with a sequence of low, medium, and high data
values, regardless of the order in which the colors are
3.2 Visual Variables for Qualitative Phenomena associated with the data.
Now let’s consider the visual variables for qualitative 3.3 Some Considerations in Working
phenomena, where visual variables should reflect only a with Visual Variables
nominal level of measurement.
You should bear in mind that we have illustrated only a
Orientation and Shape fraction of the many symbols that could be used to
As with the size visual variable, the character of the depict visual variables; for example, either circles or
orientation visual variable is a function of the kind of squares might be used to depict point phenomena for
spatial phenomena. For linear, areal, and true 3-D the size visual variable. A major group of symbols not
phenomena, orientation refers to the direction of indi- shown in the figures are pictographic symbols, which are
vidual marks making up the symbol. In contrast, for intended to look like the phenomenon being mapped
point phenomena, orientation refers to the direction (as opposed to geometric symbols such as circles). For
of the entire point symbol (Figure 4). (Marks of instance, Figure 5 illustrates the use of different-sized
differing direction could be applied to point symbols, beer mugs to represent the number of microbreweries
but the small size of point symbols often makes it and brewpubs in each U.S. state. Pictographic symbols
difficult to see the marks.) Because orientation is most are often used in maps intended for children.
appropriate for representing nominal data, we do not Also keep in mind that the visual variables can serve
as basic building blocks of more complex representa-
tions. For example, Figure 6 illustrates how the visual
* See Brewer (1994a) for a discussion of terminology associated with variables spacing and size might be combined.
color. MacEachren (1994a, 27) called the resulting symbol
16:18:26.
Microbreweries and Brewpubs

1996
FIGURE 7 Different “patterns” or “textures” that can be cre-

ated to portray nominal information. Note that these are not
readily described in terms of the visual variables shown in
Figure 3.
Location refers to the position of individual symbols. We

chose not to illustrate this visual variable because it is an
inherent part of mapping (e.g., each symbol shown for
point phenomena can be defined by the x and y coordi-
Number of Establishments nate values of its center). If location were illustrated, it
would be represented by constant symbols (identical
3 to 10 11 to 15 20 to 32 44 to 72 149 dots for point phenomena) that varied only in position.
FIGURE 5 Using a pictographic visual variable (beer mugs)

4 COMPARISON OF CHOROPLETH,
to represent the number of microbreweries and brewpubs
PROPORTIONAL SYMBOL, ISOPLETH,
in each U.S. state. (For similar data, see http://
AND DOT MAPPING
brewpubzone.com.)
In this section, we define and contrast four common

a form of texture. In our opinion, such an approach thematic mapping techniques: choropleth, proportional
produces a rather coarse-looking map, but it can high- symbol, isopleth, and dot. For illustrative purposes, we
light various aspects of a distribution. examine these techniques by mapping data for acres of
In the figures we have used, the term for each visual wheat harvested in counties of Kansas (Table 1).
variable appears to be a clear expression of the visual Selecting an appropriate technique is a function of both
differences that we see; for example, in the case of the the nature of the underlying phenomenon and the pur-
orientation visual variable for point phenomena pose for making the map. Here we consider a basic
(Figure 4), we see that one square is at a different introduction to these mapping techniques.
orientation than another. Moreover, if we wanted, we
could compute a mathematical expression of this dif-
4.1 Choropleth Mapping
ference (that one square is rotated 40° from a vertical,
whereas another is rotated 50°). Sometimes, describing
A choropleth map is commonly used to portray data col-
the visual difference between symbols is not so easy.
lected for enumeration units, such as counties or states.
For example, try describing the differences between the
To construct a choropleth map, data for enumeration
symbols shown in Figure 7. Such symbols are often
units are typically grouped into classes and a gray tone or
referred to as differing in “pattern” or “texture” and
color is assigned to each class. The choropleth map is
are frequently used to symbolize nominal data.
clearly appropriate when values of a phenomenon
It also should be noted that the visual variable loca-
change abruptly at enumeration unit boundaries, such as
tion was not explicitly depicted in the illustrations.
for state sales tax rates. Choropleth maps might also be
appropriate when you want the map reader to focus on
“typical” values for individual enumeration units, even
though the underlying phenomenon does not change
abruptly at enumeration unit boundaries. For example,
politicians and government officials might use this
approach when stressing how one county or state
compares with another. Although choropleth maps
FIGURE 6 Combining the visual variables spacing and size. are commonly used in this fashion, it is important to
(After MacEachren 1994a, p. 26.) recognize two major limitations: (1) Such maps do not
95
16:18:26.
TABLE 1 Wheat harvested in Kansas counties 1993

County Acres Harvested Area of County (in acres) % of Land in Wheat
Allen 35200 324735 10.8
Anderson 36600 373075 9.8
Atchison 12300 278937 4.4
Barber 150200 727904 20.6
Barton 199400 576390 34.6
Bourbon 14800 409210 3.6
Brown 26600 365894 7.3
Butler 80700 926707 8.7
Chase 15000 497369 3.0
Chautauqua 15500 412610 3.8
Cherokee 78000 377881 20.6
Cheyenne 147800 651910 22.7
Clark 93100 626576 14.9
Clay 99400 420435 23.6
Cloud 107700 461204 23.4
Coffey 31400 419001 7.5
Comanche 92000 506214 18.2
Cowley 125600 725741 17.3
Crawford 41100 380157 10.8
Decatur 133200 572289 23.3
Dickinson 153600 546094 28.1
Doniphan 9000 254242 3.5
Douglas 15700 304118 5.2
Edwards 108400 397693 27.3
Elk 11600 416442 2.8
Ellis 132800 575971 23.1
Ellsworth 97700 463260 21.1
Finney 224800 833658 27.0
Ford 226100 701190 32.2
Franklin 14500 369251 3.9
Geary 17700 259175 6.8
Gove 162400 685862 23.7
Graham 96400 575048 16.8
Grant 123300 368538 33.5
Gray 190700 555865 34.3
Greeley 175900 498307 35.3
Greenwood 11100 736333 1.5
Hamilton 160000 637046 25.1
Harper 281100 512292 54.9
Harvey 125000 345324 36.2
Haskell 126800 369004 34.4
Hodgeman 136900 551464 24.8
Jackson 17500 421717 4.1
Jefferson 8900 354557 2.5
Jewell 126300 585604 21.6
Johnson 11300 306957 3.7
Kearny 144100 557439 25.9
Kingman 216800 554529 39.1
Kiowa 116300 464185 25.1
Labette 66200 421057 15.7
Lane 155200 459018 33.8
Leavenworth 11400 303572 3.8
Lincoln 106000 460879 23.0
Linn 16000 387873 4.1
Logan 179800 688192 26.1
Lyon 16000 549098 2.9
Marion 142200 610675 23.3
Marshall 92500 578662 16.0
Mcpherson 226300 576316 39.3
Meade 121000 627167 19.3
Miami 10800 378094 2.9
Continued
16:18:26.
TABLE 1 Continued
County Acres Harvested Area of County (in acres) % of Land in Wheat
Mitchell 150000 459022 32.7
Montgomery 54000 413768 13.1
Morris 46000 449145 10.2
Morton 78100 466625 16.7
Nemaha 31000 459537 6.7
Neosho 45000 369610 12.2
Ness 171600 687891 24.9
Norton 139400 563491 24.7
Osage 19500 459872 4.2
Osborne 107400 572513 18.8
Ottawa 103400 460748 22.4
Pawnee 153400 482433 31.8
Phillips 103700 572563 18.1
Pottawatomie 17300 551123 3.1
Pratt 163100 469889 34.7
Rawlins 160100 686637 23.3
Reno 292500 813939 35.9
Republic 96600 459717 21.0
Rice 160400 466673 34.4
Riley 17800 397742 4.5
Rooks 71500 573454 12.5
Rush 154000 459719 33.5
Russell 98100 576143 17.0
Saline 121500 459756 26.4
Scott 170500 459183 37.1
Sedgwick 262300 646618 40.6
Seward 98400 410803 24.0
Shawnee 14500 354420 4.1
Sheridan 120300 573394 21.0
Sherman 185400 675725 27.4
Smith 126000 573619 22.0
Stafford 160200 508439 31.5
Stanton 134100 436392 30.7
Stevens 101600 465830 21.8
Sumner 443700 758976 58.5
Thomas 235400 687149 34.3
Trego 119500 575635 20.8
Wabaunsee 15000 511601 2.9
Wallace 123800 584778 21.2
Washington 88000 575534 15.3
Wichita 152300 460170 33.1
Wilson 52400 367770 14.2
Woodson 18800 322672 5.8
Wyandotte 700 99196 0.7
portray the variation that might actually occur within and U. Assume that regions S and T have equal-sized
enumeration units, and (2) the boundaries of enumera- enumeration units, each 16 acres in size. In contrast,
tion units are arbitrary, and thus unlikely to be associated assume that region U has enumeration units four times
with major discontinuities in the actual phenomenon.* the size of those in S and T, or 64 acres in size.
An important consideration in constructing choro- Let’s further assume that the number of acres of
pleth maps is the need for data standardization, in which wheat harvested from enumeration units in each
raw totals are adjusted for differing sizes of enumera- region is as follows: 0 in S, 16 in T, and 64 in U (these
tion units. To understand the need to standardize, con- numbers are shown within each enumeration unit in
sider map A in Figure 8, which portrays a hypothetical Figure 8A). The acres of wheat harvested from each
distribution consisting of three distinct regions: S, T, enumeration unit represent raw totals. Mapping these
raw totals with the choropleth method produces the
result shown in Figure 8B (note that higher data val-
* See Langford and Unwin (1994) for a more detailed discussion of ues are depicted by a darker gray tone). A user exam-
these limitations. ining this map would likely conclude that because
97
16:18:26.
S T U S T U
0 0 0 0 0 0 0 0 16 16 16 16 16 16 16 16
64 64 64 64
0 0 0 0 0 0 0 0 16 16 16 16 16 16 16 16
0 0 0 0 0 0 0 0 16 16 16 16 16 16 16 16
64 64 64 64
0 0 0 0 0 0 0 0 16 16 16 16 16 16 16 16
0 0 0 0 0 0 0 0 16 16 16 16 16 16 16 16
64 64 64 64
0 0 0 0 0 0 0 0 16 16 16 16 16 16 16 16
0 0 0 0 0 0 0 0 16 16 16 16 16 16 16 16
64 64 64 64
0 0 0 0 0 0 0 0 16 16 16 16 16 16 16 16
0 0 0 0 0 0 0 0 16 16 16 16 16 16 16 16
64 64 64 64
0 0 0 0 0 0 0 0 16 16 16 16 16 16 16 16
0 0 0 0 0 0 0 0 16 16 16 16 16 16 16 16
64 64 64 64
0 0 0 0 0 0 0 0 16 16 16 16 16 16 16 16
A B
S T U S T U
C D
FIGURE 8 A hypothetical illustration of the effect of data standardization: (A) raw totals—number of acres of wheat harvested
in each enumeration unit; (B) a choropleth map of the raw totals; (C) a choropleth map of standardized data achieved by dividing
the raw totals by the area of the corresponding enumeration unit; and (D) a proportional circle map of the raw totals.
region U is the darkest, it must have more wheat data. The difference between the maps is not as
grown in it. Unfortunately, this conclusion would be dramatic as the hypothetical example because the areas
inappropriate because we have not accounted for the of counties in Kansas are similar in size, but we can still
size of enumeration units. One approach to adjust (or note areas where the patterns differ.
standardize) the size of enumeration units is to divide Standardization not only adjusts for the differing sizes
each raw total by the area of the associated enumera- of counties but also provides a very useful attribute,
tion unit; the resulting values are 0/16, or 0 for region namely, the proportion (or percentage) of each county
S; 16/16, or 1 for region T; and 64/64, or 1 for region U. from which wheat was harvested. Such an attribute pro-
Mapping these values with the choropleth method vides an indication of the probability that one might see
results in Figure 8C; note that regions T and U are now wheat being cut at harvest time while driving through the
identically shaded and thus have the same density of county. The standardized map is shown in Figure 10A
wheat harvested. for comparison with the other methods of mapping dis-
In a similar fashion, the Kansas wheat data can be cussed in this section. (Note that the lightness visual vari-
standardized by dividing the acres harvested within able has been used for the choropleth map.)
each county by the area of the corresponding county.
Because the wheat harvested data and the area of the 4.2 Proportional Symbol Maps
county are both in acres, dividing the two produces a
proportion (or percentage, when we multiply the result A proportional symbol map is constructed by scaling
by 100; see Table 1). Figure 9 portrays maps of both the symbols in proportion to the magnitude of data occurring
unstandardized and the standardized Kansas wheat at point locations. These locations can be true points, such
16:18:26.
A Wheat Harvested in Kansas, 1993 B Wheat Harvested in Kansas, 1993

Thousands of Acres Percentage of Land Area
0.7 to 54 66 to 108 116 to 142 144 to 191 199 to 444 0.7 to 9.8 10.2 to 19.3 20.6 to 28.1 30.7 to 40.6 54.9 to 58.5
FIGURE 9 Standardizing wheat harvested in Kansas counties in 1993: (A) a map of the number of acres harvested; and (B) a stan-
dardized map resulting from dividing the number of acres harvested by the area of each county. (Data Source: Kansas Agricultural
Statistics 1994.)
as an oil well, or conceptual points, such as the center of appropriate), but in terms of the density of wheat
an enumeration unit for which data have been collected; harvested, regions T and U are identical.
the latter is the case with the wheat harvested data. In
contrast to the standardized data depicted on choropleth 4.3 Isopleth Map
maps, proportional symbol maps are normally used to
display raw totals. Thus, the magnitudes for acres of An isarithmic map (or contour map) is created by inter-
wheat harvested are depicted as proportional circles in polating a set of isolines between sample points of
Figure 10C. (Note that the visual variable used here known values; for example, we might draw isolines
is size.) between temperatures recorded for individual weather
The raw totals depicted on proportional symbol stations. The isopleth map is a specialized type of
maps provide a useful complement to the standardized isarithmic map in which the sample points are associ-
data shown on choropleth maps. Raw totals are impor- ated with enumeration units. It is an appropriate alter-
tant because a high proportion or rate might not be native to the choropleth map when one can assume that
meaningful if there is not also a high raw total. As an the data collected for enumeration units are part of a
example, consider counties of the same size having smooth, continuous 21/2-D phenomenon. For example,
populations of 100 and 100,000, in which 1 and 1,000 in the case of the wheat data, it might be argued that
people, respectively, have some rare form of cancer. the proportion of land with wheat changes in a rela-
Dividing the number of cancer cases by the population tively gradual (smooth) fashion, as opposed to changing
yields the same proportion of people suffering from just at county boundaries (as on the choropleth map).
cancer (0.01), but the rate for the less populous county In a fashion similar to a choropleth map, an isopleth
would be of lesser interest to an epidemiologist. map also requires standardized data. Referring again
Although the proportional symbol map is a better to the hypothetical raw totals shown in Figure 8A,
choice than the choropleth map for depicting raw totals, imagine drawing contours through such data. High-
care should be taken in using it. To illustrate, consider valued contour lines would tend to occur in region U,
map D in Figure 8, which displays the hypothetical where there are high values in the data; but as has
wheat data using proportional circles. Note that all cir- already been shown for the choropleth case, region U
cles in region U are larger than those in region T. This is really no different from region T. Dividing the raw
could lead to the mistaken impression that counties in totals by the area of each enumeration unit would
region U are more important in terms of wheat produc- result in standardized data that could be appropriately
tion than those in region T. Counties in region U might contoured.
be more important to a politician in assigning tax dollars The isopleth map resulting from contouring the stan-
(i.e., more wheat harvested indicates that a greater tax is dardized Kansas wheat data is shown in Figure 10B.
99
16:18:26.
Thematic Mapping Techniques

Wheat Harvested in Kansas, 1993
A Choropleth B Isopleth
10
30
20
20
30
30
30
10
30 10
40 50
20
20
Percentage of Land Area
Contour Lines Represent
Percent of Land Area
0.7 to 9.8 10.2 to 19.3 20.6 to 28.1 30.7 to 40.6 54.9 to 58.5
C Proportional Symbol D Dot
Thousands of Acres
Each Dot Represents
0.7 222.2 443.7 12,000 Acres of Wheat
FIGURE 10 A comparison of basic thematic mapping techniques: (A) choropleth, (B) isopleth, (C) proportional symbol, and (D)
dot maps. The choropleth and isopleth maps are based on the percentage of land area from which wheat was harvested, whereas
the proportional symbol and dot maps are based on the total acres of wheat harvested.
(Again, note that the visual variable lightness has been higher than any of the surrounding counties. Note that
used.) Although this map might be more representative a similar problem occurs within two northern counties
of the general distribution of wheat harvested than (Figure 10B). The dot map could be a solution to this
the choropleth map, the assumption of continuity and type of problem.
the use of county-level data produce some questionable
results. For example, note the island of higher value 4.4 Dot Mapping
near the center of the extreme southeastern county
(Cherokee). In reality, it seems unlikely that you would To create a dot map, one dot is set equal to a certain
find a higher value here; the high value is more likely a amount of a phenomenon, and dots are placed where
function of the fact that the centers of counties were that phenomenon is most likely to occur. The phenome-
used as a basis for contouring and Cherokee’s value was non might actually cover an area or areas (e.g, a field or
16:18:26.
fields of wheat), but for the sake of mapping, the the option of showing several of them, thus providing
phenomenon is represented as located at points. the user with various perspectives on the distribution of
Constructing an accurate dot map requires collecting wheat harvested in Kansas.
ancillary information that indicates where the phenome-
non of interest (wheat, in our case) is likely found. For
the wheat data, this was accomplished using the cropland 5 SELECTING VISUAL VARIABLES FOR CHOROPLETH
category of a land use/land cover map. The resulting dot MAPS
map is shown in Figure 10D. (In this case, the visual vari-
able location is used.) Clearly, the dot map is able to rep- In the preceding section, the visual variable lightness
resent the underlying phenomenon with much more was utilized for the choropleth map. From section 3,
accuracy than any of the other methods we have dis- we know that there are a number of other visual vari-
cussed. Also note that parts of the distribution exhibit ables that might be used to represent a quantitative
sharp discontinuities that would be difficult to show with phenomenon that is treated as areal in nature. This
the isopleth method (which presumes smooth changes). section considers how we might select among these
visual variables, and how we might avoid selecting
4.5 Discussion improper qualitative visual variables. The basic solu-
tion is to select a visual variable that appears to con-
An examination of Figure 10 reveals that each of the ceptually “match” the level of measurement of the
four maps provides a quite different picture of wheat data. For illustrative purposes, we again use the Kansas
harvested in the state of Kansas. Which method is used wheat data.
should depend on the purpose of the map. If the purpose The specific visual variables we discuss are illustrated
is to focus on “typical” county-level information, then the in Figure 11 and Color Plate 4. In examining these fig-
choropleth and proportional symbol maps are appropri- ures, note that they depict classed maps using
ate. The choropleth map provides standardized informa- maximum-contrast symbolization, which means that
tion, whereas the proportional symbol map provides raw symbols for classes have been selected so that they are
total information. It must be emphasized that neither maximally differentiated from one another. An alterna-
map depicts the detail of the underlying phenomenon, tive approach would be to create an unclassed map in
which is unlikely to follow enumeration unit boundaries. which symbols are directly proportional to the value for
When data are collected in the form of enumeration each enumeration unit (as in Figure 6B). The maximum-
units, the dot and isopleth methods should be considered contrast approach is used here because it is common
as two possible solutions for representing an underlying and more easily constructed (particularly in the case of
phenomenon that is not coincident with enumeration the size visual variable).
unit boundaries. In the case of the wheat data, the dot In addition to discussing Figure 11 and Color
method is probably the more appropriate approach Plate 4, we also consider Figure 12, which summarizes
because it can capture some of the discontinuities in the the use of visual variables for various levels of mea-
phenomenon.The isopleth method, however, could prob- surement. Note that the body of this figure is shaded
ably be improved with a finer grid of enumeration units and labeled to indicate various levels of acceptability:
(e.g., townships);* of course, this would also be true of the Poor (P), Marginally effective (M), and Good (G).
choropleth and proportional symbol maps. MacEachren (1994a, 33) developed a similar figure,
It must be noted that we have considered only four which he appeared to apply to all kinds of spatial phe-
of the more common methods of thematic mapping. nomena. We use Figure 12 only for areal phenomena;
One alternative would be a dasymetric map, which, like as an exercise, you might consider developing such a
the dot map, can show very detailed information, but figure for other kinds of phenomena.
uses standardized data. Another alternative would be to We’ll consider the perspective height and size visual
modify the proportional symbol map by making the variables first because they have the greatest potential
area of the circle that is filled in proportional to the per- for logically representing the numerical data depicted
cent of land area from which wheat is harvested—this on choropleth maps. Use of perspective height produces
creates what is called a pie chart. Finally, we should what is commonly termed a prism map (Figure 11A). In
keep in mind that if maps are to be viewed in an inter- Figure 12, note that perspective height is the only visual
active graphics environment, the mapmaker will have variable receiving a “good” rating for numerical data.
The justification is that an unclassed map based on per-
spective height can portray ratios correctly (a data value
twice as large as another will be represented by a prism
*Data at the township level are not released to the general public to twice as high), and that readers will perceive the height
protect the confidentiality of individual farm production. of resulting prisms as ratios (Cuff and Bieri 1979).
101
16:18:26.
Visual Variables
A Perspective Height B Size
N Percentage of Land Area
0 to 9.8 10.2 to 19.3 20.6 to 28.1 30.7 to 40.6 54.9 to 58.5
C Spacing D Lightness
Percentage of Land Area Percentage of Land Area
0 to 9.8 10.2 to 19.3 20.6 to 28.1 30.7 to 40.6 54.9 to 58.5 0.7 to 9.8 10.2 to 19.3 20.6 to 28.1 30.7 to 40.6 54.9 to 58.5
E Orientation
Percentage of Land Area FIGURE 11 Visual variables for representing the percentage
of wheat harvested in Kansas counties: (A) perspective height,
0.7 to 9.8 10.2 to 19.3 20.6 to 28.1 30.7 to 40.6 54.9 to 58.5 (B) size, (C) spacing, (D) lightness, and (E) orientation. For
color visual variables, see Color Plate 4.
16:18:26.
Nominal Ordinal Numerical
c c
Spacing P M M
Size P M M
a b
Perspective
Height P M G
Orientation G P P
Shape G P P
Arrangement G P P
Lightness P G M
d d
Hue G G M
Saturation P M M
P = Poor M = Marginally Effective G = Good

a
Since height differences are suggestive of numerical differences, use with caution for
ordinal data.
b
Hidden enumeration units and lack of a north orientation are problems.
c
Not aesthetically pleasing.
d
The particular hues selected must be carefully ordered, such as yellow, orange, red.
FIGURE 12 Effectiveness of visual variables for each level of measurement for areal phenomena. (After MacEachren 1994a, 33.)
There are two problems, however, that complicate other portions can be seen. A third solution is to use
the extraction of numerical information from prism the perspective height variable but also to symbolize
maps. One is that tall prisms sometimes block smaller the distribution with another visual variable; for
prisms. A solution to this problem is to rotate the example, Figure 11D might be displayed in addition
map so that blockage is minimized; for example, to Figure 11A.
the map in Figure 11A has been rotated so that the Another problem with prism maps is that rotation
view is from the lower valued northeast. A second might produce a view that is unfamiliar to readers who
solution to the blockage problem is to manipulate the normally see maps with north at the top. This problem
map in an interactive graphics environment. If a flex- can be handled by showing a second map (as suggested
ible program is available, it might even be possible to earlier) or by using an overlay of the base to show the
suppress selected portions of the distribution so that amount of rotation (as in Figure 2C).
103
16:18:26.
The size visual variable is illustrated in Figure 11B; data, and other hues work better for ordinal data. For
note that here the size of individual marks making up example, to display different soil types (alfisols, entisols,
the areal symbol has been varied. Size might be consid- mollisols), red, green, and blue hues might be deemed
ered appropriate for representing numerical relations appropriate (one of these hues does not inherently
because circles can be constructed in direct proportion represent a higher data value than another). For ordinal
to the data (a data value twice another can be repre- and higher-level data, logically ordered hues are neces-
sented by a circle twice as large in area). However, a sary; for example, a yellow, orange, and red scheme
correction factor might have to be implemented to (Color Plate 4B) is one possibility because orange is
account for underestimation of larger circles. seen as a mixture of yellow and red.
Although some cartographers (most notably The remaining visual variables (orientation, shape,
Bertin) have promoted the use of the visual variable and arrangement) are appropriate only for creating
size on choropleth maps, two problems are apparent. nominal differences. As an example, consider the orien-
First, it is questionable whether map users actually tation variable. You might try to create a logical pro-
consider the sizes of circles when used as part of an gression of symbols by starting with horizontal lines for
areal symbol. Users might analyze circle size when try- low values and then gradually changing the angle of the
ing to acquire specific information, but it seems lines so that the highest class is represented by vertical
unlikely that they would do so when analyzing the lines, but an examination of Figure 11E reveals that this
overall map pattern. Rather, it is more likely that they approach is not effective; the changing angle of the lines
would perceive areas of light and dark, in a fashion appears to create nominal differences, and the resulting
similar to the lightness visual variable. Second, many map is “busy.”
cartographers (and presumably map users) find the It should be noted that cartographers are not in com-
coarseness of the resulting symbols unacceptable— plete agreement on the ratings displayed in Figure 12,
they would prefer the fine tones shown in Figure 11D. so you might develop slightly different ratings. For
The latter problem in particular caused us to give the example, one of our students rated the orientation vari-
size variable only a moderate rating for portraying able “poor” (even for nominal data) because he felt it
numerical data (Figure 12). lacked aesthetic quality and that it was difficult to dis-
Note also that we have given both perspective height criminate among different orientations.
and size only moderate ratings for portraying ordinal
data. The logic is that if such variables are used to illus-
trate numerical relations, users might perceive such SUMMARY
relations when only ordinal relations are intended.
Obviously, both variables are inappropriate for nominal In this chapter, we have covered basic principles for
data because different heights and sizes suggest quanti- symbolizing geographic phenomena. We have discov-
tative rather than qualitative information. ered that the nature of the underlying phenomenon is
Although other visual variables can be manipu- an important consideration in selecting an appropriate
lated mathematically to create proportional (ratio) symbology. For example, if the underlying phenomenon
relationships, the resulting symbols cannot be inter- is smooth and continuous (e.g., yearly snowfall for
preted easily in a ratio fashion. For example, consider Russia), then a contour map would be appropriate, but
the lightness variable shown in Figure 11D. It is easy a choropleth map would not be.
to see that one shade is darker or lighter than another, Another important consideration in selecting sym-
but it is difficult to establish proportional relations bology is the level of measurement of the data. Ideally,
(e.g., that one shade is twice as dark as another). there should be a logical match between the level of
Similar comments can be made for the visual vari- measurement and the symbology (or visual variable)
ables spacing, saturation, and hue, with the following used to represent the data. For instance, if data are
caveats: First, note that we have given spacing only a numerical (e.g., the magnitude of electrical generation
moderate rating for ordinal information because, in at power plants in kilowatt hours), then the symbology
our opinion, the symbols are not aesthetically pleas- should be capable of enabling a map reader to visualize
ing, and there is the implication that low data values numerical relations (e.g., a proportional symbol map
are qualitatively different from high data values. would be appropriate) and in certain cases to obtain
Second, we have given saturation only a moderate rat- exact data values. Keep in mind, however, that we gen-
ing for ordinal information because it is our experi- erally do not expect readers to acquire precise numeri-
ence that people have a difficult time understanding cal information from maps; rather, maps are primarily
what a “greater” saturation means. used to show spatial patterns.
We have rated hue as “good” for both nominal and Although the underlying phenomenon is an impor-
ordinal data because some hues work well for nominal tant consideration in selecting symbology, we have seen
16:18:26.
that map purpose can also play an important role. For We have also learned that for some mapping methods,
example, if the mapmaker wishes to show “typical” val- data standardization is important, and that raw totals
ues for enumeration units, then a choropleth map might need to be adjusted to account for the area over which
be appropriate even when the underlying phenomenon the data have been collected (typically an enumeration
is not coincident with enumeration units, as was the case unit). The simplest form of adjustment is to divide raw
with wheat harvested in Kansas. The mapmaker should totals by the areas of enumeration units (e.g., we could
realize, however, that in this instance a choropleth map divide the number of people in counties by the areas of
might lead to incorrect perceptions of the underlying counties to create a map of population density).
phenomenon.
FURTHER READING
Bertin, J. (1983) Semiology of Graphics: Diagrams, Networks, Chrisman, N. (2002) Exploring Geographic Information
Maps. Madison, WI: University of Wisconsin Press. Systems. New York: Wiley.
(Translated by W. J. Berg) Discusses several extensions to the four standard levels of mea-
Chapter 2 of this widely cited text focuses on visual variables. The surement (nominal, ordinal, interval, and ratio). For a theoretical
text is a bit difficult to read because it has been translated from discussion, see Chrisman (1998).
French to English. Forrest, D. (1999b) “Geographic information: Its nature, clas-
Brewer, C. A. (1994a) “Color use guidelines for mapping and sification, and cartographic representation.” Cartographica
visualization.” In Visualization in Modern Cartography, ed. 36, no. 2:31–53.
by A. M. MacEachren and D. R. F. Taylor, pp. 123–147. Presents a sophisticated approach to selecting symbology as a
Oxford: Pergamon. function of the phenomena, the spatial data, and the level of mea-
Pages 124–126 cover terminology for using color. We consider surement.
Brewer’s work more fully in Chapter 14. MacEachren, A. M. (1994a) Some Truth with Maps: A Primer
Burrough, P. A., and McDonnell, R. A. (1998) Principles of on Symbolization and Design. Washington, DC: Association
Geographical Information Systems. Oxford: Oxford Univer- of American Geographers.
sity Press. Spatial arrangement of geographic phenomena, levels of measure-
Chapter 2 provides an alternative view of geographic phenomena ment, and visual variables are covered on pages 13–34.
by contrasting entity and continuous field approaches.
GLOSSARY
abrupt phenomena: phenomena that change abruptly over choropleth map: a map in which enumeration units (or data-
geographic space, such as sales tax rates for each state in the collection units) are shaded with an intensity proportional to
United States. the data values associated with those units (e.g., census tracts
abstract sound variables: analogous to visual variables, these shaded with gray tones that have an intensity proportional to
are variables used to depict abstract sounds (e.g., the location population density).
and loudness of sound). chroma: see saturation.
areal phenomena: geographic phenomena, such as a constrained ratio: a level of measurement for data sets that
forested region, that are two-dimensional in spatial extent, are constrained to a fixed set of numbers, such as probability
having both length and width; data associated with enumera- (the range is 0 to 1) or percentages (the range is 0 to 100).
tion units can also be considered areal phenomena because continuous phenomena: geographic phenomena that occur
each unit is an enclosed area. everywhere, such as the yearly distribution of snowfall in
arrangement: a visual variable in which the marks making Wisconsin or sales tax rates for each state in the United States.
up symbols are arranged in various ways, such as breaking up contour map: see isarithmic map.
lines into dots and dashes to create various forms of political
counts: a level of measurement in which individual objects,
boundaries.
such as people, are counted.
balanced data: two phenomena coexisting in a complemen-
cyclical: a level of measurement appropriate for phenomena
tary fashion, such as the percentage of English and French
that have a cyclical character, such as angular measurements
spoken in Canadian provinces.
(the cycle is 360°).
bipolar data: data characterized by either a natural or a
dasymetric map: like choropleth maps, area symbols are
meaningful dividing point, such as the percentage of popula-
used to represent zones of uniformity, but the bounds of zones
tion change.
need not match enumeration unit boundaries.
105
16:18:26.
discrete phenomena: geographic phenomena that occur at numerical data: a term used to describe the data associated
isolated locations, such as water towers in a city. with the interval and ratio levels of measurement, as numeri-
dot map: a map in which small symbols of uniform size cal values are assigned to the data.
(typically, solid circles) are used to emphasize the spatial pat- ordinal: a level of measurement in which data are ordered
tern of a phenomenon (e.g., one dot might represent 1,000 or ranked but no numerical values are assigned, such as rank-
head of cattle). ing states in terms of where you would like to live.
fishnet map: a map in which a fishnet-like structure provides orientation: (1) the indication of direction on a map, nor-
a three-dimensional symbolization of a smooth, continuous mally taking the form of an arrow or a graticule; (2) a visual
phenomenon. variable in which the directions of symbols (or marks making
fuzzy categories: a level of measurement in which category up symbols) are varied, such as changing the orientation of
memberships are fuzzy, such as some remote sensing classifi- short line segments to indicate the direction from which the
cation procedures. wind blows.
gamma function: the relation between the voltage of a color perspective height: a visual variable involving a perspective
gun and the associated luminance of the CRT display. three-dimensional view, such as using raised sticks (or lol-
lipops) to represent oil production at well locations.
geometric symbols: symbols that do not necessarily look like
the phenomena being mapped, such as using squares to depict pictographic symbols: symbols that look like the phenom-
barns of historical interest; contrast with pictographic symbols. ena being mapped, such as using diagrams of barns to depict
the locations of barns of historical interest; contrast with geo-
GeoWall: a specialized form of wall-sized display that metric symbols.
includes three basic components: a computer with a dual-head
graphics card, two DLP projectors on an adjustable stand, and pie chart: a modification of the proportional circle in which
polarization equipment, which includes filters for the projec- a portion of the circle is filled in (e.g., to represent percentage
tors, a display screen capable of preserving polarization, and of land area from which wheat is harvested).
glasses for users. point phenomena: geographic phenomena that have no spa-
hue: along with lightness and saturation, one of three com- tial extent and can thus be termed “zero-dimensional,” such as
ponents of color; it is the dominant wavelength of light, such the location of oil well heads.
as red versus green. prism map: a map in which enumeration units are raised to
interval: a level of measurement in which numerical values a height proportional to the data associated with those units.
can be assigned to data, but there is an arbitrary zero point, proportional symbol map: a map in which point symbols are
such as for SAT scores. scaled in proportion to the magnitude of data occurring at
isarithmic map: a map in which a set of isolines (lines of point locations, such as circles of varying sizes representing
equal value) are interpolated between points of known value, city populations.
as on a map depicting the amount of snowfall in a region. ratio: a level of measurement in which numerical values are
isopleth map: a form of isarithmic map in which control assigned to data and there is a nonarbitrary zero point, such
points are associated with enumeration units (e.g., assigning as with the percentage of forested land.
population density values for census tracts to the centers of raw totals: raw numbers that have not been standardized to
those tracts and then interpolating between the centers). account for the area over which the data were collected (e.g.,
large-scale map: a map depicting a relatively small portion we might map the raw number of acres of tobacco harvested
of the Earth’s surface. in each county, disregarding the size of the counties).
level of measurement: refers to the different ways that we remote sensing: the acquisition of information about the
measure attributes; we commonly consider nominal, ordinal, Earth (or other planetary body) from high above its surface
interval, and ratio levels. via reflected or emitted radiation; most commonly done via
satellites.
lightness: along with hue and saturation, one of three com-
ponents of color; refers to how dark or light a color is, such as saturation: along with hue and lightness, one of three com-
light blue versus dark blue. ponents of color; can be thought of as a mixture of gray and a
pure hue: very saturated colors have little gray, whereas desat-
linear phenomena: geographic phenomena that are one-
urated colors have a lot of gray.
dimensional in spatial extent, having length but essentially no
width, such as a road on a small-scale map. shape: a visual variable in which the form of symbols (or
marks making up symbols) is varied, such as using squares
location: a visual variable that refers to the possibility of
and triangles to represent religious and public schools.
varying the position of symbols, such as dots on a dot map.
size: (1) a visual variable in which the magnitudes of
maximum-contrast shades: on a choropleth map, areal
symbols (or marks making up symbols) are varied (e.g., using
shades are selected so that they have the maximum possible
circles of different magnitudes to depict city populations); (2)
contrast with one another.
how large a typeface is; normally expressed in points, where a
nominal: a level of measurement in which data are placed point is 1/72 of an inch.
into unordered categories, such as classes on a land use/land
small-scale map: a map depicting a relatively large portion
cover map.
of the Earth’s surface.
16:18:26.
smooth phenomena: phenomena that change gradually over phenomena: a form of volumetric phenomena in which each
geographic space, such as the yearly distribution of snowfall in point on the surface is defined by longitude, latitude, and a value
Wisconsin. above (or below) a zero point (e.g., the Earth’s topography).
spacing: a visual variable in which the distance between unipolar data: data that have no dividing point and do not
marks making up a symbol is varied, such as shading counties involve two complementary phenomena, such as per capita
on a choropleth map with horizontal lines of varied spacing. income in African countries; contrast with bipolar and bal-
spatial dimension: a term that describes whether a anced data.
phenomenon can be conceived of as points, lines, areas, or vol- value: see lightness.
umes. visual variables: the perceived differences in map symbols
true 3-D phenomena: a form of volumetric phenomena in that are used to represent spatial phenomena.
which each longitude and latitude position has multiple
attribute values depending on the height above or below a
zero point (e.g., the level of ozone in the Earth’s atmosphere
varies as a function of elevation above sea level).
REFERENCES
Bertin, J. (1983) Semiology of Graphics: Diagrams, Networks, Kansas Agricultural Statistics. (1994) Kansas Farm Facts.
Maps. Madison: University of Wisconsin Press. Topeka: Kansas State Board of Agriculture.
Brewer, C. A. (1994a) “Color use guidelines for mapping and Langford, M., and Unwin, D. J. (1994) “Generating and
visualization.” In Visualization in Modern Cartography, ed. mapping population density surfaces within a geographical
by A. M. MacEachren and D. R. F. Taylor, pp. 123–147. information system.” The Cartographic Journal 31, no.
Oxford, England: Pergamon. 1:21–26.
Castner, H. W., and Robinson, A. H. (1969) Dot Area MacEachren, A. M. (1992) “Visualizing uncertain informa-
Symbols in Cartography: The Influence of Pattern on Their tion.” Cartographic Perspectives, no. 13:10–19.
Perception. Washington, DC: American Congress on
MacEachren, A. M. (1994a) Some Truth with Maps: A Primer
Surveying and Mapping.
on Symbolization & Design. Washington, DC: Association
Chrisman, N. (2002) Exploring Geographic Information of American Geographers.
Systems (2nd ed.). New York: Wiley.
MacEachren, A. M., and DiBiase, D. (1991) “Animated maps
Cuff, D. J., and Bieri, K. R. (1979) “Ratios and absolute of aggregate data: Conceptual and practical problems.”
amounts conveyed by a stepped statistical surface.” The Cartography and Geographic Information Systems 18, no.
American Cartographer 6, no. 2:157–168. 4:221–229.
DiBiase, D., Krygier, J., Reeves, C., MacEachren, A., and McCleary, G. F. (1983) “An effective graphic ‘vocabulary.’”
Brenner, A. (1991) “Elementary approaches to carto- IEEE Computer Graphics & Applications 3, no. 2:46–53.
graphic animation.” Videotape. University Park, PA: Deasy
Morrison, J. L. (1984) “Applied cartographic communication:
GeoGraphics Laboratory, Department of Geography,
Map symbolization for atlases.” Cartographica 21, no.
Pennsylvania State University.
1:44–84.
Eastman, J. R. (1986) “Opponent process theory and syntax
South Carolina State Budget and Control Board. (1994)
for qualitative relationships in quantitative series.” The
South Carolina Statistical Abstract. Columbia, SC: Office of
American Cartographer 13, no. 4:324–333.
Research and Statistical Services.
Forrest, D. (1999b) “Geographic information: Its nature,
classification, and cartographic representation.”
Cartographica 36, no. 2:31–53.
107
16:18:26.
16:18:26.
Scale and Generalization
109
16:24:23.
OVERVIEW
This chapter presents basic concepts in scale and general- or areal feature, thus retaining the critical points defining
ization. Scale is a fundamental geographic principle, that feature. Section 5 illustrates a practical application of
but there is often confusion about the exact meaning generalization using the National Historical Geographic
of geographic scale, cartographic scale, and data Information System (NHGIS; http://www.nhgis.org), and
resolution. Geographers think about scale as area covered, section 6 describes MapShaper, a free Web-based general-
where large-scale studies cover a large region such as a ization service (http://mapshaper.org).
state. Cartographers think about scale mathematically and
use the representative fraction (RF) to express the rela-
tionship between map and Earth distance. For instance, 1 GEOGRAPHIC AND CARTOGRAPHIC SCALE
most national map series are established at a specific scale,
such as the French 1:25,000 BD Topo, and the USGS’s Scale is a fundamental concept in all of science and is of
1:24,000 series. Data resolution refers to the granularity of particular concern to geographers, cartographers, and oth-
the data, such as the pixel size of a remote sensing image. ers who are interested in geospatial data. Astronomers
Directly related to the concept of scale are the ideas of gen- work at a spatial scale of light years, physicists work at the
eralization and modifying the information content so that atomic spatial scale in mapping the Brownian motion of
it is appropriate at a given scale. It would not be possible, atoms, and geographers work at spatial scales from the
for instance, to depict the street network for the entire human to the global. Within the fields of geography and
United States with the country mapped at a scale that would cartography, the terms geographic scale and cartographic
fit on one page—only major highways could be depicted. scale are often confused. Geographers and other social
Section 1 introduces the concepts of geographic and car- scientists use the term scale to mean the extent of the study
tographic scale and covers how scale controls the amount area, such as a neighborhood, city, region, or state. Here,
of map space and thus the appropriate information con- “large scale” indicates a large area—such as a state—
tent. Section 2 provides some basic definitions of gen- whereas “small scale” represents a smaller entity—such as
eralization, including a discussion of some fundamental a neighborhood. Climatologists, for instance, talk about
generalization operations. Section 3 discusses several con- large-scale global circulation in relation to the entire
ceptual models of the generalization process. One of the Earth; in contrast, urban geographers talk about small-
more complete models divides the process into “why,” scale gentrification of part of a city. Alternatively, carto-
“when,” and “how” components of generalization. Section graphic scale is based on a strict mathematical principle:
4 describes the many operations that have been designed the representative fraction (RF). The RF, which expresses
for the generalization process, including simplifi- the relationship between map and Earth distances, has
cation, smoothing, aggregation, amalgamation, collapse, become the standard measure for map scale in cartogra-
merging, refinement, exaggeration, enhancement, and phy. The basic format of the RF is quite simple: RF is
displacement. We also discuss in detail simplification, expressed as a ratio of map units to Earth units (with the
which involves “weeding” unnecessary data from a linear map units standardized to 1). For example, an RF of
16:24:23.
1:25,000 indicates that 1 unit on a map is equivalent to This assumes that one can generate, from a master data-
25,000 units on the surface of the Earth. The elegance of base, additional versions at a variety of scales. The need
the RF is that the measure is unitless—in our example, the for such multiple-scale databases results from the require-
1:25,000 could represent inches, feet, or meters. Of course, ments of users. For instance, when mapping census data
in the same way that 1/2 is a larger fraction than 1/4, at the county level, a user might wish to see significant
1:25,000 is a larger scale than 1:50,000. Similarly, a scale of detail along the boundaries. Alternatively, when using
1:25,000 depicts relatively little area but does so in much the same boundary files at the state level, less detail is
greater detail, whereas a scale of 1:250,000 shows a larger needed. Because the generation of digital spatial data is
area in less detail. Thus, it is the cartographic scale that extremely expensive and time-consuming, one master
determines the mapped space and the level of geographic version of the database is often created, with smaller scale
detail possible. At the extreme, architects work at very versions being generated from this master database.
large scales, perhaps 1:100 (where individual rooms and A good example of this is the standard mapping of
furniture can be depicted), whereas a standard globe the National Historical Geographic Information System
might be constructed at a scale of 1:30,000,000, allowing (NHGIS). The master database for the NHGIS is at a
for only the most basic of geographic detail to be provided. scale of 1:150,000, which is the original scale of the
Certain design issues have to be considered when repre- TIGER files and the digital line graphs (DLGs) of
senting scales on maps, as do a variety of methods for the USGS. From this base scale, two additional databases,
representing scale, including the RF, the verbal statement, one at 1:500,000 and one at 1:1,000,000, are being created
and the graphical bar scale. through the application of generalization operations. All
Let’s assume that one wishes to produce a map at 1 inch three versions of this multiple-scale database will be
equals 10 miles. What would be the representative frac- accessible via the NHGIS Web site (http://www.nhgis.org).
tion, or RF? Since 1 mile is equivalent to 63,360 inches, 10
miles equals 10 × 63,360, or 633,600 inches. So the exact
RF would be 1:633,600. Of course, the standard mapping 2 DEFINITIONS OF GENERALIZATION
scales (representative fractions) for many countries differ,
with the U.S. Geological Survey having established the 2.1 Definitions of Generalization in the Manual
base map at 1:24,000, or 1 inch equals 2,000 feet. European Domain
mapping agencies have used larger scales—1:10,000 and
1:20,000, since the size of the countries allows for many Generalization is the process of reducing the information
fewer map sheets even at the larger scale. content of maps because of scale change, map purpose,
The term data resolution, which is related to scale, indi- intended audience, and/or technical constraints. For
cates the granularity of the data that are used in mapping. instance, when reducing a 1:24,000 topographic map (large
When mapping population characteristics of a city (an scale) to 1:250,000 (small scale), some of the geographical
urban scale), the data can be acquired at a variety of resolu- features must be either eliminated or modified because the
tions, including census blocks, block groups, tracts, and amount of map space is significantly reduced. Of course, all
even minor civil divisions (MCDs). Each level of resolution maps to some degree are generalizations, as it is impossible
represents a different “grain” of the data. Likewise, when to represent on a map all features from the real world, no
mapping biophysical data using remote sensing imagery, matter what the scale.A quote from Lewis Carroll’s (1893)
a variety of spatial resolutions are possible, based on the Sylvie and Bruno Concluded nicely illustrates this concept:
sensor. Common grains are 79 meters (Landsat Multi-
Spectral Scanner), 30 meters (Landsat Thematic Mapper), “That’s another thing we’ve learned from your Nation,”
20 meters (SPOT HRV multispectral), and 1 meter (Ikonos said Mein Herr, “map making. But we’ve carried it much
panchromatic). “Low resolution” refers to coarser grains further than you. What do you consider the largest map
(e.g., counties), and “high resolution” refers to finer that would be really useful?”
grains (e.g., blocks). Cartographers must be sure that they “About six inches to the mile.”
understand the relationships among geographic scale, “Only about six inches!” exclaimed Mein Herr. “We very
cartographic scale, and data resolution, and how these rela- soon got to six yards to the mile. Then we tried a hundred
yards to the mile. And then came the grandest idea of all!
tionships influence the information content of the map.
We actually made a map of the country on a scale of a mile
to the mile!”
1.1 Multiple-Scale Databases “Have you used it much?” I enquired.
“It has never been spread out yet,” said Mein Herr. “The
Increasingly, cartographers and other geographic infor- farmers objected: they said it would cover the whole country,
mation scientists require the creation of multiscale/ and shut out the sunlight! So now we use the country itself,
multiresolution databases from the same digital data set. as its own map, and I assure you it does nearly as well.”
111
16:24:23.
Cartographers have written on the topic of cartographic maps were made with pen and ink. When changing scale,
generalization since the early part of the twentieth cartographers would carefully study a feature at the
century. Max Eckert, the seminal German cartographer original scale, and then decide how the feature needed to
and author of Die Kartenwissenschaft, wrote in 1908, “In be modified for the new scale. Did it need to be moved
generalizing lies the difficulty of scientific map making, (displaced) to avoid colliding with another feature? Did
for it no longer allows the cartographer to rely merely it need to be simplified and/or smoothed? A series of tech-
on objective facts but requires him to interpret them niques were developed over hundreds of years to help car-
subjectively” (p. 347). Other cartographers also have tographers generalize maps and map features. Today, we
struggled with the intrinsic subjectivity of the generaliza- attempt to replicate manual generalization with comput-
tion process as they have attempted to understand and ers; thus the term digital generalization is now appropriate.
define cartographic generalization. For instance, in 1942
John K. Wright argued that “Not all cartographers are
3 MODELS OF GENERALIZATION
above attempting to make their maps seem more accu-
rate than they actually are by drawing rivers, coasts, form
lines, and so on with an intricacy of detail derived largely To better understand the complexity of generalization,
from the imagination” (p. 528). Wright identified two researchers have attempted to design conceptual models
major components of the generalization process: simplifi- of the process. Some researchers have focused on funda-
cation, which is the reduction of raw information that is mental operations and the relationships among them,
too intricate, and amplification, which is the enhancement whereas other researchers have created complex models.
of information that is too sparse. The idea that gener-
alization could be broken down into a logical set of 3.1 Robinson et al.’s Model
processes such as simplification and amplification has
become a common theme in generalization research. Arthur Robinson and his colleagues (1978) developed
Erwin Raisz (1962), for example, identified three one of the first formal models or frameworks to better
major components of generalization: combination, omis- understand the generalization process. They separated
sion, and simplification. Arthur Robinson and his col- the process into two major steps: selection (a preprocess-
leagues (1978) identified four components: selection, ing step) and the actual process of generalization, which
simplification, classification, and symbolization. In involves the geometric and statistical manipulation of
Robinson et al.’s model, selection was considered a pre- objects. Selection involves the identification of objects to
processing step to generalization itself. Selection retain in (or eliminate from) the map or database. For
allowed for the identification of certain features and fea- instance, in developing content for a thematic map, often
ture classes, whereas generalization applied the various a minimal amount of base material is selected, such as
operations such as simplification. We discuss the major roads, political boundaries, or urbanized areas.
Robinson model in more detail in section 3.1. Detailed base information, such as place-names and
hydrologic networks, are often left off a map because this
2.2 Definitions of Generalization in the Digital Domain base information is not deemed critical. On the other
hand, considerable base information is often selected for
Robert McMaster and Stuart Shea (1992) noted that the detailed topographic maps, as this information is deemed
generalization process itself accomplishes the following: critical. Generalization involves the processes of simplifi-
reduces the amount of data that needs to be stored; helps cation, classification, and symbolization. Simplification is
to modify features with changing map scale; and assists in the elimination of unnecessary detail from a feature; clas-
simplifying (for instance, classifying) and symbolizing sification involves categorizing objects; and symboliza-
data. One definition of generalization is the process of tion is the graphic encoding.
deriving, from an original data source, a symbolically or
digitally encoded cartographic data set through the appli- 3.2 McMaster and Shea’s Model
cation of spatial and attribute transformations. Of course,
in applying these transformations (through generaliza- Attempting to create a comprehensive conceptual
tion operations), we try to be mindful of the map’s pur- model of the generalization process, McMaster and
pose and intended audience. For instance, is the map Shea (1992) identified three significant components: the
designed for a transportation analysis or for a bike route, conceptual objectives, or why to generalize; the carto-
and what is the impact on the generalization operations? metric evaluation, or when to generalize; and the funda-
Students of modern cartography should realize that mental operations, or how to generalize (Figure 1). We
much of what we know today about the generalization will now discuss why and when we generalize; in section
process is the result of what cartographers over the cen- 4 we will consider a number of issues related to how we
turies learned through manual cartography, where generalize.
16:24:23.
Cartographic Generalization
Conceptual Objectives Cartometric Evaluation Fundamental Operations FIGURE 1 An overview of

(Why to generalize) (When to generalize) (How to generalize)
McMaster and Shea’s model of gen-
eralization.
Why Generalization Is Needed: The Conceptual • An urban neighborhood at a scale of 1:10,000

Objectives of Generalization • A small city at a scale of 1:100,000
The conceptual objectives of generalization include reduc- • A large metropolitan area at a scale of 1:1,000,000
ing complexity, maintaining spatial accuracy, maintaining • Several states at a scale of 1:10,000,000
attribute accuracy, maintaining aesthetic quality, maintain- • Most of a hemisphere at a scale of 1:100,000,000
ing a logical hierarchy, and consistently applying the rules of • The entire world with plenty of room to spare at a
generalization. Reducing complexity is perhaps the most scale of 1:1,000,000,000
significant goal of generalization. The difficulty for the He explained that these examples, which range from
cartographer is relatively straightforward: How does one largest (1:102) to smallest (1:109) span eight orders of
take a map at a scale of, perhaps, 1:24,000 and reduce it to a magnitude and a logical geographical spectrum of
scale of 1:100,000? More important, the cartographer faces scales. Geographers work at a variety of scales, from the
the difficulty of reducing the information content so that it very large—the neighborhood—to the very small—the
is appropriate for the scale. Obviously, the complexity of world. Generalization is a key activity in changing the
detail that is provided at a scale of 1:24,000 cannot be rep- information content so that it is appropriate for these
resented at 1:100,000; some features must be eliminated different scales. However, a rough guideline that car-
and some detail must be modified. For centuries, through tographers follow is that scale change should not exceed
considerable experience, cartographers developed a sense 10 * . Thus, if you have a scale of 1:25,000, you should
of what constituted appropriate information content. use the scale only for generalizations up to 1:250,000.
Figure 2 nicely illustrates this point. The top portion of this Beyond 1:250,000, the original data are “stretched”
figure depicts the same feature—a portion of Ponta Creek beyond their original fitness for use.
in Kemper and Lauderdale Counties, Mississippi—at four Two additional theoretical objectives that are important
different scales: 1:24,000, 1:50,000, 1:100,000, and 1:250,000. in generalization are maintaining the spatial accuracy of
These features, digitized by Philippe Thibault (2002) in his features and maintaining the attribute accuracy of features.
doctoral dissertation, effectively show the significantly dif- Spatial accuracy deals primarily with the geometric shifts
ferent information content that is visible as one reduces the that necessarily take place in generalization. In simplifica-
scale from 1:24,000 to 1:250,000. The bottom part of the tion, for instance, coordinate pairs are deleted from the
illustration depicts an enlargement of the smaller scale fea- data set. By necessity, this shifts the geometric location of
tures to match the features at 1:24,000. Note, for instance, the features, creating “error.” The same problem occurs
the enlargement of the 1:250,000 scale line by 1,041.67 per- with feature displacement, in which two features are pulled
cent to match Ponta Creek at 1:24,000. In this case, the apart to prevent a graphical collision. Goals of the gener-
cartographer has manually generalized Ponta Creek alization process are to minimize this shifting and to
through a series of transformations—including simplifica- maintain as much spatial accuracy as possible. Attribute
tion, smoothing, and enhancement (as described later)—as accuracy deals with the subject being mapped, such as
a holistic process, unlike current computer approaches that population density or land use. For instance, classification,
require a separation of these often linked processes.The set a key component of generalization, often degrades the
of decisions required to generalize cartographic features original “accuracy” of the data through data aggregation.
based on their inherent complexity is difficult if not impos-
sible to quantify, although as described next, several When Generalization Is Required
attempts have been made over the past two decades.
In a digital cartographic environment, it is necessary to
Clearly, there is a direct and strong relationship
identify those specific conditions when generalization
among scale, information content, and generalization.
will be required. Although many such conditions can be
John Hudson (1992) explained the effect of scale by
identified, four fundamental conditions include:
indicating what might be depicted on a 5 by 7-inch map:
• A house at a scale of 1:100 1. Congestion
• A city block at a scale of 1:1,000 2. Coalescence
113
16:24:23.
Lines Represented at Original Scale
Line A
1:24,000
Line B
1:50,000
Line C
1:100,000
Line D
1:250,000
Enlarged to a Common 1:24,000
Line A
100%
Line B
208.33%
Line C
416.67%
Line D
1,041.67%
FIGURE 2 Depiction of Ponta
Creek (in Mississippi) at four differ-
Portion of Ponta Creek, Kemper and Lauderdale Co., Mississippi ent scales. (Courtesy of Philippe
Thibault.)
3. Conflict
4. Complication
As explained by McMaster and Shea (1992), congestion
refers to the problem, under scale reduction, of too many
objects being compressed into too small a space,
which results in overcrowding due to high feature density
(Figure 3). Significant congestion results in decreased
communication of the mapped message—for instance,
when too many buildings are in close proximity.
Coalescence refers to the condition in which features
graphically collide due to scale change. In these situations,
features actually touch. This condition thus requires
the implementation of the displacement operation, as FIGURE 3 How a change in scale can create congestion.
16:24:23.
discussed shortly. Conflict results when, due to generaliza- TABLE 1 A framework for generalization operations. (After
tion, an inconsistency between or among features occurs. McMaster and Monmonier 1989, and McMaster 1989b.)
For instance, if generalization of a coastline eliminates a Raster-mode generalization Vector-mode generalization
bay with a city located on it, either the city or the coastline Structural generalization Point feature generalization
must be moved to ensure that the urban area remains on Simple structural reduction Aggregation
the coast. Such spatial conflicts are difficult both to detect Resampling Displacement
and to correct. The condition of complication is depen- Numerical generalization Line feature generalization
dent on the specific spatial configurations that exist in a Low-pass filters Simplification
High-pass filters Smoothing
defined space. An example is a digital line that changes in
Compass gradient masks Displacement
complexity from one part to the next, such as a coastline
Vegetation indices Merging
(like Maine’s) that progresses from very smooth to very Enhancement
crenulated. In this context, Barbara Buttenfield (1991) Numerical categorization Areal feature generalization
demonstrated the use of line-geometry-based structure Minimum-distance to means Amalgamation
signatures as a means of controlling the tolerance values, Parallelopiped Collapse
based on complexity, in the generalization process. Later, Maximum-likelihood classification Displacement
we provide details on other techniques for detecting Categorical generalization Volume feature generalization
Merging (of categories) Smoothing
changes in linear complexity.
Aggregation (of cells) Enhancement
One issue important to determining when to generalize Nonweighted Simplification
is measuring the amount of generalization that either has Category-weighted Holistic generalization
been done or is planned. Two basic types of measures Neighborhood-weighted Refinement
can be identified: procedural and quality assessment. Attribute change
Procedural measures are those needed to invoke and
control the process of generalization. Such measures might processing. Geographical features are normally repre-
include those that: (1) select a simplification algorithm, sented in either a “vector” or a “raster” format inside of
given a certain feature class; (2) modify a tolerance value a computer. The vector representation uses x-y coordinate
along a feature as the complexity changes; (3) assess the pairs to represent point features such as a house (a single
density of a set of polygons being considered for agglom- x-y coordinate pair), a line feature such as a river (a string
eration; (4) determine whether a feature should undergo a of connected x-y coordinate pairs), or an areal feature
type change (e.g., area to point) due to scale modification; such as a park boundary (a string of x-y coordinate
and (5) compute the curvature of a line segment to invoke pairs in which the first pair matches the last pair). The
a smoothing operation. Quality assessment measures eval- raster approach uses a matrix of cells of a given resolution
uate both individual operations, such as the effect of (e.g., 30 meters) to represent features. Many standard GIS
simplification, and the overall quality of the generalization books (e.g., Lo and Yeung 2007) describe these two data
(e.g., poor, average, excellent). Several studies have structures in more detail. Vector-based operators require
discussed mathematical and geometric measures, includ- more complicated strategies because they operate on
ing those by Buttenfield (1991), McMaster (1986; 1987), strings of x-y coordinate pairs and require complex search-
and Plazanet (1995). ing strategies.The next section will provide a more detailed
discussion of individual vector-based operations (raster-
based operations are not covered in this text). Figure 4 pro-
4 THE FUNDAMENTAL OPERATIONS vides graphic depictions of some key operations.
OF GENERALIZATION
4.2 Vector-Based Operations
4.1 A Framework for the Fundamental Operations
Simplification
In the McMaster and Shea model discussed earlier, the Simplification is the most commonly used generalization
third major component involves the fundamental operator. The concept is relatively straightforward,
operations, or how to generalize. Most of the research because at its most basic level, it involves a “weeding” of
in generalization assumes that the process can be broken unnecessary coordinate data. The goal is to retain as
down into a series of logical operations that can be classi- much of the geometry of the feature as possible, while
fied according to the type of geometry of the feature. For eliminating the maximum number of coordinates. Below,
instance, a simplification operation is designed for linear we provide more detail on the simplification process.
features, whereas an amalgamation operator works
on areal features. Table 1 provides a framework for Smoothing
the operations of generalization, dividing the process Although often assumed to be identical to simplification,
into those activities needed for raster- and vector-mode smoothing is actually a much different process. The
115
16:24:23.
Spatial Operator Original Map Generalized Map

Simplification
Selectively reducing the number of
points required to represent
an object 15 points to represent line 13 points to represent line
Smoothing
Reducing angularity of angles
between lines
Aggregation
Grouping point locations and
representing them as areal objects Sample points Sample areas
Amalgamation
Grouping of individual areal
features into a larger element Individual Small lakes
small lakes clustered
Collapse Airport Airport

Replacing an object’s physical
details with a symbol
representing the object
School
School
Presence of city
City boundary
Merging
Grouping of line features
All railroad yard rail lines Representation of railroad yard
Refinement
Selecting specific portions of an
object to represent the entire object All streams Only major streams
in watershed in watershed
Exaggeration Bay Bay

To amplify a specific portion of
an object
Inlet Inlet
Enhancement
To elevate the message imparted
by the object
Roads cross Roads cross; one bridges the other
Stream Stream
Displacement
Separating objects
Road Road
FIGURE 4 Fundamental operations of generalization. (Courtesy of Philippe Thibault.)
16:24:23.
smoothing operation shifts the position of points to smaller scale. A continuous example is census tract
improve the appearance of the feature (Figure 4). data, where several tracts with similar statistical attrib-
Smoothing algorithms relocate points in an attempt to utes can be joined together.
plane away small perturbations and capture only the most
significant trends of the line (McMaster and Shea 1992). Collapse
As with simplification, there are many approaches for the The collapse operation involves the conversion of
process—a simple classification is provided in Table 2. geometry. For instance, it might be that a complex urban
Research has shown that a careful integration of simplifi- area is collapsed to a point due to scale change, and is
cation and smoothing routines can produce a simplified, resymbolized with a geometric form such as a circle. A
yet aesthetically acceptable, result (McMaster 1989a). complex set of buildings might be replaced by a simple
rectangle—which might also involve amalgamation.
Aggregation
As depicted in Figure 4, aggregation involves merging Merging
multiple point features, such as a cluster of buildings. This Merging involves fusing together groups of linear fea-
process involves grouping point locations and represent- tures, such as parallel railway lines or edges of a river or
ing them as areal units. The critical problems in this oper- stream (Figure 4). Merging can be viewed as a form of
ation are determining the density of points needed to collapse in which an areal feature is converted to a line.
identify a cluster to be aggregated and specifying the A simple solution is to average the two or multiple sides
boundary around the resulting cluster. The most common of a feature, and to use this average to calculate the new
approach is to triangulate the points (i.e., create triangles feature’s position.
among neighboring points) and determine the density of
the triangles (e.g., a grouping of smaller triangles might Refinement
represent a cluster for aggregation) (Jones et al. 1995). Refinement is another form of resymbolization that is
much like collapse (Figure 4). However, refinement is an
Amalgamation operation that involves reducing a multiple set of fea-
Amalgamation is the process of fusing together nearby tures such as roads, buildings, and other types of urban
polygons, a process needed for both noncontinuous and structures to a simplified representation, rather than a
continuous areal data (Figure 4). A noncontinuous conversion of geometry. The key to refinement is that
example is a series of small islands in close proximity complex geometries are resymbolized to a simpler form,
that have size and detail that cannot be depicted at the thus creating a “typification” of the objects. The example
TABLE 2 A classification of algorithms used to smooth cartographic features. (After McMaster and Shea 1992,
Generalization in Digital Cartography, pp. 86–87, copyright Association of American Geographers.)
Category 1: Weighted averaging
Calculates an average value based on the positions of existing points and neighboring points, with only the endpoints
remaining the same; maintains the same number of points as the original line; algorithms can be easily adapted to
different smoothing conditions by adjusting tolerance values (e.g., the number of points used in smoothing); all
algorithms use local or extended processors.
Examples: Three-point moving average
Five-point moving average
Other moving average methods
Distance-weighted averaging
Slide averaging
Category 2: Epsilon filtering
Algorithm uses certain geometrical relationships between the points and a user-defined tolerance to smooth the carto-
graphic line; endpoints are retained, but the absolute number of points generated for the smoothed line is algorithm-
dependent; approaches are local, extended local, and global.
Examples: Epsilon filtering
Brophy algorithm
Category 3: Mathematical approximation
Develop a mathematical function or series of mathematical functions that describe the geometrical nature of the line;
the number of points on the smoothed line is variable and is controlled by the user; retention of the endpoints and of
the points on the original line is dependent on the choice of algorithms and tolerances; function parameters can be
stored and used later to regenerate the line at the required point density; approaches are local, extended local, and
global.
Examples: Local processing: cubic spline
Extended local processing: b-spline
Global processing: bezier curve
117
16:24:23.
of refinement shown in Figure 4 is a selection of part of smaller islands offshore. In the process of scale reduction,
a stream network that depicts the “essence” of the dis- all features would tend to coalesce. The operation of dis-
tribution in a simplified form. placement would pull the features apart to prevent this
coalescence. What is critical in the displacement operation
Exaggeration is the calculation of a displacement hierarchy because one
Exaggeration is one of the more commonly applied gener- feature will likely have to be shifted away from another
alization operations. Often it is necessary to amplify a spe- (Nickerson and Freeman 1986; Monmonier and McMaster
cific part of an object to maintain clarity in scale reduction. 1990). A description of the mathematics involved in
The example in Figure 4 depicts the exaggeration of the displacement can be found in McMaster and Shea (1992).
mouth of a bay, which would close under scale reduction.
The Simplification Process
Enhancement
Most simplification routines utilize complex geometrical
Enhancement is a symbolization change that empha- criteria (distance and angular measurements) to select
sizes the importance of a particular object. For instance, critical points, those points that are significant in defining
the delineation of a bridge under an existing road is the structure of a linear or areal feature. A general classi-
often portrayed as a series of cased lines, which assists fication of simplification methods consists of five
in emphasizing that feature over another. approaches: independent point routines, local processing
routines, constrained extended local processing routines,
Displacement unconstrained extended local processing routines, and
Displacement is perhaps the most difficult of the general- global methods. Independent point routines select coordi-
ization operations because it requires complex measure- nates based on their positions along a line, nothing more.
ment (Figure 4). The problem can be illustrated with a For instance, a typical nth point routine might select every
series of cultural features in close proximity to a complex third point to quickly weed out unnecessary coordinate
coastline. Assume, for example, that a highway and a rail- data. Although computationally efficient, these algo-
road follow a coastline in close proximity, with a series of rithms are crude in that they do not account for the
TABLE 3 A classification of algorithms used to simplify cartographic features. (After

McMaster and Shea 1992, Generalization in Digital Cartography, p. 73, copyright Association
of American Geographers.)
Category 1: Independent point algorithms
Do not account for the mathematical relationships with the neighboring pairs; operate
independently of topology.
Examples: nth point routine
Random selection of points
Category 2: Local processing routines
Utilize the characteristics of the immediate neighboring points in determining significance.
Examples: Distance between points
Angular change between points
Jenks’s algorithm (distance and angular change)
Category 3: Constrained extended local processing routines
Search continues beyond immediate coordinate neighbors and evaluates sections of a line.
Extent of search depends on distance, angular, or number of points criterion.
Examples: Lang algorithm
Opheim algorithm
Johannsen algorithm
Deveau algorithm
Roberge algorithm
Visvalingam algorithm
Category 4: Unconstrained extended local processing routines
Search continues beyond immediate coordinate neighbors and evaluates sections of a line.
Extent of the search is constrained by the geomorphological complexity of the line, not
by algorithmic criterion.
Example: Reumann–Witkam algorithm
Category 5: Global routines
Considers the entire line, or specified line segment; iteratively selects critical points.
16:24:23.
true geomorphological significance of a feature. Local the geomorphological complexity of the line, not by
processing routines utilize immediate neighboring points algorithmic criteria. Finally, global algorithms process the
to assess the significance of the point.Assuming a point to entire line feature at once and do not constrain the
be simplified as xn , yn , these routines evaluate the point’s search to subsections. The most commonly used simplifi-
significance based on its relationship to immediate neigh- cation algorithm—the Douglas–Peucker—takes a global
boring points xn - 1 , yn - 1 , and xn + 1 , yn + 1. This relationship approach and processes a line “holistically.” Details of the
is normally determined by either a distance or an angular Douglas–Peucker algorithm can be found in McMaster
criterion, or both. Constrained extended local processing (1987) and McMaster and Shea (1992), and comparisons
routines search beyond the immediate neighbors of the various algorithms can be found in McMaster
and evaluate larger sections of lines, again, normally (1987). Table 3 provides details on algorithms that can be
determined by distance and angular criteria. Certain found in each of the five categories.
algorithms search around a larger number of points, per- The effect of the Douglas–Peucker algorithm can
haps two, three, or four in either direction, whereas other be seen in Figures 5 and 6, where the algorithm
algorithms use more complex criteria. Unconstrained is applied to Hennepin County, Minnesota, using a
extended local processing routines also search around 350-meter tolerance value. A tolerance value allows the
larger sections of a line, but the search is terminated by user to decide how much detail the algorithm eliminates.
Effects of Uniform Generalization

Douglas-Peucker Algorithm
FIGURE 5 Overview of effects of

Original map Douglas–Peucker simplification in
Hennepin County, Minnesota. (After
Generalized McMaster, R. B., and Sheppard, E.
to 350m (2004) “Introduction: Scale and
Desired Geographic Inquiry,” p. 9. In Scale and
appearance for scale Geographic Inquiry: Nature, Society
Approximately 1:250,000 when and Method, ed. by E. Sheppard and
Critical points lost displayed on a 8.5" x 11" page R. B. McMaster. Courtesy of Blackwell
Publishing.)
119
16:24:23.
Critical points retained
Critical points lost
FIGURE 6 Enlargements of areas

outlined in Figure 5. (Courtesy of
National Historical Geographic
Information System.)
The original spatial data, taken from the U.S. Bureau of A significant problem of the generalization process
the Census TIGER files, is in light gray, whereas the gen- involves the identification of appropriate tolerance
eralized feature is in black. Note that many of the origi- values for simplification. Unfortunately, this is mostly
nal points have been eliminated, thus simplifying the an experimental process in which the user must test and
features. Unfortunately, the effects of this approach—as retest values until an acceptable solution is empirically
with most generalization processes—are not consistent, found. As explained previously, cartographers often
given that the algorithm behaves differently depending turn to measurements to ascertain the complexity of a
on the geometric character or geomorphological signifi- feature and to assist in establishing appropriate toler-
cance of the feature. In areas that are more “natural,” ance values.
such as streams and rivers, the simplification produces a Figure 7 depicts the calculation of one specific mea-
relatively good approximation. For instance, note that a surement, the trendlines, for the Hennepin County data
reasonable set of critical points seems to have been set. The trendlines for a digitized curve are based on a
retained for the northern boundary of Hennepin calculation of angularity, or where the lines change direc-
County, which follows the Mississippi River. However, tion. Where a curve changes direction—for example,
in urban areas in the central portion of the map, where from left to right—a mathematical inflection point
the census geography follows the rectangular street is defined (theoretically, the point of no curvature). The
network, the results are less satisfactory. In many cases, connection of these inflection points, which indicates
logical right angles are simplified to diagonals. the general “trend” of the line, is called the trendline. The
16:24:23.
More Complex:
-Trendline intersects original line
multiple times
-Trendline is shorter
FIGURE 7 Comparison of the origi-

nal unsimplified line and a trendline,
which is an indicator of the degree of
simplification. The image is of the
northern portion of Hennepin
Less Complex: County shown in Figure 5. (Courtesy
-Trendline nearly matches original line
of National Historical Geographic
Information System.)
complexity of a feature can be approximated by looking space. The NHGIS-produced generalization is depicted in
at the trendlines for the entire feature or for the entire Figure 8B. In particular, note the reduction of detail in the
data set. A simple measure of complexity derived from three inset boxes (i, ii, and iii), where the coastline is sim-
the trendline is the trendline/total length of a line, or the plified, islands are eliminated, and the overall complexity
sinuosity of a feature. Along relatively straight line seg- is reduced. The third illustration (Figure 8C) shows a gen-
ments, with little curvilinearity, the trendline will be very eralization produced by the U.S. Bureau of the Census.
close to the curve, and the trendline/total length ratio will Although also a simplified version, this map might also be
be nearly equal to 1.0 (e.g., the relatively straight line considered an overgeneralization of the features in which
near the middle of Figure 7). However, a highly complex too much of the critical geomorphological character has
curve, such as the northern border of Hennepin County, been eliminated. Note in particular inset ii, where the
will deviate significantly from the trendline, and the ratio important peninsula has been eliminated.
will be distinctly less than 1.0. Thus, the greater the dif- A second set of illustrations (Figure 9) depicts the
ference between the actual digitized curve and the trend- NHGIS’s capability of generating generalized boundaries
line, the more complex the feature. for a variety of target scales. The scales selected for
these illustrations are 1:150,000 (Figure 9A), 1:400,000
(Figure 9B), and 1:1,000,000 (Figure 9C), each depicting
5 AN EXAMPLE OF GENERALIZATION the coastline around Charlotte Harbor, Florida. Note that
the main map on the left of each figure superimposes the
To illustrate a practical application of generalization, generalized data over the ungeneralized data at a scale of
we will consider how information is generalized when 1:400,000 for each of the target scales. The related maps
using the National Historical Geographic Information on the right of each figure illustrate how the ungeneral-
Systems (NHGIS) database. Much of this database was ized and generalized data would appear at the target
acquired from U.S. Geological Survey (USGS) digital line scales of 1:150,000, 1:400,000, and 1:1,000,000. Note how
graphs that were digitized at a scale of 1:150,000. Consider the detail along the coastline has been generalized for the
Figure 8, which depicts three different representations of given scale.
coastlines and county boundaries along the Florida Gulf The NHGIS generalization system (Schroeder and
Coast from Tampa Bay northward. Figure 8A shows the McMaster 2007) works as follows: The main processing
NHGIS raw data at a scale of 1:2,000,000. There we can is divided into two parts. The first part eliminates small
see excessive detail along the coast, where lines coalesce areas (e.g., islands, small parts of multipart features, and
and converge, a problem that is created by having far too slivers caused by historical boundary changes) accord-
much information (too many coordinate pairs) for the ing to measures of area and area/perimeter (McMaster
121
16:24:23.
A B C
FIGURE 8 Generalization of the Florida Gulf Coast, assuming a target scale of 1:2,000,000: (A) the NHGIS raw
data (based on USGS digital data at a scale of 1:150,000; (B) an NHGIS-produced generalization; (C) a general-
ization produced by the U.S. Bureau of the Census. (Courtesy of the International Cartographic Association.)
1987). The second part generalizes boundaries in four individual edges, which are boundaries that have
steps: either no neighbors or one neighbor on each side in
each feature class (e.g., a boundary between two
1. Join feature parts that touch each other at only one census tracts or a coastline between a census tract
node in order to simplify topological relationships. and the ocean). Step 3 will therefore remove long,
Conceptually, this is similar to transforming a narrow “extensions” that occur along a single edge,
figure-8 into an hour-glass figure. but will not remove such features if they lie between
2. Apply two simple line simplification algorithms to two connected edges.
remove insignificant points, or those vertices that An important component within each of these steps is
contribute little to the geographical character of the maintenance of correct topology, which requires
a boundary. The first algorithm removes vertices additional operations to prevent generalized bound-
connecting nearly parallel segments, effectively aries from intersecting one another. Additional details
“straightening” lines that are nearly straight any- on the NHGIS generalization process can be found in
way. The second algorithm is the Douglas–Peucker Schroeder and McMaster (2007).
(1973) algorithm, implemented using a low toler-
ance. This is primarily a preprocessing step that has
little visual impact at the target scale but reduces 6 MAPSHAPER: A FREE WEB-BASED
line complexity in order to speed up the following, GENERALIZATION SERVICE
more complex operations.
3. Complete line generalization using an altered ver- MapShaper (http://mapshaper.org/) is a generalization
sion of the Visvalingam–Whyatt (1993) algorithm program that is available for free via the Web. It grew
with many modifications designed to maintain out of Mark Harrower and Matthew Bloch’s (2006)
boundary smoothness, avoid topological conflicts, “frustration with the relatively limited vector general-
and prevent overreduction of small features. ization capabilities and the prohibitive cost of commer-
4. Eliminate node wedges, which are narrow spaces cial GIS software” (p. 23). Some key characteristics of
found where multiple feature edges intersect. MapShaper are as follows: (1) users can automatically
This requires a separate step because the above construct topology; (2) users can specify critical points;
line generalization operations are applied only to (3) users can easily change the level of generalization;
16:24:23.
A
FIGURE 9 Generalization of the coastline along Charlotte Harbor, Florida, using the NHGIS database:
(A) target scale 1:150,000; (B) target scale 1:400,000; and (C) target scale 1:1,000,000. (Courtesy of the
International Cartographic Association.)
123
16:24:23.
(4) users can see the results of generalization immedi- the Douglas–Peucker and the Visvalingam–Whyatt
ately; and (5) users can see the effects of various simpli- methods (note that these are also used in NHGIS), as well
fication algorithms with a mere click of the mouse. as a modified version of the Visvalingam–Whyatt method
Harrower and his colleagues stress that many vector file “that is intended to create linework with an even less
formats do not encode topological relationships. To jagged appearance” (Bloch and Harrower 2006). Given
solve this problem, MapShaper automatically builds any specified level of generalization, these three methods
topology (using an ESRI .shp file type) “so that shared can be compared by simply clicking on the desired
borders and critical shared points are preserved regard- simplification method. To illustrate, Figure 10 compares
less of how much generalization is applied” (p. 24). two screen captures for a portion of the World Countries
Remember that the NHGIS system also emphasized the database using the Douglas–Peucker (Figure 10A) and
importance of topology. Although sophisticated simplifi- the modified Visvalingam–Whyatt (Figure 10B) simplifi-
cation algorithms are intended to select critical points, cation methods. The black line shows the results of apply-
sometimes these algorithms fail to select the points that ing the algorithms, and the gray lines show the original
seem logical from a visual standpoint. To handle this ungeneralized data. Note especially how the two algo-
problem, MapShaper allows users to zoom to a large rithms behave differently along the coast of Florida in the
scale where individual points can be specified as critical southeastern United States. One limitation of this purely
(or, alternatively, critical points can be deselected). visual approach is that it ignores any differences between
The level of generalization within MapShaper can be the lines that might be measured numerically (see section
changed via a slider bar; moreover, this change is instan- 3.2); but there is the advantage of being able to see the
taneous: Using the sample World Countries database of effects of generalization on any desired subarea immedi-
208,444 points provided with MapShaper, generalization ately. We encourage you to experiment with MapShaper
to any desired level can be accomplished in a fraction of a so that you can see the effects of using these different
second. Algorithms provided within MapShaper include algorithms.
FIGURE 10 Using MapShaper to

compare the effects of (A) the
Douglas–Peucker and (B) the modi-
fied Visvalingam–Whyatt simplifica-
tion algorithms. (Courtesy of Mark
Harrower.)
16:24:23.
SUMMARY multiple point features to create an areal feature);

amalgamation (combining individual areal features to
Scale is a fundamental process in geography and car- create a larger areal feature); collapse (replacing an
tography, as it requires the cartographer to select the object’s physical details with a symbol representing the
appropriate information content, given the map’s pur- object); merging (grouping linear features); refinement
pose and intended audience. The processes used to (selecting portions of a feature to represent the entire
manipulate (i.e., change the scale of) the spatial infor- feature); exaggeration (amplifying a specific portion of a
mation are collectively known as generalization. For feature, as in the mouth of a bay); enhancement (chang-
centuries cartographers have tackled the problem of ing the symbolization to emphasize the importance of a
generalization in the manual domain, but conversion to feature, as in using cased lines for a bridge); and
the digital world has created many new challenges. This displacement (moving features apart to prevent coales-
chapter has provided a discussion of the various forms cence). We considered a variety of simplification algo-
of scale including cartographic scale, which is depicted rithms that have been developed, ranging from local pro-
with a representative fraction (RF) such as 1:24,000. It cessing routines that utilize only immediate neighboring
should be noted, however, that geographers and other points to global algorithms that consider an entire linear
spatial scientists often conceptualize scale very differ- feature.The most commonly used algorithm has been the
ently, such as climatologists and ecologists use of the Douglas–Peucker routine, a global algorithm.
term large scale to describe global processes. We provided an example of generalization using
We reviewed major definitions and models of the ongoing work at the National Historical Geographic
generalization process. Generalization models include Information System (NHGIS), housed at the
those by Robinson and his colleagues and those by University of Minnesota. We saw that creating an
McMaster and Shea. The McMaster and Shea model, acceptable generalization at a particular target scale
which was designed for the generalization process in a requires the application of not just one, but a variety,
digital environment, includes three significant compo- of algorithms and a consideration of the linkages
nents: the conceptual objectives, or why to generalize; between lines that constitute areal features (i.e., one
the cartometric evaluation, or when to generalize; and must carefully consider the topology of spatial fea-
the fundamental operations, or how to generalize. tures). We also discussed MapShaper, a free Web-based
In the context of how to generalize, we considered a generalization service. Interestingly, MapShaper uti-
broad range of fundamental operations, including lizes some of the same algorithms the NHGIS system
simplification (“weeding” out unnecessary coordinate uses, including the Douglas–Peucker and the
data, thus retaining the critical points that define a linear Visvalingam–Whyatt routines. A distinct advantage of
or areal feature); smoothing (shifting points to improve MapShaper is the user’s ability to see the results of
the appearance of a feature); aggregation (merging various simplification methods instantaneously.
FURTHER READING
Buttenfield, B. P. (1990) “NCGIA Research Initiative 3: This edited volume explores scale from a variety of perspectives,
Multiple Representations.” http://www.ncgia.ucsb.edu/ including spatial and temporal statistical analysis, multiple-scale data
research/initiatives.html#i3. for the analysis of biophysical phenomena, and landscape ecology.
This NCGIA working paper summarizes the results of Initiative 3, Jensen, J. R. (1996) Introductory Digital Image Processing: A
on multiple representations held in 1989, and includes a compre- Remote Sensing Perspective, 2nd ed. Upper Saddle River, NJ:
hensive bibliography of earlier works on the subject. Prentice Hall.
Douglas, D. H., and Peucker, T. K. (1973) “Algorithms for the Provides a detailed description of many image processing tech-
reduction of the number of points required to represent a dig- niques, including those considered to be part of raster-based gen-
itized line or its character.” The Canadian Cartographer eralization.
10(2):112–123. João, E. M. (1998) Causes and Consequences of Map
A classic paper that reviews the Douglas–Peucker algorithm, Generalization. Bristol, PA: Taylor and Francis.
which has become the standard simplification technique used in Focuses on the quantitative effects of map generalization. The
cartography and GIS. study uses a series of European maps and also provides a method-
Goodchild, M. F., and Quattrochi, D. A. (1997) “Scale, multi- ology for studying the effects of generalization.
scaling, remote sensing, and GIS.” In Scale in Remote Sensing Mackaness, W., Ruas, A., and Sarjakoski, L. T., eds. (2007)
and GIS, ed. by D. A. Quattrochi and M. F. Goodchild, pp. Generalization of Geographic Information: Cartographic
1–11. New York: Lewis. Modelling and Applications. Amsterdam: Elsevier Press.
125
16:24:23.
Reviews the major international research on generalization, W. Mackaness, A. Ruas, and L. T. Sarjakoski, pp. 37–66.
including both theory and applications. Amsterdam: Elsevier Press.
McMaster, R. B., and Lindberg, M. (2003) “The National Reviews the current research in generalization operators and
Historical Geographic Information System (NHGIS).” algorithms.
Proceedings of the 21st International Cartographic Sheppard, E., and McMaster, R. B., eds. (2004) Scale and
Conference, Durban, South Africa, pp. 821–828. Geographic Inquiry: Nature, Society, and Method. Oxford:
Reviews the NHGIS project, which has built a spatial database of Blackwell.
the U.S. Census at the tract and county levels from 1790 to the pre- This edited volume reviews recent research in scale from the
sent. Users can find census tract and county boundaries and human-social, biophysical, and methodological perspectives.
accompanying statistics at http://www.nhgis.org.
Visvalingam, M., and Whyatt, D. (1993) “Line generalization
McMaster, R. B., and Shea, K. S. (1992) Generalization in by repeated elimination of points.” The Cartographic Journal
Digital Cartography. Resource Publications in Geography. 30(1):46–51.
Washington, DC: Association of American Geographers. Reports on a popular simplification method that uses an areal tol-
This monograph reviews the history of generalization, including erance to remove coordinate pairs.
many of the basic models, and provides details of the fundamental Weibel, R. (1995) Cartography and Geographic Information
operations. Systems 22, no. 4 (special issue on “Map Generalization”).
Regnauld, N., and McMaster, R. B. (2007) “A synoptic view of This special issue contains a series of papers on both the data mod-
generalisation operators.” In Generalization of Geographic els for generalization and generalization algorithms. Several of the
Information: Cartographic Modelling and Applications, ed. by papers focus on road generalization.
GLOSSARY
aggregation: a generalization operation that involves the geographic scale: the extent of a study area, such as a neigh-
joining together of multiple point features, such as a cluster of borhood, city, region, or state; here, “large scale” indicates a
buildings. large area (contrast with cartographic scale).
amalgamation: a generalization operation that combines global algorithms: simplification algorithms that process an
nearby polygons (e.g., a series of small islands in close prox- entire line at once (e.g., the Douglas–Peucker method).
imity having a size and amount of detail that cannot be independent point routines: simplification algorithms that
depicted at a small scale). select coordinates based on their position along the line, nothing
cartographic scale: based on the representative fraction; a more (e.g., an nth point routine might select every third point).
large cartographic scale portrays a small portion of the Earth local processing routines: simplification algorithms that uti-
(contrast with geographic scale). lize immediate neighboring points in assessing whether a
collapse: a generalization operation involving a conversion of point should be deleted.
geometry; for instance, a complex urban area might be collapsed merging: a generalization operation involving the fusing of
to a point due to scale change, and symbolized with a circle. groups of line features, such as parallel railway lines or edges
constrained extended local processing routines: simplifica- of a river or stream.
tion algorithms that search beyond the immediate neighbor- refinement: a generalization operation that involves reduc-
ing points and evaluate larger sections of lines. ing a multiple set of features such as roads, buildings, and
critical points: those points that are significant for defining other types of urban structures to a simplified representation
the structure of a linear or areal feature. or “typification” of the objects.
data resolution: indicates the granularity of the data that are representative fraction (RF): expresses the relationship
used in mapping (e.g., 1-meter remote sensing data are con- between map and Earth distances and is expressed as a ratio
sidered high resolution). of map units/earth units (e.g., an RF of 1:25,000 indicates that
displacement: a generalization operation that involves 1 unit on the map is equivalent to 25,000 units on the surface
moving features apart to prevent coalescence (e.g., when a of the Earth).
highway and a railroad follow a coastline in close proximity). simplification: involves the elimination or “weeding” of
enhancement: a generalization operation that involves a unnecessary coordinate data when generalizing a linear or
symbolization change to emphasize the importance of a par- areal feature.
ticular object (e.g., the delineation of a bridge over an existing smoothing: shifts the position of points making up a line to
road is often portrayed as a series of cased lines). improve the appearance of the line.
exaggeration: a generalization operation in which the char- unconstrained extended local processing routines: simpli-
acter of an object is amplified (e.g., exaggerating the mouth of fication algorithms that search large sections of a line, with
a bay that would close under scale reduction). the search terminated by the geomorphological complexity of
generalization: the process of reducing the information the line, not by n algorithmic criterion.
content of maps due to scale change, map purpose, intended
audience, or technical constraints.
16:24:23.
REFERENCES
Bloch, M., and Harrower, M. (2006) “MapShaper.org: A map McMaster, R. B., and Monmonier, M. (1989) “A conceptual
generalization Web service.” Proceedings of AutoCarto framework for quantitative and qualitative raster-mode
2006, Vancouver, Washington, CD-ROM. generalization.” GIS/LIS’89 Proceedings, Volume 2,
Buttenfield, B. P. (1991) “A rule for describing line feature Orlando, FL, pp. 390–403.
geometry.” In Map Generalization: Making Rules for McMaster, R. B., and Shea, K. S. (1992) Generalization in
Knowledge Representation, ed. by B. P. Buttenfield and R. Digital Cartography. Resource Publications in Geography.
B. McMaster, pp. 150–171. London: Longman. Washington, DC: Association of American Geographers.
Carroll, L. (1893) Sylvie and Bruno Concluded. New York: Monmonier, M. (1989) “Geographic brushing: Enhancing
Dover Publications (1988). exploratory analysis of the scatterplot matrix.”
Douglas, D. H., and Peucker T. K. (1973) “Algorithms for the Geographical Analysis 21, no. 1:81–84.
reduction of the number of points required to represent a Monmonier, M. S., and McMaster, R. B. (1990) “The sequen-
digitized line or its character.” The Canadian Cartographer tial effects of geometric operators in cartographic line
10(2):112–123. generalization.” International Yearbook of Cartography
Eckert, M. (1908) “On the nature of maps and map logic.” 30:93–108.
Bulletin of the American Geographical Society 40, Nickerson, B. G., and Freeman, H. R. (1986) “Development
no. 6:344–351. Translated by W. Joerg. of a rule-based system for automated map generalization.”
Harrower, M., and Bloch, M. (2006) “MapShaper.org: A map Proceedings, Second International Symposium on Spatial
generalization web service.” IEEE Computer Graphics and Data Handling, Williamsville, NY: International
Applications 26, no. 4:22–27. Geographical Union Commission on Geographical Data
Sensing and Processing, Seattle, WA, pp. 537–556.
Hudson, J. C. (1992) “Scale in space and time.” In
Geography’s Inner Worlds: Pervasive Themes in Plazanet, C. (1995) “Measurement, characterization and
Contemporary American Geography, ed. by R. F. Abler, classification for automated line feature generalization.”
M. G. Marcus, and J. M. Olson, pp. 280–300. New AUTO-CARTO 12 (Volume 4 of ACSM/ASPRS ’95
Brunswick, NJ: Rutgers University Press. Annual Convention & Exposition Technical Papers),
Bethesda, MD, pp. 59–68.
Jones, C. B., Bundy, G. L., and Ware, J. M. (1995) “Map gener-
alization with a triangulated data structure.” Cartography Raisz, E. (1962) Principles of Cartography. New York:
and Geographic Information Systems 22, no. 4:317–331. McGraw-Hill.
Lo, C. P., and Yeung, A. K. W. (2007) Concepts and Schroeder, J. P., and McMaster, R. B. (2007) “The creation of
Techniques of Geographic Information Systems (2nd ed.). a multiscale National Historical Geographic Information
Upper Saddle River, NJ: Pearson Prentice Hall. System for the United States Census.” Proceedings, 23rd
International Cartographic Conference, Moscow, Russia,
McMaster, R. B. (1986) “A statistical analysis of mathemati- CD-ROM.
cal measures for linear simplification.” The American
Cartographer 13, no. 2:330–346. Sheppard, E., and McMaster, R. B. (eds.). (2004) Scale and
Geographic Inquiry: Nature, Society, and Method. Oxford,
McMaster, R. B. (1987) “The geometric properties of numeri- England: Blackwell.
cal generalization.” Geographical Analysis 19, no.
4:103–116. Thibault, P. (2002) “Cartographic generalization of fluvial
features.” Unpublished PhD dissertation, University of
McMaster, R. B. (1989a) “The integration of simplification Minnesota, Minneapolis, MN.
and smoothing routines in line generalization.”
Cartographica 26, no. 1:101–121. Visvalingam, M., and Whyatt, D. (1993) “Line generalization
by repeated elimination of points.” The Cartographic
McMaster, R. B. (1989b) “Introduction to ‘Numerical Journal 30(1):46–51.
Generalization in Cartography.’” Cartographica 26, no.
1:1–6.
127
16:24:23.
16:24:23.
The Earth and Its
Coordinate System
OVERVIEW
This chapter provides fundamental material related to the England, as the Prime Meridian that would be the origin
shape and size of the Earth and the nature of its coordi- for longitude values dividing the world into the Western
nate system. Knowledge of these concepts is essential for and Eastern Hemispheres.
understanding the material on map projections. Section 3 explains the field of geodesy, the science of
Section 1 introduces the basic characteristics of the understanding and explaining the Earth’s size and shape.
Earth’s coordinate system, which is comprised of an Approximately 2,000 years ago, the Greek Eratosthenes
imaginary network of lines called the graticule. The was able to calculate the circumference of the Earth as
graticule is further composed of lines of latitude and 40,500 kilometers (25,170 miles), a figure not far from our
longitude, which criss-cross the Earth’s surface, allow- present-day value of 40,075 kilometers (24,906 miles).
ing point locations to be uniquely described in terms of Determining the correct shape of the Earth has proved
a sexagesimal system (degrees, minutes, and seconds). more problematic, as it wasn’t until about 300 years ago
The arrangement of this coordinate system is based on that Sir Isaac Newton suggested that the Earth is an oblate
the Earth’s position relative to the sun as well as its spheroid or ellipsoid, bulging at the Equator due to
alignment with celestial objects such as the North Star. centrifugal force. This ellipsoid concept extends to the
This alignment establishes the Equator as the reference reference ellipsoid as a solid body that more accurately
line for latitude values, which start at the Equator and defines the Earth’s shape than a simple spherical assump-
run 90° to the North Pole and 90° to the South Pole. tion. Advances in satellite technology have led to the
On the other hand, the Prime Meridian, passing concept that the Earth is not a smooth, mathematically
through Greenwich, England, serves as the reference definable surface (as described by the reference ellipsoid),
line for longitude values that run 180° east and 180° but due to differences in gravitational forces, can be
west of this line. modeled as a geoid—a shape that the Earth would take on
Section 2 discusses the history of accurately determin- if the world’s oceans were allowed to flow over land, adjust-
ing latitude and longitude. Determining longitude proved ing to the gravitational differences across its surface and
to be particularly difficult because there is no convenient creating a single water body.
reference line dividing the Earth into Eastern and Western
Hemispheres (the location of the Prime Meridian is arbi-
trary). Two events were critical to providing accurate lon- 1 BASIC CHARACTERISTICS
gitude locations. First, in the 1700s John Harrison invented OF THE EARTH’S GRATICULE
the chronometer—a timepiece using spring mechanisms
rather than a pendulum for movement, an important fact Addresses, zip codes, and telephone area codes are just
when sailing at sea because pendulum clocks were unus- some of the many ways that places on the Earth’s surface
able at sea due to the constant rocking motion. The second can be located. Although useful, such approaches are lim-
event was the International Meridian Conference, which ited because they have no uniformity (e.g., each country
established the meridian running through Greenwich, has its own system of postal codes), they locate areas
129
16:27:49.
The Earth and Its Coordinate System
rather than specific points, and they do not completely 1.1 Latitude
cover the Earth (no area codes exist over the world’s
oceans). To overcome these limitations, a system com- Describing a point on the Earth’s surface requires that a
posed of latitude and longitude known as the graticule was location’s latitude and longitude be known with respect
developed and is a critical foundation for understanding to an origin. In the case of latitude, the Equator serves
the concepts on map projections. as a convenient origin because it divides the Earth into
To understand the graticule, we start by recalling that two equal halves. Figure 2A illustrates lines of latitude,
the Earth rotates about its axis, called the axis of rotation, shown as thin gray lines, with the Equator represented
which is aligned with the North Star (Polaris) and passes by the thicker gray line. Because lines of latitude are
through the North and South Poles (Figure 1). parallel to each other, they are often called parallels.
Perpendicular to this axis and positioned halfway Latitude values are reported in angular measurements
between the poles, we can envision an imaginary plane of degrees, minutes, and seconds. Similar in concept to
passing through the Earth’s center. The trace of this plane units of time, this sexagesimal system has a base unit of
on the Earth’s circumference creates a circle called the 60 (the decimal system’s base unit is 10) where each
Equator, which divides the Earth into the Northern and degree is divided into 60 minutes and each minute is
Southern Hemispheres. divided again into 60 seconds. In this system, the ° sym-
It is interesting to note that the alignment of the bol denotes the number of degrees, a single quote (’)
Earth’s axis of rotation with the North Star is not exact indicates minutes, and double quotes (”) specify the
and is purely coincidental. In fact, the North Star is number of seconds. There are 90° of latitude north and
actually a small distance away from the Earth’s axis of south of the Equator (designated as 0°) for a total of
rotation. Moreover, the distance between them is 180° from pole to pole. It is customary to apply the terms
changing from year to year due in part to the Earth’s North and South to designate latitude locations above
change in orbital path and axial tilt. Because of this con- or below the Equator. In some cases, plus (+) and minus
stant change, the North Star will someday no longer (-) signs are attached as a prefix to the degree values,
provide utility as a pole star. indicating latitude locations above and below the
North
Star
Axis of
North Rotation
Pole
Northern
Hemisphere
Direction of Imaginary
Earth’s Plane
Rotation
tor
Equa
Southern
FIGURE 1 The Equator is formed by
Hemisphere
the intersection of an imaginary plane
perpendicular to and bisecting the
South Earth’s axis of rotation, which is
Pole
aligned with the North Star. The
Equator divides the Earth into
Northern and Southern Hemispheres.
16:27:49.
A B C
North North
Pole Pole 160º W 180º 160º E
140º W 140º E
120º W ºN 120º E
80
60º N
100º W 100º E
Prime Meridian
P r i m e M e r id i a n
40º N
80º W 80º E
20º N
Eq
Equator uat
or
0º
60º W 60º E
20º S
40º W 40º E
20º W 0º 20º E
Lines of Latitude Lines of Longitude The Graticule

Lines of latitude run in an Lines of longitude run in a north The graticule is a
east to west direction, but to south direction converging at combination of latitude
are labeled either north the poles, but are labeled and longitude; here we
or south of the Equator. either east or west from the see a 20-degree
Prime Meridian. graticule spacing.
Because lines of latitude
never intersect one another, The term meridian applies to any
they are often referred to line of longitude and can be
as parallels. thought of as a line passing
through both poles dividing the
Earth equally into day and night.
FIGURE 2 Latitude and longitude on the Earth’s surface.
Equator, respectively. Thus, the latitude of the or ⫺77° 02’ 7”. The 180th meridian is directly opposite
Washington Monument can be specified as 38° 53’ 22” N the Prime Meridian and partially coincides with the
or + 38° 53’ 22”. International Dateline (the line dividing days of the
Earth’s rotation). Figure 2C shows the graticule—lines
of latitude and longitude in combination with a spacing
1.2 Longitude
of 20°.
To utilize latitude and longitude when locating
Longitude, using the same sexagesimal system as lati-
positions on the Earth’s surface, two quantities must
tude, measures location east or west of an origin called
be known: the angular distance from the Equator to a
the Prime Meridian, which has an associated value of 0°
given location (the latitude) and the angular distance
longitude coinciding with the Royal Observatory in
from the Prime Meridian to a given location (the lon-
Greenwich, England. In Figure 2B, lines of longitude
gitude). Figure 3 illustrates how such angles are
(also called meridians) are shown as thin gray lines, with
determined. To illustrate, assume the point in ques-
the Prime Meridian represented by the thicker gray
tion is positioned at 60° North latitude and 45° East
line. Lines of longitude run north–south from pole to
longitude. The latitude value of 60° is the angle
pole, but are measured east or west of the Prime
formed between (1) a line passing through the point
Meridian. There are 180° of longitude east and west of
on the Earth’s surface and the center of the Earth,
the Prime Meridian for a total of 360° of longitude. As
and (2) a plane passing through the Equator and the
with latitude, plus and minus signs are attached as a pre-
center of the Earth (the plane of the Equator).
fix to indicate location east or west of the Prime
Latitude is usually symbolized by the Greek letter f
Meridian, respectively.* Thus, the longitude of the
,† as shown in Figure 3A. Similarly, Figure 3B shows
Washington Monument can be specified as 77° 02’ 7” W
that a longitude value of 45° is the angle formed
* The “ + = east” and “ - = w est” convention is not universal.

Different countries and mapping organizations sometimes assign
“ + = w est” and use a complete 360° system of longitude rather †The adoption of f is not universal as a symbol for latitude. The
than dividing the world into two 180° halves. Greek letters c and w are also used to reference latitude.
131
16:27:49.
A B
(60, λ) (φ, 45)

60º N
Latitude
Prime 45º E
φ = 60º Meridian Longitude
0º
λ = 45º
Equator Equator
Determining Latitude Determining Longitude

The angular distance φ The angular distance
between the plane of the λ between the
Equator and a line passing Prime Meridian and
through the point and the meridian in question. FIGURE 3 Determining the angular
the center of the Earth. values of (A) latitude and (B) longi-
tude.
between (1) a plane passing through the point and the plane passes through the Earth’s center. Because
the center of the Earth, and (2) a plane passing lines of longitude are traces of planes that intersect the
through the Prime Meridian. Longitude is usually Earth’s surface and its center, all meridians are great cir-
symbolized by the Greek letter l. The conventional cles. It is important to note, however that not all great cir-
manner of referencing a location on the Earth’s sur- cles coincide with meridians. The Equator is also a great
face is to state the latitude value first and the direc- circle because its plane intersects the Earth’s center, but
tional counterpart (i.e., North or South), then the lon- no other parallels intersect the Earth’s center. Rather, all
gitude value along with its directional counterpart other parallels form circles on the Earth’s surface called
(i.e., East or West). For example, the U.S. population small circles. However, not all small circles coincide with
center for the 2000 Census is located in Phelps a parallel. A small circle results when a plane passes
County, Missouri, at 37° 41’ 49” North latitude and through the Earth’s curved surface, but does not inter-
91° 48’ 34” West longitude.* sect the Earth’s center. Figure 4 illustrates great and
small circles on the Earth’s surface.
1.3 Distance and Directions on the Earth’s Great circles also establish an important relationship
Spherical Surface regarding measurement of directions on the Earth’s sur-
face. In Figure 5 an arc of a great circle is shown as a solid
The arrangement of the parallels and meridians on the black line between points A (Fairbanks, Alaska) and B
Earth’s surface sets up two important relationships: dis- (Tokyo, Japan). As the great circle arc crosses the Earth’s
tances and directions. These relationships are important surface, it also intersects each sequential meridian
to cartographers because various types of maps and the between A and B. At a, b, and c, the angle made between
projections that serve as their frameworks are used for each meridian and the great circle arc is called an azimuth
accurately representing distances and directions. (or direction).† Along a great circle arc, the azimuth at
The shortest distance between any two points is a each meridian intersection constantly changes, which is
straight line. On the Earth’s curved surface, however, the problematic for navigators. For instance, airplanes ideally
shortest distance between two points is an arc of a great fly along great circles, the more direct route, but their
circle. A great circle results from the trace of the inter- flight paths must be constantly adjusted because the
section of any plane and the Earth’s surface as long as
† Azimuths are customarily measured in a clockwise fashion starting with
* The Getty Thesaurus of Geographic Names allows you to look geographic north as the origin and passing through 360°. This definition of
up the latitude and longitude of places throughout the world azimuth is not universally followed. For instance, azimuths can be mea-
(http://www.getty.edu/research /conducting_research/vocabularies/tgn/). sured from magnetic north rather than from geographic north.
16:27:49.
North
Pole
Small Circle
a
A
b
c
a'
Earth’s b'
Center c'
B Great Circle Path
Loxodrome Path
Equa Navigated Path
tor
Small
Great Circle
Circles
South FIGURE 5 A great circle arc from point A to point B creates

Pole a constantly changing angle (or azimuth) along each merid-
ian that the great circle arc intersects. The angle at a is differ-
FIGURE 4 Examples of great and small circles on the Earth’s ent from the angles at b and c. The loxodrome from point A
surface. to point B crosses each meridian at the same angle at a', b',
and c'. The navigated course (thick gray line) results from a
series of short loxodromes that approximate the great circle
path.
azimuths along the great circle arc are also constantly
changing. From a navigation standpoint, having to con-
stantly alter a course is not very practical.
Lines on the Earth’s surface that have constant
direction can also be described. Loxodromes (or rhumb
lines) are special lines that intersect all meridians at
the same angle. A simple example will illustrate this
concept. The Equator and all small circles that coincide
with parallels that intersect meridians at right angles
are loxodromes. A loxodrome can be constructed, how-
ever, that is not aligned with a parallel or meridian; for
North
instance, Figure 6 shows that a line constructed on the Pole φ = 30º
Earth’s surface from A to B so that it intersects all B
meridians at the same constant angle will, if continued
beyond point B at the same angle, spiral toward the φ = 30º
North Pole. This constant angle relationship is due to
the convergence of the meridians at the poles. In a
more practical example, Figure 5 shows the loxodrome
path between points A and B as a dashed line. Note A
that as the loxodrome path intersects each meridian, φ = 30º
the angle at a' is the same as the angle at b' and c'.
Loxodromes are a preferred course for navigation due
to the minimum number of alterations to the course. FIGURE 6 A line drawn from point A to point B crossing
Although navigating along a great circle route across each meridian at a constant angle is called a loxodrome. If
the Earth’s surface is the most direct route, it is not extended, this line will continue to spiral toward the North
practical for navigators to constantly change their Pole.
133
16:27:49.
course. In most instances, the great circle route will be

approximated by a series of short loxodromes that nec-
Sun
essarily lengthen the travel time, but also facilitate the
navigation process.
2 A BRIEF HISTORY OF LATITUDE Mirror B

AND LONGITUDE
Moveable Arm
The concept of latitude can be traced back to early
Mirror A
Greek civilization where Dicaerchus (a student of Eyepiece
Aristotle) placed the first line of latitude on a map of Horizon
the known world running from the Pillars of Hercules
(Strait of Gibraltar) through the Mediterranean Sea.
||
|
|
Other lines of latitude were identified and later placed ||
||
|
|| e ||
on maps according to direct observation of nature. The ||
||| S cal | | | |
|||||||||||||||
seasonal progression of the sun, for example, provided
the latitude values for the Equator, Tropic of Cancer (at
23° 30’ 0” N), and Tropic of Capricorn (at 23° 30’ 0” S).
Expanding on the concept of latitude, Strabo (1st cen- FIGURE 7 The sextant, a device for measuring latitude.
tury BCE) wrote in his Geography a detailed descrip-
tion of the world. His published world map in 18 CE
illustrated the oikumene (inhabitable Earth) that was The North Star facilitates determining latitude posi-
divided into climate regions called climata. The climata tion in the Northern Hemisphere because its location in
zones included the frigid, temperate, and torrid, which the sky is fixed and can easily be found because the Big
were equally placed above and below the Equator and Dipper (Ursa Major) constellation points to it. In the
were aligned with various parallels that ran through Southern Hemisphere, however, there is no South Star.†
notable cities.* Therefore, an alternative method must be used to find
The desire to accurately fix a location on the Earth’s the South Pole, and the Southern Cross constellation is
surface with respect to latitude came easily to early nav- the key. Figure 8 illustrates the Southern Cross and its
igators, but required the development of various instru- relationship to the South Pole. By simply extending a
ments, such as the astrolabe. Developed by the ancient line from the top of the Southern Cross through its bot-
Greeks, the astrolabe’s chief utility was to measure the tom 4.5 times its length, one reaches the Southern
angular height of the sun or a star above the horizon, Celestial Pole; this process works regardless of the time
which provided one’s latitude. The astrolabe relied on of night or season.
the fact that the North Star remained at a fixed position Unlike the Equator for latitude, there is no natural
above the Earth’s surface. By aligning the instrument origin for longitude. This once created considerable dif-
with the horizon and the North Star, the angle (or lati- ficulty in determining longitude positions. Moreover,
tude) of the position could be read from the instrument. there is no fixed celestial body (e.g., the North Star) that
The modern-day sextant (Figure 7) is a direct descen- provides information for determining longitude. Those
dant of the astrolabe. The sextant makes use of two mir- who pursued this problem knew that, similar to latitude,
rors, which allow light to reach the eyepiece, and a a fixed entity needed to be known, and then the value
moveable arm. The observer aligns the sextant with the of that entity at a specific location could be compared
horizon by looking through the eyepiece and mirror A. to the fixed value to arrive at a longitude estimate.
Mirror B is attached to a moveable arm, which, once As a way to solve this problem, attention turned to
the sextant is aligned with the horizon, is positioned so the concept of time. It was known that the Earth makes
that the sun’s rays are reflected off mirror B and pass one complete revolution about its axis in a 24-hour
onto mirror A. The resulting image cast through the period. Thus, each location on the Earth’s surface makes
eyepiece is one object (the sun) superimposed on the a complete 360° rotation during this 24-hour time
other (the horizon). The angle between the two objects period. Through simple division, each hour of rotation
is then read off the scale, giving the latitude of the corresponds to 15° of longitude change. Knowing this,
observer. the task was then to compare the current local time with
* For more information, see http://www.henry-davis.com/MAPS/ †A star (Sigma Octantis) is almost as close to the South Pole as the
AncientWebPages/115mono.html. North Star is to the North Pole, but this star is barely visible.
16:27:49.
should serve as the origin—similar to the Equator

being the origin for latitude. Prior to the mid-1800s,
there was no universally agreed-on meridian that was
assigned 0° longitude. Rather, it was customary for
Length of the
Southern Cross
each country to place the origin of longitude in that
1x nation’s capital. This practice, although demonstrating
2x a sense of national pride, proved incredibly confusing
Southern
3x Seasonal rotation to those who utilized maps of various countries. In an
Celestial Pole
4x path of the
Southern Cross
attempt to set an internationally recognized Prime
4.5x Meridian, 41 delegates from 25 nations met at the
International Meridian Conference in Washington,
DC, in 1884. By the end of the conference, Greenwich
had won the position of being designated 0° longitude
by an overwhelmingly favorable vote. There were two
main reasons for choosing Greenwich. First, the
United States had already chosen Greenwich as the
basis for its own national time zone system. Second, in
FIGURE 8 Determining the location of the South Pole using the late nineteenth century, almost 75 percent of the
the Southern Cross constellation. world’s commerce depended on sea charts, which used
Greenwich as the Prime Meridian. Today, most
nations adopt the meridian running through the
some agreed-on fixed time. The difference in time could London Observatory at Greenwich, England, as the
easily be converted into degrees of longitude locating Prime Meridian.*
the position in question. For example, if a ship carried a One final point about longitude is that the meridian
clock set to London time, it could compare local noon concept also played an important role in the develop-
at its current position with the time on the clock set to ment of the meter as a standard of measure. When, in
London time and simply calculate the time difference 1791, the French Académie Royale des Sciences was
and thus determine its longitude relative to London. instructed to create a new system of units, it was
Although simple in theory, it proved extremely difficult decided that this new system should be set on a base
to develop a reliable timepiece (known as a 10 system and that the fundamental measuring units
chronometer) that would keep accurate time, especially should be based on values that were unchanging. To
while at sea. At the time, pendulum clocks were in com- this end, l’Académie Royale des Sciences calculated
mon use on land, but were unusable on ships due to the the distance of an imaginary line that began at the
constant rocking and the gravitational differences expe- North Pole, ran through Paris, and ended at the
rienced across the oceans. Equator. They then divided this line into exactly 10
In 1714 England established the Board of Longitude million identical pieces. The length of one of these
to promote the development of a practical and useful pieces became the base unit for the new system of
method for finding longitude at sea. The problem was measurement—the meter. This was the beginning of
so pressing that the Board offered a monetary award of the metric system.
£20,000 to anyone who could make a timepiece capable
of longitudinal accuracy within one-half a degree
(2.5 minutes of time) on a journey from England to the 3 DETERMINING THE EARTH’S SIZE AND SHAPE
West Indies. Over the six weeks of the voyage, the clock
would be allowed a daily error of less than three sec- One of the more important historical scientific pursuits
onds, something barely possible for the best clocks on has been determining the Earth’s size and shape—a
land. John Harrison stepped up to the challenge, devel- pursuit that is still ongoing in the field of geodesy. You
oping four different timepieces over nearly 30 years, might think that after 2,000 years of scientific inquiry,
finally arriving at a design he called H4 that satisfied scientists would have a firm grasp on the Earth’s size
the Board of Longitude’s requirements on time accu- and shape. To some extent they do, but the Earth’s inte-
racy. Interestingly, Harrison did not receive the full rior and exterior constantly change, making it difficult
monetary award until after an act of the British to pin down its exact size and shape.
Parliament rewarded him the prize at the age of 79. He
died shortly thereafter. * The Royal Observatory has an interesting Web page full of informa-
Another problem that plagued those desiring pre- tion about the history of the Prime Meridian (see http://www.nmm.
cise longitude was deciding which line of longitude ac.uk/server/show/nav.2857).
135
16:27:49.
3.1 The Earth’s Size L P1

Parallel Rays
to the Sun
There is a rich history on the efforts involved in deriv- Sun’s Rays at
ing estimates of the Earth’s size. One early investigator Alexandria
P2
was Eratosthenes (276–195 BCE), who worked at the φ = 7º 12'
Alexandria library in Egypt. To determine the Earth’s Ki
Alexandria lo 810
circumference, Eratosthenes applied the following for- m
et
mula for calculating the circumference of a circle (C): e rs
Syene
C = 360° . d/f
where d is the distance between two locations on the

Earth’s surface, and f is the angular (or latitude) differ-
ence separating the two locations. Eratosthenes
selected Syene (present-day Aswan, Egypt) and a city
to the north, Alexandria, Egypt, as the two locations
φ = 7º 12'
because these cities were aligned along a meridian,
allowing the distance to be computed more accurately. Earth’s Center
Modern scholars are unsure of how Eratosthenes
arrived at an estimate of distance; some think that the FIGURE 9 Method employed by Eratosthenes to estimate the
distance was an educated guess, whereas others think Earth’s size. Through geometry, a line L that intersects another
he based it on the number of days it took for a camel to line P1 creates an angle f. If that line continues to the Earth’s
walk between the two cities! In any case, he arrived at a center and intersects another line P2 parallel to P 1,the angle
value of approximately 810 kilometers (500 miles). To formed at the second intersection is the same as the angle at the
obtain the angular or latitude difference between these first intersection.
cities, Eratosthenes knew that at noon on a day that has
the greatest amount of daylight in the Northern
between modern-day miles or kilometers and the units
Hemisphere (the Summer Solstice), the sun was directly
Eratosthenes used.* Regardless of his results, or modern-
overhead a local well in Syene. On that same day and
day interpretations thereof, his measurement should be
time, he measured an angle cast by a shadow in
regarded as one of the first to scientifically investigate
Alexandria and calculated it to be 7° 12’. Through a sim-
the Earth’s size.
ple geometric relationship, the angle he measured on
the Earth’s surface at Alexandria was the same as the
3.2 The Earth’s Shape
one that would be measured at the Earth’s center
(Figure 9). Substituting these values into the equation,
Similar to investigations of the Earth’s size, there has been
he arrived at an estimated circumference of 40,500 kilo-
a constant quest to arrive at a definitive description of the
meters (25,170 miles):
Earth’s shape. Recorded evidence of these investigations
comes from various Greek scholars, such as Pythagoras
C = 360° . 810 kilometers/7°12’ = 40,500 kilometers
(6th century BCE), who relied on philosophy to deduce
the spherical nature of the Earth’s surface. Pythagoras
This result is very close to contemporary measurements
argued that the Earth must be a sphere because the
of the Earth’s size—40,075 kilometers (24,906 miles).
sphere was considered the perfect shape and the Greeks
Although a remarkable feat for his time, there is con-
inhabited a perfect world. Other scholars, such as
siderable doubt placed on the accuracy of Eratosthenes’s
Aristotle (4th century BCE) and Archimedes (3rd cen-
measurement, as his method introduced many unac-
tury BCE), made estimates of the Earth’s size through
counted errors. We have already mentioned the problem
direct observation. For instance, Aristotle noticed that as
of determining distance, but there were other problems.
a ship sailed toward the horizon, the ship’s hull disap-
For instance, Syene and Alexandria are not along the
peared first rather than became smaller and smaller.
same meridian—Aswan is located at 32° 53’ 56” E, and
From our everyday experience, we, just as individu-
Alexandria is located at 29° 55’ 09” E—an east–west
als have for over 2,000 years, see evidence suggesting
angular difference of 2° 58’ 47”. We also know that Syene
is located at 24° 5’ 15” North latitude and not along the * The units in which Eratosthenes reported distance were stadia,
Tropic of Cancer, which is at 23° 30’ 0” North latitude, which originated in ancient Greece. See How Many? A Dictionary of
and thus the sun’s rays were not perfectly vertical. Units of Measurements (http://www.unc.edu/~rowlett/units/dictL.
Moreover, there is no agreement on the exact conversion html).
16:27:49.
North Prolate Spheroid Oblate Spheroid

Star
east - west east - west
north - south
north -
south
φ
φ
FIGURE 11 The prolate spheroid, shown on the left, has an

Equator east–west axis that is compressed or flattened relative to the
north–south axis. The oblate spheroid, on the right, shows the
opposite relationship. Both figures are greatly exaggerated to
show the differences.
Angle between
φ Horizon
and North Star developed technique called triangulation.† During their
Direction to survey, the Cassinis encountered the fact that one degree
North Star of latitude in the northern part of France was not the
Earth’s polar axis points Tangent to same length as one degree of latitude in the southern part
toward North Star Horizon
of France. They concluded that the Earth is not a perfect
sphere, but rather a prolate spheroid—an ellipse that
FIGURE 10 Positions on the Earth’s surface suggesting a bulges in the north–south direction and is compressed in
spherical surface. Note that the position of the North Star is the east–west direction (Figure 11).
relative and not to scale. About the same time period, other pieces of evidence
regarding the Earth’s gravitational forces were being
the Earth is a sphere. One simple way to see this shape investigated that would later be paramount in providing
is to note that during a lunar eclipse, the edge of the a more exact description of its shape. For instance, Jean
Earth’s shadow is a circular arc. A more involved proof Richer (1630–1696) examined the periods, or motions, of
is the relative position of the North Star with respect to pendulums at different locations on the Earth. While on
locations on the Earth’s surface. At the Equator, the expedition to Cayenne, French Guyana, he found that a
apparent position of the North Star is very low with pendulum beat more slowly there than it did in Paris.
respect to the horizon,* but as one progresses toward From this, Richer theorized that gravity must be weaker
the North Pole, the apparent position with respect to at Cayenne, and therefore that Cayenne is further from
the horizon rises, until at the North Pole, where the the Earth’s center than Paris, which contradicted the
North Star is directly overhead (see Figure 10). This Cassinis’ prolate spheroid assumption.
phenomenon can be explained only if the Earth has a In concert with Richer and other gravity investiga-
spherical surface—a fact that early navigators utilized tions, Sir Isaac Newton (1642–1727) was promoting his
for locating their latitude in the Northern Hemisphere. laws of gravitational motion. His 1687 treatise stated
that every object in the universe attracts every other
The Prolate versus Oblate Spheroid Controversy object with a force that is equal to the product of their
masses divided by the square of the distance between
The notion of the Earth as a perfect sphere existed for
them. He applied his gravitational concept to the Earth’s
more than 1,000 years after the early Greeks first pro-
posed the idea. The idea was not challenged until the late
† Triangulation is a method of surveying where two points having
1600s, when the Cassini family undertook the first com-
known locations are established and the distance between them is
prehensive, large-scale survey of France through a newly
measured. This line is called the baseline. Next, using one end of the
baseline, an angle is measured to a distant point. At the other end of
* Although there are many different definitions of horizons, the the baseline, an angle is measured to the same distant point closing
meaning here is the general term, which means the apparent or the triangle. Using the angles of this triangle, the length of any
visible junction of the Earth and sky as seen from any position on unknown side might be computed. Other triangles are established
Earth’s surface. from this initial triangle, creating a triangulated network.
137
16:27:49.
E q ua t o r
Semiminor
Axis
b
Semimajor
Axis
a
North a b c
Pole
A Full
Revolution FIGURE 13 The notion of a reference ellipsoid and associ-
ated parameters.
deviates from a circle is called the flattening constant f

FIGURE 12 The notion of centrifugal force is supported by and is computed as:
differing speeds at different latitudes—positions near the
Equator (c) must travel farther and therefore faster than f = (a - b)>a
those near the poles (a).
A two-dimensional ellipse can be rotated about its semi-
shape and suggested that the Earth’s rotating mass minor axis, arriving at a 3-D figure called a reference
creates a centrifugal force that essentially pushes mass ellipsoid, which can be used to describe the Earth’s size
away from its center. He substantiated his assumption and shape. On a reference ellipsoid, the semimajor axis
by noting that different latitudes on the Earth’s surface coincides with the equatorial radius, which is the dis-
travel at different speeds. For example, in Figure 12 we tance from the Earth’s center to a point on the Equator,
can see that each latitudinal position completes one full and the semiminor axis coincides with the polar radius,
rotation per day, but that positions at the Equator have which is the distance from the Earth’s center to the
much farther to travel and therefore must travel faster. North Pole.
Thus, according to Newton’s law, the centrifugal force Armed with the idea of a reference ellipsoid, vari-
must be stronger at the Equatorial regions, which, he ous countries undertook national surveys to create
proposed, creates an oblate spheroid where the Equator ellipsoids that best fit local areas. Figure 14 illustrates
bulges but the poles are flattened (Figure 11).
Whether the Earth is a prolate or an oblate spher-
Earth’s
oid became the focus of a heated debate among the apparent
A poor ellipsoid
scientific community in the early 1700s. To resolve the fit for Germany
elevational
debate, l’Académie Royale des Sciences in Paris orga- surface
nized two expeditions—one to the equatorial region
of Ecuador and another to Lapland (the border b
between Sweden and Finland) to measure the length
of one degree of latitude on the Earth’s surface. The
expeditions took more than nine years to complete, a
but finally reported that one degree of latitude at the
Equator is 111.321 kilometers, whereas at Lapland
the same distance is 111.900 kilometers. Thus, these
findings proved Newton correct—the Earth is an A good ellipsoid fit
for Australia
oblate spheroid.
The fact that the Earth was described as an oblate
Ellipsoid
spheroid provided scientists a new figure to describe
the Earth’s shape in simple mathematical terms.
Basically, an oblate spheroid can be described as an FIGURE 14 A comparison between the Earth’s irregular sur-
ellipse, which has a semimajor axis—a—and a semimi- face and a smooth reference ellipsoid surface. (This illustration
nor axis—b (Figure 13). The degree to which the ellipse is greatly exaggerated to show the effects.)
16:27:49.
TABLE 1 Several common ellipsoids developed for national mapping programs (From NIMA at http://earth-info.nga.mil/
GandG/publications/tm8358.1/8358010.gif ).
Ellipsoid Date of Semimajor Semiminor

Name Inception Axis (meters) Axis (meters) Flattening Where Used
Airy 1830 6,377,563.44 6,356,256.91 1/299.33 Great Britain

Everest 1830 6,377,276.35 6,356,075.42 1/300.8 India
Clarke 1866 6,378,206.4 6,356,583.8 1/294.98 North America
International 1924 6,378,388.0 6,356,911.95 1/297.0 Select areas
Krassovsky 1940 6,378,245.0 6,356,863.03 1/298.3 Soviet Union
WGS 1972 6,378,135.0 6,356,750.5 1/298.26 U.S. Defense Dept.
GRS 1980 6,378,137.0 6,356,752.31 1/298.257 World
WGS 1984 6,378,137.0 6,356,752.31 1/298.257 World
this concept, where a reference ellipsoid’s equatorial latitudes are commonly used for global phenomena,
and polar radii have been defined so as to “fit” such as satellite ground tracking maps.
Australia, and Table 1 lists several prominent histori- On a spherical surface, geodetic and geocentric lati-
cal reference ellipsoids. In total, dozens of reference tudes are computed using the same methods previously
ellipsoids have been defined, and currently there are outlined. Note, however, in Figure 15B that geodetic
more than 30 in use by various countries around the
world.
A Latitude Determination
Reference Ellipsoid and the Graticule on a Reference Ellipsoid
Although defined in general terms at the beginning Polar
of this chapter, the same location can take on differ- Axis
ent latitude values, depending on the shape of the
Earth’s surface used and the map purpose involved.
Figure 15A shows geodetic and geocentric latitudes
on a reference ellipsoid. Geodetic latitude computed φ’ φ
Equatorial
on a reference ellipsoid is measured by an angle that Earth’s Axis
results when a perpendicular line at the surface is Center
drawn toward the Earth’s center. In every instance,
except when the location in question is directly over
the Equator or one of the poles, the perpendicular
line will never intersect the Earth’s center. Rather, φ’ Geocentric Latitude
the line will pass through the equatorial plane at some φ Geodetic Latitude
other location. The angle measured at the intersec-
tion of the line and the equatorial plane is called the B Latitude Determination
geodetic latitude. On the other hand, geocentric on a Sphere
latitude computed on a reference ellipsoid is mea-
Polar
sured by an angle that results when a line at the Axis
Earth’s surface is drawn intersecting the plane of the
Equator at the Earth’s center. Given the same point
on the reference ellipsoid, the geocentric and geo-
detic latitudes will be off by a small amount. * This
φ’ φ Equatorial
amount is significant for creating accurate maps (e.g.,
Earth’s Axis
large-scale topographic maps) of local areas. For Center
instance, surveyors commonly use geodetic latitude
when referencing their survey positions because their
surveys are tied to a specific reference ellipsoid for
the country in which they are surveying. Geocentric φ’ Geocentric Latitude
φ Geodetic Latitude
* The difference between geodetic and geocentric as measured on the

reference ellipsoid reaches a maximum of approximately 11’ 45” at FIGURE 15 A comparison between geocentric and geodetic
45° North and South latitude (Van Sickle 2004). latitudes on (A) a reference ellipsoid and (B) a sphere.
139
16:27:49.
and geocentric latitude computations produce the same TABLE 2 The length of one degree of latitude as
result on a spherical surface—both latitude computa- measured along a meridian based on the GRS80
tions measure the angle that results when a line at the reference ellipsoid
Earth’s surface is drawn intersecting the plane of the Latitude Kilometers Statute Miles
Equator at the Earth’s center. In other words, geodetic
0° 110.574 68.708
and geocentric latitude computations produce the 10° 110.608 68.729
same latitude value when using a spherical model of 20° 110.704 68.789
the Earth. 30° 110.852 68.881
Because the ellipsoid is not perfectly spherical, the 40° 111.035 68.994
length of one degree of geodetic latitude is not con- 50° 111.229 69.115
60° 111.412 69.228
stant—the flattening at the poles means that one
70° 111.562 69.321
degree of latitude here is longer than one degree at 80° 111.660 69.382
the Equator. It is customary to measure one degree of 90° 111.694 69.403
latitude beginning at the Equator and moving toward
Note. Values were computed using INVERSE, a free program available from
a pole, with the measurement centered on a multiple the National Geodetic Survey at http://www.ngs.noaa.gov/TOOLS/
of, for example, 10° (Figure 16A). Table 2 lists lengths Inv_Fwd/Inv_Fwd.html. Values are rounded to the nearest thousandths
place.
of one degree of latitude for the GRS80 reference
ellipsoid at 10° intervals, and the Appendix lists the
lengths of one degree of latitude for the GRS80 refer- Due to the convergence of meridians at the poles,
ence ellipsoid in one-degree increments. the length of one degree of longitude becomes progres-
On the sphere, distance between successive degrees sively smaller as one travels from the Equator to the
of latitude is constant and is derived by first calculating poles, regardless of whether we assume a spherical
the Earth’s circumference and then dividing the result Earth. This notion is illustrated in Figure 16B. Table 3
by 360°. So, assuming the Earth to have a circumference lists lengths of one degree of longitude for the GRS80
of 40,075 kilometers (24,906 miles), the length of one reference ellipsoid, and the Appendix lists the lengths
degree of latitude is approximately 111.2 kilometers of one degree of longitude for the GRS80 reference
(69.1 miles). ellipsoid in one-degree increments.
A B
111.694 km 0 km
90º N 0 mi 90º N
69.403 mi
111.412 km 55.800 km
60º N 60º N
69.228 mi 34.673 mi
110.852 km 96.486 km FIGURE 16 (A) The different

30º N 30º N
68.881 mi 59.954 mi
lengths of one degree of latitude as
measured along the meridian cen-
tered at 0° 30’ West longitude, on
the GRS80 reference ellipsoid. (B)
The different lengths of one degree
110.574 km 111.320 km of longitude as one travels from the
0º 0º
68.708 mi 1º W 69.171 mi Equator to the North Pole as com-
0º 1º W 0º
puted on the GRS80 reference
ellipsoid.
16:27:49.
TABLE 3 Lengths of one degree of longitude as

measured along a specific parallel
Longitude Length in Kilometers Statute Miles Satellite’s
Orbit
0° 111.320 69.171 North
10° 109.639 68.127 Pole
20° 104.647 65.025
30° 96.486 59.954
40° 85.394 53.061 Earth’s
50° 71.696 44.550 Rotation
60° 55.800 34.673
Meridians
70° 38.187 23.728
80° 19.394 12.051 E qu
ato r
90° 0 0
Node Node
Note. Values were computed using INVERSE, a free program available from 2
the National Geodetic Survey at http://www.ngs.noaa.gov/TOOLS/ 1
Inv_Fwd/Inv_Fwd.html. Values are rounded to the nearest thousandth
place. South
Pole
The Geoid
Up to the late 1950s, suitable reference ellipsoids were
developed on the Earth’s surface utilizing surveying
instruments that required visual contact between the FIGURE 17 The precession of the nodes. Due to Earth’s
instruments. As such, these surveys were very localized elliptical nature, the orbital path of a satellite is in the oppo-
in nature, and thus no single reference ellipsoid could site direction of Earth’s gravity, causing the satellite’s path to
describe the entire Earth in very precise terms. In addi- predictably cross a different meridian with each orbit (after
tion, each new reference ellipsoid had its own center, Maling 1992).
which made it difficult for other countries to adopt
existing reference ellipsoids. Satellite measurements As satellite measurements have continued, a more
have changed this, as models describing the Earth’s size descriptive and informative term has been used to
and shape are now computed based on the Earth’s cen- describe the Earth’s shape—the geoid (meaning Earth-
ter of mass for the whole planet. This new idea allows a like). Understanding the geoid concept is somewhat dif-
single reference ellipsoid to be used that gives a better ficult because it is not directly observable. Generally
fit to the entire Earth than previously possible. speaking, the geoid is defined as a smooth, undulating
The first U.S. satellite measurements of the Earth’s surface the Earth would take on if the oceans were
shape were from the National Aeronautical and Space allowed to flow freely over the continents—without cur-
Administration’s (NASA) Vanguard rocket program in rents, tides, waves, and so on—which would create a sin-
the late 1950s. If the Earth was perfectly spherical, satel- gle, undisturbed water body. This water body would then
lites would operate in an orbital path around the Earth, be free to adjust to differences in the gravitational forces
and that path would coincide with a plane. Deviations (i.e., centrifugal force caused by the Earth’s rotation) as
from this plane due to variations in the Earth’s gravita- well as the uneven distribution of the Earth’s mass that
tional forces resulted in a predictable motion called exists at its surface. Thus, this new surface would have
precession of the nodes. Figure 17 illustrates this con- undulations (i.e., peaks and valleys) that reflect the influ-
cept. The amount of change that occurred with each ence of gravity. One can think of this “new” surface as
orbital path was measured very precisely, providing an coinciding with mean sea level, which varies in height by
indication of the gravitational forces acting on the some 100 meters (328 feet) worldwide.
Earth’s surface and thus a measure of the Earth’s shape. Unlike the reference ellipsoid, the geoid is impossi-
In fact, the Vanguard rocket program showed a slight ble to define in simple mathematical terms. To compute
bulge in the Northern Hemisphere’s polar areas and a a geoidal surface requires knowing, among other para-
slight depression around the Southern Polar areas, giv- meters, the specific nature of the gravitational forces
ing rise to the concept of a pear-shaped Earth. The that exist across the Earth’s surface. These forces, called
North Pole was reported to be about 10 meters (32.8 gravitational anomalies, are not evenly distributed, nor
feet) farther from the Equator than previously thought, are they equal in their intensity, because the Earth’s
and the South Pole was found to be resting approxi- crust contains different rock densities (e.g., metamor-
mately 30 meters (98.4 feet) closer to the Equator than phic and igneous rocks are more likely to contain the
an elliptical Earth model suggested (Maling 1992, 14). iron-rich mineral magnetite and thus influence gravity
141
16:27:49.
point. This single point defines the origin of a datum. By

definition, a datum is any quantity that serves as a base
Ear
th’s value (a value of zero) by which other values are refer-
Su enced. A geodetic datum, then, is a model that describes
R e f e re rfa
nce ce the location, direction, and scale relationships with
Elli
ps Perpendicular respect to an origin on the Earth’s surface. With respect
oid
’s to Geoid’s
S ur to geodesy, there are two types of datums: horizontal
fa Surface
and vertical. A horizontal datum specifies locations in
ce
terms of geodetic latitude and longitude as computed
Perpendicular to from a chosen reference ellipsoid relative to an origin,
Reference Ellipsoid’s
90º
which coincides with the Equator for latitude values
Surface 90º and the Prime Meridian for longitude values. The hori-
Ge
zontal datum provides a precise description of the para-
oid’
Reference
meters for the reference ellipsoid so that the semimajor
s Su rf
Ellipsoid’s Geodetic axis (equatorial radius) and flattening constant for the
Center Latitude Earth are known, thus providing an accurate means to
ac e
establish latitude and longitude. A vertical datum pro-
Earth’s Center vides a base value that enables elevations to be deter-
mined, and as mentioned earlier is tied to the concept
of mean sea level, which relates to the geoid. Although
FIGURE 18 The relationship among surfaces representing these two datum concepts have been historically
the Earth, a reference ellipsoid, and a geoid. Note that when separate because the surfaces to which they refer are
determining geodetic latitude, a line perpendicular to the different (i.e., reference ellipsoid and geoid), they are
reference ellipsoid is not perpendicular to the geoid. commonly tied together today because they are con-
ceptually related. Choosing an appropriate datum is
necessary, for example, in surveying, to accurately pro-
to a greater extent than sedimentary rocks), but they
vide latitude and longitude based on a specific horizon-
can be measured and mapped. At every location on the
tal datum and a derived elevation value determined
Earth’s surface, a line can be placed so that it is perpen-
through the geoid.
dicular to the local gravitational anomaly. As shown in
Historically, the United States used the Clarke
Figure 18, the line perpendicular to the geoid will not
Ellipsoid of 1866 as the reference ellipsoid and the
coincide with a line that is perpendicular to the chosen
geoid whose origin is located at Meades Ranch in
reference ellipsoid’s surface. A geoidal surface (or
Kansas (located near the geographic center of the
equipotential* surface) can be created from these per-
United States) for its large-scale topographic mapping
pendicular gravitational measurements. In general
program. Taken together, the Clarke 1866 reference
terms, the Earth’s crust is thicker over continents, where
ellipsoid and the geoid origin of Meades Ranch com-
the geoid typically rises, but falls over the oceans, where
prise the North American Datum of 1927 (NAD27).
the crust is thinner. However, even the smooth undula-
However, since the advent of satellite geodesy, large
tion in the geoid surface has minor highs and lows that
errors have been encountered when comparing the
can rise as much as 60 meters (196.8 feet) in some areas.
results of today’s highly accurate survey techniques
Color Plate 1 shows one geoid model for the contermi-
with NAD27 values. To remedy this discrepancy, a new
nous United States developed by the National Geodetic
reference ellipsoid, the Geodetic Reference System
Survey.
(GRS), was recommended by the International
Association of Geodesy and subsequently adopted by
Geodetic Datum the National Geodetic Survey,† creating the North
The relationship between the geoid and a chosen refer- American Datum of 1983 (NAD83). Table 4 contrasts
ence ellipsoid is important for a variety of mapping NAD27 and NAD83 on several key parameters.
applications, especially large-scale mapping projects NAD83 utilizes the Geodetic Reference System of 1980
such as aeronautical charting. Because the geoid is a as the reference ellipsoid, and the geoid is now centered
complex undulating surface while the reference ellip-
soid is a smooth, regular, mathematically defined sur- † See the National Geodetic Survey (NGS) home page (http://www.
face, the two typically do not coincide, except at one ngs.noaa.gov/) for further information on this agency and its activities.
For more detailed documents concerning horizontal and vertical
* The term equipotential surface is defined as a surface having the datums, geodesy, and geodetic models, see the listing of NGS publica-
same potential of gravity everywhere. tions at http://earth-info.nga.mil/GandG/publications/.
16:27:49.
TABLE 4 A listing of parameter values comparing the would be represented by a line 12.75 centimeters
reference ellipsoids of NAD27 and NAD83 (7.9 inches) long. At this size, the difference between
Datum Parameters NAD27 NAD83 the equatorial and polar radii of the GRS80 reference
ellipsoid parameters would be about 0.21 mm. Because
Reference ellipsoid Clarke 1866 Geodetic Reference
0.21 mm is also the approximate width of a line that can
name System 80
Semimajor axis (a) 6,378,206.4 6,378,137.0 be used to represent fine detail on maps, the flattening
meters meters would be hidden by the width of the pen used to draw
Semiminor axis (b) 6,356,583.8 6,356,752.3 world outlines. So, the selection of a more accurate
meters meters model of the Earth using the GRS80 reference ellipsoid
Flattening 1/294.9786982 1/298.2572221 would not produce a visual difference in the final
Intended use North America Worldwide
appearance of the map when compared to the spherical
Note: Values taken from Snyder (1987). model. This is an important conclusion because it per-
mits the assumption that the Earth can be regarded as
at the Earth’s center of mass. As a result of changing spherical for thematic applications. This assumption
from NAD27 to NAD83, most locations within North drastically simplifies the mathematics involved when
America have new latitude and longitude values. This creating a map projection.
readjustment has resulted in the publication of new
coordinate data for approximately 250,000 geodetic SUMMARY
control survey points throughout the United States,
yielding coordinate shifts that exceed 400 meters
(1,312.3 feet) in some parts of the country.* In this chapter, we have examined the basic concepts
behind the size and shape of the Earth and its geo-
graphic coordinate system. We learned that the
Geodetic Datums and Thematic Cartography graticule serves as the framework for the imaginary net-
Although reference ellipsoids, geoids, geodetic datums, work of latitude and longitude lines, allowing any point
and geodesy are certainly worthwhile topics of investi- location on the Earth’s surface to be uniquely identi-
gation, they are limited in significance to large- and fied. The Equator serves as the dividing line for the
medium-scale mapping applications that involve accu- Northern and Southern Hemispheres and references all
rate measurements of distances, directions, or areas. For latitude locations north or south of that line. On the
example, a ship’s navigator uses a nautical chart to accu- other hand, the Prime Meridian divides the Earth into
rately plot a ship’s intended course so that its precise Eastern and Western Hemispheres and is the reference
location can be known at any instant. In the realm of line for all longitude locations east or west of that line.
thematic mapping, the focus is on examining spatial pat- Although the graticule is important for referencing spa-
terns and distributions, and so knowledge of precise tial activity that takes place on the Earth’s surface, a
direction, distance, and area is not critical. more important concept, especially with regard to map-
To illustrate the limited applicability of datums to making, is the changes to the graticule’s appearance
thematic mapping, consider a small-scale thematic map- that occur as a result of map projection.
ping exercise that requires symbolizing data for each The Earth’s size and shape are important considerations
country of the world. Assume for this problem that the in determining locational accuracy. Historically, beginning
model of the Earth is the GRS80 reference ellipsoid as with the work of Eratosthenes, the Earth’s shape evolved
opposed to a spherical model of the Earth. In this case, from a simple spherical assumption to a more complex
the actual difference in length between the two refer- shape called an oblate spheroid. This figure of the Earth
ence ellipsoids’ semiaxes is approximately 21.3 kilome- was developed after Newton proposed his gravitational
ters (13.2 miles), or the circumference associated with laws in which a rotating body produces centrifugal force,
the polar axis is about 135 kilometers (83.3 miles) causing a bulging of the Earth at the Equator.
smaller than that for the equatorial axis When creating a model of the Earth, geodesists fre-
(40,075 - 39,940 = 135km), and the flattening is very quently utilize a reference ellipsoid, which is a
close to 1/298. If you were to draw a page-sized map of smooth, mathematically definable figure that flattens
the world with the GRS80 parameters, the Equator at the poles and bulges at the Equator. Numerous ref-
erence ellipsoids have been developed, each defined
* The USGS and NGS have collaborated to provide these tables for
by different parameters for the semimajor and semi-
the determination of the datum shifts as they relate to the use of
minor axes, with the objective of trying to accurately
mapping and charting products at scales of 1:10,000 and smaller (U.S. map specific regions of the Earth. In recent years, the
Geological Survey 1989). reference ellipsoid concept has been updated to
143
16:27:49.
reflect discoveries of gravitational variation across The discussions of the graticule and the Earth’s size
the Earth’s surface. This variation has been modeled and shape within this chapter are important to all facets
as a geoid, the two-dimensional curved surface the of cartography. The graticule serves as the framework
Earth would take on if the oceans were allowed to around which locations on the Earth’s surface are
flow freely over the continents. A datum combines the recorded and subsequently analyzed with maps. The
geoid and a specific reference ellipsoid to produce a Earth’s size and shape are equally important to map-
reference for horizontal locations defined by latitude ping. Although the reference ellipsoid, geoid, and
and longitude (established by the reference ellipsoid) datum concepts are more appropriate to large-scale
and vertical elevations (defined by the geoid). The topographic mapping applications, every cartographer
North American Datum of 1927 (NAD27) served as needs to be aware of these concepts’ importance, espe-
the basis on which the U.S. topographic mapping pro- cially given the increased reliance on and utilization of
gram was established, but has been updated to digital spatial data.
NAD83 to reflect the contributions of gravitational
measurements made by satellites.
FURTHER READING
Ariel, A., and Berger, N. A. (2006) Plotting the Globe: Stories National Oceanic and Atmospheric Administration. (1989)
of Meridians, Parallels, and the International Date Line. North American Datum of 1983. Washington, DC: U.S.
Westport, CT: Praeger Publishers. Department of Commerce.
An interesting historical overview of the human endeavors and A comprehensive overview of the North American Datum
controversies that resulted in the establishment of meridians, par- of 1983.
allels, and the International Date Line. National Ocean Service. (1983) Geodesy for the Layman. 5th
Bowditch, N. (1995) The American Practical Navigator. ed. Washington, DC: National Oceanic and Atmospheric
Bethesda, MD: National Imagery and Mapping Agency. Administration. Available at http://earth-info.nga.mil/GandG/
Chapter 2 provides an overview of geodesy, datums, and map publications/geolay/toc.html.
projections. A classic nonmathematical treatment of geodesy. Many illustra-
Brown, L. A. (1990) The Story of Maps. Dover, DE: Dover tions and simple explanations provide the novice with a clear
Publications. understanding of geodesy.
A historical account focusing on maps and the people who Robinson, A. H., Morrison, J. L., Muehrcke, P. C., Kimerling,
made them, from the earliest Greeks through the European A. J., and Guptill, S. C. (1995) Elements of Cartography. 6th
Renaissance. ed. New York: Wiley.
Defense Mapping Agency. (1981) Glossary of Mapping, Chapter 4 presents an overview of basic geodesy.
Charting, and Geodetic Terms. 4th ed. Washington, DC: U.S. Smith, J. R. (1997) Introduction to Geodesy: The History and
Department of Defense. Concepts of Modern Geodesy. New York: Wiley.
A glossary of cartographic terms relating to geodesy, datums, and Provides readers with a basic overview and understanding of geo-
map projections. desy without the complex mathematical formulas and equations
Iliffe, J. C. (2000) Datums and Map Projections. Boca Raton, necessary to actually practice geodesy.
FL: CRC Press. Snyder, J. P. (1987) Map Projections: A Working Manual.
Chapters 1 through 6 cover aspects of the geoid, datums, and posi- Washington, DC: United States Geological Survey.
tioning on the Earth. Although much of the first half of the text is A comprehensive overview of map projections including worked-
readable for the introductory student, a grasp of mathematics is out examples of mathematical equations for many common map
essential to make full use of this text. projections.
Maling, D. H. (1992) Coordinate Systems and Map Projections. Sorbel, D. (1995) Longitude: The True Story of a Lone Genius
New York: Pergamon. Who Solved the Greatest Scientific Problem of His Time. New
A readable book for the introductory student; the focus is on all York: Viking Penguin.
aspects of map projections. Chapters 1 through 4 cover the basics Presents an overview of basic science, cultural history, and person-
of the Earth’s shape and size. ality conflicts that encompassed John Harrison during his efforts
National Imagery and Mapping Agency (NIMA). (1997) to solve the question of longitude determination.
Department of Defense World Geodetic System 1984, Its United States Geological Survey. (1989) North American
Definition and Relationships With Local Geodetic Systems, Datum of 1983, Map Data Conversion Tables. Denver, CO:
Technical Report TR8350.2 Third Edition. Washington, DC: United States Geological Survey.
U.S. Department of Defense.
These conversion tables, produced cooperatively by the U.S.
A discussion of the development of the World Geodetic System. Geological Survey and the National Oceanic and Atmospheric
16:27:49.
Administration, are designed to allow users of maps and charts based Woodward, D., and Lewis, M. (1998) The History of
on the NAD27 to convert scaled or digitized information to NAD83. Cartography: Cartography in the Traditional African, American,
Van Sickle, J. (2004) Basic GIS Coordinates. Boca Raton, FL: Arctic, Australian, and Pacific Societies. Vol. 2. Chicago:
CRC Press. University of Chicago Press.
A detailed examination of ellipsoids, geoids, datums, map projec- Takes a detailed and critical look at the history of cartography, its
tions, and local coordinate systems. technologies, and advances made in the African, American, Arctic,
Australian, and Pacific societies. Also see Woodward and Harley
Wilford, J. N. (1982) The Mapmakers: The Story of the Great
(1987; 1994) and Harley and Woodward (1992).
Pioneers in Cartography from Antiquity to the Space Age. New
York: Vintage Books.
Presents a history of cartography from antiquity to the space age,
focusing on technology’s impacts on cartography.
APPENDIX: LENGTHS OF ONE DEGREE LATITUDE AND LONGITUDE

This appendix provides two tables listing the lengths of one value using conversions where 1 meter = 0.00062137119
degree of latitude and longitude. Table A-1 lists lengths of statute mile and 1 meter = 0.0005399568 nautical mile.
one degree of latitude as measured along any meridian at The lengths of one degree of latitude and longitude were
one-degree intervals. Table A-2 lists the lengths of one de- derived through a Visual Basic 6.0 program using double
gree of longitude at one-degree intervals as measured along precision, but have been rounded here to two decimal
a given parallel. In both tables, the lengths in statute miles places. The equations used in the program were provided
and nautical miles were derived directly from the meter courtesy of daan Strebe.
TABLE 1 Lengths of one degree of latitude along a meridian at one-degree intervals

Latitude Statute Miles Nautical Miles Meters Latitude Statute Miles Nautical Miles Meters
0 68.7077 59.7053 110,574.28 28 68.8600 59.8377 110,819.45

1 68.7079 59.7055 110,574.61 29 68.8702 59.8465 110,835.76
2 68.7085 59.7061 110,575.63 30 68.8805 59.8555 110,852.44
3 68.7096 59.7070 110,577.32 31 68.8911 59.8647 110,869.46
4 68.7110 59.7082 110,579.68 32 68.9019 59.8741 110,886.81
5 68.7129 59.7099 110,582.71 33 68.9128 59.8836 110,904.46
6 68.7152 59.7119 110,586.41 34 68.9240 59.8933 110,922.39
7 68.7179 59.7142 110,590.77 35 68.9353 59.9031 110,940.57
8 68.7210 59.7169 110,595.79 36 68.9467 59.9131 110,959.00
9 68.7246 59.7200 110,601.45 37 68.9583 59.9231 110,977.64
10 68.7285 59.7234 110,607.76 38 68.9700 59.9333 110,996.48
11 68.7328 59.7272 110,614.71 39 68.9818 59.9436 111,015.48
12 68.7375 59.7313 110,622.29 40 68.9937 59.9539 111,034.63
13 68.7426 59.7357 110,630.49 41 69.0057 59.9643 111,053.91
14 68.7481 59.7404 110,639.29 42 69.0177 59.9748 111,073.28
15 68.7539 59.7455 110,648.70 43 69.0298 59.9853 111,092.74
16 68.7601 59.7509 110,658.69 44 69.0419 59.9958 111,112.24
17 68.7667 59.7566 110,669.26 45 69.0541 60.0064 111,131.78
18 68.7736 59.7626 110,680.39 46 69.0662 60.0169 111,151.32
19 68.7809 59.7689 110,692.07 47 69.0784 60.0275 111,170.84
20 68.7885 59.7755 110,704.29 48 69.0905 60.0380 111,190.32
21 68.7964 59.7824 110,717.03 49 69.1025 60.0485 111,209.74
22 68.8046 59.7896 110,730.27 50 69.1145 60.0589 111,229.06
23 68.8131 59.7970 110,744.01 51 69.1265 60.0693 111,248.28
24 68.8220 59.8047 110,758.22 52 69.1383 60.0796 111,267.35
25 68.8311 59.8126 110,772.89 53 69.1501 60.0898 111,286.27
26 68.8405 59.8207 110,787.99 54 69.1617 60.0999 111,305.00
27 68.8501 59.8291 110,803.52 55 69.1732 60.1099 111,323.53
Continued
145
16:27:49.
TABLE 1 Continued
Latitude Statute Miles Nautical Miles Meters Latitude Statute Miles Nautical Miles Meters
56 69.1846 60.1198 111,341.83 74 69.3501 60.2636 111,608.25
57 69.1958 60.1295 111,359.88 75 69.3564 60.2691 111,618.38
58 69.2069 60.1391 111,377.65 76 69.3624 60.2743 111,627.93
59 69.2177 60.1486 111,395.13 77 69.3679 60.2791 111,636.87
60 69.2284 60.1578 111,412.29 78 69.3731 60.2836 111,645.19
61 69.2388 60.1669 111,429.11 79 69.3779 60.2877 111,652.88
62 69.2491 60.1758 111,445.57 80 69.3823 60.2915 111,659.94
63 69.2591 60.1845 111,461.66 81 69.3863 60.2950 111,666.35
64 69.2688 60.1929 111,477.34 82 69.3898 60.2981 111,672.11
65 69.2783 60.2012 111,492.61 83 69.3930 60.3009 111,677.21
66 69.2875 60.2092 111,507.44 84 69.3958 60.3033 111,681.64
67 69.2964 60.2170 111,521.81 85 69.3981 60.3053 111,685.40
68 69.3051 60.2245 111,535.71 86 69.4000 60.3070 111,688.49
69 69.3134 60.2317 111,549.12 87 69.4015 60.3083 111,690.89
70 69.3214 60.2387 111,562.03 88 69.4026 60.3092 111,692.60
71 69.3291 60.2454 111,574.40 89 69.4032 60.3097 111,693.64
72 69.3365 60.2518 111,586.24 90 69.4034 60.3099 111,693.98
73 69.3435 60.2578 111,597.53
TABLE 2 Lengths of one degree of longitude along a parallel at one-degree intervals

Parallel Statute Miles Nautical Miles Meters Parallel Statute Miles Nautical Miles Meters
0 69.1707 60.1077 111,319.49 33 58.0691 50.4607 93,453.21
1 69.1603 60.0986 111,302.65 34 57.4052 49.8838 92,384.79
2 69.1289 60.0713 111,252.13 35 56.7238 49.2917 91,288.17
3 69.0766 60.0259 111,167.95 36 56.0251 48.6845 90,163.69
4 69.0034 59.9623 111,050.13 37 55.3093 48.0625 89,011.67
5 68.9093 59.8805 110,898.71 38 54.5766 47.4257 87,832.46
6 68.7943 59.7806 110,713.72 39 53.8272 46.7745 86,626.40
7 68.6585 59.6626 110,495.23 40 53.0613 46.1090 85,393.86
8 68.5020 59.5266 110,243.28 41 52.2792 45.4294 84,135.19
9 68.3247 59.3726 109,957.97 42 51.4811 44.7358 82,850.76
10 68.1267 59.2005 109,639.36 43 50.6672 44.0286 81,540.97
11 67.9081 59.0106 109,287.56 44 49.8378 43.3079 80,206.19
12 67.6690 58.8027 108,902.65 45 48.9932 42.5739 78,846.84
13 67.4093 58.5771 108,484.76 46 48.1335 41.8268 77,463.30
14 67.1292 58.3337 108,033.99 47 47.2590 41.0670 76,056.00
15 66.8288 58.0726 107,550.49 48 46.3700 40.2945 74,625.35
16 66.5081 57.7939 107,034.39 49 45.4668 39.5096 73,171.79
17 66.1672 57.4977 106,485.83 50 44.5497 38.7126 71,695.75
18 65.8063 57.1841 105,904.98 51 43.6188 37.9037 70,197.68
19 65.4254 56.8531 105,292.01 52 42.6745 37.0832 68,678.02
20 65.0247 56.5049 104,647.09 53 41.7171 36.2512 67,137.23
21 64.6042 56.1395 103,970.40 54 40.7469 35.4081 65,575.77
22 64.1641 55.7571 103,262.15 55 39.7641 34.5541 63,994.13
23 63.7046 55.3577 102,522.54 56 38.7691 33.6894 62,392.77
24 63.2256 54.9416 101,751.77 57 37.7621 32.8144 60,772.19
25 62.7275 54.5087 100,950.09 58 36.7435 31.9292 59,132.86
26 62.2103 54.0592 100,117.71 59 35.7135 31.0342 57,475.30
27 61.6741 53.5934 99,254.89 60 34.6725 30.1296 55,800.00
28 61.1192 53.1112 98,361.87 61 33.6208 29.2157 54,107.48
29 60.5457 52.6128 97,438.91 62 32.5588 28.2928 52,398.25
30 59.9538 52.0984 96,486.28 63 31.4866 27.3611 50,672.82
31 59.3436 51.5682 95,504.26 64 30.4048 26.4210 48,931.74
32 58.7153 51.0222 94,493.14 65 29.3135 25.4727 47,175.53
Continued
16:27:49.
TABLE 2 Continued
Parallel Statute Miles Nautical Miles Meters Parallel Statute Miles Nautical Miles Meters
66 28.2132 24.5166 45,404.73 79 13.2412 11.5063 21,309.60

67 27.1041 23.5529 43,619.88 80 12.0506 10.4716 19,393.49
68 25.9867 22.5818 41,821.53 81 10.8562 9.4338 17,471.35
69 24.8612 21.6038 40,010.23 82 9.6585 8.3930 15,543.78
70 23.7280 20.6191 38,186.54 83 8.4577 7.3496 13,611.39
71 22.5875 19.6280 36,351.02 84 7.2544 6.3039 11,674.77
72 21.4399 18.6308 34,504.24 85 6.0488 5.2562 9,734.52
73 20.2858 17.6278 32,646.76 86 4.8413 4.2069 7,791.25
74 19.1253 16.6194 30,779.15 87 3.6323 3.1563 5,845.56
75 17.9589 15.6058 28,902.01 88 2.4221 2.1048 3,898.05
76 16.7869 14.5874 27,015.89 89 1.2113 1.0526 1,949.33
77 15.6097 13.5645 25,121.40 90 0.0000 0.0000 0.00
78 14.4277 12.5373 23,219.10
GLOSSARY
axis of rotation: an imaginary line running through the freely over the continents, creating a single, undisturbed water
North and South Poles about which the Earth rotates. body; the surface would have undulations reflecting the influ-
azimuth: refers to specifying direction on a map and is ence of gravity.
customarily measured in a clockwise fashion starting with graticule: latitude and longitude taken in combination.
geographic north as the origin and passing through 360°. great circle: represents the line along which a plane inter-
centrifugal force: the tendency for an object that is posi- sects the Earth’s surface when the plane passes through the
tioned on and in motion along a curved path to move away Earth’s center.
from the center of that rotating body. horizontal datum: a specific reference ellipsoid to which
chronometer: a highly accurate mechanical timekeeping accurate latitude and longitude values are referenced.
device that was developed to determine longitude. International Dateline: the line coinciding approximately
degrees: a basic unit of measure utilized in specifying longi- with the 180th meridian; when crossing it, a day is lost or
tude and latitude; the Earth encompasses 360° of longitude gained depending on the direction of travel.
and 180° of latitude. International Meridian Conference: a conference held at
ellipsoid: see oblate spheroid. Washington, DC, in 1884 that established Greenwich,
Equator: the origin for the system of latitude (0° latitude); England, as the origin (0°) for the system of longitude.
serves as the dividing line between the Northern and latitude: an imaginary line that crosses the Earth’s surface in
Southern Hemispheres. an east-to-west fashion; latitude is measured in degrees,
flattening constant: expresses the degree to which an ellipse minutes, and seconds north or south of the Equator.
deviates from a circle; computed as (a – b)/a, where a and b are longitude: an imaginary line originating at the poles that
the semimajor and semiminor axes of the ellipse, respectively. crosses the Earth’s surface in a north-to-south fashion; longi-
geocentric latitude: latitude as computed on a reference tude is measured in degrees, minutes, and seconds east or west
ellipsoid is measured by an angle that results when a line at of the Prime Meridian.
the Earth’s surface is drawn intersecting the plane of the loxodromes: lines that intersect all meridians at the same
Equator at the Earth’s center. angle.
geodesy the field of study investigating the Earth’s size and meridian: see longitude.
shape. minutes: a unit of measurement of the sexagesimal degree
geodetic datum: a combination of a specific reference system; each degree contains 60 minutes.
ellipsoid and a geoid (also commonly referred to as a datum). normal distribution: a term used to describe the bell-shaped
geodetic latitude: latitude as computed on a reference ellip- curve formed when data are distributed along the number line.
soid is measured by an angle that results when a perpendicu- North American Datum of 1927 (NAD27): a common
lar line at the reference ellipsoid’s surface is drawn toward the datum for accurate maps of North America defined by the
Earth’s center. Clarke 1866 reference ellipsoid and a geoid specification hav-
geoid: describes the two-dimensional curved surface that ing Meades Ranch, KS, as the point where the geoid and the
would result on the Earth if the oceans were allowed to flow reference ellipsoid contact each other.
147
16:27:49.
North American Datum of 1983 (NAD83): a datum for satellite geodesy: the use of Earth-orbiting satellites to
accurate maps of North America defined by the Geodetic derive very accurate information regarding the Earth’s size
Reference System (GRS80) reference ellipsoid and a geoid and shape.
located at the Earth’s center of mass. seconds: a unit of measurement in the sexagesimal degree
North Pole: the geographic location in the Northern system; each minute contains 60 seconds.
Hemisphere with a latitude of 90° North that serves as a point sexagesimal system: the system for specifying locations on
around which the Earth’s axis rotates. the Earth’s surface in degrees, minutes, and seconds (a base 60
North Star: the star (called Polaris) that happens to be system).
located directly above the North Pole. sextant: a mechanical device used to determine latitude on
Northern Hemisphere: the portion of the Earth extending the Earth’s surface.
from the Equator to the North Pole. small circle: represents the line along which a plane inter-
oblate spheroid: refers to the notion that the Earth bulges at sects the Earth’s surface when the plane does not pass through
the Equator and compresses at the poles due to the centrifu- the Earth’s center.
gal force caused by rotation. South Pole: the geographic location in the Southern
parallel: see latitude. Hemisphere with a latitude of 90° South that serves as the
Prime Meridian: the internationally agreed-upon meridian point around which the Earth’s axis rotates.
with a value of 0° that runs through the Royal Observatory Southern Hemisphere: the portion of the Earth extending
located in Greenwich, England. from the Equator to the South Pole.
prolate spheroid: refers to the historical notion that the triangulation: (1) a method of surveying where two points
Earth bulges at the poles and is compressed at the Equator; having known locations are established and the distance
today it is known that an oblate spheroid (bulging at the between them is measured; angles from this baseline are then
Equator) is the actual shape. measured to a distant point, creating a triangle, which can then
reference ellipsoid: a smooth, mathematically defined figure serve as a basis for other triangles; (2) a method of interpolat-
with a semimajor axis that is longer than the semiminor axis ing data for isarithmic maps in which original control points
and is designed to approximate the Earth’s true shape to a are connected by a set of triangles.
better degree than would a spherical model. vertical datum: a reference surface (often mean sea level)
rhumb lines: see loxodromes. for elevations on the Earth’s surface.
REFERENCES
Harley, J., and Woodward, D. (1992) The History of National Imagery and Mapping Agency (NIMA). (1997)
Cartography: Cartography in the Traditional Islamic and Department of Defense World Geodetic System 1984, Its
South Asian Societies. Chicago: University of Chicago Definition and Relationships With Local Geodetic Systems,
Press. Technical Report TR8350.2 Third Edition. Washington
Maling, D. H. (1992) Coordinate Systems and Map DC: U.S. Department of Defense.
Projections (2nd ed.). Oxford, England: Pergamon. U.S. Geological Survey. (1989) North American Datum of
1983, Map Data Conversion Tables. Denver, CO: Author.
16:27:49.
Elements of Map Projections
149
16:30:31.
OVERVIEW
The process of projecting the spherical Earth onto a flat general projection equations, explains the parameters
surface is accomplished through a map projection. One of necessary to create a projection, and illustrates some
the more complex processes in cartography, the map pro- different projections that result.
jection plays an important role in the map’s appearance In addition to understanding the mathematical
(e.g., shapes of landmasses and the arrangement of the equations associated with projections, you need to com-
graticule) as well as the kinds of map uses that are possible prehend the various characteristics that a given projection
(e.g., accurately measuring distances as opposed to visual- possesses. Section 4 describes the characteristics of class,
izing the spatial variation of a data set symbolized by dots). case, and aspect of a projection. The projection’s class
In this chapter, we consider the fundamental elements refers to the developable surface used in creating the pro-
of map projections. In section 1, the general concept of jection. Thus, the cone produces the conic class, the plane
the map projection is presented. Section 2 discusses the produces the planar class, and the cylinder produces the
projection process, which begins conceptually with the cylindrical class of map projections. The developable sur-
reference globe (a model of the Earth at some chosen face can be positioned so that it touches the reference
scale) and a developable surface (a mathematically globe along one point or line creating the tangent case, or
definable surface onto which the land masses and gratic- two lines creating the secant case. The decision to use the
ule are projected from the reference globe). Traditionally, tangent or secant case depends, in part, on the level of
the cartographer would set the scale of the reference accuracy desired for the map. The developable surface can
globe and then choose one of the developable surfaces also be centered over any location on the reference globe.
(cone, plane, or cylinder) onto which the landmass and If centered over one of the poles, a polar aspect results; if
graticule would be projected. Once projected, the devel- the center is somewhere along the Equator, an equatorial
opable surface was “unrolled,” producing a map. Today, aspect results; or, if the center is not at either pole or along
computers and mathematical equations have replaced the Equator, an oblique aspect results.
this laborious task with software capable of producing Regardless of how carefully a projection is constructed,
hundreds of projections. Although the reference globe distortion is an inevitable consequence. Section 5 explains
and developable surface are no longer relied on when distortion and illustrates its consequences. One way to
creating projections, they help in conceptually under- understand distortion is to realize that, unlike on the
standing the projection process and are referred to Earth, scale varies considerably across a projection and
throughout the chapter. thus a map’s surface. This variation in scale can be ana-
Computers make creating projections a rather trivial lyzed across a projection through the scale factor. This
process compared to what it was just a few years ago, but numeric assessment explains how much departure there is
understanding the mathematical process involved in creat a given location from the scale that is found at that same
ating a projection remains important if projection soft- location on the Earth’s surface. The scale factor is also a
ware is to be utilized effectively. Section 3 reviews some useful underlying concept for Tissot’s indicatrix, which
16:30:31.
provides a visual means of showing how distortion varies consequence that results when the Earth’s two-
at point locations across a projection. In addition, the indi- dimensional curved surface is projected to a map—
catrix provides a quantitative analysis of distortion that that is, when a map projection is created. The map
describes the amount and type of distortion that occurs at projection is one of the most important concepts in
points across the map. When the indicatrix is mapped cartography, because the way in which the Earth’s
across a projection, certain distortion patterns become two-dimensional curved surface is projected directly
apparent and can be classified according to the projec- impacts the appearance of the graticule (lines of lati-
tion’s class. Examining these distortion patterns provides tude and longitude in combination) and landmasses,
insight on how suitable a particular projection’s class, case, and ultimately the kinds of uses for which a map can
and aspect are for minimizing distortion over a specific be applied.
geographic area.
Although no projection can avoid some form of distor- 2 THE REFERENCE GLOBE
tion, most projections minimize distortion in either areas, AND DEVELOPABLE SURFACES
angles, distances, or direction. In section 6, we consider
classes of projections that have been named according to
The techniques employed to create map projections have
these properties. Equivalent projections preserve areal
changed considerably over the past 30 years. Prior to the
relationships, whereas conformal projections preserve
advent of computers, map projections were laboriously
angular relations. Preserving distance and directional rela-
created through manual drafting techniques. Two key
tionships are accomplished on equidistant and azimuthal
concepts from that era include the reference globe and
projections, respectively. Compromise projections pre-
the developable surface.These conceptual aids were used
serve no specific property, but rather strike a balance
to assist the mapmaker in envisioning how the Earth’s
among various projection properties.
two-dimensional curved surface would be projected to
create a map. Although computers have replaced manual
1 THE MAP PROJECTION CONCEPT drafting techniques, these concepts are still useful in
explaining how a map projection is created.
When the cartographer begins creating a map projec-
When trying to create a model of the Earth, a globe is an
tion, an initial step involves conceptually reducing the
obvious possibility. Globes are particularly suitable for
Earth’s size to a smaller, imaginary globe. The final size
representing the Earth because they preserve areal,
of this globe is set according to the principal map scale,
angular, distance, and directional relationships.
which is usually dictated by the map purpose. Figure 1
Unfortunately, globes of any size are rather inconvenient
shows the relationship between the Earth and the refer-
to carry around and store, and detailed representations
ence globe, which is a model of the Earth at a reduced
of even large countries would require rather sizable
scale that is used to project the graticule and landmasses
globes. Using portions of a globe to reproduce a specific
onto the map. The reference globe thus serves as a model
country would reduce some of the bulkiness, but measur-
of the Earth, sized to the principal scale of the final map.
ing, for example, distances across a globe’s curved surface
A developable surface is a simple mathematical
is difficult.
surface that can be flattened to form a plane without
Maps have four distinct advantages over globes.
First, the two-dimensional nature of maps makes many
cartometric activities (e.g., measuring areas, plotting a
course, calculating direction, etc.) much easier than on Earth
the curved surface of a globe. Second, maps can show
Reference Globe
considerable detail for a given landmass. Consider the set to
United States Geological Survey’s (USGS) 1:24,000 Final Map Scale
topographic map series. This series maps portions of
the United States in greater detail than could be found
on even the largest globe. Third, most maps are of a
dimension that makes them easy to work with and
very portable—most 1:24,000 sheets are roughly
56 cm * 66 cm (22 in. * 26 in.). Fourth, maps are less
costly to produce and purchase. A single USGS
1:24,000 map sheet costs about $7.00, whereas a 30 cm
(12 in.) diameter globe is priced around $75.00. FIGURE 1 The relationship between the Earth and the ref-
Although maps have numerous advantages over erence globe. Note that the sizes of the Earth and reference
globes, they suffer from distortion, which is a natural globe are not to scale.
151
16:30:31.
compressing or stretching any part of the original 3 THE MATHEMATICS OF MAP PROJECTIONS
surface. There are three developable surfaces: cylinders,
cones, and planes. To project the graticule from the ref- The reference globe and developable surfaces are con-
erence globe to a map, first a developable surface is ceptual aids that help illustrate the projection process,
conceptually placed over the reference globe, touching but they are not used to create projections today.
it along either a meridian or parallel or at one point Rather, the field of mathematics is utilized to create
(Figure 2); second, the graticule and geographic land- projections, and so it is important to understand some
masses are projected onto the developable surface. of the basic mathematical manipulations used to pro-
Afterward, the developable surface is “unrolled,” ject the Earth onto a map.
revealing the graticule and landmasses. For illustrative purposes, assume that you want to
Figure 3 provides a detailed view of how the refer- transform a point on the Earth’s curved surface to a flat
ence globe and developable surface can be used to map and that the point in question is defined by the lati-
create a map projection. Conceptually speaking, a pro- tude and longitude values of 40° N and 60° W, as shown
jection is created by first shining an imaginary light on in Figure 4. Our goal is to project this coordinate pair to
the reference globe. The light source, called the point of a corresponding set of x and y Cartesian coordinates.
projection, can be located in one of several positions At a minimum, all map projections must involve at
with respect to the reference globe. For example, in least two mathematical equations: one defining the x
Figure 3 the point of projection changes from the center value and another defining the y value. A simple map
of the reference globe (Figure 3A) to a position on the projection involving two equations is as follows:
side of the reference globe opposite where the devel-
opable surface touches the reference globe, to a point at
x = R*(l - l0)
infinity (Figure 3C). The light casts an image of the
graticule and landmasses onto the developable surface y = R*f
(in this case, a plane), which touches the reference globe
at one point, creating a projection. Note how changing where l is the longitude value, f is the latitude value, l0
the location for the point of projection influences the is the value of the central meridian (the location of the
spacing of the graticule. projection’s east–west center), and R is the radius of the
Cylinder
Cone
Plane
Developable Surface
Developable Map FIGURE 2 How developable sur-
and the
Surface Projection
Refererence Globe faces and the reference globe are
used to create map projections.
16:30:31.
A B C
Map
Projection
Developable
Surface: Plane
Reference
Globe
Point of contact
between Point of
Developable Surface Projection
and Reference Globe
FIGURE 3 A detailed view of how the reference globe and developable surface can be used to create a map projection. From A
to C, the point of projection changes from the center of the reference globe to a point on the reference globe opposite where the
developable surface touches the reference globe to a point at infinity.
reference globe.The x and y coordinate values, commonly

referred to as plotting coordinates, specify the longitude
and latitude values in a Cartesian coordinate system.
To compute the x and y values using these equations,
four steps are necessary. First, the longitude value for the
central meridian must be selected. Any longitude value
can be selected, but if 0° is chosen, the central meridian
will coincide with the Prime Meridian. Second, a value of
R must be specified to indicate the radius of the reference
globe (ultimately setting the final scale of the map). For
example, if a world map was to be created with a principal
scale of 1:30,000,000, then a reference globe with a radius
(R) of 21.24 cm (36 in.) would be used. To simplify the
computations that follow, we assume an R value of 1.0.
Third, all latitude and longitude degree values must
be converted into radians. This conversion is especially
necessary when using map projection software or com-
puter programming languages, as neither can compute
trigonometric functions specified in degree values.
Converting degree values to radians involves multiply-
ing the degree measurement by the constant p/180,
FIGURE 4 A point on Earth’s curved surface described by the where p is approximately 3.1415. For instance, 90° in
graticule. In this case, the point is located at 60° W longitude radians equals 90° * (p/180) = 1.5707 or p/2, and 180°
and 40° N latitude. equals p radians. Because longitude values on the Earth
153
16:30:31.
range from -180° to + 180° and all x values represent side of Figure 5). The plate carrée projection is one of the
longitude degree values, then x values range from oldest projections, having been developed by the ancient
- p to + p. On the other hand, because latitude values Greeks. Note that the lines of latitude and longitude for
on the Earth range from -90° to + 90° and all y values the plate carrée are spaced at equal intervals.
represent latitude values, then y values range from Furthermore, although all lines of longitude are equal in
-p/2 to +p/2. For instance, 60° W converts to - 1.047 length (as they should be), they do not converge at the
radians and 40° N converts to 0.698 radians. poles, as they ideally should.
The fourth step in computing x and y values involves We can expand on the simple pair of equations just
inserting longitude and latitude radian values into the introduced by adding trigonometric functions. For
equations. Using our values of -1.047 and 0.698, we have: example, consider what happens if we compute the sine
of the latitude—we would have the following:
x = 1.0* (-1.047 - 0) = - 1.047
y = 1.0* 0.698 = + 0.698 x = R* (l - l0)
y = R*sinf
If all remaining point locations on the Earth’s curved
surface defined by latitude and longitude were computed This mathematical transformation produces the
(e.g., every 15°) using the above equations, the plate car- Lambert cylindrical projection developed in 1772 by
rée projection would be generated (as shown on the right Johann H. Lambert (Figure 6). Note that the meridians
160º W 180º 160º E

140º W 140º E
120º W 120º E
ºN
80
6 ºN
0
100º W 100º E
Prime Meridian
40º N
80º W 20º N 80º E

qu
E
ato
r
0º
60º W 60º E
20º S
40º W 40º E
20º W 0º 20º E
FIGURE 5 The mathematics of a map projection transforms a point on the Earth’s curved surface represented by a latitude and
longitude pair to x and y coordinate pairs on a map. In this case, the plate carrée projection results. The ranges of x and y coordi-
nate values in radians are shown in gray along the sides of the map projection.
FIGURE 6 The Lambert cylindrical projection with a 15° graticule spacing.
16:30:31.
are equally spaced as they are on the plate carrée projec- This set of mathematical equations produces the poly-
tion in Figure 5, but that the spacing of the parallels cylindrical projection shown in Figure 8. In this case, some
decreases as their distance from the Equator increases.As of the graticule characteristics of the previous projections
an alternative, we can change just the spacing of the x val- should be apparent. For instance, the meridians are
ues by introducing the cosine function to our simple equa- curved as in the sinusoidal projection and the parallels
tions. This time, we take the cosine of the latitude value: are parallel as in the other two cylindrical projections.
x = R* (l - l0)*cos f
4 MAP PROJECTION CHARACTERISTICS
y = R* f
The sinusoidal pseudocylindrical projection results, as Through manipulation of the mathematical equations,
shown in Figure 7. The term pseudocylindrical is applied numerous projections are possible, where each projec-
here because this projection shares only some of the tion has specific characteristics, making each one useful
visual characteristics of the graticule with the plate car- for a particular mapping purpose. This section discusses
rée cylindrical projection, namely that the lines of lati- the various characteristics of map projections based on
tude are parallel and, in this case, equally spaced. class, case, and aspect.
However, the meridians of this projection are curved
lines converging at the North and South Poles. 4.1 Class
Now, combining the cosine and sine functions
together yields the following equations: Earlier, we saw that the developable surface concept
was useful in illustrating the manner in which the gratic-
x = R*(l - l0)*cos f ule and landmasses on the reference globe are pro-
jected to the map. This concept is also a constructive
y = R* sinf
FIGURE 7 The sinusoidal projec-

tion with a 15° graticule spacing.
FIGURE 8 A polycylindrical pro-

jection with a 15° graticule spacing.
155
16:30:31.
way to describe the overall shape and appearance of graticule onto the cylinder, and then unrolling the
the graticule after the projection process is complete, cylinder. On cylindrical projections, lines of longitude
referred to as the projection’s class. The three common typically appear as straight, equally spaced, parallel
map projection classes are cylindrical, conic, and planar. lines, whereas lines of latitude appear as straight paral-
The cylindrical class of projections results from lel lines that intersect the lines of longitude at right
wrapping the developable surface of a cylinder around angles (Figure 9A and B). The spacing of the parallels
the reference globe, projecting the landmasses and distinguishes one cylindrical projection from another.
Cylindrical Class
A Tangent
B Secant
Conic Class
C Tangent
D
Secant
Planar Class
Tangent
E
FIGURE 9 A comparison of the dif-

ferent classes of projections: (A–B)
Secant
F cylindrical, (C–D) conic, and (E–F)
planar. Also shown are the tangent
and secant cases for each class.
16:30:31.
The conic class of projections results from wrapping typically appear as straight, equally spaced, parallel lines
the developable surface of a cone around the reference that radiate from the center (when the center of the
globe, projecting the landmasses and graticule onto the projection is one of the poles), and lines of latitude
cone, and then unrolling the cone. On conic projections, appear as equally spaced concentric circles centered
lines of longitude typically appear as straight lines of about a point, for example, one of the poles (Figure 9E
equal length radiating from a central point (usually one and F). The spacing of the parallels distinguishes one
of the poles), whereas lines of latitude appear as con- planar projection from another.
centric circular arcs centered about one of the poles There are many variations of the three developable
(Figure 9C and D). The overall shape of most conic pro- surfaces that create other interesting projections. Like
jections can be described as a pie wedge, where a pie the cylindrical projections, the pseudocylindrical class
would be a full circle. Note that in the right-hand por- of projections shows parallels as straight, nonintersect-
tion of Figure 9C, the pie wedge is slightly more than ing lines. However, pseudocylindrical projections are
one-half, whereas in Figure 9D, the pie wedge is slightly visually distinguished from cylindrical projections by
less than one-half. The angular extent of the pie wedge curved meridians that are equally spaced along every
and the spacing of the parallels distinguish one conic parallel and converge at the poles, which are repre-
projection from another. sented by points or lines. Figure 10A illustrates the
The planar class of projections results from position- Craster parabolic pseudocylindrical projection with the
ing the developable surface of a plane next to the refer- meridians converging to points, and Figure 10B shows
ence globe and projecting the landmasses and graticule the Eckert III pseudocylindrical projection with merid-
onto the plane.* On planar projections, lines of longitude ians converging to lines representing the poles.
The polyconic class of projections conceptually
*Some texts refer to the planar class of projections as azimuthal. We
employs multiple cones rather than one cone. Polyconic
have chosen to use planar because it is easier to relate to the devel- projections display curved lines (nonconcentric circular
opable surface concept (you can see that a plane is used). arcs) representing the parallels in a fashion similar to
FIGURE 10 A comparison of the

(A) Craster parabolic and (B)
Eckert III pseudocylindrical projec-
tions. Each projection has a 15°
graticule spacing.
157
16:30:31.
the conic projections, but the meridians curve toward For a given projection class, the mapmaker usually
the central meridian. Figure 11 illustrates the rectangu- has a variety of choices for the desired specific line(s)
lar polyconic projection. There are many other varia- or point of tangency. For instance, with the conic class,
tions of the basic developable surface concept. any line of latitude can be selected as the tangent line.
However, unlike the cylinder, cone, and plane, each In the secant case of the conic class, any two lines of lat-
variation requires a mathematical rather than a visual itude can be selected. If these lines are nonequally
approach to explain how each projection is developed. spaced on opposite sides of the Equator, equally spaced
on the same side of the Equator, or nonequally spaced
4.2 Case on the same side of the Equator, the parallels and
meridians take on the familiar cone shape. If, however,
The case of a projection relates to how the developable the two lines are equidistant from and on opposite sides
surface is positioned with respect to the reference of the Equator, the parallels and meridians appear as
globe and is either tangent or secant. A brief example straight parallel lines.
illustrates the tangent idea. Conceptually speaking, Secant lines and points of tangency each have the
imagine a ball rolling across the floor. At any given same scale as the principal scale of the reference
time, there is exactly one point in common between the globe. Thus, secant lines are called standard lines,† and
ball and the floor. This point of contact is called the points of tangency are called standard points. All other
point of tangency. If all points of contact were con- lines and points will have either a larger or a smaller
nected together (either on the ball or on the floor), a scale than the principal map scale of the reference
line of tangency would result. In the tangent case of a globe. Figure 12 illustrates the concept of a standard
map projection, the reference globe touches the devel- line and its impact on scale variation across a map. In
opable surface along only one line or at one point. the figure, a portion of the reference globe is repre-
Figure 9A, C, and E illustrates the tangent case for the sented by the dashed line, and the developable surface
cylindrical, conic, and planar map projections, respec- is represented by the solid line of gray values. Note
tively. The secant case of a projection occurs when the that the developable surface cuts the reference globe,
developable surface passes through the reference creating two standard lines that have the same scale as
globe. In the secant case of the cylindrical and conic the reference globe. The area between the standard
map projections (Figure 9B and D), there are two lines projected to the developable surface (shown by a
secant lines, whereas in the case of the planar projec- solid light gray line) has a compressed scale, whereas
tion, there is one secant line (Figure 9F).* the areas beyond the standard lines (shown by the
† In many cases, standard lines coincide with lines of latitude and are
*In the strictest sense, a secant case for the planar class is not possible, referred to as standard parallels. Note that standard line is a general
but it is included here for conceptual completeness. term—a standard line can also coincide with small circles or meridians.
FIGURE 11 The rectangular poly-

conic projection with a 15° graticule
spacing.
16:30:31.
standard lines at 25° S and 50° S. Note that when plac-

ing the standard lines in the southern latitudes, the
parallels are curves that concave toward the South
Pole and the meridians converge to the South Pole.
The choice of the tangent or secant case, as well as
their placement with respect to the reference globe,
impacts the shape of the landmasses and the arrange-
Standard
Line ment of the graticule. For instance, in Figure 13A, the
Reference convergence of the meridians at the North Pole is not
Globe as great as the meridian convergence in Figure 13B. As
a result, landmasses in the northern latitudes are not
as stretched in an east–west direction in Figure 13B.
Figures 13A and B also illustrate that the placement of
the standard lines in the northern latitudes greatly dis-
torts the landmasses in the extreme Southern
Developable Hemisphere (e.g., Antarctica is stretched out the
Surface length of the map). In contrast, when both standard
lines are placed in the Southern Hemisphere (as in
Figure 13C), the landmasses there are not as distorted,
but the landmasses in the Northern Hemisphere are
greatly distorted.
4.3 Aspect
The aspect of a projection concerns the placement of

the projection’s center with respect to the Earth’s sur-
face. In general terms, a projection can have one of
three aspects: equatorial, oblique, and polar. An equa-
torial aspect is centered somewhere along the Equator,
Standard
Line
a polar aspect is centered about one of the poles, and
an oblique aspect is centered somewhere between a
pole and the Equator. The aspect of the projection can
be defined more precisely in terms of the placement of
Scale the central meridian and the latitude of origin.
Exaggerated Figure 14 shows four different aspects for the equidis-
tant cylindrical projection. For these maps, the central
Scale meridian is located at -96° W, and the latitude of
Compressed
origin begins at 0° (an equatorial aspect) and moves
northward every 30° of latitude (oblique aspects)
until the projection is centered over the North Pole
FIGURE 12 The effect of standard lines on scale variation (a polar aspect).
on a map.
5 DISTORTION ON MAP PROJECTIONS
solid darker gray lines) have an exaggerated scale. An important consequence of map projections is the
Figure 13 shows the effect of selecting different tan- distortion created in the resulting map. More specifi-
gent and secant cases for the Euler conic projection, cally, we can define distortion as the alteration of the
developed by Leonhard Euler in 1777. Figure 13A size of the Earth’s landmasses and the arrangement
shows the tangent case with only one standard line at of the Earth’s graticule when they are projected to
25° N, and Figure 13B shows the secant case with two the two-dimensional flat map. There are numerous
standard lines located at 25° N and 50° N. Note that ways in which distortion can be examined and ana-
for both figures, the meridians converge toward the lyzed on a projection, many of which are covered in
poles and the parallels form curves concave toward a thorough review by Karen Mulcahy and Keith
the North Pole. Figure 13C shows the secant case with Clarke (2001).
159
16:30:31.
FIGURE 13 The Euler conic projection on which the location of the standard lines is represented by solid black line(s). A has the
standard line tangent to 25° N. B has standard lines secant at 25° N and 50° N. C has standard lines secant at -25° S and -50° S.
5.1 A Visual Look at Distortion projection presenting a view over the North Atlantic
Ocean similar to what one would see if looking at the
One way to examine distortion is to visually compare Earth from space. Note the size of Greenland relative
the sizes of landmasses and the arrangement of the to the United States in this figure. Now compare the
graticule on the spherical Earth to how the landmasses size of Greenland to the United States on the Mercator
and graticule arrangement appear on the map projec- projection in Figure 15B; obviously, Greenland is now
tion. To illustrate, Figure 15A shows an orthographic relatively much larger than the United States. As a
16:30:31.
Equatorial Aspect Oblique Aspect

Centered -96º W, 0º Centered -96º W, 30º N
Oblique Aspect Polar Aspect

Centered -96º W, 60º N Centered -96º W, 90º N FIGURE 14 Different aspects of a
cylindrical projection.
point of reference, Greenland has a land area of 5.2 Scale Factor

2,166,086 km2, which is smaller than the United States
at 9,158,960 km2, but on the Mercator projection, To understand the details of Tissot’s indicatrix, we
Greenland is represented as larger than the United begin with an introduction of the scale factor (SF),
States. In general, you can see that landmasses in the which is a numerical assessment of how the map scale
upper latitudes on the Mercator projection are consid- at a specific location on a map compares with the map
erably exaggerated—a limitation when using this pro- scale at a standard point or along standard line(s).
jection for maps of the world. Figure 9 showed that the scale on the map is the same
Depending on the way in which a projection is devel- as the reference globe only where the developable sur-
oped, distortion can also cause the Earth’s graticule to face comes in contact with the reference globe at a
take on a number of different appearances. For standard point or along standard line(s). In all other
instance, on the Earth’s surface, all meridians converge locations, depending on the way the projection is
to the poles. This relationship is preserved on the ortho- mathematically created, the scale is either exaggerated
graphic projection, but on the Mercator projection in or compressed.
Figure 15B, all meridians are equally spaced straight The SF at any given location is computed using the
lines. As a result of this nonconvergence, the graticule following formula:
on the Mercator projection is stretched, compared to its
representation on the orthographic projection. Local scale
A visual inspection of a map projection is useful in Scale factor =
Principal scale
providing a general overview of the distortion present
on the map, but a more sophisticated approach is needed where local scale is the scale computed at a specific
to quantitatively analyze the amount and kind of distor- location and principal scale is the scale computed at a
tion throughout a projection. Moreover, once the quan- standard point or along standard line(s). Using this for-
titative analysis is complete, there remains the need to mula, we can compute how much deviation exists
visualize the distribution of distortion. There have been between the local scale and the principal scale at any
several approaches to quantitative distortion analysis location on a given projection. To illustrate, assume that
that have been developed, but our discussion focuses on we are working with the quartic authalic projection
one of the more common—Tissot’s indicatrix. (suitable for world maps) shown in Figure 16, and that
161
16:30:31.
A B
FIGURE 15 (A) The orthographic projection gives the appearance of the Earth as if viewed from space. (B) On the
Mercator projection, note the exaggerated size of the landmasses, especially in the upper latitudes. Also, the meridians do
not converge at the poles, and the spacing of the lines of latitude increases from the Equator to the poles. Although it might
not appear so, both projections are shown at the same scale.
FIGURE 16 The quartic authalic projection with a 15° graticule spacing. The letter a references the location at which the
scale factor is computed.
16:30:31.
this projection has a principal scale of 1:235,043,988 local scale along the 30° N latitude (1:222,660,646), we
along the Equator, which is the standard line. Now can compute the SF at this location. Substituting the dif-
assume you wish to compute the SF at some point along ferent scale values into the SF equation, we arrive at the
the parallel at 30° N, say, between 0° and 15° W (shown following:
on Figure 16 as a). You would first calculate the local 1:222,660,646
map scale between 0° and 15° W, which can be found SF = = 1.056
1:235,043,988
using the following steps: Measuring along the parallel
at 30° N between 0° and 15° W, assume you arrive at a This SF value suggests that the local scale has been exag-
map distance of 0.65 cm (0.256 in.). To compute the cor- gerated, meaning that the distances on the Earth’s curved
responding Earth distance for that same distance, look surface along 30° N latitude have been enlarged with
in the Appendix and find the distance for one degree of respect to its true distance. On the other hand, if a local
longitude for 30°, which is 96,486.28 meters. Because scale was computed to be 1:240,000,000 and the principal
the Earth distance between 0° and 15° is 15°, we need to scale remained at 1:235,043,988, then the SF would be
multiply 96,486.28 by 15°. Doing so, we arrive at an 0.979, and a compressed scale would result. Obviously, if
Earth distance of 1,447,294.2 meters, or 1447.2942 km. we had a location on the same projection with a local scale
Next, substituting the 0.65 cm and 1447.2942 km into of exactly 1:235,043,988, the SF would be 1.0, indicating
the basic scale equation, we obtain the following local that there has been no change in scale at that location.
scale:
5.3 Tissot’s Indicatrix
Local scale = Earth Distance/Map distance
Local scale = 1447.2942 km/0.65 cm Nicolas Tissot, a French mathematician, was one of the
pioneers in analyzing distortion found on map projec-
Before solving, the units in the equation must be the tions. Tissot developed the indicatrix, which provides a
same, so we convert km to cm: graphical representation of distortion at various points
across a projection. The indicatrix, as found on the ref-
1447.2942 km* 100,000 cm = 144,729,420 cm erence globe’s surface where all spatial relationships
are preserved, results in no distortion, and is called the
Therefore, unit circle.* In other words, at each and every point on
the reference globe, an infinitely small circle exists that
Local scale = 144,729,420 cm/0.65 cm = 1:222,660,646 can be described as a unit circle having a radius (or SF)
of 1.0, as shown in Figure 17. Even though the indicatrix
The local scale along the 30° parallel is thus computed as is displayed as a unit circle of some areal extent, it is
1:222,660,646, which is larger than the principal scale
along the Equator of 1:235,043,988. Knowing the princi- * In mathematics, a unit circle radius equals 1 unit, which takes on
pal scale of the map projection (1:235,043,988) and the special meaning when solving trigonometric functions.
Indicatrix
Indicatrix
r = 1.0
FIGURE 17 The indicatrix and its

characteristics as shown on the refer-
Reference Globe ence globe. The radius of the indica-
trix on the reference globe is 1.0.
163
16:30:31.
important to remember that the appearance of the indi- this ellipse is equal to the area of the unit circle. In this
catrix on the reference globe is restricted to measure- instance, there is an increase in the SF along a, but the SF
ments made around infinitely small points and is not along b decreases proportionately (Figure 18B). This
considered valid when discussing landmasses or water leads to a change in the angle 2v, and so the projection is
bodies of any great extent. said to possess angular distortion. Note, however, that the
After projection, the indicatrix has two characteris- area of the unit circle is preserved, as found on the refer-
tics that allow the amount and kind of distortion to be ence globe, so there is no areal distortion. In the third
analyzed. First, the indicatrix is defined by two specific instance, the SFs along a and b change in the same man-
radii called the semimajor (a) and semiminor (b) axes. ner (e.g., a = 2.0 and b = 2.0 ). Here, the indicatrix is still
By convention and through mathematical manipula- a circle, but it is larger (or smaller) in size than the unit
tion, the semimajor axis is aligned in the direction of circle found on the reference globe (Figure 18C), indicat-
the maximum SF, and the semiminor axis is aligned in ing areal distortion. Because the angle 2v has been pre-
the direction of the minimum SF. These two axes are served, there is no angular distortion. The fourth instance
always perpendicular to one another. Second, on the is where a and b change unequally, producing an ellipse,
indicatrix there exist two lines called l and m that inter- and the area of the ellipse does not equal the unit circle
sect, forming an angle called 2v (Figure 18). The spe- (Figure 18D). In this case, there is both areal and angular
cific changes in the SFs along a and b as well as the distortion.
alteration of 2v after projection provide a useful means When the indicatrix is plotted on a projection
to analyze distortion. (e.g., every 15°), the type and amount of distortion can be
There are four general instances of the indicatrix that visualized across the projection. For example, Figure 19
result after a projection is created. In the first instance, shows the indicatrix for the Winkel Tripel projection. In
there is no change in the SFs along a and b, so the indi- this case, note that the central meridian at approximately
catrix is still a unit circle (Figure 18A), indicating no dis- 44° 28' 25'' N/S is the only location where the indicatrix
tortion. The second instance is where the SFs along a and is a circle and the SFs along both a and b are 1.0. At all
b change unequally, creating an ellipse, but the area of other locations, the indicatrix takes on an elliptical shape,
a=b a≠b
A B
b
No Change a 2ω a b 2ω
in Area
a = 1.0 a = 2.5
b = 1.0 b = 0.4 FIGURE 18 The possible configura-
tions of the indicatrix after projec-
C D
tion. The indicatrix in (A) is the unit
circle as found on the reference
b b globe, and thus there is no distortion
in scale factors, areal, or angular rela-
Change a 2ω a 2ω tions. In (B), the scale factors are not
in Area equal; the area of the original indica-
trix has been preserved, and note the
change in 2v, suggesting angular dis-
a = 2.0 a = 2.5 tortion. The scale factors in (C) are
b = 2.0 b = 1.7 greater than 1.0, but equal to one
another and thus the area is exagger-
No Change Change ated, but there is no angular distor-
in Angles in Angles tion. In (D), there is both angular and
areal distortion.
16:30:31.
FIGURE 19 The indicatrix on the

Winkel Tripel projection. Note that
the indicatrix is a circle only along
the central meridian at approxi-
mately 44° 28' 25'' N/S. The indicatrix
takes on a variety of elliptical shapes
and sizes throughout the remainder
of the projection, suggesting that this
projection preserves neither areas
nor angles. The indicatrix is plotted
every 15°.
suggesting that there is angular distortion throughout the Minimum error projections possess no specific prop-
projection. In a similar light, the ellipses do not appear to erty, but by mathematically optimizing SFs, they achieve
have the same areal extent, indicating that areal relations lower overall distortion across a projection than can be
are not preserved in the Winkel Tripel projection. achieved when one property is preserved. Next, we con-
sider each of the projection properties in detail.
5.4 Distortion Patterns
6.1 Preserving Areas
Figure 9 showed that the developable surface contacts
the reference globe at either a single standard point or Equivalent projections preserve landmasses in their
along one or two standard lines. Where the standard true proportions, as found on the Earth’s surface. To
point or standard line(s) contact the reference globe, no illustrate equivalency, consider the landmasses in
distortion in scale is present—the SF at these locations Figure 21, in which we see four samples of Greenland
is 1.0. However, we learned that at other locations that have been taken from different projections (each
throughout the map projection, scale varies. This scale projection has the same approximate principal scale).
variation can be mapped, producing characteristic Figure 21A through 21C are from equivalent projec-
distortion patterns that are unique to the specific class tions, whereas Figure 21D is from a nonequivalent pro-
and case of projections, as shown in Figure 20, where jection. To preserve areal relations in Figure 21A
darker grays indicate greater distortion. through C, the shapes of landmasses have been
distorted (e.g., in Figures 21A and C, the northern por-
tion of Greenland is expanded in an east-to-west direc-
6 PROJECTION PROPERTIES tion compared to its representation on the Earth).
To ensure that areas are preserved on equivalent pro-
A map projection is said to possess a specific property jections, the SFs must be controlled so that each indica-
when it preserves one of the spatial relationships (i.e., trix contains the same area. For instance, Figure 22
areas, angles, distances, and directions) found on the shows the indicatrix on the Albers equivalent conic
Earth’s surface. The preservation of a particular prop- projection, which possesses two standard lines at 30° N
erty is achieved by controlling the SFs throughout the and 45° N latitude. On this projection, the indicatrix
projection. For instance, projections that preserve either appears as a circle along each standard line. At all other
areas or angles throughout the projection are called locations, the indicatrix takes on an elliptical shape, sug-
equivalent (equal area) and conformal, respectively. gesting that the scale factors a and b change across this
Projections also are capable of preserving several projection. In the upper latitudes, the ellipses display a
special properties. When all distances from a particular contorted appearance, elongated in an east–west direc-
location are correct, then the projection is said to be tion and compressed in a north–south direction. Recall
equidistant. Azimuthal projections, which were intro- that for equivalence, b must decrease proportionately as
duced earlier as planar projections, preserve directions a increases. The difference in lengths of a and b indicates
or azimuths from one central point to all others. angular distortion at these locations. Regardless of the
165
16:30:31.
Secant
Tangent
Standard
CYLINDRICAL Lines
Tangent Standard Secant

Lines
CONIC
Tangent Secant
Standard
Line
PLANAR
Standard
Point
FIGURE 20 The relationship between the projection class, case, and distortion pattern. In the tangent case of each projection class,
the standard line or point is the location where the scale factor is 1.0, shown by the thicker white line or point. In the secant case of
each projection class, the standard lines (note that the planar class only has one standard line) have a scale factor equal to 1.0. Darker
gray shading indicates increasing distortion.
appearance of the ellipses, all are of the same size, ensur- some explanation because the term conformal has been
ing that areas are preserved across the projection. misunderstood and misused by many. It often has been
interchanged with orthomorphic, meaning “correct
6.2 Preserving Angles shape,” and this has conveyed the assumption that con-
formal projections preserve shapes of entire landmasses
Conformal projections preserve angular relationships (both large and small). Unfortunately, conformal maps do
around a point by uniformly preserving scale relations not preserve shapes of landmasses per se. Rather, preser-
about that point in all directions. This concept requires vation of shapes is found only at infinitely small points.
16:30:31.
Conformal projections preserve angular relations by 6.3 Preserving Distances

ensuring that SFs change along a and b at the same rate.
As Figure 23 shows, the indicatrix throughout the In simple terms, projections that maintain the principal
Lambert conformal conic projection is displayed as a scale from two points on the map to any other point on
circle, with a and b necessarily changing uniformly in the map are said to be equidistant. For instance, if the
size. At all locations, a and b remain equal to one two points are the poles,* then all meridians are straight
another, permitting 2v to maintain the same angle as lines that have the same principal scale; an example is the
found on the reference globe. plate carrée projection (Figure 5). Recall that equivalent
and conformal projections achieve their properties by
varying the SFs during the transformations. With these
A B projections, scale cannot be preserved along both a and
b. Scale on a projection can be preserved only along a
single line from one point to any other point (usually
placed at the center of the projection), between two
points, or along standard line(s). For instance, the
azimuthal equidistant projection (Figure 24A) portrays
correct scale from the projection’s center along a straight
line to any other point on the map. Examples of other
C D equidistant projections include the equirectangular cylin-
drical and Euler equidistant conic (Figures 24B and
24C), where scale is correct along all meridians; and the
doubly equidistant azimuthal (Figure 24D), which pre-
serves scale along a straight line from either of two cen-
tral points—in this case between Washington, DC, and
London, England.
In some cases, a mapmaker might wish to focus on a
certain area of the map, and show the relationship of
FIGURE 21 Greenland depicted on three different equiva-
lent projections—(A) the Eckert IV pseudocylindrical, (B) * In technical terms, any two points that are exactly opposite each
the Lambert azimuthal equivalent, and (C) the Albers equiv- other on the Earth’s surface are called antipodes. For example, a loca-
alent conic—and one nonequivalent projection, (D) the tion at 39° N and -83° W would have its antipode located at - 39° S
Miller cylindrical projection. (the opposite hemisphere) and 97° E (180° - 83°).
Standard
Lines
FIGURE 22 Indicatricies for the

Albers equivalent conic projection,
which has standard lines at 30° N and
45° N latitude (shown by the thicker
black lines). Indicatricies are circles
along the standard lines; for all other
locations, indicatricies are elliptical
in shape, but retain the same area as
indicatricies along the standard lines.
167
16:30:31.
Standard
Lines
FIGURE 23 Indicatricies for the

Lambert conformal conic projection,
which has standard lines at 15° N and
60° N latitude. All indicatricies are cir-
cles (indicating no angular distortion),
but the indicatricies change in size
when moving away from the standard
lines (indicating areal distortion).
that area to the larger geographical context. In other gnomonic. Each azimuthal projection is conceptually
words, the central area to be emphasized is shown at a developed in a similar fashion by placing a plane
larger scale and the peripheral area is shown at a tangent to one point on the reference globe. As such,
smaller scale. Such a variable scale map projection pre- the SF on these azimuthal projections is only true at the
sents an interesting view of the Earth and, as the name center point and is greater than 1.0 at all other loca-
suggests, the SFs change considerably across the pro- tions. It is useful to note that the azimuthal projections,
jection but can be controlled to yield interesting and in addition to preserving direction, each preserve
useful results. For example, Figure 25 shows the another property: The stereographic is conformal,
Hagerstrand logarithmic projection centered at 96° W, the Lambert is equal area, the azimuthal equidistant is
40° N. On this projection, notice that the center shows equidistant only along the meridians (here the SF is
the U.S. Midwest enlarged compared to the rest of the set to 1.0 so that one can measure distance from the
landmass, which becomes increasingly smaller toward center of the projection to any other point), and the
the map edge. gnomonic shows all great circle routes as straight lines.
The orthographic displays a spherical appearance sug-
6.4 Preserving Directions gesting the roundness of the Earth, which is a correct
representation of the Earth if viewed from a distant
Directions or azimuths can be preserved on azimuthal point in space.
projections. On this kind of projection, directions are
preserved from the center of the map to any other 6.5 Compromise Projections
point on the map. When measuring azimuths from the
center of an azimuthal projection, all straight lines In both equivalent and conformal mapping, the sizes
drawn or measured to distant points also represent and shapes of landmasses are often visually distorted,
great circle routes. As such, azimuthal projections have sometimes to the point of being unrecognizable.
been used extensively for navigation and pinpointing A solution to this problem is compromise projections,
locations. which manipulate the SFs so that the extreme angular
Azimuthal projections are among the oldest known and areal distortion found on equivalent and confor-
projections, some having been developed by the ancient mal projections is not present. Thus, a compromise
Greeks. Figure 26 shows five common azimuthal pro- projection strikes a balance between the distortion in
jections: (A) Lambert equivalent, (B) stereographic area present on conformal projections and the angular
conformal, (C) equidistant, (D) orthographic, and (E) distortion that is common on a purely equivalent
16:30:31.
A B
C D
FIGURE 24 Four equidistant projections: (A) the azimuthal equidistant, (B) the equirectangular cylindrical, (C) the Euler
equidistant conic, and (D) the two-point (or doubly) equidistant azimuthal.
projection. With compromise projections, even though compromise projections, the National Geographic
no projection property is completely preserved, the Society believed the Robinson presented a better
combined areal and angular distortion is usually less visual impression of the shape of the landmasses, and
than if a single property was preserved, and the result- so it was employed up to 1998, when it was replaced
ing map generally gives a better visual representation with the Winkle Tripel (Figure 27C), another compro-
of landmasses. mise projection.
Probably the most notable compromise projection A variation on the compromise projection concept
is the Robinson. Its popularity increased dramatically is the minimum error projection. As previously dis-
when the National Geographic Society in 1988 made cussed, each projection property has the objective of
the decision to replace the Van der Grinten I projec- preserving one of the spatial relationships (area,
tion (Figure 27A) with the Robinson (Figure 27B) as angles, directions, and distances) on the final map.
the projection for their world maps. Even though both Although this is often advantageous for specific map-
the Van der Grinten I and the Robinson are ping applications, in some cases, the preservation of
169
16:30:31.
FIGURE 25 The Hagerstrand logarithmic projection with a 30°

graticule spacing. Here, the appearance of the map is determined
by taking the logarithmic distance from the center to all other
map locations.
A B C
D E
FIGURE 26 Five common azimuthal projections each centered at 96° W and 40° N: (A) Lambert equivalent, (B) stereographic
conformal, (C) equidistant (scale is preserved from the center to any point on the map), (D) orthographic, and (E) gnomonic
(allows all great circles to be drawn as straight lines).
16:30:31.
conceptual aids in understanding how the Earth’s

A curved two-dimensional surface is projected onto a
map. Today, computer software and mathematics are
heavily involved in developing projections, and so we
illustrated how mathematical equations could be used
to produce a map projection.
Using software to create a projection is not diffi-
cult, but it does require understanding the importance
of the basic projection characteristics of class, case,
and aspect. The projection’s class is directly related to
the developable surface used in creating a projec-
tion—cones, planes, and cylinders produce conic,
planar, and cylindrical projections. The case of a pro-
jection relates to the way in which the developable
surface is positioned with respect to the reference
globe. When the developable surface touches the ref-
B erence globe, the tangent case results, where only one
line (or point) of contact is shared between the two
surfaces. When the developable surface passes
through the reference globe, the secant case results,
where contact is along either one or two lines. The
projection’s aspect relates to the way in which the
developable surface can be positioned anywhere on
the reference globe. Changing the aspect allows the
projection to be centered at one of the poles (a polar
aspect), along the Equator (an equatorial aspect), or
somewhere in between a pole and the Equator
C (an oblique aspect).
Distortion is an inevitable consequence of the
projection process and involves the alteration of scale
about a point on a map compared to that same point
on the Earth. An important consideration in deter-
mining the distortion at each point on the map is the
scale factor, which is the ratio of the local scale to the
principal scale. Closely associated with the scale
factor is Tissot’s indicatrix, which is one of the more
common graphical devices used to visualize distor-
tion across the map. If the indicatrix is mapped across
the projection, definite distortion patterns emerge
FIGURE 27 Three compromise projections: (A) Van der
that in part reflect the nature of the developable
Grinten I, (B) Robinson, and (C) Winkel Tripel.
surface that was conceptually used to create the
projection.
one property often leaves considerable areas, angles,
Important spatial relationships on the Earth include
or distances exaggerated in other parts of the map. In
the ability to measure distances, directions, areas, and
some mapping applications, such as general reference
angles. Each projection has certain properties that
maps, there is no need to preserve a specific property.
allow these spatial relationships to be preserved on a
Rather, the desire might be to lessen the consequences
map. Preserving areal and angular relations is made
of having landmasses modified from how they appear
possible through equivalent and conformal projec-
on the reference globe, and so a minimum error
tions. Equidistant and azimuthal projections permit
projection is used.
distances and directions to be measured in specific
ways across a projection. One consequence of preserv-
SUMMARY ing a specific property is that the other properties are
often distorted (e.g., on conformal projections, areal
In this chapter we examined the fundamental elements relations are distorted). Therefore, compromise
of map projections. We began with a discussion of the projections have been developed that do not preserve
reference globe and the developable surface as any specific property.
171
16:30:31.
FURTHER READING
Bugayevskiy, L. M., and Snyder, J. P. (1995) Map Projections: Maling, D. H. (1992) Coordinate Systems and Map Projections.
A Reference Manual. London: Taylor & Francis. 2nd ed. Oxford: Pergamon.
Provides a mathematical treatment of the projection process. Covers a broad range of topics on map projections, including chap-
Canters, F., and Decleir, H. (1989) The World in Perspective: ters devoted to mathematical principles, Tissot’s indicatrix, and
quantitative analysis.
A Directory of World Map Projections. New York: Wiley.
McDonnell, P. W. (1991) Introduction to Map Projections. 2nd
Includes a useful discussion of the principles of the map projec-
tion process, mathematical derivation of map distortion, and ed. Rancho Cordova, CA: Landmark Enterprises.
illustrations and a historical overview of many world map Explains, using spherical trigonometry, the various equations for
projections. numerous map projections; involves less complex mathematics
Chamberlin, W. (1947) The Round Earth on Flat Paper: Map than the Bugayevskiy and Snyder or Yang, Snyder, and Tobler
texts.
Projections Used by Cartographers. Washington, DC: National
Geographic Society. Snyder, J. P. (1987) Map Projections: A Working Manual.
A broad introduction to fundamental map projection elements, an
Washington, DC: U.S. Geological Survey.
explanation of the importance of map projections in cartography, A useful overview of basic map projection concepts, numerous
and many insightful illustrations. worked-out examples of projection equations (both spherical and
ellipsoidal forms), and a brief history of each projection.
Deetz, C. H., and Adams, O. A. (1945) Element of Map
Projections with Applications to Map and Chart Construction. Snyder, J. P. (1993) Flattening the Earth: Two Thousand Years
5th ed. Washington, DC: U.S. Department of Commerce Coast of Map Projections. Chicago: University of Chicago Press.
and Geodetic Survey. A concise and readable overview of the history of map projec-
Provides a historical perspective on the methods used to manually tions, their development, key individuals, and mathematical evolu-
construct projections. tion as a key factor in the evolution of increasingly sophisticated
map projections.
Fenna, D. (2006) Cartographic Science: A Compendium of
Map Projections with Derivations. Boca Raton, FL: CRC Snyder, J. P., and Steward, H. (1988) Bibliography of Map
Press. Projections. Washington, DC: U.S. Geological Survey.
A comprehensive technical work dealing with the derivation and Includes 2,551 entries of various texts and articles relating to map
computation of dozens of map projections. projections.
Freeman, T. G. (2002) Portraits of the Earth: A Mathematician Snyder, J. P., and Voxland P. M. (1989) An Album of Map
Looks at Maps. Providence, RI: American Mathematical Projections. Washington, DC: U.S. Geological Survey.
Society. Illustrates numerous projections suitable for world maps. Also
Provides a summary of the map projection process without a provides a useful overview of distortion analysis, including a dis-
heavy reliance on mathematical equations and their derivations. cussion of Tissot’s indicatrix.
Iliffe, J. C. and Lott, R. (2008) Datums and Map Projections. Tissot, N. A. (1881) Memoire sur la Representation des
2nd Ed. Dunbeath: Whittles Publishing. Surfaces et les Projections des Cartes Geographiques. Paris:
Gautier-Villars.
Covers fundamental aspects and mathematical computations of
the map projection process. Discusses the development of his indicatrix.
Maling, D. H. (1989) Measurements from Maps: Principles and Yang, Q., Snyder, J. P., and Tobler, W. R. (2000) Map Projection
Methods of Cartometry. Oxford: Pergamon. Transformation: Principles and Applications. London: Taylor
Reviews the nature of measurements taken from a map, and
& Sons.
consequently the distortion that results, as scale varies across Presents a thorough treatment of the mathematics of the map pro-
a map. jection process.
GLOSSARY
aspect: (1) placement of a projection’s center with respect to Cartesian coordinates: a coordinate system in which the
the Earth’s surface; common aspects include equatorial, polar, locations of points in a plane are referenced with respect to
and oblique; (2) the direction in which a topographic slope the x and y distance from two perpendicular intersecting axes
faces (e.g., a northwest aspect would face the northwest). that form an origin.
azimuthal projection: a projection in which all directions or case: describes how a developable surface touches a refer-
azimuths along straight lines radiating from the map projec- ence globe’s surface; the two common cases are tangent and
tion’s center represent great circles on the Earth; directions secant.
are preserved from the center of the map to any other point central meridian: the line of longitude at the center of a map
on the map. about which the projection is symmetrical.
16:30:31.
class: the type of developable surfaces used to create a oblique aspect: a map projection with a center that is aligned
projection, including cylindrical, conic, and planar. somewhere between the Equator and a pole.
cluster analysis: a mathematical method for grouping observa- planar projection: a map projection in which the graticule is
tions (say, counties) based on their scores on a set of attributes. conceptually projected onto a plane that is tangent or secant
compromise projection: a map projection that does not to the Earth’s surface.
possess any specific property but, rather, strikes a balance plane: one of the three developable surfaces; produces the
among various projection properties. planar class of projections.
cone: one of three developable surfaces; produces the conic plotting coordinates: a coordinate system used to draw lati-
class of projections. tude and longitude on a planimetric surface.
conformal projection: a map projection in which angular point of projection: where an imaginary light source is
relationships around any point correspond to the same angu- located when creating a projection.
lar relationship on the Earth’s surface. polar aspect: a map projection with a center that coincides
conic projection: a map projection in which the graticule is with one of the poles.
conceptually projected onto a cone that is either tangent or polyconic projection: a class of projections that conceptu-
secant to the developable surface; in most instances, conic pro- ally involves the placement of many cones over the reference
jections represent meridians as straight lines and parallels as globe, each with its own standard line producing curved
smooth curves about the North Pole. meridians (except for a straight central meridian) and paral-
cylinder: one of three developable surfaces; produces the lels represented as circular arcs.
cylindrical class of projections. programming languages: how users write computer code to
cylindrical projection: a map projection in which the gratic- develop software applications.
ule is conceptually projected onto a cylinder that is either tan- pseudocylindrical projection: a projection that represents
gent or secant to the developable surface; in most instances, lines of latitude as straight, parallel lines (similar to cylindri-
cylindrical projections represent meridians and parallels as cal projections), but the meridians are curved lines
straight lines intersecting each other at 90°. converging to the poles, which are represented as either lines
developable surface: a surface that can be flattened to form or points.
a plane without compressing or tearing any part of it; three R: the stated radius of the reference globe.
commonly used developable surfaces for map projections are
the cylinder, cone, and plane. reference globe: a conceptual spherical model of the Earth
reduced to the same scale as the final map where the scale of
distortion: involves altering the size of the Earth’s land- the reference globe is defined by its radius, R.
masses and the arrangement of the Earth’s graticule when
they are projected to the two-dimensional flat map. scale factor (SF): the ratio of the scale found at a particular
location on a map compared to the stated scale of the map;
equatorial aspect: positioning the developable surface over departure from a scale factor of 1.0 indicates distortion.
the reference globe such that the Equator becomes the central
latitude. secant case: describes a developable surface that intersects
the reference globe along two separate lines, usually two
equidistant projection: a map projection for which distances parallels of latitude.
from one point (usually the projection’s center) to all other
points are preserved compared to those distances on the standard line: a line on a map projection with a scale factor
Earth’s surface. of 1.0; thus distortion is zero.
equivalent (equal-area) projection: a map projection for standard point: a point on a map projection with a scale
which areas are preserved compared to the same areas on the factor of 1.0; thus distortion is zero.
Earth’s surface. tangent case: describes a developable surface that intersects
latitude of origin: the latitude that is set as the center of a the reference globe along one line, usually a parallel of lati-
map projection. tude.
map projection: a systematic transformation of the Earth’s Tissot’s indicatrix: a graphical device plotted on a map pro-
graticule and landmass from the curved two-dimensional jection that illustrates the kind and extent of distortion at
surface to a planimetric surface. infinitesimally small points.
minimum-error projection: a projection having the least variable scale projection: a map projection with a scale that
amount of total distortion of any projection in that same class is manipulated in specific ways to emphasize certain land
according to a specified mathematical criterion. areas at different scales.
REFERENCES
Mulcahy, K. A., and Clarke, K. C. (2001) “Symbolization and Geographic Information Science 28, no.
of map projection distortion: A review.” Cartography 3:167–181.
173
16:30:31.
16:30:31.
Selecting an Appropriate
Map Projection
OVERVIEW
This chapter presents an overview of methods for select- symbol and dot maps so that we can depict both urban
ing an appropriate projection for a variety of thematic and rural population, respectively. Here we need to select
maps. We begin by reviewing several common a projection that preserves areal relations so that the map
approaches to selecting map projections. We focus on a user can properly compare dot distributions in different
projection selection guideline developed by John geographic areas; thus, we focus on the selection of vari-
Snyder because it offers the greatest utility to beginning ous parameters for conic projections: standard parallels,
map designers. Specifically, Snyder’s guideline utilizes central meridian, and central latitude.
three separate tables that initially recommend projec- The third data set examines migration from Europe and
tions based on the extent of the geographic area to be Asia to the United States. This map’s objective is twofold.
mapped: (1) world, (2) hemisphere, and (3) continent, First, we want to show the general migration route over
ocean, or smaller region. Each table is further orga- which immigrants traveled; second, we want the map user
nized according to additional considerations made dur- to see the spherical nature of the Earth over which the
ing the projection selection process. For instance, the immigrants traveled. To meet the first requirement, we
recommended projections for the world category are choose graduated flow lines as the symbology. The map
arranged according to the desired property (e.g., con- user can thus visualize the number of immigrants from the
formal) and any special characteristic (e.g., the projec- different Asian and European countries, as well as the gen-
tion’s center). eral routes taken. To meet the second requirement, it is
To explore the utility of Snyder’s selection guideline, important for the map to communicate the spatial proxim-
in section 2 we discuss how an appropriate projection ity between the United States, Europe, and Asia; thus, we
can be selected for four sample data sets. For our first focus on planar map projections that present a hemisphere.
data set, we imagine creating a world map of literacy The fourth data set focuses on a small geographic area:
rates by country. In this case, we choose to create a choro- Kansas. Specifically, the data set entails the paths of F4 and
pleth map and thus need to select a projection that pre- F5 tornadoes across Kansas during the past 50 years. In
serves areal relations. Here we introduce the advantages this map, we symbolize tornado paths as arrows—the
of using an interrupted projection for world maps, on length of the arrow indicates the distance the tornado
which the graticule has been “cut,” creating lobes that traveled across the ground, and the direction of the arrow
encompass specific geographic areas (e.g., South indicates the general direction the tornado took. Selecting
America). On an interrupted projection, lobes showing projections for this small geographic area highlights the
specific geographic areas defined by longitude bounds fact that regardless of the projection property chosen (con-
are created that have lower distortion than if those same formal vs. equidistant), small geographic areas are equally
areas were shown on a noninterrupted world map. well represented by most projections.
The second data set focuses on creating a map show- The chapter concludes with a discussion of five key
ing the population distribution of Russia at the oblast objectives that should be emphasized when selecting a
and kray levels. In this case, we combine proportional map projection for a thematic data set. First, in most cases,
175
17:32:08.
Selecting an Appropriate Map Projection
the mapmaker should select a projection with the lowest generally is not very useful, as it does not consider the
distortion. Second, amounts of distortion can be kept map’s purpose.
small by aligning the geographic area (or data set) under Arthur Robinson and his colleagues (1995)
consideration with the standard line(s) or by positioning described another simple guideline for selecting projec-
the map’s center with the standard point. Third, as the tions based on the relationship between projection
amount of geographic area under consideration increases properties and the intended map purpose. For instance,
(e.g., mapping Kansas compared to mapping the United they recommended conformal projections for analyz-
States), distortion becomes a more important considera- ing, measuring, or recording angular relationships, as in
tion. Fourth, just because a projection has seen consider- navigation, piloting, and surveying; suitable projections
able exposure (e.g., it has been used in prominent atlases) include the Mercator, transverse Mercator, Lambert
does not mean that the projection is suitable for your spe- conformal conic, and stereographic. They recommended
cific application. Fifth, an often overlooked aspect of the equivalent projections for geographic comparison
map projection is its influence on the overall map design— across a map, a common purpose of thematic maps. For
a topic that is not well studied by cartographers. example, a dot map’s primary goal is to visually com-
pare different dot distributions across geographic areas,
and this comparison is greatly facilitated if the geo-
1 POTENTIAL SELECTION GUIDELINES graphic areas are represented in their correct propor-
tions. Equivalent projections that they recommended
Selecting an appropriate map projection is one of the include the cylindrical equivalent, Lambert’s azimuthal,
more complex tasks in cartography. In essence, when Albers and Lambert conic, and most equivalent
selecting a projection, a match must be made between pseudocylindrical projections. Robinson et al. also
the map’s purpose and the various projection properties noted that when recording and tracking direction of
and characteristics. This task is difficult because of the movement, planar projections are useful, including the
many variables involved when creating a map, such as orthographic (useful for showing the Earth as if viewing
the map’s scale, the amount of the Earth to be mapped, it from space), azimuthal equidistant (useful for show-
the level of generalization, and the thematic symbology ing correct directions and distances from the center to
used. Similarly, any given projection has numerous char- any other point), and the gnomonic (useful for showing
acteristics (i.e., class, case, and aspect) and associated that all great circles are represented by straight lines).
properties (i.e., equivalent, conformal, equidistant, and Robinson et al. also mentioned that some projections
azimuthal). Rarely will a single projection have all the have been designed for specific applications. One such
characteristics and properties necessary to satisfy all example is the space oblique Mercator, which is used to
variables involved in the mapmaking process. record the ground track of satellites orbiting above the
In an attempt to recommend projections for specific Earth’s surface.
map purposes, various projection selection guidelines
have been developed. Although the format of each 1.1 Snyder’s Hierarchical Selection Guideline
guideline differs, the general purpose of each is to pro-
vide the mapmaker with one or more projections that Although the preceding approaches are conceptually
can reasonably be applied to a particular purpose. For useful, we feel that neither guideline provides the level
example, a simple selection guideline discussed by of detail or the flexibility needed to adequately select
Frederick Pearson (1984) related the choice of projec- an appropriate projection. Thus, we focus on a guide-
tion to the range of latitude depicted on the map. line developed by John Snyder (1987) that permits the
Equatorial regions lying 30° on either side of the equa- mapmaker to select a map projection for many map-
tor are mapped with cylindrical projections; midlatitude ping purposes. Snyder presented a hierarchical list of
regions between 30° and 65° are mapped with conic pro- suggested projections that is organized according to
jections; and polar areas above 65° are mapped with pla- the region of the world to be mapped, the projection
nar projections. This guideline’s logic rests on the fact property (e.g., equivalent, conformal, equidistance,
that the location of the standard point or line(s) of a pro- azimuthal), and the characteristic (e.g., class, case, and
jection class lies within the geographic area recom- aspect). The advantage of a hierarchical approach is
mended for that class, hence ensuring that distortion is that the mapmaker can begin with a broad question
low throughout the region of interest. For instance, those (e.g., size of geographic region to be mapped) and log-
areas lying along the Equator are matched with cylindri- ically proceed to more detailed questions about the
cal projections that have one standard line coinciding mapping situation (e.g., an oblique, polar, or equator-
with the Equator or two standard lines equally spaced ial aspect). In so doing, the mapmaker is led down a
on either side of the Equator. Although Pearson’s guide- particular path through the hierarchy until an appro-
line provides a starting point for selecting projections, it priate projection is recommended.
17:32:08.
Snyder’s hierarchy begins with dividing geographic and oblique Mercator projections should be chosen,
areas into three categories: (1) world, (2) hemisphere, respectively. If constant scale is not a requirement, then
and (3) continent, ocean, or smaller region. Snyder also the Lagrange, August, and Eisenlohr projections are
drew a distinction between the appropriate models of recommended (Figure 1). The Lagrange conformal pro-
the Earth for each geographic area. For instance, for jection shows the world within a circle—an advantage
maps of the world’s hemispheres, continents, and over the stereographic conformal projection, which
oceans, a spherical model of the Earth is reasonable. commonly shows only one hemisphere. The August and
With these large geographic extents, the map uses do Eisenlohr projections, both conformal epicycloidal*
not call for highly accurate maps such as large-scale projections, are useful because they are conformal
topographic ones. For smaller regions, an ellipsoidal throughout—on many conformal projections, the prop-
model should be used, as smaller regions often form the erty of conformality is not maintained at the poles. A
base of maps on which highly accurate measurements shortfall of the epicycloidal conformal projections is the
are made (e.g., navigational charts). rapidly increasing scale distortion from the projection’s
center. For instance, note that Antarctica’s appearance
World Map Projections in the conformal projections in Figure 1 is greatly exag-
gerated (in reality, South America’s area extent is
Table 1 presents Snyder’s map projection guideline for
roughly 19 million km2, whereas Antarctica’s is only
creating a map of the entire world. The mapmaker
14 million km2).
begins by choosing from among the following projec-
If an equivalent projection is desired, there are three
tion properties: conformal, equivalent, equidistant,
general categories: noninterrupted, interrupted, and
straight rhumb lines, and compromise distortion. If a
oblique aspect. An interrupted projection shows the
conformal projection is desired, then the mapmaker has
graticule as “cut” along specific meridians creating one or
several projections from which to choose according to
more lobes, as shown by the Eckert V pseudocylindrical
the scale variation across the projection. For instance,
when the map purpose requires that a constant scale be * An epicycloidal projection is conceptually developed as a curve
shown along the Equator, a meridian, or an oblique traced by a point on a circle’s circumference of radius a that rolls
great circle, then the Mercator, transverse Mercator, along the outside of a fixed circle having a radius of 2a.
TABLE 1 Snyder’s map projection guideline showing projections for mapping the world
Region Mapped Property Characteristic Named Projection
World Conformal Constant scale along Equator Mercator

Constant scale along a meridian Transverse Mercator
Constant scale along an oblique Oblique Mercator
great circle
No constant scale anywhere on Lagrange
the map August
Eisenlohr
Equivalent Noninterrupted Mollweide
Eckert IV & VI
McBryde or McBryde–Thomas
Boggs Eumorphic
Sinusoidal
Other miscellaneous pseudocylindricals
Hammer (a modified azimuthal)
Interrupted Any of the above except Hammer
Goode’s Homolosine
Oblique aspect Briesemeister
Oblique Mollweide
Equidistant Centered on a pole Polar azimuthal equidistant
Centered on a city Oblique azimuthal equidistant
Straight Mercator
rhumb lines
Compromise Miller cylindrical
distortion Robinson pseudocylindrical
177
17:32:08.
A B C
FIGURE 1 Different conformal projections of the world. The Lagrange (A) presents the world in a circle, whereas the August (B)
and Eisenlohr (C) are epicycloids.
projection in Figure 2. Each lobe has its own central have lower overall distortion than projections that pre-
meridian that creates an area of lower distortion than if serve a single property. In many cases, compromise pro-
the projection was not interrupted. The number of lobes jections are developed specifically to match the shape of
and placement of each cut is determined by the map- landmasses to their appearance on a globe, as in the
maker. In Figure 2A, two hemispheres are created (the Robinson projection.
cut is placed along the Prime Meridian and 180°). In con-
trast, in Figure 2B, the cuts are placed along the Atlantic Map Projections for a Hemisphere
and Indian Oceans so that the continents have lower dis- Table 2 presents Snyder’s projection guidelines for
tortion. In Figure 2C, the cuts are again placed along the maps of a hemisphere. Planar projections are suited for
oceans, but an additional cut is made in the southern mapping hemispheres because a standard point is
Pacific Ocean. Although the cuts in Figure 2B and Figure placed at the center of the projection and the projection
2C are positioned over water to provide continuity of the can be easily recentered over any area of importance.
landmasses, the mapmaker, depending on the map’s pur- Planar projections are also interesting in that many of
pose, might choose to place the cuts over land, preserving them combine the azimuthal property (showing direc-
the continuity of the oceans. tions correctly from the projection’s center to all other
The oblique aspect (a projection with a center that is points) with an additional property (e.g., the stereo-
neither the pole nor the Equator) is warranted when geographic projection is both conformal and azimuthal),
graphic areas not aligned along the Equator or centered whereas most projections possess only one useful prop-
at the pole must be brought to the center of the projec- erty (e.g., the Albers conic projection is exclusively
tion. For example, Figure 3 shows the Briesemeister mod- equivalent). When showing a hemisphere, the following
ified azimuthal projection,* which is equivalent and is cen- properties are appropriate: conformal, equivalent,
tered over northwestern Europe at 10° E and 45° N, thus equidistant, and global look. For each property, note
bringing northern Europe, the Norwegian Sea, and that a specific planar projection is recommended.
Scandinavia to the map’s center.
In situations where equidistance is required, Snyder Map Projections for a Continent, Ocean,
suggested the azimuthal equidistant projection. Snyder or Smaller Region
also included projections in his guideline that require Table 3 presents Snyder’s guidelines for maps of a con-
straight rhumb lines or compromise distortion. Straight tinent, ocean, or smaller region. Note that this hierarchy
rhumb lines are a special property that only a few pro- includes a subdivision that distinguishes between the
jections possess; for instance, on the Mercator projection,
all rhumb lines (i.e., lines of constant compass bearing) TABLE 2 Snyder’s projection selection guideline showing
appear as straight lines; this is obviously needed in cer- planar projections for mapping a hemisphere
tain mapping applications such as navigation. Some pro- Region Mapped Property Named Projection
jections do not have a specific property and are described
Hemisphere Conformal Stereographic conformal
as compromise projections. These projections typically
Equivalent Lambert azimuthal
equivalent
* The
Equidistant Azimuthal equidistant
Briesemeister projection, always centered at 10° E and 45° N, is Global look Orthographic
a special oblique case of the Hammer modified azimuthal projection.
17:32:08.
FIGURE 2 The Eckert V pseudo-

cylindrical projection showing differ-
ent placement of cuts.
predominant directional extent of the landmass: standard lines can be positioned to coincide with the
east–west, north–south, oblique (not aligned in an east–west extent. In contrast, Antarctica, a landmass
east–west or north–south direction), and equal extent in that is approximately of equal extent in all directions,
all directions. This subdivision matches the position of will be more appropriately mapped with a single stan-
the standard line(s) or point to the directional extent of dard point, as distortion increases radially with distance
the landmass, ultimately leading to a reduction of dis- from this point.
tortion for the geographic area under consideration. For After examining directional extent of the land-
instance, a map of Canada, which has considerable mass, the mapmaker should consider the location of the
east–west extent, will have lower distortion when the landmass to be mapped. For instance, given a landmass
179
17:32:08.
FIGURE 3 The Briesemeister modi-

fied azimuthal projection centered at
10° E and 45° N.
of predominant east–west extent on the Earth’s surface, centered on the landmass is recommended. For land-
the region can be located along the Equator or away masses with extents that are not aligned north–south or
from the Equator. If the landmass is located along the east–west, Snyder recommended an oblique Mercator
Equator, one of the cylindrical projections that have the or oblique cylindrical equivalent projection with a stan-
Equator as a standard line is suitable (e.g., Mercator or dard line aligned with the predominant extent of the
cylindrical equivalent). If the landmass’s east–west landmass. If a landmass has no predominant directional
extent is located away from the Equator (e.g., the mid- extent, then a planar projection having a standard point
latitudes), then a conic projection with one or two stan- is recommended (e.g., stereographic or Lambert
dard lines is a suitable choice (e.g., Lambert conformal azimuthal equivalent). The last step is to examine which
conic or Albers equivalent conic). For landmasses hav- projection property (i.e., conformal or equivalent) is
ing a predominant north–south extent, a transverse appropriate—topographic maps usually require confor-
Mercator or transverse cylindrical equivalent projec- mal projections, whereas thematic maps usually require
tion with a standard line corresponding to a meridian equivalent projections.
TABLE 3 A portion of Snyder’s map projection guideline, showing projections for mapping
a continent, ocean, or smaller region
Region Mapped Directional Extent Location Property Named Projection
Continent, ocean, East–West Along the Equator Conformal Mercator

or smaller region
Equivalent Cylindrical
equivalent
Away from the Equator Conformal Lambert conformal
conic
Equivalent Albers equivalent
conic
North–South Aligned anywhere Conformal Transverse Mercator
along a meridian
Equivalent Transverse
cylindrical equivalent
Oblique Anywhere Conformal Oblique Mercator
Equivalent Oblique cylindrical

equivalent
Equal extent Polar, Equatorial, or Conformal Stereographic
Oblique
Equivalent Lambert azimuthal
equivalent
17:32:08.
2 EXAMPLES OF SELECTING PROJECTIONS more lobes. Two factors to consider when deciding on
the appropriateness of an interrupted projection are the
In this section, we utilize Snyder’s guideline as a foun- intended audience and where the data exist on the map.
dation for selecting projections for four fictitious map- Recall that interrupted projections remove the continu-
ping situations. For each mapping situation, we discuss ity of the oceans or landmasses, which could make the
the logic of Snyder’s guideline and why the projections map difficult to comprehend. If the audience is assumed
he recommended are appropriate. Additional insights to have poor geographic knowledge (e.g., elementary
into selecting projections not covered by Snyder’s school children), an interrupted projection would be
guidelines are also explored. inappropriate. On the other hand, if the audience is
assumed to have good geographic knowledge (e.g., pol-
icymakers), then an interrupted projection would be
2.1 A World Map of Literacy Rates acceptable. In deciding whether an interrupted projec-
tion is appropriate, a second factor is the distribution of
For our first example, imagine that we wish to create a the data across the map. In the literacy rate example,
world map illustrating the distribution of literacy rates the data are concentrated over the land and not over
by country. We begin our selection process by consider- the oceans. Because the data are concentrated over
ing the data, symbolization method, intended audience, land, the continuity of the oceans can be sacrificed to
and overall purpose for the map. Literacy rate is nor- create an interrupted projection. Therefore, the cuts can
mally computed as the percentage of all people who can be placed over the oceans, with each region of interest
read. Although the underlying phenomenon of literacy being placed in a different lobe, as in Figure 4C.
rates is arguably spatially continuous in nature, we pre- Comparing Figure 4C with the noninterrupted
sume that we wish to examine the phenomenon at the Mollweide projection in Figure 4A, we see that the cuts
level of countries for which data are available (in the along the oceans create a discontinuity that might make
CIA Factbook*). Given our desire to portray data at the the map difficult for those with limited geographic
country level, choropleth symbolization is suitable. We knowledge to interpret. However, it is also important to
further presume that the purpose is to create a map that note that in Figure 4C, the landmasses toward the pro-
allows for a comparison of literacy rates by country jection edges (northwest Canada, Chile, Argentina, and
around the world—noting the concentrations of high Australia) appear less distorted than they do in Figure
and low literacy rates, and permitting us to examine the 4A.
spatial pattern. We also presume that we wish to high- Having examined the possibility of an interrupted or
light those areas of the world with low literacy rates. noninterrupted projection, another consideration is
With our purpose in mind, we now consider Snyder’s whether or not to specify an oblique aspect. Most world
projection guideline. Because our focus is on the world, maps employ an equatorial aspect, where the map is
we start with Table 1. Our first step is to select a desired centered at the Equator along a chosen central merid-
projection property. Given our choice of choropleth ian––usually selected so that either the Atlantic or
symbolization, a critical goal is to preserve the relative Pacific Ocean is central to the map. In cases where the
areas of each enumeration unit as each would appear geographic area of interest is not along the Equator, the
on a globe. If this is not done, then areas appearing projection’s center must be moved to allow the geo-
larger than they should will result in greater visual graphic area of interest to be brought to the center of
emphasis on the associated data. This sort of thinking the map––an advantage of the oblique aspect. Recall
leads us to choose an equivalent projection such as the that on most projections, lower amounts of distortion
Mollweide shown in Figure 4A. For comparison, Figure are usually found at a projection’s center. This is espe-
4B shows a Mercator conformal projection, which is cially true on world maps where, for example, polar
inappropriate for this situation. Note how the Mercator areas suffer the greatest distortion when a point some-
projection emphasizes the high literacy rate category in where along the Equator is the projection’s center. To
the upper latitudes, suggesting that it dominates the determine whether an oblique projection is necessary,
world. we need to again examine the geographic distribution of
Once the property of equivalence is decided, the the data and determine if a particular geographic area of
mapmaker has to choose the characteristic of the pro- the Earth has high or low literacy rates that are impor-
jection: noninterrupted, interrupted, or oblique aspect. tant and, if so, where they are. Based on our original
As mentioned earlier, an interrupted projection reduces statement, we wish to focus on those areas with low lit-
distortion across a geographic area by creating one or eracy rates in an attempt to highlight their situation. In
reviewing the literacy data, Central Africa and portions
*The CIA World Factbook is available at https://www.cia.gov/cia/ of Central Asia have the lowest literacy rates and should
publications/factbook/index.html. be brought to the center of the map. These geographic
181
17:32:08.
Comparison of Map Projections

Literacy Rates, 2003
A Mollweide Equivalent B Mercator Conformal

(centered on the Prime Meridian)
Percentage of Adults Who are Literate Percentage of Adults Who are Literate
No Data 17.6 to 53.0 53.1 to 70.6 70.7 to 85.1 85.2 to 94.1 94.2 to 100 No Data 17.6 to 53.0 53.1 to 70.6 70.7 to 85.1 85.2 to 94.1 94.2 to 100
C Mollweide Equivalent D Mollweide Equivalent

(interrupted form) (centered on 40º E)
FIGURE 4 World literacy rates mapped on the (A) Mollweide equivalent projection, (B) Mercator conformal projection,
(C) an interrupted Mollweide projection, and (D) a Mollweide projection that has been recentered along the Equator at 40° E.
areas are situated along the Equator, and thus an new central meridian is chosen to coincide with 40° E, as
oblique aspect is not called for. However, we should con- shown on the Mollweide projection in Figure 4D.
sider recentering the projection over the area of interest An additional thought on selecting an appropriate
by selecting a new central meridian. In this case, the area projection not directly addressed by Snyder is the idea
of low literacy rates primarily falls near 40° E, and so a of the poles being represented by a point or a line. In
17:32:08.
Figure 5, the quartic authalic and Eckert IV equivalent map. We assume that this map is to be included in a
projections are shown. Notice that the quartic authalic textbook for high school students who are taking an
projection (Figure 5A) represents the poles as points, introductory course in geography. In this context, the
whereas the Eckert IV projection (Figure 5B) repre- map will serve a supporting role in the discussion of
sents the poles as lines. On the quartic authalic projec- Russia and its cultural geography. Population can be
tion, landmasses in the upper latitudes are compressed thought of as discrete entities (individual people), but
and difficult to see due to the convergence of the merid- in some cases can be assumed to be more continuous
ians to a point (especially at smaller scales). On the in nature (e.g., a visit to New York City can give the
other hand, on the Eckert IV projection, the landmasses impression that people are everywhere). Thus, in con-
in the upper latitudes are stretched in an east–west ceptual terms, population distribution is highly vari-
direction due to the nonconvergence of the meridians, able. In Russia, there are very dense urban centers and
making recognition of the landmasses more apparent, vast expanses of emptiness. To capture this variation, a
but distorting their overall shapes. combined proportional symbol–dot method seems
To help determine whether a point or a line should appropriate, as it facilitates the observance of high
represent the poles, we can examine the map’s geo- population concentrations (urban areas) and expanses
graphic focus. Because our area of focus is Africa and of low population (e.g., in central Siberia).
Central Asia and not the upper latitude areas, it might Starting with Snyder’s guideline, we turn toward
make sense to use a projection where the poles are rep- the continent, ocean, or smaller region selection
resented by a point. According to Snyder’s guideline, guideline (Table 3). The first selection criterion to be
the Mollweide, Boggs, and Sinusoidal projections all determined is the directional extent of the landmass.
represent the poles as a point. On the Mollweide pro- Russia clearly has a greater east–west than a
jection, the meridians curve to the poles more gently north–south extent (Figure 6): The east–west extent is
than on the Boggs or Sinusoidal, and the landmasses approximately 5,163 miles (ranging from 27° E along
are not as compressed in the upper latitudes, thus mak- the Baltic border to 170° W in the Bering Sea),
ing it a suitable choice for the literacy rate data. whereas its north–south extent is 2,764 miles (ranging
from 40° N along the Azerbaijan border to 80° N in
2.2 A Map of Russian Population Distribution the Arctic Ocean). Moreover, Russia is positioned
away from the Equator—60° N is the approximate
For our second example, we consider mapping the central latitude of the country’s north–south extent.
population distribution in Russia for 1993. As in the Based on this discussion and according to Table 3, we
previous example, we begin the process of selecting a see that Russia has a considerable east–west direc-
projection by considering the data, symbolization tional extent and is located away from the Equator in
method, intended audience, and overall purpose of the the midlatitudes, calling for a conic projection.
A Literacy Rates, 2003 B Literacy Rates, 2003

(Quartic Authalic) (Eckert IV)
FIGURE 5 The literacy rate data shown on the (A) quartic authalic and (B) Eckert IV equivalent pseudocylindrical projections.
The quartic authalic represents the poles as points, and the Eckert IV represents the poles as lines.
183
17:32:08.
not change (each map has 15 dots), there obviously is a

difference in the density of dots over the given area.
Snyder recommended the Albers equivalent conic
projection for thematic maps of landmasses that have a
predominant east–west extent and are located away
from the Equator because most conic projections have
one or two standard lines that coincide with a line of lat-
itude that can be placed to minimize the distortion of
the mapped area. Although there are no universally
accepted rules on the placement of standard parallels on
conic projections, the overall goal is to select latitudes
that will result in the lowest distortion over the mapped
area. One approach is to place one standard parallel at
one-sixth the latitudinal distance from the southern lati-
tude limit of the map and the other standard parallel at
one-sixth the latitudinal distance from the northern lati-
tude limit of the map.* For example, Russia’s southern
latitude limit is approximately 41° 43'N (found between
the border of Russia and Azerbaijan), and the northern
latitude limit is approximately 81° 53'N (along the
northern border of Ostrov Komsomolec––an island in
FIGURE 6 An orthographic projection showing the consid- the Arctic Ocean). This latitude range is close to 40° 10'.
erable east–west extent of Russia. One-sixth this latitudinal range of 40° 10' is 6.7° or
6° 42'. Given the latitudinal range and the general place-
ment guideline presented earlier, the two standard par-
To select a specific conic projection, we need to con-
allels would be placed at 48° 25'N (i.e., 41° 43' + 6° 42')
sider the appropriate projection property. Snyder’s
and 75° 11'N (i.e., 81° 53' - 6° 42'). You should realize
guideline (Table 3) offers conformal or equivalent
that the approach for standard parallel placement dis-
properties for the conic projections. Because we are
cussed here is only a rule of thumb; more sophisticated
creating a combined graduated symbol–dot map, an
mathematical approaches to selecting standard parallels
equivalent projection is necessary to ensure that map
have been developed. For details, see Snyder (1993) and
users will correctly interpret the relationship between
Maling (1992).
the phenomenon represented by the dots and the geo-
In addition to the selection and placement of suit-
graphic area in which it is contained. To demonstrate
able standard parallels, most conic projections have
the importance of an equivalent projection, examine
other parameters that can be modified. One parame-
Figure 7, which shows a fictitious dot map for some
ter is the central meridian, which should be chosen so
enumeration unit. In Figure 7A the enumeration unit is
that it is at the center of the longitudinal range. The
shown on an equivalent projection, whereas in Figure
longitude limits of Russia are approximately 163°
7B the same enumeration unit is projected on a con-
(ranging from 27° E, Ostorov Gogland—a small island
formal projection. Although the number of dots does
in the Gulf of Finland—to 170° W, the boundary
between Russia and Alaska). Given this range, the
central meridian for Russia would fall at approxi-
A B mately 108° 30'E. Another parameter for most conic
projections is the central latitude, which is the lati-
tude that falls halfway between the northern and
southern latitudinal limits.† Assuming the northern
* Although the “one-sixth” rule will not yield the lowest distortion over
the mapped area per se, the procedure will result in low distortion
without the complex mathematics associated with finding the standard
parallels that result in the lowest distortion over the mapped area.
† Note that although the central latitude definition might seem obvious,
it is a parameter that is available when defining or changing projection

FIGURE 7 A fictitious data set represented through dot sym-
parameters in many Geographic Information System (GIS) software
bolization on (A) an equivalent projection, and (B) a confor- programs, because the central latitude is used to vertically center the
mal projection. projection on the page.
17:32:08.
and southern latitude values of 41° 43'N and 81° 53'N, atlases of the former Soviet Union (Snyder 1993). An
respectively, the central latitude would be placed at interesting comparison of the graticule spacing of the
61° 48'N. Using these values for the Albers conic pro- parallels between the Albers equivalent conic and the
jection parameters, a combined graduated symbol–dot Kavrayskiy IV equidistant conic is shown in Figure 9.
map of the population distribution of Russia is shown On the Albers equivalent projection (Figure 9A), the
in Figure 8. spacing of the parallels decreases from the standard
It is interesting to note that the Albers equivalent lines toward the polar regions (a characteristic appear-
projection was introduced by Heinrich Albers in 1805, ance of lines of latitude on equivalent projections),
but was not frequently employed until the 1900s, when it whereas the Kavrayskiy IV equidistant projection
became the primary projection used by the U.S. (Figure 9B) shows the parallels as equally spaced along
Geological Survey for maps of the conterminous United a meridian (a characteristic appearance of lines of lati-
States.* The standard parallels for these maps were tude on equidistant projections).
placed at 29° 30'N and 45° 30'N and resulted in a maxi-
mum scale error throughout the map of no more than 2.3 A Map of Migration to the United States
11/4 percent (Deetz and Adams 1945). Other countries,
such as Russia, also have utilized a single conic projec- For our third example, we focus on an appropriate
tion for atlas mapping of their entire country. For exam- projection for a series of maps (a small multiple)
ple, the Bol’shoy Sovetskiy Atlas Mira relied heavily on showing migration to the United States. We assume
the Kavrayskiy IV equidistant conic projection for the map will be included as part of an atlas describing
a history of U.S. ethnicity and will be published by the
* The National Atlas, published by the U.S. Geological Survey in 1970, U.S. Census Bureau for the general public. Although
contains numerous maps drawn on the Albers equivalent projection. the atlas will have maps showing migration from all
Population of Russia, 1993

Values in Millions
8.5
6.5
4.5 25,000 People
2.5
0.5
FIGURE 8 A combined proportional symbol–dot map showing the population distribution across Russia on an Albers
equivalent projection. Data obtained from United Nations Educational Scientific and Cultural Organization (UNESCO),
1987, through United Nations Environment Program (UNEP)/Global Resource Information Database (GRID)-Geneva at
http://www.grid.unep.ch/index.php.
185
17:32:08.
A B
FIGURE 9 The (A) Albers equivalent and (B) Kavrayskiy IV equidistant conic projections. Both projections are at the same
scale and have standard parallels at 47° N and 62° N.
populated continents, this specific series of small, mul- data. Conformality preserves angles at infinitesimally
tiple maps is designed to show the number of individ- small points throughout the mapped area, but this prop-
uals who migrated to the United States from erty is not required for our data. Equivalent projections
European and Asian countries during each 10-year preserve areal relations throughout the mapped area. In
period between 1960 and 2000. A goal is not to simply the previous two examples (world literacy rates and
show the number of immigrants per country, but to Russian population distribution), there was a justifiable
provide a general sense of the route by which the need for equivalent projections. However, the enumera-
migration took place. Thus, we argue that the flow tion units of the migration data are not directly associ-
map is an appropriate symbolization method. ated with the flow line symbol (i.e., unlike a choropleth
Turning to Snyder’s selection guideline, we come map, the enumeration units on a flow map are not a sym-
across the problem that the mapped area does not bol); thus, there is no direct need to preserve areal rela-
neatly fit any of his predefined geographic categories. The tions. An equidistant projection preserves distances from
closest geographic category that our mapped area fits into one point to all other points, but also is not directly
is a hemisphere. In this case, given the source areas for the related to our map purpose. Global look projections (the
immigrants (European and Asian countries) and their orthographic being among them) give the appearance of
ultimate destination (the United States), the Northern the Earth as if looking down on it from outer space,
Hemisphere is the logical geographic area to be mapped. which often produces an eye-catching map. However,
Because the United States should figure prominently in global look projections are typically limited to showing a
the center of the map and because it is located neither hemisphere and have considerable distortion near the
along the Equator nor at a pole, the projection calls for an map edges.
oblique aspect. Moreover, the longitudinal range of our Although no specific property appears to be an obvi-
data encompasses approximately 180°—a hemisphere. ous choice for a projection for the migration data, a
Therefore, we can proceed with Snyder’s scenario that mapping situation such as this provides an opportunity
considers projections for a hemisphere (in our case, the to be creative and make use of the flexibility that planar
Northern Hemisphere). Examining Table 2, we now must projections provide. One planar projection appropriate
determine which projection property is desirable for our for the migration data is the azimuthal equidistant.
17:32:08.
Although discounted earlier on the basis that our data Central America; notice the dramatic effect this projec-
do not require directions or distances to be measured tion has in emphasizing the United States as the
correctly, the projection does have interesting qualities surrounding area fades into the background. Obviously,
that make it suitable for the migration data. For instance, with a projection such as this, the scale changes dramati-
the azimuthal equidistant projection has lower overall cally from the center outward. Applying the vertical per-
scale distortion than projections that preserve a single spective azimuthal projection to the migration data, we
property, such as equivalence. As a result, equidistant produced the map shown in Figure 13. Here, the map’s
projections tend to minimize the distortion of shapes of focus is the United States, which is clearly the conver-
landmasses. For this reason, Maling (1992, 109) indicated gence point of the flow lines. The shapes of the land-
that “equidistant map projections are often used in atlas masses are shown similar to their appearance on the
maps, strategic planning maps and similar representa- globe so as to not confuse potential map users about the
tions of large parts of the Earth’s surface.” Another geographic area. The projection also captures the user’s
interesting quality of the azimuthal equidistant projec- attention by providing a compelling view of the data.
tion is that it displays the entire world (in its default con-
figuration), whereas most planar projections are capable 2.4 A Map of Tornado Paths Across Kansas
only of showing a single hemisphere. Although the (1950–2000)
azimuthal equidistant projection can show the entire
world (Figure 10A), it can easily be cropped to focus on In the fourth projection selection example, we focus on
specific geographic areas, as shown in Figure 10B. the paths that category F4 and F5* tornadoes took
Another useful planar projection that would allow the across Kansas from 1950 to 2000. As before, we begin
migration routes to be effectively shown, as well as pro- by considering the data, symbolization method,
vide an interesting perspective view of the Earth, is the intended audience, and overall map purpose. We
vertical perspective azimuthal projection. With this pro- assume that this map is to be created for weather
jection, the amount of land that is shown can be manipu- experts who are interested in learning about counties in
lated—note that Figure 11A shows much more land Kansas that historically have been hit by tornadoes.
within the border of the projection than Figure 11B, Specifically, they are interested in using this map to
which focuses on North and South America. The vertical visualize the precise track and distance along the
perspective azimuthal projection also can be used to
zoom in on a portion of the Earth’s surface, which is sim- * The Fujita scale is used to indicate the intensity of a tornado, where
ilar to using a camera to zoom in on an object. For F0 is the least intense and F5 is the most intense according to the
instance, Figure 12 illustrates a zoom for North and damage inflicted by the tornado.
A B
FIGURE 10 (A) The azimuthal equidistant planar projection showing the entire world and (B) the same projection cropped to
focus on the migration data.
187
17:32:08.
A B
FIGURE 11 Two perspective views of the Earth using the vertical perspective azimuthal projection.
selecting a projection for a continent, ocean, or smaller

region as the geographic area to be mapped, will provide
suitable guidance for selecting a projection for Kansas.
The first step in applying Table 3 is to determine the
directional extent of the region to be mapped. In this
case, Kansas has a considerably greater east–west than
north–south extent. In fact, Kansas is about twice as long
in an east–west extent (about 400 miles vs. about
200 miles). In the second step, we determine the posi-
tional characteristic of Kansas. Lying between 37° N and
40° N, Kansas clearly is positioned away from the
Equator. For the third step, we decide on the appropriate
projection property. Notice that for an east–west extend-
ing landmass that is located away from the Equator,
Snyder’s table lists conic projections that are either con-
formal or equivalent. Because the data focus on the trac-
ings of tornado paths across Kansas, angles and distances
would be the appropriate projection properties to con-
sider. We can eliminate equivalent projections because
this property is not appropriate—we are not interested
in preserving areal relations. Recall that conformal
FIGURE 12 The vertical perspective azimuthal projection projections preserve accurate angles about individual
zoomed into North and Central America. point locations (e.g., where a tornado touches down).
However, conformal projections do not preserve scale,
and therefore distances shown on the projection would
ground each tornado took. This map requirement not be the same as the distances found on the Earth.
focuses on a geographic area to be mapped that is com- Since one of the requirements for this mapping situa-
paratively smaller than the other map examples we tion was to measure distances, we decided to go outside
have considered. of Snyder’s guidelines and see whether an equidistant
Next, we consider which of Snyder’s categories to use projection would be a suitable choice. Recall that
for our mapping situation. Because we are dealing with a equidistant projections allow for distances to be mea-
smaller region for this data set, Table 3, which handles sured as they would be found on the Earth, but do not
17:32:08.
Migration to the United States from Europe and Asia

1961 to 1970 1971 to 1980
1981 to 1990 1991 to 2000
Population in Thousands
441 to 550 1,650
331 to 440
221 to 330
1,250
111 to 220
0 to 110 1,100
FIGURE 13 The vertical perspective azimuthal projection showing migration to the United States from Europe and Asia dur-
ing different time periods.
189
17:32:08.
preserve angles. However, equidistant projections restrict Figure 15A does indeed show the indicatrix as a series of
the way in which distances are preserved (e.g., all dis- circles on the Lambert conformal conic projection, but
tances are correct from the center of the map to any Figure 15B also portrays the indicatrix as a series of cir-
other point). Thus, as in the previous example, there is no cles on the equidistant projection, suggesting that for this
single suitable map projection for this particular map- data set, the choice of projection property does not have
ping situation. A conformal or equidistant conic projec- a significant impact on the visual display of the data.
tion might be sufficient, but trade-offs are associated with We can determine whether the geometric shapes in
each projection property and resulting distortion. Figure 15 are actually circles or ellipses by examining
To help solve this dilemma between the conformal Table 4, which shows various distortion values for the
and equidistant properties, using various distortion mea- Lambert conformal and equidistant conic projections.
sures we will compare the differences in the amount of Beginning on the left side, the first two columns in the
distortion between the Lambert conformal conic and table list every degree of latitude from 33° N to 45° N
the equidistant conic projections. To begin our distortion along 98° W longitude. Also listed for each projection are
analysis, examine Figure 14, where we see the Lambert a, the scale factor for the semimajor axis; b, the scale fac-
conformal conic (Figure 14A) and the equidistant conic tor for the semiminor axis; area distortion; and angular
(Figure 14B) projections for Kansas. The north–south distortion values.
latitudinal limits range from approximately 35° N to 42° In examining the table, note that the scale factors
N, and the east–west longitudinal limits range from for a and b on the Lambert conformal conic projection
about 90° W to 104°W. Each projection shares the same at any given latitude are equal, a characteristic of con-
conic parameters of a central meridian placed at 98° W, formal projections. The scale factors equal 1.0 at the
a central latitude at 38° 30'N and two standard paral- location of the standard parallels. On this projection,
lels—one at 40° 50'N and the other at 36° 10'N.* the areal distortion is not severe—ranging from 1.012
Although the projections differ in their respective prop- (a very slight exaggeration of areas) to 0.998 (a very
erties, there appears to be nothing that visually distin- slight compression of areas). There is no angular dis-
guishes the mapped appearance of Kansas on the two tortion on this conformal projection—only values of
projections. 0.0° are reported in the Angular Distortion column.
Now examine Figure 15, which shows Tissot’s indica- For the equidistant conic projection, the values for a
trix for the same mapped area. Recall that the indicatrix and b range from 0.999 around the standard parallels
appears as a circle of different sizes on conformal projec- to 1.007 toward the top of the map. The amount of areal
tions (i.e., circles are increasingly larger the greater their distortion is slightly less (ranging from 1.006–0.999)
distance from the standard point or lines), but it gener- than on the Lambert conformal conic projection
ally appears as ellipses on an equidistant projection. (1.012–0.998). There is also no more than 0.34°
(approximately one-third) of a degree of angular dis-
*The location of these standard parallels was set at one-sixth the tortion on the equidistant conic projection over this
distance from the limiting parallels. geographic area, which is rather low.
A B
40º 40º
38º 38º
36º 36º
102º 100º 98º 96º 94º 102º 100º 98º 96º 94º
FIGURE 14 The (A) Lambert conformal conic and (B) equidistant conic projections of Kansas.
17:32:08.
A B
40º 40º
38º 38º
36º 36º
102º 100º 98º 96º 94º 102º 100º 98º 96º 94º
FIGURE 15 (A) The Lambert conformal conic and (B) equidistant conic projections of Kansas, showing Tissot’s indicatrix at
a 1° spacing.
After reviewing the distortion values, we can see that Based on the preceding discussion, we selected the
neither projection preserves scale throughout the geo- Lambert conformal conic projection to map the tor-
graphic area of interest, which means that regardless of nado data (Figure 16). We should stress again that
which projection is selected, there will be some distor- given the small geographic area of interest, the
tion present in any distance measurement taken from intended map scale, and the tornado data set, the
the map, but that the measurement error at this scale choice of projection is rather inconsequential.
will be negligible. Because neither the Lambert confor- However, we chose conformality over equidistance
mal nor the equidistant projection has a substantial because on conformal projections, angular relations
amount of scale error across the mapped area, comput- are preserved at every point, and the errors from any
ing distances on either projection will result in the same distance measurement on this projection are negligi-
approximate result. In terms of angular distortion, the ble. Note in Figure 16 that the data are represented as
Lambert conformal conic projection preserves angular flow lines, where the origin of the flow line indicates
relations, which facilitates measuring the direction of the location where each tornado touched down. The
each tornado’s path. flow line extent is the distance over which the tornado
TABLE 4 Values for a, b, area distortion, and angular distortion for the Lambert conformal and equidistant conic projections
between latitudes 33° and 45°
Lambert Conformal Conic Equidistant Conic
Area Angular Area Angular

Latitude a b Distortion Distortion a b Distortion Distortion
33 1.004 1.004 1.007 0.0° 1.004 1.000 1.004 0.21°

34 1.002 1.002 1.004 0.0° 1.002 1.000 1.002 0.12°
35 1.000 1.000 1.000 0.0° 1.001 1.000 1.001 0.06°
36 1.000 1.000 1.000 0.0° 1.000 1.000 1.000 0.0°
37 1.000 1.000 1.000 0.0° 1.000 0.999 0.999 0.03°
38 0.999 0.999 0.998 0.0° 1.000 0.999 0.999 0.05°
39 0.999 0.999 0.998 0.0° 1.000 0.999 0.999 0.05°
40 1.000 1.000 0.999 0.0° 1.000 1.000 1.000 0.02°
41 1.000 1.000 1.000 0.0° 1.001 1.000 1.001 0.00°
42 1.001 1.001 1.002 0.0° 1.002 1.000 1.001 0.06°
43 1.002 1.002 1.005 0.0° 1.003 1.000 1.002 0.14°
44 1.004 1.004 1.008 0.0° 1.005 1.000 1.004 0.23°
45 1.006 1.006 1.012 0.0° 1.007 1.000 1.006 0.34°
191
17:32:08.
Destructive Tornadoes of KS IA
1950 - 2000
NE
CO
MO
FIGURE 16 The Lambert confor-

mal conic projection showing the
paths of F4 and F5 tornadoes across
Kansas from 1950 to 2000. Data from
F4 Tornado Path the National Climate Data Center
OK F5 Tornado Path found at http://www.ncdc.noaa.gov/
TX oa/climate/climatedata.html.
traveled, and the arrowhead indicates the direction of edge of each projection, especially on the Lambert con-
the tornado’s path. The point of the arrowhead indi- formal conic. Thus, when mapping smaller geographic
cates where the tornado dissipated. Examining the areas such as a state, variation in distortion values is less
pattern of tornado paths, we see that the general trend of a concern than for larger geographic areas.
is from the southwest to the northeast, and that the In addition to these key objectives learned from our
south central portion of the state appears to have a examples, there are two important considerations to keep
concentration of the most intense tornado activity for in mind when selecting map projections. First, a projec-
the time period. tion’s having been used many times (e.g., in prominent
atlases) does not mean that the projection is suitable for
2.5 Discussion your specific application. In many cases, a projection’s
usage was dictated by its availability and not its suitability.
The four projection examples each provided interesting One example is the overuse of the Mercator projection as
opportunities to see how map projections can be the basis for thematic maps (e.g., showing world popula-
selected. In summary, there are three key objectives from tion distribution), an obvious misuse of the Mercator
our examples that we would like to emphasize. First, in projection. Second, an often overlooked aspect of map
most cases, it is the objective of the mapmaker to select a projections is their influence on overall map design.
projection for which distortion is the lowest. Of course, Although this idea is more difficult to grasp, there are
the kinds of distortion that should be considered depend clear enhancements that projections can make with
on the purpose of the map (e.g., for many thematic activ- regard to map design. For example, Richard Edes
ities, equivalent projections are suitable). Second, Harrison in his Look at the World atlas used the ortho-
amounts of distortion can be kept small by aligning the graphic projection to present some powerful images of
geographic area (or data set) under consideration with World War II to illustrate the spatial relationships
the standard line(s) or by positioning the map’s center between the United States and the rest of the world. For
with the standard point. Third, as the extent of the geo- instance, Color Plate 1A shows the spatial proximity of
graphic area under consideration increases (e.g., from North America to Europe, and Color Plate 1B illustrates
country to hemisphere), distortion becomes a more the proximity of South America and Antarctica.
important consideration. For example, we noted in the Harrison’s use of the orthographic and other azimuthal
map of Kansas tornado paths that the variation in the projections throughout Look at the World was a signifi-
scale factors was very small on either projection. Had we, cant departure from atlases that relied on the Mercator
however, used the Lambert conformal or equidistant projection. By using azimuthal projections, especially the
conic projection to map the entire world, we would have orthographic, Harrison attempted to show how spatially
noted extreme distortion of the scale factors toward the connected the continents are.
17:32:08.
SUMMARY The third data set focused on migration patterns from

Europe and Asia to the United States. The objective for
In this chapter, we examined several commonly refer- this map was to show the routes by which immigration
enced map projection selection guidelines, noting that took place, as well as the spatial proximity between the
they are limited in helping novice cartographers through sources of migration and the United States—essentially
the often confusing process of selecting a projection. In encompassing a hemisphere. To help us select a projec-
contrast, we feel that Snyder’s guideline is well organized tion for this data set, we utilized Snyder’s table listing
and presents a logical hierarchy for selecting an appropri- various planar projections suitable for representing a
ate projection. Snyder’s guideline begins with a focus on hemisphere. We noted in our discussion that the planar
the geographic area to be mapped: world; hemisphere; projections offer a considerable range of design options
and continent, ocean, or smaller region. Once the geo- not found with other map projection classes. In the end,
graphic area to be mapped has been selected, projection we selected the vertical perspective azimuthal projec-
properties and characteristics are closely examined. For tion to highlight the migration paths and the spatial
example, under the world category, the specific projection proximity of Europe and Asia to the United States, as
properties (conformal) and specific characteristics well as the spherical nature of the Earth.
(e.g., constant scale along the Equator) are examined. The last data set involved a historical look at the
We relied on Snyder’s guideline to assist in the selec- most destructive tornado paths across Kansas. Here, we
tion of appropriate map projections for four different utilized Snyder’s selection guideline for a continent,
thematic data sets. The first data set focused on world lit- ocean, or smaller region. Using his guideline, we
eracy rates. In selecting a projection for this data set, we selected a conic projection. However, the small size of
focused on the world category, chose to preserve areal the geographic area prompted us to investigate the
relationships (to ensure that areas were preserved in nature of the distortion pattern across the mapped area.
their correct proportions and that no area visually dom- We were curious to see what the impact would be of
inated the map), and chose an interrupted projection (to using projections with different properties––equidistant
reduce distortion by placing cuts over the oceans where versus conformal, which are two properties useful for
data were not present). The result was an interrupted our tornado data. In the distortion analysis, it became
Mollweide projection. In addition, discussion was raised clear that neither projection offered any substantial
regarding the impression of selecting a projection with benefit because such a small geographic area was under
poles that were represented by points or lines, a topic consideration.
not directly addressed by Snyder’s guideline. The chapter concluded with a brief overview of five
In the second data set, we looked at the population key criteria that you should consider when selecting map
distribution of Russia. In this instance, we felt it desir- projections. First, in most cases, it is the objective of the
able to create a combined proportional symbol–dot map mapmaker to select a projection for which distortion is
that would represent the spatial variation of Russia’s the lowest. Second, amounts of distortion can be kept
population––some areas have high population densities small by aligning the geographic area (or data set) under
whereas others are vast areas of emptiness. Because the consideration with the standard line(s) or by positioning
geographic area of interest was Russia, we utilized the map’s center with the standard point. Third, as the
Snyder’s guidelines for a continent, ocean, or smaller amount of geographic area under consideration increases,
region. Noting that Russia has a considerable east–west distortion becomes a more important consideration.
extent, we selected a conic projection. Moreover, we Fourth, just because a projection has seen considerable
chose an equidistant conic projection to enable the user exposure (e.g., use in prominent atlases) does not mean
to properly compare areas of differing densities. This that the projection is suitable for your specific applica-
selection also allowed us to discuss how to select several tion. Fifth, an often overlooked aspect of the map projec-
parameters associated with conic projections: standard tion is its influence on overall map design––a topic that is
parallels, central meridians, and the central latitude. not well studied by cartographers.
FURTHER READING
American Cartographic Association. (1986) Which Map Is Presents a general overview of map projections, with special atten-
Best: Projections for World Maps. Bethesda, MD: American tion given to choosing world map projections.
Congress on Surveying and Mapping. American Cartographic Association. (1991) Matching the
Presents a general overview of map projections. Map Projection to the Need. Bethesda, MD: American
Congress on Surveying and Mapping.
American Cartographic Association. (1988) Choosing a
World Map: Attributes, Distortions, Classes, Aspects. Bethesda, Discusses several map projections and how their properties and
MD: American Congress on Surveying and Mapping. characteristics are appropriate for specific data sets.
193
17:32:08.
Bugayevskiy, L., and Snyder, J. P. (1995) Map Projections: A Maling, D. (1992) Coordinate Systems and Map Projections.
Reference Manual. London: Taylor & Francis. 2nd ed. Oxford, England: Pergamon.
Chapter 7 takes a mathematical look at selecting projections. Chapters 11 and 12 cover various methods used to select map
Canters, F. (2002) Small-Scale Map Projection Design. New projections.
York: Taylor & Francis. Nyerges, T., and Jankowski, P. (1989) “A knowledge base for
Chapter 6 focuses on automated methods for selecting projections
map projection selection.” The American Cartographer 16,
for small-scale maps. no. 1:29–38.
Canters, F., and Decleir, H. (1989) The World in Perspective: A Describes the development of a knowledge base for selecting map
Directory of World Map Projections. West Sussex, England: projections.
Wiley. Pearson, F. (1984) Map Projection Methods. Blacksburg, VA:
Part 1 examines selecting map projections from the standpoint of
Sigma Scientific.
distortion analysis. Chapter 10 presents a simple map projection selection guideline.
Deetz, C. H., and Adams, O. A. (1945) Element of Map Peters, A. (1983) The New Cartography. New York:
Projections with Applications to Map and Chart Construction. Friendship Press.
5th ed. Washington, DC: U.S. Department of Commerce Coast Discusses development of the Peters projection, which is designed
and Geodetic Survey. to better represent the Earth’s equatorial regions than other cylin-
Briefly describes selecting projections for a variety of purposes. drical projections such as the Mercator. The claims that Peters
made in this text caused quite a stir among cartographers.
De Genst, W., and Canters, F. (1996) “Development and
implementation of a procedure for automated map projection Robinson, A. (1974) “A new projection: Its development and
selection.” Cartography and Geographic Information Systems characteristics.” International Yearbook of Cartography 14:
23, no. 3:145–171. 145–155.
Discusses the development of an automated approach for select- Discusses the development of the Robinson projection.
ing map projections. Robinson, A. H., Morrison, J. L., Muehrcke, P. C., Kimerling,
Harrison, R. E. (1944) Look at the World. New York: Knopf. A. J., and Guptill, S. C. (1995) Elements of Cartography. 6th
ed. New York: Wiley.
Presents numerous hemispheric maps based on the orthographic
azimuthal projection. Chapter 5 briefly describes some commonly used map projections.
Hsu, M. (1981) “The role of projections in modern map Snyder, J. P. (1993) Flattening the Earth: Two Thousand Years
design.” Cartographica 18, no. 2:151–186. of Map Projections. Chicago: University of Chicago Press.
Reviews selecting projections based on map design issues. Useful discussion throughout the text on the selection and appli-
cation of map projections.
Jankowski, P., and Nyerges, T. (1989) “Design considerations
for MaPKBS-Map Projection Knowledge-Based System.” The Snyder, J. P. (1994) Map Projections: A Working Manual.
American Cartographer 16, no. 2:85–95. Washington, DC: U.S. Government Printing Office.
Presents MaPKBS, an expert system for the selection of a suitable Provides some limited discussion on Snyder’s rationale for his
map projection. selection guideline.
GLOSSARY
graduated symbol map: see proportional symbol map. lobe: A portion of an interrupted projection described by a
interrupted projection: a projection having several central central meridian usually displaying a specific landmass, such
meridians, where each central meridian (usually located over as South America.
a landmass) creates a lobe. projection selection guideline: a set of expert-developed
rules and suggestions that are intended to help novices select
an appropriate map projection.
REFERENCES
Deetz, C. H., and Adams, O. S. (1945) Element of Map Snyder, J. P. (1987) Map Projections: A Working Manual.
Projections with Applications to Map and Chart Washington, DC: U.S. Geological Survey.
Construction (5th ed.). Washington, DC: U.S. Department Snyder, J. P. (1993) Flattening the Earth: Two Thousand Years
of Commerce Coast and Geodetic Survey. of Map Projections. Chicago: University of Chicago Press.
Maling, D. H. (1992) Coordinate Systems and Map U.S. Geological Survey. (1970) The National Atlas of the
Projections (2nd ed.). Oxford, England: Pergamon. United States of America. Washington, DC: U.S. Geological
Pearson, F. (1984) Map Projection Methods. Blacksburg, VA: Survey.
Sigma Scientific.
17:32:08.
Principles of Color
OVERVIEW
The increasing use of color on maps necessitates that familiar with, such as hue, lightness, and saturation. This
mapmakers understand the proper use of color. To section also considers the CIE color model, which, in
assist you in developing this understanding, this theory, allows a mapmaker to reproduce colors specified
chapter covers some basic principles of color. How by others (e.g., a friend tells you that a particular color
color is processed by the human visual system is the progression is very effective, and you want to be sure that
topic of section 1. There we discuss the nature of you use the same progression).
visible light, the structure of the eye, theories of color
perception (focusing on opponent-process theory),
1 HOW COLOR IS PROCESSED
simultaneous contrast (i.e., how color perception
BY THE HUMAN VISUAL SYSTEM
is influenced by its surroundings), color vision
impairment, and the visual processing that takes place
behind the eye. Some of this material (e.g., details of 1.1 Visible Light and the Electromagnetic Spectrum
the structure of the eye) might seem far removed from
cartography, but developing rules for sound map We see maps as visible light, whether it is reflected
design requires knowledge of how our visual system from a paper map or emitted from a computer screen.
processes information. Visible light is a type of electromagnetic energy,
Section 2 describes hardware considerations relevant which is a waveform having both electrical and mag-
to the production of soft-copy color maps on graphics netic components (Figure 1).* The distance between
displays (those produced using a computer screen). In two wave crests is known as the wavelength of light.
the realm of graphics displays, this chapter focuses on Because visible wavelengths are small, they are typi-
how cathode ray tubes (CRTs) and liquid crystal dis- cally expressed in nanometers (nm), which are 1 bil-
plays (LCDs) are able to produce millions of colors by lionth of a meter. Visible wavelengths range from 380
using the additive colors red, green, and blue (RGB). to 760 nm. Figure 2 relates visible light to other forms
Specifying appropriate colors is a common problem of electromagnetic energy that humans deal with; the
faced by cartographers. For example, you might like complete continuum of wavelengths is called the
to create a smooth progression of colors extending from electromagnetic spectrum.
a light desaturated green to a dark saturated green. We have all seen or read about how a prism splits
Section 3 covers numerous color models that have been sunlight into the color spectrum (red, orange, yellow,
developed for specifying colors. Some of these models green, blue, indigo, and violet). This phenomenon occurs
are hardware-oriented (RGB and CMYK) and therefore because the visible portion of sunlight consists of a
of limited use to the mapmaker. Others are user-oriented
(HSV, Munsell, and Tektronix’s HVC) and thus permit *Light also consists of photons (packets of energy), which behave as
color specification in terms mapmakers are apt to be particles when light strikes a surface (Birren 1983, 20).
From Chapter 10 of Thematic Cartography and GeoVisualization, Third Edition. Terry A. Slocum, Robert B. McMaster,
Fritz C. Kessler, Hugh H. Howard. Copyright © 2009 by Pearson Education, Inc. All rights reserved.
195
17:34:54.
Principles of Color
broad range of wavelengths, instead of being Figure 2 are arranged from short to long wavelength
concentrated at a particular wavelength. Different (from violet to red), and thus match the colors we might
colors arise in a prism as a function of how much each see using a prism.
wavelength is bent, with shorter wavelengths (e.g., blue)
bent more than longer wavelengths (e.g., red). Note that 1.2 Structure of the Eye
the colors in the visible portion of the spectrum in
The basic features of the eye that concern cartographers
are shown in Figure 3. After passing through the cornea
(a protective outer covering) and the pupil (the dark
E area in the center of our eye), light reaches the lens,
which focuses the light on the retina. Changing the shape
of the lens, an automatic process known as
accommodation, focuses images. As we age, our lenses
become more rigid, and our ability to accommodate thus
S o u rc e weakens. Generally, around the age of 45, our ability to
accommodate becomes so weak that corrective lenses
Distan (glasses or contacts) are necessary. The fovea is the por-
ce
tion of the retina where our visual acuity is the greatest.
M The optic nerve carries information from the retina to
the brain and creates what is commonly termed our
blind spot.
An enlargement of the retina is shown in Figure 4.
FIGURE 1 Electromagnetic energy is a waveform having Note that it consists of three major layers of nerve cells
both electrical (E) and magnetic (M) fields. Wavelength is the (rods and cones, bipolar cells, and ganglion cells), along
distance between two crests. with two kinds of connecting cells (horizontal and
Wavelengths in Meters
-14 -12 -10 -8 -6 -4 -2 0 2 4 6 8
10 10 10 10 10 10 10 10 10 10 10 10
Visible Light
Gamma UV Infrared
X Rays Radar Radio Waves AC Circuits
Rays Rays Rays
Orange
Yellow
Violet Blue Green Red
400 500 600 700
Wavelengths in Nanometers
FIGURE 2 Relation of visible light to other forms of electromagnetic energy.
17:34:54.
Principles of Color
glion cell corresponds to a group of rods or cones, or

what is termed a receptive field. These receptive fields
are circular in form and overlap one another.
Retina
Cornea 1.3 Theories of Color Perception
Fovea Psychology textbooks (e.g., Goldstein 2007) generally

consider two major theories of color perception:
Lens trichromatic and opponent process. The trichromatic
theory, developed by Thomas Young (1801) and cham-
pioned by Hermann von Helmholtz (1852), presumes
Pupil that color perception is a function of the relative stimu-
lation of the three types of cones (blue, green, and red).
If only one type of cone is stimulated, that color is per-
Optic ceived (e.g., a red light would stimulate primarily red
Nerve cones, and thus red would be perceived). The percep-
tion of other colors is a function of the relative ratios of
stimulation (a yellow light would stimulate green and
FIGURE 3 Basic features of the eye relevant to cartography. red cones, and so yellow would be perceived).
The opponent-process theory, originally developed
by Ewald Hering (1878), states that color perception is
based on a lightness–darkness channel and two oppo-
amacrine cells) that enable cells within the major layers nent color channels: red–green and blue–yellow. Colors
to communicate with one another. Rods and cones are within each opponent color channel are presumed to
specialized nerve cells that contain light-sensitive chem- work in opposition to one another, meaning that we do
icals called visual pigments, which generate an electrical not perceive mixtures of red and green or blue and yel-
response to light. The concentration of cones is greatest low; rather, we see mixtures of pairs from each channel
at the fovea, and the highest concentration of rods is (red–blue, red–yellow, green–blue, and green–yellow).
about 20° on either side of the fovea. Overall there are For many years, proponents of the two theories of
about 120 million rods and 6 million cones. color perception hotly debated each theory’s merits,
Cones function in relatively bright light and enable presuming that only one theory could be correct. It is
color vision, whereas rods function in dim light and play now apparent, however, that both can help explain the
no role in color vision. The cones are of primary inter- way we see color. The trichromatic theory is correct in
est to cartographers because most maps are viewed in the sense that our color vision system is based on three
relatively bright light (an exception would be maps types of cones and that information from these cones
viewed in the dim light of an aircraft cockpit). combines to produce the perception of color. The man-
Physiological examination of cones taken from the eye ner, however, in which information from the cones com-
of a person with normal color vision reveals three dis- bines is based on opponent-process theory.
tinct kinds based on the wavelength to which they are There is both psychophysical and physiological evi-
most sensitive: short (blue), medium (green), and long dence in support of opponent-process theory. The psy-
(red) (Bowmaker and Dartnall 1980).* Alan chophysical evidence comes from the seminal work of
MacEachren (1995, 56) noted that the distribution and Leo Hurvich and Dorothea Jameson (1957), which
sensitivity of these three kinds of cones vary in the showed that a color of an opposing pair could be elim-
retina: Although blue cones cover the largest area, they inated by adding light for the other color in the pair;
are least sensitive, thus making blue inappropriate for for example, when yellow light is added to blue light,
small map features. the blue eventually disappears. The physiological evi-
The major function of the bipolar and ganglion cells dence is based on an analysis of how electrical signals
(Figure 4) is to merge the input arriving from the rods pass through cells in the nervous system. In this
and cones. Although there are about 126 million rods regard, an important concept is that nerve cells fire at
and cones, there are only about 1 million ganglion cells. a constant rate even when they are not stimulated.
Considerable convergence must take place between the Firing above this constant rate is termed excitation,
rods and cones and the ganglion cells; each single gan- and firing below it is termed inhibition. By studying
electrical activity in cells, physiologists have noted
* Hubel (1988, 163–164) indicated that technically, the terms violet, linkages between the blue, green, and red cones and
green, and yellowish-red would probably be more appropriate. the bipolar and ganglion cells; for example, a red light
197
17:34:54.
Principles of Color
R R
R R R R
Rods and Cones (R) R R
R
Horizontal Cells (H) H
Bipolar Cells (B) B B B

B B B
A
Amacrine Cells (A) A
Ganglion Cells (G) G G G G
Optic
Nerve
Fibers
Light Light
FIGURE 4 Major layers of cells found in the retina. (Adapted from J. E. Dowling and B. B. Boycott, 1966, “Organization of the
primate retina: electron microscopy,” Proceedings of the Royal Society of London, 166, Series B, pages 80–111, Figure 23; courtesy
of the Royal Society and John E. Dowling.)
might excite red cones, which in turn excite bipolar ganglion cells is unknown. One model that has been
and ganglion cells (Derrington et al. 1983; De Valois suggested is shown in Figure 5. In this model, the
and Jacobs 1984). blue–yellow channel (BY) is excited by green and red
Although experts in human vision are reasonably cones and inhibited by blue cones; the red–green chan-
sure that certain colors are in opposition to one another nel (RG) is excited by red cones and inhibited by green
and that excited and inhibited nerve cells play a role, cones; and red, green, and blue cones stimulate the
the precise linkage between the cones and bipolar and lightness–darkness channel (WBK).
17:34:54.
Principles of Color
black shifts toward a lighter tone, whereas the same

Ganglion Cone tone surrounded by white shifts toward a darker tone.
Cells Cells Note that in this case, the shifts are toward the opposite
side of the lightness–darkness channel in the opponent-
BY Blue process model.
When different hues are used, the apparent color of
an area will tend to shift toward the opponent color
RG Green of the surrounding color. For example, Color Plate V
of Hurvich (1981) illustrates an example in which a
gray tone is surrounded by either green or blue. When
the surround is green, the gray tone appears reddish;
WBK Red in contrast, when the surround is blue, the gray tone
appears yellowish.
Optic Simultaneous contrast is believed to come about
Nerve Inhibition because of the receptive fields mentioned earlier.
Excitation Receptive fields are not uniform; rather, distinctive
centers and surrounds characterize them, with visual
information in the surround having an impact on the
FIGURE 5 A model of how color information reaching the
information found in the center. For a detailed discus-
cones can be converted to opponent processes. The blue–yel-
sion of how simultaneous contrast operates, see
low channel (BY) is excited by green and red cones and inhib-
Hurvich (1981).
ited by blue cones; the red–green channel (RG) is excited by
red cones and inhibited by green cones; and red, green, and
blue cones stimulate the lightness–darkness channel (WBK, 1.5 Color Vision Impairment
for white to black). (After Eastman, 1986. First published in
The American Cartographer, 13(4), p. 326. Reprinted with per- Up to this point, we have assumed map readers with
mission from the American Congress on Surveying and normal color vision. Actually, a substantial number of
Mapping.) people have some form of color vision impairment. The
highest percentages are found in the United States and
Europe (approximately 4 percent, primarily males), and
1.4 Simultaneous Contrast the lowest incidence (about 2 percent overall) appears
in the Arctic and the equatorial rainforests of Brazil,
One problem sometimes encountered when reading Africa, and New Guinea (Birren 1983).
maps is that the perceived color of an area might be The bulk of the color vision–impaired can be split
affected by the color of the surrounding area, a prob- into two broad groups: anomalous trichromats and
lem known as simultaneous contrast, or induction dichromats. These groups are distinguished on the basis
(Brewer 1992). This concept is illustrated for lightness of the number of colors that must be combined to
in Figure 6. Here the gray tones in the central boxes are match any given color; anomalous trichromats require
physically identical, but the one on the left appears three colors, whereas dichromats use two. For both
lighter. This occurs because a gray tone surrounded by groups, the most common problem is distinguishing
FIGURE 6 An illustration of simultaneous contrast for a black-and-white image. The central gray strips are physically identical, but
the one surrounded by black appears lighter.
199
17:34:54.
Principles of Color
between red and green; for anomalous trichromats, opponent cells similar to those found within the retina
there is some difficulty, whereas for dichromats, the are also found in the LGN (De Valois and Jacobs
colors cannot be distinguished. 1984). Interpretation of the visual information begins
Two hypotheses have been proposed for color vision in the primary visual cortex, the first place where all of
impairment: (1) a change in the colors to which cone the information from both eyes is handled. As with the
cells are sensitive and (2) changes in one of the oppo- LGN, our knowledge of processing in this area is
nent-process channels (normally the red–green one). largely a function of physiological experiments with
Assuming that changes in the cone cells are the cause, animals. Probably the most significant of these is the
the two major groups have been divided into sub- work of David Hubel and Torsten Wiesel, who
groups: protanomalous and deuteranomalous for anom- received the 1981 Nobel Prize for their efforts. They
alous trichromats and protanopes and deuteranopes for found three kinds of specialized cells in the primary
dichromats. The two subgroups differ on the basis of the visual cortex: simple cells, which respond best to lines
types of cones affected; for example, protanopes and of particular orientation; complex cells, which respond
deuteranopes are presumed to be missing red and blue to bars of particular orientation that move in a partic-
cones, respectively. ular direction; and end-stopped cells, which respond to
moving lines of a specific length or to moving corners
1.6 Beyond the Eye or angles. Not only did Hubel and Wiesel discover
these different kinds of cells, but they also mapped out
It is important to realize that the eye is part of the where they occur within the primary visual cortex
larger visual processing system shown in Figure 7. (Goldstein 2007).
Note first that information leaving the eyes via the Although such findings are certainly significant,
optic nerves crosses over at the optic chiasm; up to that researchers have not yet been able to explain how the
point, information from each eye is separate, but path- brain handles a complex real-world situation, such as a
ways beyond this point contain information from both map. Functional magnetic resonance imaging (fMRI)
eyes. After passing through the optic chiasm, each may ultimately provide some insight into this issue.
pathway enters the lateral geniculate nucleus (LGN). fMRI utilizes an MRI scanner, which generates strong
Physiological experiments with animals reveal that magnetic fields that pass through the human body; you
Optic Nerves
Optic Chiasm
Lateral Geniculate
Nucleus (LGN)
FIGURE 7 An overview of the visual

processing system viewed from under-
neath the brain. (From Sensation &
Perception 3rd edition by Goldstein.
1989. Reprinted with permission of
Wadsworth, a division of Thomson
Primary Visual Cortex
Learning: www.thomsonrights.com.
Fax 800-730-2215.)
17:34:54.
Principles of Color
might be familiar with MRI scanners because they are

commonly used to diagnose sports-related injuries. In
the case of maps, we are interested in what goes on in
the human brain, where changes in the metabolic
demands of nerve cells lead to changes in blood oxy-
genation levels, which can be measured by an MRI
scanner. Since changing metabolic demands of nerve
cells are an indication of which area of the brain is
activated, it is, at least theoretically, possible to associ-
ate a particular map use task with a particular area
of the brain.
Amy Lobben, Judy Olson, and Jie Huang (2005)
are the first cartographers to begin experimenting
with fMRI. Although the full results of their experi-
ments are not yet published, their preliminary analy-
sis is promising. For instance, in a map rotation task,
they observed more brain activation in the lateral
FIGURE 8 The order of refresh on a CRT.
occipital gyri, a finding consistent with studies of brain
activation associated with the rotation of non-map
objects. Olson (personal communication, 2006) argues
2.2 CRTs
that fMRI research is exciting because “there is . . .
the prospect of finding which areas of the brain are
Images on CRT screens are created by firing electrons
used to perform various map related tasks . . . .
from an electron gun at phosphors, which emit light
Ultimately, there is the possibility of its affecting how
when they are struck. Monochrome CRTs contain a sin-
we teach things . . . so that appropriate brain activity
gle electron gun, whereas color CRTs contain three
takes place.”
guns, normally designated as R (red), G (green), and B
(blue). The names for the guns have nothing to do with
2 HARDWARE CONSIDERATIONS the type of electrons they fire, but are a function of
IN PRODUCING COLOR MAPS FOR which type of phosphor the electrons strike on the
GRAPHICS DISPLAYS screen. In Figure 9,* we see two common arrangements
of electron guns and phosphors (delta and in-line). Note
This section considers some hardware aspects of pro- that a shadow mask or aperture grill (termed a “Metal
ducing color maps for graphic displays. The term mask” in the figure) is positioned so that each electron
graphic display is commonly used to describe the com- gun can hit only one type of phosphor.
puter screen (and associated color board) on which a Different colors on a CRT screen result from the
map is displayed in soft-copy form; examples include principle of additive color: The colored phosphors are
cathode ray tubes (CRTs), liquid crystal displays visually added (or combined) to produce other colors.
(LCDs), plasma displays, and electroluminescent dis- This principle is normally demonstrated with overlap-
plays. We focus on CRTs and LCDs because they are ping colored circles (as in Color Plate 1), but phosphors
the most commonly used. on a CRT screen do not actually overlap. Rather, we see
a mixture of color because the human eye cannot
2.1 Vector versus Raster Graphics resolve the very fine detail of individual phosphors; the
concept is analogous to pointillism techniques used in
Images on graphic displays can be generated using two nineteenth-century paintings.
basic hardware approaches: vector and raster. In the Together, three phosphors compose a pixel. One
vector approach, images are created much like drawing measure of the resolution of a monitor is the number
a map by traditional pen-and-ink methods: The hard- of addressable pixels, normally specified as the number
ware moves to one location and draws to the next loca- of pixels displayable horizontally and vertically.†
tion. In contrast, in the raster approach, the image is Common resolutions range from 800 * 600 up to
composed of pixels (or picture elements), which are cre- *Foley/Feiner/Hughes/Van Dam, COMPUTER GRAPHICS: PRIN-
ated by scanning from left to right and from top to bot- CIPLES AND PRACTICE, Fig. 4.14 p. 159 and Figure 4.15 p. 160,
tom (Figure 8). Prior to about 1980, the vector approach © 1994, 1991, 1982 Addison Wesley Longman Inc. Reproduced by
was more common, but today virtually all graphic dis- permission of Pearson Education, Inc. All rights reserved.
plays use a raster approach. † See Peddie (1994, 7–10) for a variety of definitions for resolutions.
201
17:34:54.
Principles of Color
Delta In-line
FIGURE 9 Cross-sectional view of a portion of a color CRT, illustrating two methods of phosphor arrangement: delta and
in-line.
1280 * 1024 Problems with lower resolution systems

include jaggies or aliasing (a staircase appearance of
diagonal lines), an inability to smoothly vary the size of
small symbols and a difficulty in creating crisp text.* Fluorescent
Lamp
2.3 LCDs TFTs for red,
Liquid Crystal
Material
green, and
LCDs consist of a complex sandwich of a light source, blue TFT Layer (Thin
Film Transistors)
glass plates, polarizing film, liquid crystals, a source of
electrical power (e.g., transistors), and color filters
(Figure 10). The basic principle is that the liquid crystals
found at each pixel location initially are all in the same Polarized
Green
Red
Blue
orientation. When polarized light passes through the Glass

crystals, the crystals direct the light so that it passes
through another polarized filter; we see the result as a Pixel composed of Intensity of light is altered as
bright light. If, however, an electrical charge is applied red, green, and TFTs apply electrical current
blue subpixels to liquid crystals
to a liquid crystal, the crystal is bent and some of the
light does not pass through—the intensity of the light
that we see is a function of how strong a charge is FIGURE 10 Diagram of a liquid crystal display (LCD) sys-
applied. Color is created by using red-, green-, and blue- tem. (After Morgenstern and Seff, 2002, and Marshall Brain’s
colored filters that are associated with separate liquid HowStuffWorks (http://www.howstuffworks.com/lcd.htm).)
crystals at each pixel location.†
LCDs have been used in laptop computers and
computer projection devices since the mid-1990s. Today dominance—the desktop computer.Advantages of LCDs
they have overtaken the CRT in its traditional area of over CRTs include LCDs’ light weight, small depth (or
small footprint), absence of flicker, low power consump-
* Aliasing can be handled using antialiasing routines (Foley et al. 1996, tion, and the absence of potentially harmful x-rays and
132–142), but such routines are not common in mapping and design low-frequency magnetic fields. Disadvantages of LCDs
software. include a smaller range of available color, optimal
†For more on how LCDs work, see Marshall Brain’s HowStuffWorks performance at only one resolution (say 1280 * 1024),
(http://www.howstuffworks.com/lcd.htm). difficulty of viewing from a wide angle (although this
17:34:54.
Principles of Color
problem is disappearing), degradation in performance 3.1 The RGB Model

due to changes in ambient temperature, and the greater
expense of large-format displays. In the RGB model, colors are specified based on the
One limitation of both LCDs and CRTs is that they intensity of red, green, and blue at each pixel location.
cannot handle large map displays (e.g., an entire USGS The range of intensities can be represented as a cube, as
topographic sheet). The largest CRT and LCD screens shown in Figure 11. In Figure 11A, gray tones (or com-
generally do not exceed 21 inches (along a diagonal). pletely desaturated colors) are found along the diago-
One alternative to displaying an entire large-format nal extending from “White” to “Black.” In general,
map is, of course, to use pan, scroll, and zoom functions, lighter colors are found around the White point of the
but there are times when a map user would rather cube, and darker colors are found around the Black
examine the entire map at once or compare a variety of point of the cube. As you move away from the
maps simultaneously; for example, imagine comparing White–Black line, you move toward more saturated
50 maps showing changes in wheat production for colors; for example, at the “Red” point you would be
Kansas on a yearly basis over a 50-year period. One at the maximum saturation of red. Finally, note that
solution to this problem is to use wall-sized displays. hues are arranged in a hexagonal fashion around the
White–Black line. The latter can be seen most easily if
3 MODELS FOR SPECIFYING COLOR you look directly down the diagonal from White to
Black, as shown in Figure 11B.
The RGB model has the advantage of relating
This section considers six color models that have been
nicely to the method of color production on graphic
used to specify colors appearing on maps: RGB, CMYK,
displays, but it has two major disadvantages. One is
HSV, Munsell, HVC, and CIE. The RGB and CMYK
that common notions of hue, saturation, and lightness
models are hardware-oriented because they are based on
are not inherent in the model; although we used these
hardware specifications of red, green, and blue light on
terms to describe the model, they are not used to spec-
graphic displays, or cyan, magenta, yellow, and black ink
ify colors. Another disadvantage is that equal steps in
on printed maps. In contrast, the HSV, Munsell, and HVC
RGB color space do not correspond to equal visual
models are user-oriented because they are based on how
steps; for example, the color 125, 0, 0 will not appear to
we perceive colors (using attributes such as hue, lightness,
fall midway between the colors 0, 0, 0 and 250, 0, 0.
and saturation). The CIE system is neither hardware- nor
Typically, you will find that an incremental change in
user-oriented; however, it is “optimal” in the sense that if
low RGB values represents a smaller visual difference
you provide someone with the CIE coordinates of a color
than the same incremental change in high RGB values.
you created, that person, in theory, should be able to
In spite of these disadvantages, RGB values frequently
create exactly the same color.
are used as an option for specifying color in software
A B
Blue Cyan
(0,0,255) (0,255,255)
Blue Cyan
White
Magenta (255,255,255)
(255,0,255)
White
Magenta Green
Black Green
(0,0,0) (0,255,0)
Red Yellow
Red Yellow
(255,0,0) (255,255,0)
FIGURE 11 (A) Schematic form of an RGB color cube for specifying color. (It is assumed that red, green, and blue colors each
have a maximum intensity of 255.) (B) The cube viewed looking directly down the diagonal from White to Black.
203
17:34:54.
Principles of Color
packages, presumably because of their long history

and the consequent familiarity that many users have V
with them. Blue Cyan
240º 180º
3.2 The CMYK Model

Magenta 1.0 Green
White
Because printed maps are based on reflected (as 300º 120º
opposed to emitted) light, they create color using a

subtractive (as opposed to an additive) process. The
Red Yellow
three basic subtractive primary colors are cyan, 0º 60º
magenta, and yellow (Color Plate 1); black ink, how-
ever, is utilized when a true black is desired. Together,
cyan, magenta, yellow, and black compose the CMYK
color model. If we think of cyan, magenta, and yellow
in a fashion similar to how we think of red, green, and
blue, then it is also possible to conceive of the CMY
portion of CMYK as a cube: A certain percentage of
cyan, magenta, and yellow would correspond to a par- H
0.0 S
ticular point in the cube. Black would need to be Black
added to create true shades of gray within the cube.

Given the analogy to the RGB cube, it makes sense
that CMYK will share the same disadvantages: there is FIGURE 12 The HSV (hue, saturation, and value) color sys-
a lack of relation to common color terminology, and tem represented as a hexcone.
equally spaced colors in the model will not correspond
to equal visual steps. 3.4 The Munsell Model
3.3 The HSV Model The Munsell color model is a user-oriented system that
was developed prior to the advent of computers.
In contrast to RGB and CMYK, the HSV model is Munsell colors are specified using the terms hue, value
more intuitive from a map design standpoint because it (for lightness), and chroma (for saturation). The general
allows users to work directly with hue, saturation, and structure of the model (Figures 13 and 14 and Color
value (lightness). Color space in HSV is represented as Plate 2) is similar to that of HSV (that is, hues are
a hexcone, as shown in Figure 12. The logic of the hex- arranged in a circular fashion around the center,
cone is apparent if you compare it with the color cube chroma increases as one moves outward from the
for RGB shown in Figure 11B; note that the hexagonal center, and value increases from bottom to top). Note,
structure of the hues in the cube is retained in the however, that in contrast to HSV, the Munsell model is
hexagonal structure at the base of the hexcone. Value asymmetrical; for example, if you were to hold the
changes occur as you move from the apex of the cone to model in your hands, you would note that the lightest
its base, whereas saturation changes occur as you move green would be higher on the model than the lightest
from the center to the edge of the cone. red. The asymmetry occurs because the model is per-
The intuitive notions of hue, saturation, and value in ceptually based (i.e., the lightest possible green does
HSV have led to its common use in software. Although appear brighter than the lightest possible red).
HSV is commonly used, it also has disadvantages. One Ten major Munsell hues are recognized, and these are
is that different hues having the same value (V) in HSV split into five principal (represented by a single letter,
will not all have the same perceived value. As an exam- such as Y for yellow) and five intermediate (represented
ple, consider the base of the cone, where the highest- by two letters, such as YR) hues (Figures 13 and 14).
value green and red are found. If you create such colors Each major hue is also split into 10 subhues (consider
on your monitor, green will appear lighter than red. In a the 10 subhues shown for R in Figure 14). Values range
similar fashion, different hues having identical satura- from 0 to 10 (darkest to lightest), and chromas range
tions (S) will not have the same perceived saturations from 0 to 16 (least to most saturated). Due to the asym-
(Brewer 1994b). HSV also shares a disadvantage noted metry of the model, not all values and chromas occur for
for RGB: Selecting a color midway between two colors each hue. Munsell colors are represented symbolically
will not result in a color that is perceived to be midway as H V/C; thus, 5R 5/14 is a distinct red of moderate
between those colors. value and high saturation (a crimson).
17:34:54.
Principles of Color
Munsell Book of Color

Yellow
White
10/ (y) (50y) (250)
9/ 9/2 9/4 9/6 9/8 9/10 9/12 9/14

9/ N9/ Y9/2 Y9/4 Y9/6 Y9/8 Y9/10 Y9/12 Y9/14
8/ 8/2 8/4 8/6 8/8 8/10 8/12

8/ N9/ Y8/2 Y8/4 Y8/6 Y8/8 Y8/10 Y8/12
7/ 7/2 7/4 7/6 7/8 7/10

7/ N9/ Y7/2 Y7/4 Y7/6 Y7/8 Y7/10
6/2 6/4 6/6 6/8

6/ 6/ N9/ Y6/2 Y6/4 Y6/6 Y6/8
5
PB
Value
5/ 5/2 5/4 5/6
10 5/ N9/ Y5/2 Y5/4 Y5/6
B
5
B 10
BG 5
BG 10
G 4/2 4/4
4/ 4/ N9/ Y4/2 Y4/4
3/2
3/ 3/ N3/ Y3/2
2/2
2/ 2/ N2/ Y2/2
5R
10R
10PB 1/ 5YR 1/ N1/
5PB 10YR
10B
Black 5Y
5B /0 /2 /4 /6 /8 /10 /12 /14
10Y
10BG 5GY
5BG
10G 5G 10GY Chroma
FIGURE 13 Three-dimensional representation of the Munsell color solid. A vertical slice through yellow is shown in detail.
(Courtesy of Leo M. Hurvich.)
5R 6
10RP 1 2 3 4 7 8 9 10R
5YR
5RP
/10
10YR
/8
10P
/6
/4
5Y
5P
/2
10Y
10PB
5PB 5GY
10B 10GY
5B 5G
FIGURE 14 A horizontal slice
10BG 10G
5BG through the Munsell color solid
shown in Figure 13.
205
17:34:54.
Principles of Color
An important characteristic of the Munsell model is type of model that developers should consider includ-
that equal steps in the model represent equal percep- ing in cartographic software.
tual steps. Thus, a color that is numerically midway
between two other colors should appear to be percep- 3.6 The CIE Model
tually midway between those colors. For example, color
5R 5/5 should appear midway between 5R 2/2 and 5R CIE is an abbreviation for the French Commission
8/8. Remember that this is not a characteristic of the International de l’Eclairage (International Commission
other color models described thus far. on Illumination). In theory, careful color specification
in the CIE model means that anyone in the world
3.5 The HVC Model should be able to recognize and reproduce a desired
color. CIE colors can be specified in several ways
The HVC (hue, value, and chroma) color model devel- (Yxy, L*u*v*, L*a*b*), but in all cases, a combination
oped by Tektronix was an attempt to duplicate the of three numbers is used. We consider the Yxy model
Munsell system on computer graphic displays. The (commonly referred to as the 1931 CIE model) first
similarity of HVC and Munsell is seen in the use of because it forms the basis of other CIE methods. In the
the same three terms (hue, value, and chroma), and Yxy model, the x and y coordinates define a two-
the irregular shape of the color space (compare dimensional space within which hue and saturation
Figures 13 and 15). The two models differ, however, in vary (Figure 16). Note that hues are arranged around a
color notation. Munsell uses the H V/C notation, central white point (or equal-energy point) and that sat-
whereas in HVC, hue is specified in degrees counter- uration increases as one moves outward from the white
clockwise from 0° (red), value varies vertically from 0 point. The Y portion of the model provides the third
to 100, and chroma varies from 0 to 100 from the cen- dimension—the lightness or darkness component
tral axis to the edge of the model (Figure 15). Although (Color Plate 3).
the literature suggests that HVC is effective (e.g., The structure of the Yxy model is similar to both
Taylor et al. 1991), software associated with the model HVC and Munsell (all have hues arranged in a circular
is no longer distributed. Apparently, many designers fashion, desaturated colors in the middle, and a vertical
outside the field of cartography either do not require lightness axis), but note that in CIE, hues and satura-
or do not see the advantage of specifying colors that tions are not related in a simplistic fashion to the x and
are equally spaced in the visual sense, and so Tektronix y axes. The reason for this can be found in the manner
chose not to support the software. We have included in which CIE was established. CIE was developed using
the HVC model here because it is illustrative of the the notion that most colors can be defined by a mixture
of three colors (roughly speaking, we can call these red,
green, and blue). The appropriate combination of three
colors needed to match selected colors was determined
using human observers (the average response of the
observers was termed the standard observer). To under-
stand the matching process, imagine that you are asked
to view a screen on which a single circle is projected
using a standard light source. In the top portion of the
circle, a test color appears, and in the bottom portion,
Value you manipulate the three colors to produce a color
0 - 100
identical to the test one. If you repeated this task for
many test colors, you would discover that various
amounts of the three colors would be required to make
120º 60º
appropriate matches.
Results of actual CIE matching experiments are
180º 0º
shown in Figure 17. The three curves correspond to the
240º 300º
three colors combined in the experiments. The x axis
represents the wavelength of the test color, whereas
Chroma
Hue 0 - 100 the y axis represents the relative magnitudes of the
0 - 360º
three colors needed to match the test color. For exam-
ple, a test color at 530 nm would require 0.005 of blue,
FIGURE 15 The HVC color system developed by Tektronix. 0.203 of green, and - 0.071 of red (Wyszecki and Stiles
(Taylor et al., 1991. First published in Information Display, 7 1982, 750); these are known as tristimulus values. The
(4/5), p. 21. Courtesy of Tektronix, Inc.) negative value for red is necessary because in some
17:34:54.
Principles of Color
1.000
0.900
515 520
530
0.800 535
510 540
545
0.700 550
505 555
560
0.600 YELLOWISH
565
GREEN GREEN
570 GREENISH
GR OW
500 EEN YELLOW
L
575
YEL
y 0.500 W 580 YELLOWISH

O
LL ORANGE
YE 585
FIGURE 16 The Yxy 1931 CIE sys-
590
0.400
495
BLUISH ORANGE 595 tem. Hues are arranged in a contin-
GREEN
WHITE ORANGE REDDISH
600 uum around a central white point.
PINK ORANGE
BLUE
POINT 610 Saturation is at a maximum on the
GREEN PINK 620 edge of the horseshoe and at a min-
0.300
490 RED 640
H imum at the white point. Numerical
NIS PURPLISH
EE PURPLISH
GR BL U E PINK values on the edge of the horseshoe
RED
0.200 485 RED represent wavelengths in nanome-
LE
RE DDISH
UE
BLUE
PURPLE
PURPLE
ters. Because differing lightnesses
RP
BL
LE
PU
480 cannot be shown by this diagram, a

PURP
SH
0.100
three-dimensional diagram is
ISH
PLI
475
P UR
B LU
470
required, as in Color Plate 3. (After
460
0.000
Kelly, K. L. (1943) “Color designa-
0.000 0.100 0.200 0.300 0.400 0.500 0.600 0.700 0.800 tions for lights,” Journal of the
x Optical Society of America, 33, no.
11:627–632.)
cases, you would find that you could not match the test (We get the same proportions when X, Y, and Z are all
color with any combination of the three colors; to 10 units as we do when they are all 20 units.) This prob-
achieve a match, you would have to mix one of the lem was handled in CIE by arbitrarily assigning
three with the test color, and this is recorded as a the lightness information to Y and plotting this as the
minus value in the graph. To avoid having to work with third dimension, as shown in Color Plate 3 (Hurvich
negative values, the developers of the CIE model 1981, 284).
transformed the results shown in Figure 17 to purely It should also be noted that the CIE Yxy coordinate
positive values, which are commonly referred to as X, values can be adjusted to account for the lighting con-
Y, and Z. To get coordinates for the Yxy system, the X, ditions under which the colors are viewed. For example,
Y, and Z values were converted to proportional val- you might want to consider the potential effects of
ues: natural sunlight versus fluorescent room light. This is a
capability that is not generally included in other color
models.
x = X/(X + Y + Z)
One problem with the 1931 CIE diagram is that
y = Y/(X + Y + Z) colors are not equally spaced in a visual sense, just like
z = Z/(X + Y + Z) the RGB, CMYK, and HSV models. Fortunately, two
perceptually uniform color models (or simply uniform
Because these proportions add to 1, it is not necessary color models) have been developed based on CIE:
to plot z in the Yxy system (z would be 1 - (x + y)). L*u*v* (CIELUV) and L*a*b* (CIELAB). CIELUV
One problem with plotting proportional values is the is appropriate for graphic displays, and CIELAB is
elimination of information about lightness or darkness. appropriate for printed material.
207
17:34:54.
Principles of Color
human visual system, we found that there are three

0.4 types of cones in the retina of the eye—red, green, and
Red blue—each sensitive to certain wavelengths of light; for
Blue example, red cones are sensitive to long-wavelength
0.3 red light. Our perception of color is not simply a func-
tion of which cones are stimulated, however; we must
Tri-stimulus Values
Green also consider the notion that cones excite or inhibit

0.2 other nerve cells (e.g., the bipolar and ganglion cells).
The net result is that we perceive color according to
the opponent-process theory, which states that three
0.1
channels are involved in color perception: a light-
ness–darkness channel and two opponent-color chan-
0.0
nels (red–green and blue–yellow). Perceived colors are
a function of a mix of colors in the opponent channels;
for example, we see orange as a mixture of red and
-0.1 yellow.
350 400 450 500 550 600 650 700 750 It is important to keep in mind that not everyone
Wavelength (nm) sees color in the same way. About 4 percent of the pop-
ulation in the United States and Europe (mostly males)
has some form of color vision impairment; a common
FIGURE 17 Curves representing the relative proportion of
impairment is the inability to discriminate between red
CIE primaries needed to match test colors at various wave-
and green colors. Another characteristic of color per-
lengths.
ception is simultaneous contrast—the idea that our per-
ception of a color might be a function of the colors that
3.7 Discussion surround it. Fortunately, special color schemes have
been developed to handle the problems of color vision
Given the variety of color models presented here, it is impairment and simultaneous contrast.
natural to ask which models mapmakers need to be Maps produced on graphics displays (soft-copy maps)
familiar with. At present, the RGB, CMYK, and HSV and maps printed on paper (hard-copy maps) utilize fun-
models are most frequently used in mapping software. damentally different processes. Soft-copy maps use an
This is problematic because the RGB and CMYK additive process involving red, green and blue, whereas
models do not relate easily to our common notions of hard-copy maps utilize a subtractive process involving
hue, saturation, and lightness, and all three models do cyan, magenta, yellow, and black. Traditionally, CRTs
not produce equally spaced colors (in the visual were the dominant technology for graphics displays, but
sense). The Munsell model is not generally found in LCDs have recently become more common.The empha-
mapping software, but it is useful because it relates sis of this chapter was on soft-copy maps.
well to our notion of hue, saturation, and lightness, and We considered a variety of color models that can be
it stresses the notion of creating equally spaced colors. used to specify colors displayed on maps. Some of
The HVC model is useful for illustrating how the these are hardware-oriented (RGB and CMYK), and
Munsell model can be included in software. The CIE thus of limited use to the mapmaker. Others are user-
model, although thus far not commonly used in map- oriented (HSV, Munsell, and Tektronix’s HVC), and
ping software, is important to understand because you thus permit color specification in terms mapmakers
may see references to CIE colors when reading the and map users are apt to be familiar with, such as hue,
results of cartographic research; for example, articles lightness, and saturation. We also examined the CIE
by Olson and Brewer (1997) and Brewer et al. (1997) model, which, in theory, allows a mapmaker to repro-
report colors in CIE coordinates. duce colors specified by other people. At present, the
RGB, CMYK, and HSV models are most frequently
used in mapping software. This is problematic because
SUMMARY the RGB and CMYK models do not relate easily to
our common notions of hue, saturation, and lightness,
In this chapter, we examined several basic principles of and all three models do not produce equally spaced
color. In considering how color is processed by the colors (in the visual sense).
17:34:54.
Principles of Color
FURTHER READING
Birren, F. (1983) Colour. London: Marshall Editions. Hunt, R. W. G. (1987a) Measuring Colour. Chichester,
A general text on color and its varied uses. England: Ellis Horwood.
Brown, A., and Feringa, W. (2003) Colour Basics for GIS A thorough treatment of color models.
Users. Harlow, England: Prentice Hall. Hunt, R. W. G. (1987b) The Reproduction of Colour in
Provides a succinct overview of color issues related to cartography Photography, Printing & Television. Tolworth, England:
and GIS. Fountain Press.
Fairchild, M. D. (1998) Color Appearance Models. Reading, A good reference text on color vision.
MA: Addison-Wesley. Hurvich, L. M. (1981) Color Vision. Sunderland, MA: Sinauer
Reviews principles of color vision and describes numerous color Associates.
appearance models. Covers the details of how we perceive color, with an emphasis on
Foley, J. D., van Dam, A., Feiner, S. K., and Hughes, J. F. (1996) opponent-process theory.
Computer Graphics: Principles and Practice. 2nd ed. Reading, Itten, J. (1973) The Art of Color: The Subjective Experience
MA: Addison-Wesley. and Objective Rationale of Color. New York: van Nostrand
A standard reference on computer graphics. Chapter 4 covers Reinhold.
hardware issues, and Chapter 13 covers issues related to color Discusses principles of color from an artist’s perspective.
models; also see Chapter 1 and pp. 602–656 of Rogers (1998).
Peddie, J. (1994) High-Resolution Graphics Display Systems.
Garo, L. A. B. (1998) “Color theory.” http://personal.uncc.edu/ New York: Windcrest/McGraw-Hill.
lagaro/cwg/color/index.html.
An extensive review of graphic display systems.
An online resource covering principles of color in cartography.
Travis, D. (1991) Effective Color Displays: Theory and
Goldstein, E. B. (2007) Sensation and Perception. 7th ed. Practice. London: Academic Press.
Belmont, CA: Thomson Wadsworth.
Chapters 1 to 3 deal with computer hardware, the human visual
A standard text on perception that supplements section 1 of this system, and color models, respectively.
text.
Wyszecki, G., and Stiles, W. S. (1982) Color Science: Concepts
Hubel, D. H. (1988) Eye, Brain, and Vision. New York: and Methods, Quantitative Data and Formulae. 2nd ed. New
Scientific American Library. York: Wiley.
A treatise on how the eye–brain system functions. A detailed reference on technical aspects of color.
GLOSSARY
accommodation: the process in which the eye automatically CMYK: a system for specifying colors in which cyan (C),
changes the shape of the lens to focus on an image. magenta (M), yellow (Y), and black (K) are used; common in
additive colors: colors that are visually added (or combined) offset printing.
to produce other colors; for example, red, green, and blue are color models: refers to various approaches for specifying
additive colors used in CRT displays. color (e.g., RGB vs. CMYK).
aliasing: a staircase appearance of straight lines caused by color vision impairment: the condition in which some indi-
displaying lines on a coarse raster graphics display. viduals do not see the range of colors that most people do;
anomalous trichromats: a less serious form of color vision males are most commonly affected, with red and green colors
impairment in which people use three colors to match any being most frequently confused.
given color; contrast with dichromats. cones: a type of nerve cell within the retina that functions in
bipolar and ganglion cells: cells that merge the input relatively bright light and is responsible for color vision.
arriving from rods and cones in the eyes. cornea: the protective outer covering of the eye.
blind spot: the small, visually insensitive region of the retina dichromats: a more serious form of color vision impairment
associated with where the optic nerve carries information to in which people use two colors to match any given color;
the brain. contrast with anomalous trichromats.
cathode ray tube (CRT): a graphics display in which images electromagnetic energy: a waveform having both electrical
are created by firing electrons from an electron gun at phos- and magnetic properties; refers to the way that light travels
phors, which emit light when they are struck. through space.
CIE: an international standard for color that allows electromagnetic spectrum: the complete range of wave-
cartographers to precisely specify a color so that others can lengths possible for electromagnetic energy.
duplicate that color.
209
17:34:54.
Principles of Color
electron gun: the device used within a CRT to fire electrons receptive field: a single ganglion cell corresponding to a
at phosphors, which emit light when they are struck. group of rods or cones; receptive fields are circular and over-
fovea: the portion of the retina where visual acuity is the lap one another.
greatest. refresh: a term used to describe how images on a CRT must
graphic display used to describe the computer screen (and be constantly redisplayed (refreshed) because phosphors
associated color board) on which a map is displayed in soft- making up the screen have a low persistence.
copy form. resolution: (1) for CRT displays, the number of addressable
HSV: a system for specifying color in which hue (H), satura- pixels; for printers, the number of dots per inch; (2) a visual
tion (S), and value (V) are used; although the system utilizes variable for depicting data quality; for example, coarser raster
common color terminology, colors are not equally spaced in cells indicate greater uncertainty in the data.
the visual sense. retina: the portion of the eye on which images appear after
HVC: a system developed by Tektronix for specifying color; they pass through the lens.
the terms hue (H), value (V), and chroma (C) are used, and RGB: a system for specifying color in which red (R), green
colors are equally spaced from one another in the visual sense. (G), and blue (B) components are used, as for red, green, and
induction: see simultaneous contrast. blue color guns on a CRT.
lens: the focusing mechanism within the eye. rods: a type of nerve cell within the retina that functions in
dim light and plays no role in color vision.
liquid crystal displays (LCDs): a form of graphic display
device in which liquid crystals are sandwiched between two simultaneous contrast: when the perceived color of an area
glass plates; commonly used in laptops and overhead projec- is affected by the color of the surrounding area.
tion devices. subtractive primary colors: the colors (cyan, magenta, and
Munsell: a system for specifying color in which colors are yellow) associated with printing inks that can be mixed to
identified using hue, value, and chroma; colors are equally create a wide range of colors.
spaced from one another in the visual sense. trichromatic theory: the theory that color perception is a
opponent-process theory: the theory that color perception is function of the relative stimulation of blue, green, and red
based on a lightness–darkness channel and two opponent- cones.
color channels: red–green and blue–yellow. uniform color model: refers to a variation of the CIE color
optic nerve: the nerve that carries information from the model in which colors are equally spaced in the visual sense.
retina to the brain. vector: the manner in which images are created as we would
perceptually uniform color model: see uniform color model. draw a map by hand: the hardware moves to one location and
draws to the next.
picture elements: see pixels.
visible light: the portion of the electromagnetic spectrum to
pixels: the individual picture elements composing a raster
which the human eye is sensitive.
image on a graphics display device.
visual pigments: the light-sensitive chemicals found within
primary visual cortex: the first place in the brain where all of
rods and cones.
the information from both eyes is handled.
wall-sized display: a technology in which the graphic display
pupil: the dark area in the center of the eye.
covers a large portion of a wall; this is typically accomplished
raster: an image composed of pixels created by scanning by combining multiple projected images.
from left to right and from top to bottom.
wavelength of light: the distance between two wave crests
associated with electromagnetic energy.
REFERENCES
Birren, F. (1983) Colour. London: Marshall Editions Limited. Brewer, C. A. (1997) “Spectral schemes: Controversial color
Bowmaker, J. K., and Dartnall, H. J. A. (1980) “Visual pig- use on maps.” Cartography and Geographic Information
ments of rods and cones in a human retina.” Journal of Science 24, no. 4:203–220.
Physiology 298:501–511. Brewer, C. A., MacEachren, A. M., Pickle, L. W., and
Brewer, C. A. (1992) “Review of colour terms and simultane- Herrmann, D. (1997) “Mapping mortality: Evaluating color
ous contrast research for cartography.” Cartographica 29, schemes for choropleth maps.” Annals of the Association
no. 3/4:20–30. of American Geographers 87, no. 3:411–438.
Brewer, C. A. (1994b) “Guidelines for use of the perceptual Derrington, A., Lennie, P., and Krauskopf, J. (1983)
dimensions of color for mapping and visualization.” In “Chromatic response properties of parvocellular neurons
Color Hard Copy and Graphic Arts III, ed. by J. Bares, pp. in the macaque LGN.” In Colour Vision: Physiology and
54–63. Bellingham, WA: The International Society for Psychophysics, ed. by J. D. Mollon and L. T. Sharpe, pp.
Optical Engineering. 245–251. London: Academic Press.
17:34:54.
Principles of Color
De Valois, R. L., and Jacobs, G. H. (1984) “Neural mecha- Hurvich, L. M., and Jameson, D. (1957) “An opponent-
nisms of color vision.” In Handbook of Physiology process theory of color vision.” Psychological Review 64,
(Section 1: The Nervous System), ed. by J. M. Brookhart no. 6:384–404.
and V. B. Mountcastle, pp. 425–456. Bethesda, MD: Lobben, A. K., Olson, J. M., and Huang, J. (2005) “Using fMRI
American Physiological Society. in cartographic research.” Proceedings, 22nd International
Eastman, J. R. (1986) “Opponent process theory and syntax Cartographic Conference, A Coruña, Spain, CD-ROM.
for qualitative relationships in quantitative series.” The MacEachren, A. M. (1995) How Maps Work: Representation,
American Cartographer 13, no. 4:324–333. Visualization, and Design. New York: Guilford.
Foley, J. D., van Dam, A., Feiner, S. K., and Hughes, J. F. Morgenstern, D., and Seff, J. (2002) “Macworld’s ultimate
(1996) Computer Graphics: Principles and Practice (2nd buyer’s guide: Monitors.” Macworld, February: 56–73.
ed., C ed.). Reading, MA: Addison-Wesley.
Olson, J. M., and Brewer, C. A. (1997) “An evaluation of
Goldstein, E. B. (2007) Sensation and Perception (7th ed.). color selections to accommodate map users with color-
Belmont, CA: Thomson Wadsworth. vision impairments.” Annals of the Association of
Helmholtz, H. von. (1852) “On the theory of compound American Geographers 87, no. 1:103–134.
colors.” Philosophical Magazine 4:519–534. Peddie, J. (1994) High-Resolution Graphics Display Systems.
Hering, E. (1878) Zur Lehre vom Lichtsinne. Vienna, Austria: New York: Windcrest/McGraw-Hill.
Gerold. Taylor, J., Tabayoyon, A., and Rowell, J. (1991) “Device-inde-
Hubel, D. H. (1988) Eye, Brain, and Vision. New York: pendent color matching you can buy now.” Information
Scientific American Library. Display 7, no. 4/5:20–22, 49.
Hurvich, L. M. (1981) Color Vision. Sunderland, MA: Sinauer Wyszecki, G., and Stiles, W. S. (1982) Color Science: Concepts
Associates. and Methods, Quantitative Data and Formulae (2nd ed.).
New York: Wiley.
211
17:34:54.
17:34:54.
Map Elements and Typography
213
17:35:20.
OVERVIEW
This chapter introduces common map elements that are heading that further explains the map’s theme. The
employed in the creation of thematic maps, and presents data source informs the map user of where the the-
typography and its application in cartography. The gen- matic data were obtained. The scale provides an indi-
eral goal is to provide you with guidance in the creation cation of how much reduction has taken place, or
of efficient, attractive maps that effectively communicate allows the map user to measure distances. The scale can
geographic information. Specific goals include guidance take the form of a representative fraction, a verbal
in choosing which map elements to employ and how to scale, or a bar scale. Orientation refers to the indica-
implement them, and how to apply effective typography. tion of direction, and is represented by a north arrow
Map elements and typography are closely related to the or graticule.
topic of cartographic design. Section 1 begins with a clar- Section 3 is dedicated to typography, the art or
ification of widely misunderstood terms associated with process of specifying, arranging, and designing type.
the alignment and centering of map features. A firm Type refers to the words that appear on maps. Type is
understanding of these terms is essential in understand- organized according to type family, type style, type-
ing the remainder of the chapter. face, type size, font, and also according to whether serifs
Section 2 introduces common map elements. It is are present. Type can be modified for specific purposes
the cartographer’s job to select appropriate map ele- by altering letter spacing, word spacing, kerning, and
ments, and to implement them appropriately within the leading. General and specific guidelines are provided
context of available space. Decisions regarding map for the use of type in cartography. Also included in
elements should be made according to the needs of the section 3 is a description of labeling software that
map user, who represents the map’s intended audience. employs aspects of expert systems to place type labels
Detailed descriptions of map elements are provided, automatically, according to rules and guidelines speci-
together with rules and guidelines related to their fied by cartographers.
implementation. The elements are introduced in the This chapter was not written with specific software
order we recommend they be placed when constructing applications in mind. Graphic design applications
a map. The frame line and neat line help to organize such as Adobe Illustrator typically provide the greatest
the content of a map and to define its extent. The control over graphics and type, but recent advances in
mapped area is the region of Earth being represented; the design capabilities of Geographic Information
it consists of thematic symbols and base information. Systems have narrowed the gap, allowing you to pro-
The inset is a smaller map included within the context duce high-quality maps in a user-friendly GIS envi-
of a larger map. The title and subtitle convey the map’s ronment. The examples in this chapter reflect simple
theme, plus additional information such as the region thematic maps, although the principles set forth can
or date. The legend defines all of the symbols that are also be applied to more complex thematic maps and
not self-explanatory, and normally includes a legend general reference maps.
17:35:20.
1 ALIGNMENT AND CENTERING
A
Horizontal Alignment and Centering
Before we begin, it is important to clarify terms of Vertically Distributed Features
associated with the alignment and centering of map
features. Alignment and centering of map features is Black Black Black
achieved either through visual approximation or Mountain Mountain Mountain
measurement. Terms such as visually centered refer to
the placement of map features through visual approx-
imation, so that they look centered within an area.
Alignment and centering is also achieved through
measurement, resulting in map features that are
precisely aligned or centered. If the term “visually
centered” appears, it refers to visual approximation; Horizontally Horizontally Horizontally
all other references to alignment and centering aligned to right centered aligned to left
in this chapter are based on measurement, as
described next. B Vertical Alignment and Centering
Terms such as “horizontally aligned to right” and of Horizontally Distributed Features
“vertically centered” are used repeatedly in this
chapter to describe spatial relationships between map Black Vertically aligned
features. These terms are widely misunderstood and Mountain to bottom
warrant clarification. Map features such as symbols
and type are often arranged in relation to imaginary Black Vertically
lines, as illustrated in Figure 1. Horizontal alignment Mountain centered
and centering (Figure 1A) involves the side-to-side
movement of map features in relation to a vertical line Black Vertically aligned
(the square, circle, star, and “Black Mountain” label Mountain to top
are moved in relation to the dashed line). Horizontal
alignment and centering of multiple lines of type is
referred to as left, center, and right justification. C
Adobe Illustrator
Vertical alignment and centering (Figure 1B) involves
the top-to-bottom movement of map features in rela-
tion to a horizontal line. Be aware that the terms
horizontal and vertical are often confused in this con-
text. (The terms are even misused in the manual of
one prominent graphic design software application.) Horizontal controls Vertical controls
For example, the square, circle, star, and “Black
Mountain” label in Figure 1A are vertically distrib- ESRI ArcMap
uted (arranged from top to bottom), but alignment
and centering are achieved through side-to-side Horizontal
(horizontal) movement. In Figure 1C, the alignment controls
and centering controls of popular graphic design and
GIS applications are presented. Although the termi-
nology differs slightly from application to application,
Vertical
the concepts just presented hold true. A firm grasp of controls
these concepts and terms is crucial to fully under-
standing this chapter, and will help you when creating
maps.
FIGURE 1 (A) Options for horizontal alignment and center-
2 MAP ELEMENTS ing (right, center, and left justification). (B) Options for verti-
cal alignment and centering. (C) Alignment and centering
The various purposes a map can serve, together with controls in Illustrator and ArcMap. Illustrator screenshot used
the wide variety of mapping techniques available, by permission © 2003 Adobe Systems Incorporated. All rights
can lead to the perception that every map is com- reserved. Adobe and Illustrator are registered trademarks of
pletely unique. Despite the great variety of maps Adobe Systems Incorporated in the United States and/or
in the world, it is important to recognize that most other countries. ArcMap screenshot Source: ESRI.
215
17:35:20.
are created from a common set of map elements. design, and typography that interfere with the map
These map elements represent the building blocks of user’s ability to interpret the map.
cartographic communication: the transmission of The rules and guidelines presented here are based
geographic information through the use of maps. The on convention, research, common sense, and, to a lesser
following list presents the most common map ele- degree, the opinions of the authors based on many
ments, each of which is described in detail beginning years of experience. We believe it is important to build
in section 2.1. a foundation of knowledge and experience by follow-
ing these specific rules and guidelines. After a founda-
1. Frame line and neat line tion has been built, we encourage you to consider
2. Mapped area different points of view, and to “break the rules” in
3. Inset creative ways. Always be prepared to explain or defend
4. Title and subtitle your reasoning when choosing and implementing
map elements.
5. Legend
6. Data source 2.1 Frame Line and Neat Line
7. Scale
8. Orientation The frame line and neat line help to organize the map’s
contents and to define its extent. A frame line encloses
These eight map elements are listed in order of all other map elements (Figure 2A); it is similar to a
decreasing size—elements toward the top of the list picture frame because it focuses the map user’s atten-
occupy the most space, and elements toward the bot- tion on everything within it. The frame line should be
tom occupy the least. These map elements are also the first map element placed because it defines the ini-
listed in a logical progression and represent the order tial available space that all other map elements will be
in which we recommend they be placed when con- placed within. A neat line might also appear inside a
structing a map. Larger map elements are placed first, frame line. The neat line is used to crop (i.e., clip, or
and smaller map elements are placed last. For exam- limit the extent of) the mapped area, as it does to the
ple, the frame line occupies the largest area, estab- San Francisco Bay area in Figure 2B. A frame line
lishes the size and shape of the initial available space, should be used in most situations; a neat line is used
and should be placed first. Available space refers to when the mapped area needs to be cropped. In certain
areas on a page or screen that can be used to place cases, a frame line can also act as a neat line, enclosing
map elements. It is also referred to as “white space.” all map elements and cropping the mapped area
The mapped area and inset are among the largest map (Figure 2C).
elements, and should be placed after the initial avail- The style of the frame line and neat line should be
able space has been defined by the frame line; the title, subtle. A single thin, black line should be used; slightly
subtitle, and legend are intermediate in size, and thicker lines are appropriate when working with larger
should be placed in the space that remains after plac- formats, such as wall maps and posters. Many large-
ing the mapped area and inset; smaller map elements format maps (e.g., wall maps of the world) employ
such as the data source, scale, and orientation (typi- ornate frame lines, which often detract attention from
cally a north arrow), should be placed last in the space more important map elements, such as the mapped
that remains. area. Overly thick or ornate lines should be avoided
Cartographic design requires that you decide which for the same reason that you would avoid choosing a
map elements to include on a particular map. In addition gaudy picture frame for a fine painting or photograph
to choosing appropriate map elements, you need to (Figure 2D).
decide on the most appropriate implementation of each The size (width and height) and position of the frame
element. The content, style, size, and position of each map line depend on the desired map size and page dimen-
element must be carefully considered, even before map sions. The size and position of the neat line are normally
construction begins. dictated by the frame line, the mapped area, and the
Proper choice and implementation of map elements other map elements.
is governed in large part by the purpose of the map and
its intended audience. The map user represents the 2.2 Mapped Area
intended audience; virtually every decision you make
should be made in reference to the needs of the map The mapped area is the region of Earth being repre-
user. The proper choice and implementation of map ele- sented, as illustrated by Kenya in Figure 3. It consists
ments result in a minimization of “map noise,” which of visually dominant thematic symbols that directly
refers to unnecessary or inappropriate symbolization, represent the map’s theme (we say little about thematic
17:35:20.
A Edge of page
B Edge of page
Frame line
Frame line
Neat line
TITLE
TITLE
Legend
Legend
C Edge of page
D
Frame line acting as neat line Edge of page
TITLE TITLE FIGURE 2 (A) The frame line

encloses all map elements. (B) The
neat line crops the mapped area. (C)
A frame line can also act as a neat
line. (D) Thick or ornate lines should
be avoided. (Note that the edge of
Legend Legend the page is illustrated to put the
frame and neat lines into context.
The edge of the page is not illus-
trated elsewhere in this chapter.)
symbols in this chapter). The mapped area can also city names, and the state line. Without this base infor-
include base information that provides a geographic mation, the thematic symbols (precipitation contours
frame of reference for the theme. Base information and gray-tone shades) would be virtually meaning-
includes, but is not limited to, boundaries, transporta- less—the map user would be unable to associate them
tion routes, landmarks, and place-names. In Figure 3, with a particular geographic region.
the gray-tone shades are thematic symbols represent- In some cases, the mapped area appears as a “float-
ing the percentage of all secondary schools that are ing” geographic region disconnected from neighboring
religious or private within each district of Kenya. The regions, as illustrated by Kenya in Figure 3. Use of a
railroads represent base information that provides a floating mapped area produces available space that
geographic frame of reference for people who might often eases the placement of other map elements.
be familiar with them, but might not be familiar with However, it also removes the region of interest from its
the district boundaries. In this case, the base informa- geographic context, making the map more abstract, and
tion also suggests a possible relationship between the possibly confusing the map user. In other cases, a neat
railroads and the districts having the highest percent- line is used to crop the mapped area, as illustrated by
age of religious or private secondary schools. Base the Kansas City region in Figure 7. A cropped mapped
information that illuminates the map’s theme in this area represents the region of interest within its
manner is of particular value. As another example, geographic context, but can make the placement of
consider Figure 7. Four types of base information are other map elements more difficult due to a lack of avail-
included here: interstate highways, city boundaries, able space. Consideration of the map’s purpose, the
217
17:35:20.
Secondary Schools A TITLE B TITLE

Kenya, 1984
Legend Legend
TITLE
C D TITLE
Religious
or Private (%)
20.1 - 61.8
12.7 - 20.0
5.7 - 12.6
1.3 - 5.6
0.0 - 1.2
Source: Ministry of Education, 1987. Education in
Railroad Kenya, Nairobi, Kenya: Jomo Kenyatta Foundation.
Legend Legend
FIGURE 3 The mapped area is composed of thematic sym-
bols (gray-tone shades in this example) and base information
(railroads in this example). FIGURE 4 Sizing and position of the mapped area.
(A) Appropriate sizing and position. (B) Insufficient size, but
appropriate position. (C) Excessive size, but appropriate
position. (D) Insufficient size and inappropriate position.
geographic region of interest, and other map elements (Note the inclusion of interstate highways, which act as base
will determine whether the mapped area should be information.)
floating or cropped.
The size of the mapped area is dependent on several
factors, including the page size, margins, and space (top to bottom) within the available space, as defined
needed for other map elements. A general guideline is by the frame line. The mapped areas in Figures 4A and
to make the mapped area as large as possible within the 4B are appropriately centered; the mapped area in
available space without being “too close” to the frame Figure 4D is off-center both horizontally and verti-
line and leaving ample room for the remaining map ele- cally. A properly centered mapped area can lend a
ments (Figure 4A). Maximum size is important because sense of balance to a map, but irregularly shaped geo-
the mapped area—thematic symbols in particular—is graphic regions, together with other factors, often
instrumental in communicating the map’s information. make centering impossible. The country of Chile is a
Figure 4B represents a mapped area that does not take classic example of a geographic region that is com-
full advantage of the available space, whereas Figure 4C monly positioned off-center to accommodate addi-
illustrates a mapped area that is too large: It touches the tional map elements.
frame line and leaves inadequate space for the title and
legend. 2.3 Inset
Position of the mapped area is dependent on sev-
eral factors, including the shape of the geographic An inset is a smaller map included within the context of
region, page dimensions, and the other map elements. a larger map. Insets can serve several purposes: (1) to
If possible, the mapped area should be visually cen- show the primary mapped area in relation to a larger,
tered, both horizontally (side to side) and vertically more recognizable area (a locator inset), as illustrated
17:35:20.
A Widget Factories B Widget Distributors C WIDGET SALES, 2010
1990 2000
Area of
Detail
FIGURE 5 Insets serve several purposes: (A) locating the primary mapped area; (B) enlarging an important or congested area;
and (C) showing related themes or dates. An additional purpose is showing related areas (Figure 14).
in Figure 5A; (2) to enlarge important or congested A

areas (Figure 5B); (3) to show topics that are related to Long Term Debt
the map’s theme, or different dates of a common theme
represented by smaller versions of the primary mapped AVERAGE AGE, 2001
area, as illustrated by the 1990 and 2000 insets in Figure
5C; and (4) to show areas related to the primary Museums of Modern Art
mapped area that are in a different geographic location,
or cannot be represented at the scale of the primary B
mapped area, as illustrated by Alaska and Hawaii in Population Density
Figure 14. New Hampsire, 2010
The style of the inset can vary. In Figure 5A, the
inset is relatively subtle; its only purpose is to help ori- Number of Chickens
ent the map user. In Figures 5B and 5C, the insets take Harper County
on a central focus and attract attention along with the
primary mapped area. The size and position of the
BIRTH RATE INCREASE
1950 - 2000
inset are equally variable, depending on the purpose
of the inset, the size of the map, and the other map ele- Subtitle horizontally centered below title
ments.
2.4 Title and Subtitle FIGURE 6 (A) Titles. The title of a thematic map succinctly
describes the map’s theme. (B) Titles with subtitles. The subtitle
Most thematic maps require a title, although a title is is used to further explain the title.
sometimes omitted when a map is used as a figure in a
written document, assuming that the theme is clearly For example, the title “A Map of the Population
expressed in the figure caption. A well-crafted title can Density of New Hampshire Counties in 2010” is exces-
draw attention to a map, however, and thus we recom- sively wordy, and “NH Pop. ’10” is abbreviated to the
mend using a title in virtually all situations—even when point of being cryptic; however, “Population Density”
the theme is reflected in a figure caption. The title of a or “Population Density in New Hampshire, 2010”
thematic map should be a succinct description of the clearly expresses the theme in only a few words. Notice
map’s theme (Figure 6A), whereas the title of a general in the last example that the theme, geographic region,
reference map is normally a statement of the region and year all appear in the title, and that the theme is
being represented. Unnecessary words should be omit- aptly stated before the region and year. Use of the word
ted from the title, but care should be exercised to avoid map in a title is a statement of the obvious, and should
abbreviations that the map user might not understand. be avoided.
219
17:35:20.
The subtitle, if employed, is used to further explain the between the title and the legend (Monmonier 1993).
title. The name of the region and the data collection date Many large-format wall maps make use of this technique,
are common components of the subtitle (Figure 6B). because distances between the title and legend can be
“New Hampshire, 2010” would be an appropriate subtitle great. The subtitle, if used, should be located directly
for the “Population Density” title of the previous exam- below the title and should be horizontally centered with
ple. The name of the region is often omitted when the it (center justified), as illustrated in Figure 6B. The hori-
cartographer feels that the map user will easily identify zontal centering of multiple lines of type creates a “self-
the region. For example, most residents of the United balancing” block of type, and is used frequently in car-
States will recognize the shape of the United States, and tography.
most readers of an article focusing on Japan will recog-
nize the shape of Japan. 2.5 Legend
In the name of legibility, the style of the title and sub-
title should be plain. Avoid italics and ornate type styles. The legend is the map element that defines all of the
Although the use of bold type can emphasize the title thematic symbols on a map. Symbols that are self-
and subtitle, it is normally not required if appropriate explanatory or not directly related to the map’s theme
type sizes are chosen. Use a subtle bounding box are normally omitted from simple thematic map
around the title and subtitle only if it is necessary to legends. In contrast, legends for general reference maps,
mask the underlying mapped area to improve legibility, such as Forest Visitor maps issued by the U.S. Forest
as illustrated in Figure 7. The unnecessary use of bound- Service, often define all symbols found on the map—
ing boxes around map elements is a common source of even if they are self-explanatory. General guidelines for
map noise. legend design are provided in this section; legend design
The title should generally be the largest type on a and content for particular thematic map types are
map; the subtitle should be visibly smaller than the title. described in Part III of this text.
Both the title and subtitle should be limited to one line The style of the legend should be clear and straight-
each in most cases. If possible, the title should be placed forward. Use a subtle bounding box around the legend
toward the top of the map, where the map user is accus- only if it is necessary to mask the underlying mapped
tomed to seeing titles, and be horizontally centered area, as illustrated in Figure 7. Special care should
within the frame line (Figure 7). A case can also be made be exercised to ensure that symbols in the legend are
for keeping the title directly above the legend, to relieve identical to those found within the mapped area. This
the map user from having to jump back and forth includes size, color, and orientation if possible. The
Annual Precipitation
Average, 1961-1990
29 35
TOPEKA
KANSAS 70
70 CITY
LAWRENCE
MISSOURI
KANSAS
35
Inches
44
42
40 FIGURE 7 The title and subtitle
38
Source: NRCS Water and Climate Center, 0 40
placed at top center. Note how
36
Central U.S. Average Annual Precipitation,
34 bounding boxes are used to mask the
1961-1990. Miles
underlying mapped area.
17:35:20.
legend in Figure 8A contains symbols that differ which are read from left to right; it allows the map user
significantly from those in the mapped area; the legend to view the symbol first, and then its definition. In most
symbols in Figure 8B are appropriately identical to cases, representative symbols should be vertically dis-
those in the mapped area. Additional problems with tributed (placed in sequence from top to bottom), and
the legend in Figure 8A include an excessively thick horizontally centered with one another as illustrated
bounding box, inconsistency in type characteristics, in Figure 9B.
poorly aligned symbols and type, and the express bus Symbols should be vertically centered with their
route symbol—a linear feature that is inappropriately definitions * (Figure 9B). Textual definitions such as
“trapped” inside of an areal symbol (the rectangle). “Community Garden,” and definitions consisting of
A well-designed legend is self-explanatory, and does individual numbers such as “25,000,” should be hori-
not need to be identified with “Legend” or “Key” zontally aligned to left (Figure 9B). Ranges of num-
labels like that seen in Figure 8A. bers are normally separated by a hyphen, which is
Representative symbols should be placed on the left compact, or the word “to,” which can prevent the
and defined to the right, as illustrated in Figure 9A. placement of consecutive hyphens when negative
This arrangement is customary in English dictionaries, numbers are represented (Figure 9C). For clarity,
spaces should be included to the right and left of each
separator.† Definitions containing numbers of 1,000
or greater should incorporate commas, and decimal
A E numbers smaller than 1 should incorporate a leading
A 15x RO
R O O S E V E LLT
T
zero, as illustrated, respectively, by “1,021.8” and
AC H U S E T T S
“0.4” in Figure 9C. Space considerations might

VENUE
VENUE
E require the legend to be oriented in a horizontal fash-

M A S S AC
E
AV
A ion, with the definitions horizontally centered below

AV
A
E
TR
RUUMAN
the symbols they represent and the symbols vertically
Legend aligned to bottom, as illustrated in Figure 9D. This
E approach is used frequently in this text, particularly
Redevelopment Zone 45 with mapped areas such as Kansas that are wider
A than they are tall.
Targeted Properties
OW
EISENHO WER Notice in Figure 9 that all the point and line symbols
E Escrow Properties are separated from one another, whereas the areal sym-
OTA
TA
A
MICHIGAN
M I N N E S OT
A Acquired bols (the rectangles, or “boxes”) are either separated

O
Properties
Bus Route from one another (Figure 9A), or are connected and
Express Bus Route
share a common boundary (Figure 9C). Areal symbols
DY
K E N N E DY within the mapped area are almost always connected and
share common boundaries, suggesting that legend boxes
A E should also be connected. However, the type of data rep-
B 15x RO
R O O S E V E LLT
T
resented by the boxes also influences whether they
AC H U S E T T S
should be connected or separated. We suggest that leg-

VENUE
VENUE
E end boxes should be connected when representing the

M A S S AC
E
AV
A
AV
magnitude of a single attribute (e.g., differences in per

A
E
TR
RUUMAN
capita income), because connected boxes help to empha-
size the idea that a gradation of values (associated with a
Targeted Property
E single attribute) is being represented. Legend boxes
45
E Escrow Property should be separated when representing qualitative cate-
A gories of data (e.g., forest, desert, and tundra). This is
A Acquired Property
OW
EISENHO WER
45 Bus Route
OTA
TA
A * The legend in Figure 7 is an exception. Here, definitions represent the

MICHIGAN
M I N N E S OT
O
15x Express Bus Route values of the boundaries between symbols.

† An en dash can also be used as a separator. An en is a unit of measure
Redevelopment Zone
DY
K E N N E DY equal to half the point size of the type being used. En dashes are half
the width of em dashes, and are used to separate ranges of numbers in
written documents, without the use of spaces to the right and left of the
FIGURE 8 (A) A poorly designed legend. (B) A well- dash (Romano 1997, 223). For clarity in map legends, we recommend
designed legend that accurately represents symbols found use of the hyphen (which is shorter than the en dash) or the word “to,”
within the mapped area. with spaces on both sides.
221
17:35:20.
symbol that represents one well) should be singular in

A the legend, not plural (Figure 9F). A common example
Scrub
of this problem is a map that includes a legend defini-
Representative
Symbols Swamp Definitions tion of “Roads.” In almost all cases, each road symbol in
the mapped area represents a single road, not multiple
Mangrove roads, and should be defined in the legend as “Road”
instead of “Roads.”
B A legend heading is often included, and is used to fur-
ther explain the map’s theme. The unit of measure (for
Community Garden quantitative data) and the enumeration unit are com-
Symbols
vertically Demonstration Farm mon components of the legend heading. For example, a
distributed map with the title “College Graduates” might have a leg-
Farmer’s Market end heading such as the one illustrated in Figure 10A.
Symbols The legend heading should normally be placed above
vertically Symbols Definitions the legend, and can be horizontally centered with it; mul-
centered with horizontally horizontally
definitions centered aligned to left tiple lines of type in the legend heading can be horizon-
tally centered with one another (center justified)
(Figure 10A). The horizontal centering of the legend
C heading, legend content, and related map elements such
43 - 108 -89.2 to -45.4 as the data source creates a group of objects that is self-
32 - 42 0.4 to 187.3 balancing; it is neither “left-heavy” nor “right-heavy,”
7 - 31 597.4 to 1,021.8 and can lend to an overall sense of balance in a map. The
D E
Lake A
Percentage by
30,000 60,000 100,000 Park Census Block Group
Definitions horizontally centered
Golf
with symbols; symbols vertically
Course 5 - 38
aligned to bottom
46 - 62
F Singular: implies that one
71 - 80
Well symbol represents one well
Data unavailable for areas
Plural: implies that one symbol shown in white
Wells represents multiple wells Source: U.S. Census Bureau, United
States Census, 2010.
FIGURE 9 (A) The legend is composed of representative

Legend heading, legend, no data label, and
symbols and definitions. (B) Distribution and alignment of data source horizontally centered
symbols and definitions. (C) Ranges of numbers and numeric
formatting. (D) A horizontally oriented legend. (E) Irregular
B C
polygons used to represent areal data. (F) Singular versus Number of Days Number of Days
plural definitions. Above 90o F Above 90 o F
18 - 29 18 - 29
because separated boxes help to reinforce the idea that
11 - 14 11 - 14
distinctly different entities are being represented, as 0-5 0-5
opposed to magnitudes of a single attribute. Source: Western States Source: Western States
Areal symbols are sometimes represented in legends Weather Office, 2007 Report. Weather Office, 2007 Report.
by irregular, amorphous polygons, as illustrated in

Figure 9E. Irregular polygons can look more natural,
and are most appropriate when representing qualitative FIGURE 10 (A) Typical choropleth legend with legend head-
categories of areal data such as lakes and golf courses. ing and “no data” label. The data source appears at the bot-
They should be separated because they represent tom. (B) Legend heading, legend, and data source horizon-
unique categories of data. A symbol that represents a tally aligned to left. (C) The same objects horizontally aligned
singular feature within a mapped area (e.g., a point to right.
17:35:20.
same objects are sometimes horizontally aligned to the The legend should be large enough to allow the map
left (Figure 10B) or to the right (Figure 10C), but the user to employ it easily, but should not be so large as to
self-balancing effect is lost. occupy vast areas of space, or to challenge thematic
A category that represents an absence of areal data symbols in the mapped area. The legend heading should
can be represented on the mapped area in a neutral be smaller than the subtitle, and the definitions should
color such as white, or with a subtle texture that is not be smaller than the legend heading, as illustrated in
used in any other category. A simple note below the leg- Figure 10A.
end can be used to inform the map user of the “no data” Position of the legend is dependent on available
category, as illustrated in Figure 10A. A potential prob- space, as defined by the other map elements; if possi-
lem with this method is that very light gray areas might ble, the legend should be visually centered within a
be confused with the white no-data category, and tex- larger portion of available space. For example, in
tures might draw undue attention. Another option is to Figure 12 space is available to the left and right of
place a small dot within no-data areas; the dot must southern South America. Each section of the legend,
then be identified as representing no data in the legend. which has been split in two because of the large
One problem with this method is that a point symbol is number of categories, is visually centered within its
used to represent an areal feature, possibly confusing available space. The currency map in Figure 12 is qual-
the map user. itative in nature, and its theme would be better repre-
Legend symbols are often organized into groups, sented using colors of unique hue, as illustrated in
based on a particular criterion. For example, Figure 11 Color Plate 1. Line and dot patterns should be avoided
represents three possible methods of grouping the sym- if color reproduction is feasible.
bols that might appear on a map having to do with nat-
ural gas pipeline facilities. In Figure 11A, two groups are 2.6 Data Source
formed according to whether the symbols represent nat-
ural or cultural features. In Figure 11B, three groups are The data source allows the map user to determine
formed according to the general geometric form of the where the thematic data were obtained. For example,
symbols: point, line, or area. In Figure 11C, two groups in Figure 12, the currency information was acquired
are formed according to whether the symbols are the- from the National Trade Data Bank, a publication of
matic in nature and directly related to the map’s theme the U.S. Department of Commerce. Sources of base
(the upper group), or represent base information (the information such as roads or administrative bound-
lower group). This last approach is probably the most aries are normally omitted from the data source on
appropriate for thematic maps that include base infor- thematic maps. The data source should be formatted
mation in the legend. in a manner similar to that of a standard bibliographic
A B C
Compressor Station
River Compressor Station
Major Meter Station
Forest Major Meter Station
Milepost
Milepost
Lake City or Town
HTG Dual Pipeline
HTG Relict Pipeline
Compressor Station 84 Interstate Highway SW&P Dual Pipeline
Major Meter Station 26 Federal Highway Southeast Pipeline
Milepost 55 State Highway Greenville Pipeline
City or Town HTG Dual Pipeline
84 Interstate Highway HTG Relict Pipeline

SW&P Dual Pipeline City or Town
26 Federal Highway
Southeast Pipeline 84 Interstate Highway
55 State Highway
Greenville Pipeline 26 Federal Highway
HTG Dual Pipeline
River 55 State Highway FIGURE 11 Grouping of legend
HTG Relict Pipeline
SW&P Dual Pipeline
River symbols according to different crite-
Southeast Pipeline Forest ria: (A) natural and cultural groups;
Forest
Greenville Pipeline Lake (B) point, line, and polygon groups;
Lake and (C) thematic symbol and base
information groups.
223
17:35:20.
CURRENCY Source: Census 2010, Summary File 3, U.S. Census Bureau.
Data Source: ESRI Data & Maps, 2007.
Source: United States Central Intelligence Agency.

Map File #505103 (547149) 2-82.
Data Source: National Atlas of the United States,

http://www-atlas.usgs.gov/mld/huralll.html
Source: The International Bank for Reconstruction and

Development/The World Bank.
World Development Report, 2000: Poverty.
Source: Field Survey by Nigel Tufnel and

David St. Hubbins. December, 2008.
Multiple lines are horizontally centered
Bolivar
Boliviano FIGURE 13 Data sources. The data source indicates where
Dollar thematic data were obtained. It is among the smallest type on
Franc a map.
Guilder
Guarani
Nuevo Sol
Peso is to inform the curious, not to attract attention. Opti-
Real mally, the data source is placed directly below the legend
Sucre to which it refers, and is horizontally centered with the
legend heading and legend, as illustrated in Figure 10A.
Source: U.S. Department of Commerce.
It is also common to place the data source in the lower
National Trade Data Bank, May 6, 1999. right or lower left corner (inside the frame line).
Related to the data source is information regarding
FIGURE 12 The legend is visually centered within the avail- the coordinate system and map projection used. The
able space if possible. In this example, the legend has been split type of projection can become increasingly important
into two sections, each centered within its available space. to the map user as larger geographic regions are repre-
(Compare with color version shown in Color Plate 1.) sented and as distortion of the mapped area increases.
Include a separate block of type indicating the projec-
reference, but is often more concise and less formal. tion (and information related to the projection) if you
Evaluation of the intended audience’s needs will feel it will help the map user to interpret the map more
dictate whether to include a fully detailed data accurately. Like the data source, blocks of type related
source or a simplified version. Many map users have to map authorship and map projections should be
no intention of tracking down a data source, but are among the smallest on the map.
likely to want a general idea of where the data were
obtained. Most data sources in this text are simplified 2.7 Scale
versions.
The style of the data source should be plain and The scale indicates the amount of reduction that has
subtle, as illustrated in Figure 13. The words “Data taken place on a map, or allows the map user to mea-
Source:” or “Source:” should be included to avoid ambi- sure distances.
guity: The data source indicates where the data came The representative fraction (e.g., 1:24,000) is a ratio
from, not map authorship. If necessary, a separate block of map distance to Earth distance, which indicates the
of type can be used to indicate map authorship. If the extent to which a geographic region has been reduced
data were obtained from a publication, the name of the from its actual size. For example, on a map with a scale
publication should be italicized (as in a bibliographic ref- of 1:24,000, one unit of distance on the map (e.g., 1 cm)
erence); all other type should be normal. If the data were represents 24,000 of the same units on Earth, regard-
obtained from a preexisting map, special methods of cita- less of the unit of measure used. The representative
tion should be employed, based on the type of source fraction becomes invalid if the map on which it
map used (Clark et al. 1992). Multiple lines of type in a appears is enlarged or reduced, and cannot easily be
data source should be horizontally centered with one used to measure distances. For these reasons, the rep-
another (center justified) (Figure 13). The data source resentative fraction does not appear on most thematic
should be among the smallest type on a map; its purpose maps.
17:35:20.
Although not actually spoken (it is a block of type), information is directly related to the map’s theme and
the verbal scale reads like a spoken description of the would be difficult to acquire without a bar scale.
relationship between map distance and Earth distance. Bar scales lose much of their utility when used on maps
“One inch to the mile” means that 1 inch of distance on in which the scale changes greatly from one location to
the map represents 1 mile on Earth. Like the represen- another, such as on a map of the entire world. A bar scale
tative fraction, the verbal scale becomes invalid if the should be employed with caution in such cases, because it
map on which it appears is enlarged or reduced. will allow the map user to measure distances accurately
Distances on a map can be determined with accuracy only along certain lines (e.g., standard parallels or princi-
using the verbal scale, but only in conjunction with a pal meridians). A variable bar scale is useful with certain
ruler. Rough measurements can be taken by estimating map projections (e.g., Miller Cylindrical), as it reflects
the length of an inch, centimeter, etc. changes in scale in relation to latitude (Figure 15A).
The bar scale, or scale bar, resembles a ruler that can A basic understanding of map projections and associated
easily be used to measure distances on a map. Its abil- scale changes is essential in providing the map user with a
ity to indicate distance, together with its ability to with- bar scale that is actually useful and accurate.
stand enlargement and reduction of a map, make it the The maximum distance value represented in a bar
preferred format to include on a thematic map. You scale should always be round and easy to work with.
should normally include a bar scale on a thematic map Compare the maximum values illustrated in Figure
if distance information can enhance the map user’s 15B with those in Figure 15D. A maximum value of
understanding of the theme. The map in Figure 12 lacks 400 miles is preferred to 437 miles (quick, what’s one-
a bar scale because distance information lends no quarter of 437?). Decimal values such as 327.75 are
insight into the type of currency each country uses. difficult to work with and should be avoided in favor
However, the map in Figure 14 includes bar scales of integers. Incorporate a unit of measure that is
because distance information can potentially enhance appropriate for the intended audience. For example,
the map user’s understanding of the theme. For exam- people raised in the United States will be more famil-
ple, general measurements of the state of Colorado iar with miles than with kilometers, whereas people
taken with the bar scale yield an area of approximately from outside the United States will be most familiar
93,750 square miles. Taking into consideration the min- with kilometers. Incorporate both miles and kilome-
imum number of farms in Colorado’s category ters if you determine that the map user will benefit
(26,000), it can be estimated that there is at least one from having both (Figure 15D). In addition, try to
farm per every four square miles, on average. This choose a unit of measure that is appropriate for the
Number of Farms, 2005
Thousands
0 - 25
0 500
FIGURE 14 A bar scale (or bar
26 - 65 scales, in this example) should be
Miles
included if distance information can
0 1,000 0 500 Source: U.S. Department of Agriculture, 72 - 230
National Agricultural Statistics Service. enhance the map user’s understanding
Miles Miles
of the theme.
225
17:35:20.
maximum value on the bar scale. For example, small

A 0º units of measure such as meters or feet can result in
Latitude 15º excessively large maximum values, as illustrated by
30º “640,000” in Figure 15B. A more appropriate
45º maximum value for this bar scale would be 600 kilo-
60º
meters. Similarly, a bar scale with a maximum value of
0.18 miles could also be represented in feet, with a
75º
0 100 200 300 400 500 maximum value of 1,000.
Miles
Some bar scales include an “extension scale,” which
consists of numbers and tic marks positioned to the left
B of zero, as illustrated in Figure 15C. The extension scale
can be useful when employing a particular method of
measuring map distances,* but is also a source of confu-
sion for many map users, who expect numbers to the
left of zero to be negative (as on a number line).
Potential confusion, together with the fact that small,
intermediate distance tics can be included to the right
of zero, argue against the use of extension scales on the-
C matic maps.
1 0 1 2 3 4
Like the data source, the bar scale should be subtle;
Kilometers its purpose is to inform the curious, not to attract atten-
tion. Bulky and complex designs (Figure 15B) should
D 0 400
be avoided in favor of slender, simple, and precise
designs (Figure 15D). Line weights should be fine, and
Miles type should be among the smallest on a map. Avoid the
use of bold and italic type styles. Include intermediate
0 Meters 1,000 distance tics, but only as many as necessary for making
general measurements.
The bar scale should be long enough to be useful
0 1 2 Miles but not so long that it is cumbersome. Figure 15E
illustrates three bar scales of differing length, each of
0 1 2 3 Kilometers
which is accurate. The 50-mile bar scale is too short
to be useful, and the 1,000-mile scale is cumber-
E some—it’s actually wider than the mapped area
(Iceland). The 500-mile bar scale is a good solution
because it is long enough to be useful but doesn’t
take up too much horizontal space. The optimal
length is directly tied to the size of the mapped area
and to the quest for a maximum distance value that is
round and easy to work with. If possible, the bar scale
0 50 should be placed below the mapped area, where the
Miles
map user is accustomed to finding it, as illustrated in
0 500 Figure 14.
Miles
0 1,000
2.8 Orientation
Miles
Orientation refers to the indication of north on a map.
Orientation can be indicated by a north arrow, or
FIGURE 15 (A) A variable bar scale reflects changes in scale
through the inclusion of a graticule (a system of grid
in relation to latitude. (B) Bulky, poorly designed bar scales.
(C) Bar scale incorporating an “extension scale” to the left of * Distance on a map is transferred as tic marks to a scrap of paper,
zero. (D) Slender, simple, well-designed bar scales. (E) The which is held up to the bar scale. The right tic mark on the scrap is
500-mile bar scale represents the most appropriate length in aligned with a value on the bar scale located to the right of zero, so
this example. that the left tic mark on the scrap falls within the extension scale.
Total distance is calculated by adding distances found to the right and
left of zero.
17:35:20.
lines, normally representing longitude and latitude). It A graticule indicates direction through the orientation
is a common misconception that every map needs a of grid lines, typically meridians that run north–south, as
north arrow. The orientation of maps with north at the illustrated in Figure 16C. Notice that the meridians in this
top is a long-standing tradition in the Western world, example are not parallel to one another—this indicates
and it is assumed that “north is at the top” of most that the direction of north is variable, depending on which
maps. An indication of orientation should be included part of the map is focused on. Just as caution should be
if the map is not oriented with geographic or “true” exercised when employing bar scales on maps with vari-
north at the top. This can occur when the shape of the able scale, caution should be exercised when placing
mapped area, together with the page dimensions, north arrows on maps in which the direction of north is
require the cartographer to rotate the mapped area, as variable. For example, it would be inappropriate to
illustrated by California in Figure 16A. Map users include a north arrow on the map in Figure 16C, because
expect to find north at the top of a map; they should be it would indicate north for only one particular location,
notified if north isn’t at the top. Another case in which not for the entire mapped area. In addition to orientation,
you would indicate orientation is when the map is the graticule can also provide positional information. This
intended for use in navigation, surveying, orienteering, can be important when a map’s theme is in some way
or any other function in which the determination of related to latitude or longitude. For example, a graticule,
direction is crucial. Finally, an indication of north or at least marginal tic marks, is often included on the-
should be given if geographic features such as roads matic maps of natural vegetation, because this attribute is
are oriented in a manner that might confuse the map strongly influenced by latitude. Labels indicating the val-
user. For example, a north arrow should be included ues of grid lines should be included when employing a
when a rectangular street grid is oriented to magnetic graticule in this fashion. For example, a meridian might be
north instead of geographic north (Figure 16B). labeled “60° E,” and a parallel might be labeled “20° S.”
The style of the north arrow and graticule should be
simple and subtle. Like the data source and bar scale,
these map elements should not attract attention. Line
A
N
weights should be fine, and type should be among the

smallest on a map. Bulky and complex north arrows
should be avoided (Figure 17A), and only north should
be indicated (if necessary, the map user can infer the
other cardinal directions). Subtle and simple north
B A
N
C
B N MN N
MN
22º
FIGURE 16 (A) Use of a north arrow on a map not oriented

with north at the top. (B) Indication of geographic north when FIGURE 17 (A) Bulky and complex north arrows. (B) Subtle
geographic features are oriented to magnetic north (17° E in and simple north arrows for geographic north (N), magnetic
this case). (C) Meridians of a graticule indicating direction of north (MN), and both combined (a compound north arrow).
north.
227
17:35:20.
arrows are illustrated in Figure 17B, where a five-

pointed star (representing Polaris, the North Star)
indicates geographic north, and an arrowhead (repre-
senting a compass needle) represents magnetic north.
The arrowhead is not the most logical choice for indi-
cating geographic north because compass needles point
to magnetic, not geographic, north. Regardless of this
logical inconsistency, the arrowhead is commonly used
instead of the five-pointed star to indicate geographic
north. Compound north arrows indicate both geo-
graphic and magnetic north, and include the declina-
tion (in degrees) between the two. Compound north
arrows should be reserved for use on maps intended
for navigation with a compass.
The north arrow should be relatively small; it should
be large enough to find and use, but not so large that it
attracts attention. The north arrow should be placed in FIGURE 18 (A) Type family. (B) Type style. (C) Typeface (a
an out-of-the-way location, preferably near the bar scale. particular type family and type style). (D) Type size.
3 TYPOGRAPHY
Additional type styles include condensed, expanded, light,
and extra bold. Type of a particular family and style is
Type, or text, refers to the words that appear on maps.
referred to as a typeface, such as Palatino Roman or
Typography is the art or process of specifying, arranging,
Helvetica Bold (Figure 18C). A particular typeface is fur-
and designing type. Several of the map elements described
ther differentiated by type size (Figure 18D). Type size
in the previous section are partly composed of type—the
(height) is measured in points (one point equals 1/72”).
legend, for example—and others, such as the data source,
Although type sizes are described in points, the actual
are composed entirely of type. Type can be considered a
height of a given character cannot be inferred from its
special sort of symbol, or even a map element in its own
point size; the point size refers to the height of the metal
right. Type that is well designed and smartly applied can
block on which type was created prior to the development
make a map easier to understand and more attractive.The
of digital type. A font is a set of all alphanumeric and spe-
rules and guidelines for the use of type in cartography are
cial characters of a particular type family, type style, and
derived from general rules of typography, but they have
type size, such as Palatino Italic, 12 point (Romano 1997).
been modified over time to reflect the specific purposes of
Many typographical terms are misunderstood and
mapmaking. Fortunately, these rules and guidelines are
misused. For example, the term font is commonly (and
relatively well defined. We encourage you to build a foun-
inappropriately) used to refer to a type family or typeface.
dation of typographic skills by following the specific rules
Uppercase and lowercase letters are used in cartog-
and guidelines presented here, and then to consider alter-
raphy, but lowercase letters have proved to be easier
native approaches. Always be prepared to explain or
to read. This is because lowercase letters are less
defend your typographical decisions.
blocky, and they provide more detail that helps differ-
entiate one letter from another. The majority of type
3.1 Characteristics of Type
on a map should be set in title case, as illustrated in
Figure 19A. Title case is composed of lowercase letters
Type is commonly organized according to characteristics
with the first letter of each word set in uppercase.
such as type family, type style, and type size. Type family
Conjunctions and other “linking words” (in, on, or, of,
refers to a group of type designs that reflect common
per, by, for, with, the, and, over, etc.) are set in lower-
design characteristics and share a common base name—
case. Title case is appropriate for use in titles, subtitles,
Palatino, for example (Figure 18A). Within a type family,
legend headings, legend definitions, labels for point
type is differentiated by type style: Roman (normal), bold,
and line features, and so on. Sentence case is com-
and italic are common type styles* (Figure 18B).
posed of lowercase letters with the first letter of each
sentence set in uppercase, and is appropriate when for-
* When applying type styles, it is best to use a member of a type fam-
mal sentences are used, such as in textual explanations
ily that has been specifically designed with that style (e.g., Bookman
Bold). Many software applications allow roman type to be crudely or descriptions appearing on a map. Words set in all
modified into italic or bold, resulting in type that is unsuitable for uppercase are sometimes used as titles and as labels
high-quality printing. for areal features, as described later.
17:35:20.
A
Letter Spacing and Word Spacing
Minimal spacing
Slightly increased spacing
E X A G G E R AT E D S PA C I N G
B
Kerning
New E mployees WA
Increase or
decrease
MN
C
FIGURE 19 (A) Uppercase, lowercase, title case, and Leading
sentence case. The majority of type on a map should be set in
City Park Increase or
title case. (B) Serifs are extensions at the ends of letter strokes. Baselines
County Park decrease
Serifs are short extensions at the ends of major FIGURE 20 (A) Letter spacing and word spacing can be
letter strokes, as illustrated in Figure 19B. Type fami- altered for different situations. (B) Kerning adjusts the space
lies with serifs are termed serifed; type families with- between two individual adjacent letters. (C) Leading is the
out are termed sans serif (without serifs). Serifed type space between lines of type, from baseline to baseline.
is preferred in the context of written documents
because the serifs provide a horizontal guideline that
helps to tie subsequent letters together, reducing eye restaurant on a particular map. Related to letter and
fatigue. In cartographic applications, where type is word spacing is tracking, which refers to the simulta-
used primarily as short labels and descriptions, both neous alteration of space between both letters and
serifed and sans serif type is used; neither has proved words. Tracking should be used with caution, however,
to be more effective than the other. In certain situa- because it does not allow for the independent adjust-
tions, a serifed type family can be used for one cate- ment of letter and word spacing.
gory of features (e.g., natural), and a sans serif type Kerning refers to the variation of space between two
family can be used for another category of features adjacent letters, as illustrated in Figure 20B. Different
(e.g., cultural). combinations of adjacent letters require different
Not surprisingly, letter spacing refers to the space amounts of kerning if spaces between letters are to be
between each letter in a word, and word spacing refers visually consistent. For example, the letter pair WA
to the space between words. Minimal letter and word requires the removal of space between the letters for
spacing (Figure 20A) results in compact type that is the pair to look consistent with MN. Digital type
often easier to place on complex maps (because the includes preset “kerning pairs” that automatically set
words occupy less horizontal space). Slightly the space between various letter pairs. Kerning can also
increased letter and word spacing results in type that be performed manually, after letter and word spacing
appears to be less “cramped” and is easier to read. has been specified, to adjust the space between particu-
You should employ slightly increased spacing if possi- lar letter pairs that still appear to be too close together
ble. The blocky nature of all-uppercase type normally or too far apart. Kerning is measured according to a unit
requires greater letter and word spacing than lower- called an em, which is equal to the point size of the type
case type to prevent it from looking cramped. being used.
Exaggerated letter and word spacing is often Leading (pronounced like heading), or line spacing,
employed in conjunction with all-uppercase type refers to the vertical space between lines of type accord-
when labeling areal features, as described later. Letter ing to their baselines, and is altered to place lines of type
and word spacing should be kept consistent within closer together or farther apart (Figure 20C). Leading
individual blocks of type and among labels that are should be great enough to allow multiple lines of type to
otherwise similar. For example, each line of type in a be read easily, but not so great as to result in wasted
three-line data source should have the same letter and space between lines. (An exception to this guideline is
word spacing, as should every label that identifies a presented in section 3.6.)
229
17:35:20.
3.2 General Typographic Guidelines family can be used for one category of features (e.g.,
natural), and a sans serif family can be used for
The following is a list of general guidelines for the use another category of features (e.g., cultural). This is
of type in cartography. illustrated in Figure 21, where the rivers, forest, and
lake are labeled with a serifed type family, and the
1. Avoid the use of decorative type families, and use
cities are labeled with a sans serif family. Notice that
bold and italic styles sparingly. Script, cursive, and
Lake Isabella is treated as a natural feature here. It
otherwise fancy and ornate styles are unnecessarily
could also be treated as a cultural feature, formed by
difficult to read. They should be avoided in favor of
the construction of Isabella Dam.
more practical type families, such as those used as
examples in this chapter. The overuse of bold styles 3. Choose a realistic lower limit for type size; all type
can overshadow other type and map elements, and needs to be readable by the intended audience.
bold is often not necessary if appropriate type sizes Factors for consideration include the age and visual
are chosen. If possible, italic type should be acuity of the map user, map reproduction method,
reserved for two applications: to label hydrographic anticipated lighting conditions, and the map user’s
(water) features, and to identify publications in the physical proximity to the map. Mark Monmonier
data source, as illustrated in Figure 21. Italics are (1993) recommended a lower size limit of 7 points
appropriate for hydrographic features because for lowercase type, but this value is conservative.
their slanted form resembles the flow of water. (It is Type as small as 3.2 points (usually uppercase, sans
also conventional to use the color cyan for hydro- serif) is commonly used on congested street maps
graphic labels and features.) The use of italics for where space is at a premium (Jonathan Lawton,
publications is standard bibliographic practice. The California State Automobile Association, personal
wide variety of features on general reference maps communication, 2002). Readability is ultimately tied
might require that you use bold and italic styles out- to the typeface used, crispness of reproduction, and
side of these guidelines. other factors. The only way to ensure the readability
2. Avoid using more than two type families on a given of small type is to provide a sample to members of
map. Simpler maps can be limited to one type family. the intended audience.
For the sake of consistency, map elements such as the 4. Generally speaking, type size should correspond with
title, subtitle, legend heading, legend definitions, data the size or importance of map features. For example,
source, and scale should all employ the same type type representing the names of large cities should be
family. If two type families are required (e.g., to label noticeably larger than type used to represent small
a wide variety of map features), choose families that cities (Figure 21). Type size is also partially dictated
are distinctly different—one serifed and one sans by the relative importance of map elements, as
serif, for example. As mentioned earlier, a serifed described in section 2. Because map users are not
Look Out Below

Delano The Spectre of Dam Failure
Lake
Isabella
Isabella
Dam
2,585 ft
SEQUOIA
Wasco
eek
ver
Cr
99 Poso Ri
178
FOREST
Shafter N AT I O N A L
rn
Ke
Oildale
City Population
Bakersfield 12,736 - 21,263
408 ft 27,885 - 38,824
5
58
Lamont 247,057
0 5 10 15 20 FIGURE 21 General typographic
Miles Source: ESRI Data & Maps, 2007
Arvin
guidelines appropriately applied.
17:35:20.
sensitive to slight differences in type size, avoid dif-

ferences of less than 2 points if possible (Shortridge
A
Sibleyville
1979). LAKE
EMERALD
5. Critically evaluate and apply type specifications such Forest Boundary
as type family, type style, type size, letter spacing,
word spacing, kerning, and leading. Do not passively
accept the default settings provided by software B No vos ibi rsk
applications. Instead, consider the purpose of each Novo kuzne tsk C h it a
unit of type in the context of the map, and apply type Irk ut sk a
K a ra g Ch atang
specifications accordingly. anda
r
b a a ta
6. All type should be spell-checked. Spelling errors U la a n
Ulia sta y
interfere with cartographic communication and
undermine the credibility of a map. Special atten- A lm a ti
F ru n z Urum qi
tion should be focused on the most current e
u
B a o to
spelling of place-names, which change over time Kashi
and are often controversial. Also, be aware that
certain older place-names are considered to be
offensive or derogatory by today’s standards C
Lincoln H
ighway
(Monmonier 2006).
3.3 Specific Guidelines: All Features

S
IL
VE KE
R
HOR LA
The following is a list of specific guidelines for the SESHOE
placement of type associated with point, linear, and

areal features.
1. Orient type horizontally, as illustrated in FIGURE 22 (A) Type placed horizontally. (B) Type oriented
Figure 22A. One exception is when labeling a map with a graticule. (C) Diagonal and curved type.
that includes a graticule with curved parallels, in
which case the type should be oriented with the par-
allels (Figure 22B). Another exception is when label-
ing diagonal or curved linear and areal features, in
which case the type should reflect the orientation of
A
the features, as illustrated in Figure 22C.
Wilson Capital City
2. Avoid overprinting and, when unavoidable, mini-
mize its effects. Overprinting is a phenomenon that
occurs when a block of type is placed on top of a map Vernal
feature (e.g., a river), obscuring the type and making
it difficult to read, as illustrated by the difficult-to- B
read “Wilson” in Figure 23A. The effects of over-
u te
printing can be minimized through the use of a mask, al C o mm
halo, or callout. A mask is a polygon (e.g., a white Se ason
Mikkerton
rectangle) that is placed underneath type but above
the mapped area, as illustrated by “Capital City” in
Figure 23A. As seen in this example, masks can
C
sometimes obscure too much of the mapped area,
and thus they should be used with caution. Masks Thule Park
can also be specified with the same color as the
underlying area, allowing them to blend in better, as
illustrated by “Vernal” in Figure 22A. A halo is an
extended outline of letters (Figure 23B). Halos cover
less of the underlying mapped area than masks, while FIGURE 23 (A) Masks placed underneath type. (B) Halos
still allowing the type to be read. Care should be around type. (C) A callout combines a mask with a leader line.
exercised to ensure that halos are not so thick that
231
17:35:20.
they attract undue attention, as illustrated by

“Mikkerton” in Figure 23B. Callouts are a combina-
A
(5th best)
tion of mask and leader line (Figure 23C). Callouts Capitola
are effective, but should be used with caution (3rd best) Capitola Capitola (preferred)
because they are visually dominant and can over- (4th best) Capitola Capitola (2nd best)
shadow other map elements. Capitola
(6th best)
3. Ensure that all type labels are placed so that they
are clearly associated with the features they repre- (8th best) ONTARIO ONTARIO (7th best)
sent. In pursuit of this goal, it is often useful to place
larger type labels first, followed by intermediate
and then smaller labels (Imhof 1975). B
Bicycle Barn Bicycle Barn
3.4 Specific Guidelines: Point Features
Preferred position, but label is 3rd best position, and
separated from symbol label is close to symbol
The following is a list of specific guidelines for the
placement of type associated with point features.
C
1. When labeling point features, select positions that Shoshone
avoid overprinting underlying graphics, according Shoshone
to the sequence of preferred locations illustrated Leader line does not touch point symbol
in Figure 24A. This sequence is based on the work
of Pinhas Yoeli (1972), but is modified by the
D
authors according to the idea that, if possible, the
Devil’s Grand
symbol should be placed on the left and defined Golfcourse Des Junction
to the right (as in a legend). Notice, also, that the Moines
least preferred locations for a label are directly to Lines Lines
horizontally Lines horizontally
the right and directly to the left of the symbol. aligned to horizontally aligned to
This results in an “unfavorable optical coinci- right centered left
dence” (Imhof 1975, 132), in which the point sym-
bol might be misinterpreted as a type character in
the label. FIGURE 24 (A) Sequence of preferred locations for labeling
2. Do not allow other map features to come between point features. (B) A road coming between a point symbol
a point symbol and its label (Figure 24B). and its label. (C) Leader lines are used if necessary. (D)
Emphasize the association between the label and Multiple-line labels are horizontally aligned or centered
symbol by placing the label close to the symbol, (right, center, and left justified) to imply association with the
even if it means choosing a less preferred location. symbol.
3. If the sequence of preferred locations does not pro-

vide a suitable option, consider using a mask, halo, as illustrated in Figure 25A (Wood 2000).Avoid over-
or callout. Another option is to use a simple leader printing the coastline with type (Figure 25B).
line, as illustrated in Figure 24C. Leader lines
6. These guidelines should be followed as closely
should be very thin (e.g., 0.25 point), not include an
as possible. In practice, however, it is often impossi-
arrowhead, and point to the center of the point
ble to adhere to all guidelines simultaneously.
symbol without actually touching it.
Figure 25A illustrates how these guidelines might
4. Multiple-line labels should be placed according to be applied to a map of restaurant locations in a
the sequence of preferred locations, and individual coastal region. Notice in Figure 25A that thematic
lines of type should be horizontally aligned or cen- symbols (restaurant locations) are labeled with a
tered (left, right, or center justified) to emphasize sans serif type family, while base information such
the association between the label and point symbol as streets and the bay are identified with a serifed
(Figure 24D). type family. This is another example of how two dis-
5. Point symbols on land that are close to coastlines tinctly different type families can be used to repre-
should be labeled entirely on land if possible. sent different categories of map features. Figure
Point symbols that touch coastlines should be labeled 25B illustrates how poorly labeled thematic sym-
either entirely on land or entirely on water, bols can result in a less attractive map that is more
depending on which option offers greater legibility, difficult to interpret.
17:35:20.
Fisk
Fran
A Oceanview Inn A
c is c
Filbert’s Wa y S tr a tf o rd R
Loop
H u n ti n g to n a il ro a d
o
Don’s
Drive
CRE
Tino’s
Drive
Drive
Village Romano’s
SCE
Cafe Pete’s
St re et
Jack & Jill’s
City Steak
B
Palmer
TRIMBLE C
NT B
Bobo’s OUNTY
Joe’s Buka GRANT CO
Barton Street
UNTY
AY
Mariano’s
Horne Crab Hut
Chadwick
Hig hw ay
Lee
Olde Pelicano
Tavern C
A ll Ame rica n Rive r
Fisk
Oceanview
Fran
B Inn
c is c
Don’s Loop Filbert’s

o
D
Drive
Tino’s
CRE
Long C r e e k L on g C r e e k
Drive
Romano’s
Drive
SCE
Village Cafe Street
Jack & Pa
Jill’s
lmer
Pete’s NT B
City Steak
Joe’s
Bobo’s FIGURE 26 (A) Type labels placed above linear features.
Mariano’s
Barton Street Buka
(B) Labels on both sides of a boundary. (C) Label following
AY
Horne
Chadwick
Crab Hut the general trend of a complex curve. (D) Long feature
Hig hw ay
Lee
Olde
Tavern Pelicano labeled twice.
3. Very long linear features can be labeled more than

FIGURE 25 (A) Point features appropriately labeled. (B)
once (Figure 26D). The use of multiple labels is pre-
Point features inappropriately labeled.
ferred to the exaggeration of letter and word spac-
ing to emphasize linear extent.
7. Do not exaggerate letter or word spacing when 4. Labels for linear features should be placed
labeling point features. Exaggerated spacing weak- upright, not upside down. Correctly placed labels
ens the association between a point symbol and its read from left to right, whereas incorrectly placed
label, and tends to emphasize the areal extent of labels read from right to left, as illustrated in
features (and point features have no areal extent). Figure 27. By convention, type that is absolutely
vertical should be readable from the right side of
3.5 Specific Guidelines: Linear Features the page. These rules actually apply to all type
that appears on a map, but are most closely asso-
The following is a list of specific guidelines for the ciated with linear features.
placement of type associated with linear features.
3.6 Specific Guidelines: Areal Features
1. When labeling linear features, place type above the
features, close to but not touching them, as illus-
The following is a list of specific guidelines for the
trated in Figure 26A. Descenders, such as the lower
placement of type associated with areal features.
extensions of “g” and “y” in “Huntington Way,”
should just clear the line symbol. Type is placed 1. When labeling areal features that are large enough
above linear features because it appears to rest on to fully contain a label, visually center the label
the feature instead of hang below it, and because within the feature, as illustrated in Figure 28A.
the bottom edge of lowercase type is normally less Don’t allow the label to crowd the areal symbol—
ragged than the upper edge, resulting in a more har- allow a space of at least one and one-half times the
monious relationship between label and symbol. type size (1.5 ems) between the ends of the label
One exception is when labeling areas on both sides and the boundary of the feature (Imhof 1975). As
of a boundary, in which case type appears above when labeling linear features, follow the general
and below the line (Figure 26B). trend of areal features that have complex curves.
2. When labeling linear features that have complex 2. Consider using all-uppercase type when labeling
curves, follow the general trend of the feature areal features (Figure 28A). The blocky nature of
(Figure 26C), because type that curves too much is uppercase type can help to emphasize areal
difficult to read. extent.
233
17:35:20.
A B
U K R A I N E
Incorrect
ct
t
ec
r re
C
I
Co
rr
nc
A
nco
Co
r re
SP
r re o
t
c
I
ct
IA
Correct
C
N
Incorrect
SEA
t U k r a i n e
ct
c
I
r re
r re
Co
nco
Correct
Co
o
r re
Inc
rr
ct
ec
t
D E
MUDDY RIVER
S C E N I C
FIGURE 27 Type should be placed upright, and should read Reflection
A R E A Pool
from left to right. Vertical type should be readable from the
right side of a page.
Mollisol
FIGURE 28 (A) A type label visually centered within an

3. Exaggerated letter and word spacing can be used to areal feature. (B) Uppercase type with exaggerated letter and
emphasize areal extent, and is most effective when word spacing, emphasizing the areal extent of the feature.
applied to all-uppercase type (Figure 28B); lower- (C) Lowercase type is not as well suited to exaggerated spac-
case type tends to look disjointed when exagger- ing. (D) Exaggerated leading. (E) A river labeled as an areal
ated spacing is applied (Figure 28C). Caution feature, a small areal feature (reflection pool) labeled as a
should be exercised when exaggerating letter and point feature, and a leader line used to help identify an areal
word spacing, as individual letters can become so feature.
far apart that the map user would have trouble see-
ing the label as anything other than individual let-
ters. An extreme upper limit of four times the type 3.7 Automated Type Placement
size (4 ems) should be observed for letter spacing.
Maximum word spacing may exceed this value, but Throughout this section, we have assumed that the car-
care should be exercised to ensure that the rela- tographer is making all the decisions regarding type
tionship between words is clear. placement. When using graphic design software such as
4. Leading can also be exaggerated to emphasize Illustrator, this is almost certainly the case. However,
areal extent, as illustrated in Figure 28D. Leading specialized labeling software has been developed for
should not be so great that the relationship automatically positioning type, often within the context
between lines of type is lost. of a GIS. Labeling software focuses primarily on the
placement of type associated with map features (e.g.,
5. Features typically thought of as being linear (e.g.,
streets and lakes) as opposed to positioning map ele-
rivers) that are represented at such a large scale
ments that are composed of type, such as the title.
that they appear as areas should be labeled like
Development of this software has been the focus of
areal features, as illustrated in Figure 28E.
both computer scientists and cartographers, who have
6. Areal features that are too small to contain a label created sophisticated algorithms (including heuristics,
should be labeled as if they were point symbols, or “rules of thumb”) based on established rules, guide-
as illustrated by the reflection pool’s label in lines, and conventions of cartographic type placement.
Figure 28E. Most labeling software applications incorporate aspects
7. If necessary, leader lines can be used with areal fea- of cartographic expert systems, which make decisions
tures. Leader lines should be very thin (e.g., 0.25 based on rules and guidelines obtained from carto-
point), not include an arrowhead, and just enter the graphic experts (Zoraster 1991).
areal symbol, as illustrated by the Mollisol soil poly- Labeling software is designed to approach or achieve
gon in Figure 28E. “optimal” placement of type labels, which avoids both
17:35:20.
overprinting underlying map features and creating flict. After all type labels have been placed, the type is
conflicts among type labels. Two general approaches transferred back into the GIS application that provided
have emerged: (1) placement of each label in its pre- the input map layers.
ferred position, followed by an iterative reorganization Like Label-EZ, Maplex allows the user to specify
of labels to avoid or minimize conflict among labels rules that dictate how type should be placed for indi-
(Freeman 1995); and (2) casual, suboptimal placement vidual feature classes. In addition, Maplex supports the
of labels, followed by an iterative reorganization of detailed manipulation of type, including changes in
labels until the combined placement of all labels case and language translation; automates the symbol-
approaches an optimal state (Edmondson et al. 1996; ization of feature classes; and allows the export of type
Pinto and Freeman 1996). and symbolized features to leading graphic arts appli-
The biggest potential advantage of labeling software cations such as Illustrator. In other words, Maplex is
is its ability to save time, as manual type placement more than labeling software—it is a suite of auto-
remains one of the most time-consuming aspects of mated cartographic production tools that includes
map construction, particularly when labeling linear and automated type placement (Environmental Systems
areal features (Barrault 2001). One problem associated Research Institute 1998). Unlike Label-EZ, Maplex is
with labeling software is the fact that optimal solutions designed for use primarily with ESRI GIS applica-
are often computer-intensive when a considerable tions. ESRI also distributes Maplex for ArcGIS—an
amount of type must be positioned. Another problem extension for ArcGIS that focuses exclusively on auto-
is that the wide variety of maps, together with varia- mated type placement. In contrast to Label-EZ, which
tions in map scale and complexity, make it difficult to attempts to place labels in their optimal positions first,
achieve satisfactory results in all situations. The fin- Maplex first identifies large numbers of suboptimal
ished product normally requires some interactive edit- positions for labels, and subsequently reorganizes
ing to arrive at a solution that is visually acceptable. As them until the combined placement of all labels
more sophisticated algorithms are developed and as approaches an optimal state.
the speed of computers increases, it appears inevitable While it doesn’t fit neatly into the category of label-
that type placement will become a more fully auto- ing software just described, a software application
mated process. named TypeBrewer is worthy of mention. TypeBrewer
Two leading examples of labeling software are is a free help tool that gives nonspecialist mapmakers a
Label-EZ, distributed by MapText, Inc., and Maplex, chance to explore typographic alternatives, and to see
distributed by Environmental Systems Research the impact that various aspects of type have on the
Institute (ESRI). Label-EZ, which was developed over overall look and feel of a map.
several decades by Herbert Freeman, is capable of
working with many popular GIS applications (Freeman
2004). It allows the user to specify rules that dictate SUMMARY
how type should be placed for individual feature classes
(logical groupings of points, lines, polygons, etc.). For In this chapter, we have introduced map elements as
example, the user is allowed to specify exactly how the building blocks of cartographic communication, and
labels should be placed for point features representing have described their appropriate selection and usage in
spot elevations, versus how labels should be placed for response to needs of the map user. The frame line and
soil polygons. The software applies these specifications neat line should be subtle, and used to define a map’s
to a series of map layers derived from a GIS, and extent. Thematic symbols directly represent a map’s
attempts to place each type label in its optimal posi- theme, and should be visually dominant. They are often
tion. If the optimal position is not possible due to over- used in conjunction with base information in the
printing of map features or conflict with other type mapped area. The inset is a smaller map used in the con-
labels, the software attempts suboptimal placement text of a larger map, and can serve several purposes. The
options (as initially dictated by the user) until a suit- topic of a thematic map is clearly expressed with a
able placement is achieved. This iterative procedure is prominent, concise title, and is further explained in the
an example of the first approach to achieving optimal subtitle. Map symbols that are not self-explanatory are
placement, as described previously: labels are placed in defined in the legend of a thematic map; representative
their preferred positions, and are subsequently reorga- symbols and their definitions are arranged in various
nized to avoid or minimize conflict. If no suitable ways, according to the type of map and other factors.
placement is possible, Label-EZ does not place the The map user is told where thematic data were obtained
label, but rather creates a record in a log file. At this through the use of a subtly designed data source that is
stage, a cartographer is required to intervene and similar to a bibliographic reference, but often simpler.
use human judgment to resolve the placement con- The bar scale was identified as the most practical tool to
235
17:35:20.
allow the map user to make general measurements on a effective. Blocks of type can occupy more or less space
thematic map. A subtle, easy-to-use bar scale should be by altering letter spacing, word spacing, and leading;
included if it will help illuminate a map’s theme. A sub- spaces between letter pairs can be fine-tuned through
tle north arrow or graticule also can be employed if the kerning. General guidelines for the use of type were
map is not oriented with north at the top, or if the map described, including the needs to avoid decorative type
user will require directional information. styles and to minimize the use of bold and italic; the
We have also presented aspects of typography, and need to limit a map to two type families; the need to
have identified it as being central to the utility and select a minimum type size that will be readable by
attractiveness of a map. The cartographer needs to members of the intended audience; the need to size the
ensure that type is legible, and select appropriately type to correspond with the relative size and impor-
from type families, type styles, and type sizes. tance of map features; the need to critically evaluate
Lowercase type is more legible than uppercase, and and specify all aspects of type; and the need to spell-
type set in title case (a combination of uppercase and check type that appears on a map. Specific guidelines
lowercase type) is appropriate for most cartographic were provided for the use of type associated with point,
applications. Serifed and sans serif type are both used linear, and areal features, and an overview of auto-
in cartography, as neither type has proved to be more mated type placement was given.
FURTHER READING
Barrault, M. (2001) “A methodology for the placement and A description of the cartographic type placement problem, and a
evaluation of area map labels.” Computers, Environment, and commercial solution to it.
Urban Systems 25:33–52. Imhof, E. (1975) “Positioning names on maps.” The American
Discusses automated type placement with an emphasis on evalua- Cartographer 2, no. 2:128–144.
tion of type placement and measures of fitness. The classic reference on guidelines for positioning type on maps.
Brewer, C. (2005) Designing Better Maps: A Guide for GIS Monmonier, M. S. (1993) Mapping It Out: Expository
Users. Redlands, CA: ESRI Press. Cartography for the Humanities and Social Sciences. Chicago:
A practical guide to cartographic design for GIS users. University of Chicago Press.
Clark, S. M., Larsgaard, M. L., and Teague, C. M. (1992) Introduces practical aspects of cartographic design for noncartog-
Cartographic Citations, A Style Guide. Chicago: American raphers.
Library Association, Map and Geography Round Table, Pinto, I., and Freeman, H. (1996) “The feedback approach to
MAGERT Circular No. 1. cartographic areal text placement.” In Advances in
A guide to cartographic citation. Structural and Syntactical Pattern Recognition, ed. by P.
Perner, P. Wang, and A. Rosenfeld, pp. 341–350. Berlin:
Dent, B. D. (1999) Cartography: Thematic Map Design. 5th ed.
Springer-Verlag.
Boston: McGraw-Hill.
An example of automated suboptimal positioning of type labels,
A classic cartography textbook with chapters on cartographic
which are subsequently reorganized to approach an optimal state.
design and typography.
See also Freeman (1995).
Edmondson, S., Christensen, J., Marks, J., and Shieber, S. M.
Robinson, A. H., Morrison, J. L., Muehrcke, P. C., Kimerling,
(1996) “A general cartographic labelling algorithm.”
A. J., and Guptill, S. C. (1995) Elements of Cartography. 6th
ed. New York: Wiley.
An example of automated suboptimal positioning of type labels,
which are subsequently reorganized to approach an optimal state.
design and typography. Includes essays focusing on the modern
Environmental Systems Research Institute (ESRI) (1998) role and character of cartographic design.
Maplex—Automatic Cartographic Name Placement Software. Wood, C. H. (2000) “A descriptive and illustrated guide for
ESRI, Inc. White Paper. type placement on small scale maps.” The Cartographic
A description of the cartographic type placement problem, and a Journal 37, no. 1:5–18.
commercial solution to it.
An extensive collection of typographical guidelines for use in car-
Freeman, H. (1995) “On the automated labeling of maps.” In tography.
Shape, Structure, and Pattern Recognition, ed. by D. Dori and Zoraster, S. (1991) “Expert systems and the map label place-
A. Bruckstein, pp. 432–442. Singapore: World Scientific. ment problem.” Cartographica 28, no. 1:1–9.
A seminal paper on automated type placement, in the context of
Explains the role of expert systems in the automated positioning
soil survey maps.
of type.
Freeman, H. (2004) Label-EZ—Software for Automated Text
Placement. MapText, Inc. White Paper.
17:35:20.
GLOSSARY
available space: empty areas on a map in which map mask: a method for labeling a spatial feature; the effect of
elements can be placed. overprinting is reduced by placing a polygon (e.g., a white rec-
bar scale: the map element that is a graphical expression of tangle) underneath the type, but above the underlying graphics.
scale, resembling a small ruler. neat line: the map element—normally a rectangle—that is
base information: graphical symbols and type that provide a used to crop the mapped area.
geographic frame of reference for thematic symbols. orientation: (1) the indication of direction on a map, nor-
callout: a method for labeling a spatial feature; the effect of mally taking the form of an arrow or a graticule; (2) a visual
overprinting is minimized by using a combination of a mask variable in which the directions of symbols (or marks making
and a leader line. up symbols) are varied, such as changing the orientation of
short line segments to indicate the direction from which the
data source: the map element that describes the origin of
wind blows.
the data represented on a map.
overprinting: a phenomenon that occurs when a block of
expert system: a software application that incorporates rules
type is placed on top of another graphical object.
derived from experts to make decisions and solve problems.
representative fraction (RF): expresses the relationship
font: a set of all alphanumeric and special characters of a
between map and Earth distances and is expressed as a ratio
particular type family, style, and size.
of map units/earth units (e.g., an RF of 1:25,000 indicates that
frame line: the map element—normally, a rectangle—that 1 unit on the map is equivalent to 25,000 units on the surface
encloses all other map elements. of the Earth).
framed-rectangle symbol: a point symbol that consists of a scale: provides an indication of the amount of reduction that
“frame” of constant size, within which a solid “rectangle” is has taken place on a map.
placed; the greater the data value, the greater the proportion sentence case: type composed of lowercase letters, with the
of the frame that is filled by the rectangle. first letter of each sentence set in uppercase.
graticule: latitude and longitude taken in combination. serifs short extensions at the ends of major letter strokes.
halo: an extended outline of letters in a type label. subtitle: the map element that is used to further explain the title.
inset: the map element consisting of a smaller map used in thematic symbols: graphical symbols that directly represent
the context of a larger map (e.g., a map that is used to show a a map’s theme or topic.
congested area in greater detail).
title: the map element that provides a succinct description of
kerning: the variation of space between two adjacent letters. a map’s theme.
labeling software: software developed for the automated title case: type composed of lowercase letters with the first
positioning of type. letter of each word set in uppercase.
leading: the vertical space between lines of type; equivalent type: words that appear on a map.
to line spacing.
type family: a group of type designs that reflect common
legend: the map element that defines graphical symbols. characteristics and a common base name.
legend heading: the map element that is used to explain a
type size: the size of type, expressed in points.
map’s legend and theme.
type style: variations of type design within a given type family.
letter spacing: the space between each letter in a word.
typeface: type of a particular type family and style.
map elements: graphical units that maps are composed of,
including the title, legend, and scale. typography: the art or process of specifying, arranging, and
designing type.
map user: the person for whom a particular map has been
designed and produced. verbal scale: the map element that is a textual expression of
scale, such as “one inch to the mile.”
mapped area: the map element representing the geographic
region of interest. word spacing: the space between words.
REFERENCES
Barrault, M. (2001) “A methodology for the placement and Clark, S. M., Larsgaard, M. L., and Teague, C. M. (1992)
evaluation of area map labels.” Computers, Environment Cartographic Citations, A Style Guide. Chicago: American
and Urban Systems 25:33–52. Library Association, Map Geography Table, MAGERT
Circular No. 1.
237
17:35:20.
Edmondson, S., Christensen, J., Marks, J., and Shieber, S. M. Monmonier, M. (2006) From Squaw Tit to Whorehouse
(1996) “A general cartographic labelling algorithm.” Meadow: How Maps Name, Claim, and Inflame. Chicago:
Cartographica 33, no. 4:13–23. University of Chicago Press.
Environmental Systems Research Institute (ESRI) (1998) Pinto, I., and Freeman, H. (1996) “The feedback approach to
Maplex—Automatic Cartographic Name Placement cartographic areal text placement.” In Advances in Structural
Software. ESRI, Inc. White Paper. and Syntactical Pattern Recognition, ed. by P. Perner, P. Wang,
http://www.esri.com/library/whitepapers/pdfs/maplexwp.pdf. and A. Rosenfeld, pp. 341–350. Berlin: Springer-Verlag.
Freeman, H. (1995) “On the automated labeling of maps.” In Romano, F. J. (ed.). (1997) Delmar’s Dictionary of Digital
Shape, Structure and Pattern Recognition, ed. by D. Dori Printing and Publishing. Albany, NY: Delmar.
and A. Bruckstein, pp. 432–442. Singapore: World Shortridge, B. G. (1979) “Map reader discrimination of
Scientific. lettering size.” The American Cartographer 6, no. 1:13–20.
Freeman, H. (2004) Label-EZ—Software for Automated Text Wood, C. H. (2000) “A descriptive and illustrated guide for
Placement. MapText, Inc. White Paper. http://www.map- type placement on small scale maps.” The Cartographic
text.com/ProductLiterature/Freeman-White-Paper- Journal 37, no. 1:5–18.
041027.pdf.
Yoeli, P. (1972) “The logic of automated map lettering.” The
Imhof, E. (1975) “Positioning names on maps.” The American Cartographic Journal 9, no. 2:99–108.
Young, T. (1801) “On the theory of light and colours.”
Monmonier, M. (1993) Mapping It Out: Expository Philosophical Transactions of the Royal Society of London
Cartography for the Humanities and Social Sciences. 92:12–48.
Chicago: University of Chicago Press.
Zoraster, S. (1991) “Expert systems and the map label
placement problem.” Cartographica 28, no. 1:1–9.
17:35:20.
Cartographic Design
OVERVIEW
This chapter introduces cartographic design as a Section 2 presents a case study of how an effective,
process in which the cartographer conceptualizes and attractive thematic map can be designed according to the
creates maps according to the needs of the intended concepts, rules, and guidelines set forth in this chapter. We
map user. Section 1 describes how aspects of carto- design a real estate site suitability map according to the
graphic design have been derived from the results of general steps identified in the map communication model
map design research, and have been influenced by and according to specific design procedures identified in
principles of graphic design and Gestalt principles of section 1.1. General steps include consideration of what
perceptual organization. The design process is distilled the real-world distribution of the phenomenon (suitable
to seven procedures, including determination of how residential building parcels) might look like, definition of
the map will be reproduced, selection of appropriate the map’s purpose and its intended audience, and data
scale and projection, identification of appropriate data collection. Specific procedures include determination of
classification and symbolization methods, selection and how the map will be reproduced, selection of an appro-
appropriate implementation of map elements, establish- priate scale and projection, identification of appropriate
ment of an intellectual hierarchy, creation of rough data classification and symbolization methods, selection
sketch maps, and construction of the map using a soft- and appropriate implementation of map elements, estab-
ware application—including the creation and evalua- lishment of an intellectual hierarchy that accentuates the-
tion of rough drafts. The design process culminates in matic symbols and differentiates between various levels
an evaluation of the map by members of the intended of base information, creation of sketch maps, and con-
audience. Specific aspects of cartographic design are struction of the map using a software application—
presented, including (1) visual hierarchy—a graphical including the creation and evaluation of rough drafts.
representation of the intellectual hierarchy that visually This chapter was not written with specific software
emphasizes more important map features, and is pro- applications in mind. Graphic design applications
duced by adjusting the visual weight of map elements; such as Adobe Illustrator typically provide the great-
(2) contrast—the visual differences between map ele- est control over graphics and type, but recent
ments that distinguish them and imply their relative advances in the design capabilities of GISs have nar-
importance; (3) figure-ground—the emphasis of impor- rowed the gap, allowing the mapmaker to produce
tant map features by making them appear closer to the high-quality maps in a user-friendly GIS environ-
map user, sometimes through the use of screening; and ment. Although the examples in this chapter reflect
(4) balance—the harmonious organization of map ele- simple thematic maps, the principles set forth can also
ments and empty space that incorporates the concept of be applied to more complex thematic maps and gen-
available space. eral reference maps.
239
17:35:53.
Cartographic Design
1 CARTOGRAPHIC DESIGN ance of a map; a map can have a particular design, but
design in that sense is only the end result of the design
Cartographic design is partly a mental and partly a process. Successful cartographic design results in maps
physical process in which maps are conceived and cre- that effectively communicate geographic information.
ated. The design process appears formally as step 4 of Current theories suggest that maps do not necessarily
the map communication model (Figure 1), but it encom- communicate knowledge, but, rather, stimulate and
passes aspects of all five steps, from imagining the real- suggest knowledge through the transmission of infor-
world distribution to evaluating the resulting map. The mation (Montello 2002).
design process does not end until the final map has been Cartographic design involves the conceptualization
completed. The word design also describes the appear- and visualization of the map to be created, and is driven
by two goals: (1) to create a map that appropriately
serves the map user based on the map’s intended use,
and (2) to create a map that communicates the map’s
information in the most efficient manner, simply and
clearly. Edward Tufte (1990, 53) echoed this second
goal, eloquently stating that “Confusion and clutter are
failures of design, not attributes of information.” The
Step 1 physical act of placing, modifying, and arranging map
Consider what the real-world
elements is often referred to as the separate activity of
distribution of the phenomenon map construction, or “layout.” Because of the holistic
might look like. nature of the design process, we consider map construc-
tion to be largely integrated with the cartographic
design process.
Cartographic design is directed in large part by rules,
guidelines, and conventions, but it is also relatively
Step 2
unstructured. A single, optimal solution to a given map-
Determine the purpose of the map ping problem generally does not exist. Instead, several
and its intended audience. acceptable solutions are usually possible. “Good design
is simply the best solution among many, given a set of
constraints imposed by the problem” (Dent 1999, 241).
If map design were easy and straightforward, this chap-
ter would be unnecessary, as cartographic design expert
Step 3 systems would be used to design most maps. Expert sys-
Collect data appropriate for the tems are defined as software applications that make
map’s purpose. decisions based on rules and guidelines obtained from
cartographic experts. Cartographic expert systems, how-
ever, are currently limited to narrow aspects of carto-
graphic design, and thus lack the ability to completely
and consistently design top-quality maps. As was the
Step 4 case with map elements and typography, we believe that
Design and construct the map.
it is important to build a foundation of cartographic
design skills by following the specific rules and guide-
lines presented here, and then to consider alternative
approaches. Always be prepared to explain or defend
your design decisions.
Many aspects of cartographic design have been
Step 5 guided by the results of map design research. Arthur
Robinson (1952) sparked enthusiasm for this research
Determine whether users find the
map useful and informative. with The Look of Maps, in which he emphasized the
importance of a map’s function over its form and called
for objective experimentation with regard to map
design. Much of this research has focused solely on
determining which mapping techniques are most effec-
tive, without regard for why they are effective (a behav-
FIGURE 1 The map communication model. iorist view). In contrast, cognitive map design research
17:35:53.
Cartographic Design
focuses on why certain techniques are effective by Both cartographic design and graphic design
applying knowledge structures to the ways that people incorporate Gestalt principles of perceptual organiza-
perceive maps. This research has been driven by the tion. Gestalt is a theory of visual perception developed in
idea that understanding cognitive processes, and apply- the 1920s that attempts to describe the manner in which
ing their principles to map design, can result in more humans see the individual components of a graphical
effective maps (Montello 2002). Figure 2 illustrates the image, and then organize the components into a unified
results of eye-movement studies performed by George whole. These principles represent the theoretical under-
Jenks in 1973. These studies revealed variations in eye- pinning for many cartographic design rules, guidelines,
scan paths followed by different map users, and remain and conventions (MacEachren 1995). Closure allows us
as seminal examples of map design research. Notice in to complete an image even when parts are missing, as
Figure 2 that the map user’s eyes “entered” the map at illustrated in Figure 3A. The administrative boundary is
the title; passed through the mapped area, subtitle, and seen as a complete polygon as opposed to individual
legend; returned to the mapped area for a better look; dashes. Common fate allows us to group elements that
and then exited at the title. share the same moving direction, as illustrated by the
Although most map design research represents a traffic flow arrows in Figure 3B. Continuity allows
scientific approach to understanding how maps work, us to move our eyes from one object to another. In
the “art” of maps also plays an important role in carto- Figure 3C, our eyes “continue” the railroad tracks that
graphic communication. The artistic aspect of maps is begin in the upper left as they pass under the road and
guided less by experimentation and more by intuition into the tunnel in the lower right. Figure-ground allows
and critical examination (MacEachren 1995). It is diffi- us to perceive certain objects as being closer to us than
cult to anticipate the map user’s sensitivity to the artis- other objects, and therefore more important. In Figure
tic aspects of a map. However, it seems likely that a map 3D, New Zealand acts as figure to the Pacific Ocean,
that has been created with an artistic synthesis of con- which acts as ground. Simultaneously, the city of
trast, balance, color, and so on, has a greater chance of Wellington acts as figure to New Zealand. Figure-ground
communicating information than a map that has been is described further in section 1.4. Proximity allows us to
created in the absence of an artistic sensibility. view objects that are close together as a group, as illus-
The link between cartographic design and graphic trated by the cluster of coyote sightings in Figure 3E.
design is strong. Graphic design has been described as Similarity allows us to group objects that are similar in
“problem solving on a flat two-dimensional surface . . . size, shape, color, etc., as illustrated by the land cover
to communicate a specific message” (Arntson 2003, 2). regions in Figure 3F. Smallness, or Area, allows us to view
Both cartographic design and graphic design emphasize smaller areas as figures and larger areas as ground. This
the communication of information through graphical is illustrated in Figure 3G by small lakes in the context of
means—the primary difference being that graphic a larger study area. Symmetry allows us to view symmet-
design is mainly oriented toward advertisements and rical objects as whole figures that form around their cen-
packaging, as opposed to maps. While most maps are ters (as opposed to individual marks), as illustrated by
produced by cartographers, GIS specialists, and other the three point symbols in Figure 3H. These symbols are
geoscientists, graphic designers are responsible for a sig- symmetrical because each can be bisected into two
nificant proportion of maps produced for print and identical halves. This symmetry helps us to group the
online publication. individual components of each point symbol, such as the
six individual black diamonds on the left, into a unified
whole: a six-pointed star.
It is important for you to build a mental inventory of
designs and design possibilities from maps and other
graphical images. Borden Dent (1999) referred to this
mental inventory as an image pool, which can be built
by critically viewing as much art, graphic design, and
maps as possible. Many examples of well-designed maps
appear in this text. Use these maps to help build your
image pool.
1.1 The Design Process

FIGURE 2 An eye-movement study as an example of map
design research (From Jenks, G. F. (1973) “Visual integration The design process can be distilled into the following list
in thematic mapping: Fact or fiction?” International Yearbook of procedures.This list is an expanded version of step 4 of
of Cartography 13:27–35). the map communication model (Figure 1); it assumes
241
17:35:53.
Cartographic Design
A B C D
Tu
n ne
l
Closure Common Fate Continuity Figure-Ground

E F G H
Proximity Similarity Smallness/Area Symmetry
FIGURE 3 Cartographic examples of Gestalt principles.
that steps 1 through 3 of the map communication model map and its purpose, but the following is a gen-
have been completed, and that step 5 will be completed eral hierarchy for thematic maps (from most to
toward the end of the design process. Be aware that the least important):
procedures in this list are iterative, and need to be
repeated until the map has been completed. They will • Thematic symbols and type labels that are
sometimes need to be executed simultaneously or even directly related to the theme
out of the prescribed order. • Title, subtitle, and legend
• Base information such as boundaries, roads,
1. Determine how the map will be reproduced. place-names and so on
Reproduction considerations, such as the printing • Scale and north arrow
method to be used, will impact almost every • Data source and notes
aspect of the design process, and need to be • Frame and neat lines
resolved first.
2. Select a scale and map projection that is appropri- 6. Create one or more sketch maps. A sketch map,
ate for the map’s theme. also called a thumbnail sketch, is a rough, gener-
alized hand drawing that represents your devel-
3. Determine the most appropriate methods for
oping idea of what the final map will look like, as
data classification and symbolization.
illustrated in Figure 4. The sketch map should
4. Select which map elements to employ, and decide include all selected map elements, and should
how each will be implemented. You must also reflect the intellectual hierarchy established in
decide how to implement type. the previous procedure. Methods for graphically
5. Establish a ranking of symbols and map elements emphasizing and deemphasizing map elements
according to their relative importance. This are described in sections 1.3 and 1.4. You should
ranking is referred to as an intellectual hierarchy, experiment with various arrangements of map
or a “scale of concepts” (Monmonier 1993), and elements, striving for balance between them.
usually takes the form of a list. The intellectual Methods for establishing good balance are dis-
hierarchy often varies depending on the type of cussed in section 1.5.
17:35:53.
Cartographic Design
deemphasized in favor of base information such as

roads and the river. More important map elements such
as the title and legend have also been deemphasized,
whereas less important map elements such as the bar
scale and data source have been given far too much
visual weight. Try looking at this map while squinting,
and identify the map features that stand out. The bar
scale, data source, universities, river, and roads should
certainly not be the most noticeable features on a map
that is focusing on public schools. The map in Figure 5B
has a visual hierarchy that appropriately reflects the
intellectual hierarchy listed earlier. Thematic symbols
are visually dominant, as are the title and legend; they
are dominant because of their greater visual weight,
achieved through the manipulation of contrast
(described next). Base information (universities, river,
roads, etc.) is subdued, as are the data source and bar
FIGURE 4 A rough sketch map. The map resulting from this scale. An effective visual hierarchy results in a map that
sketch is illustrated in Color Plate 1. clearly reflects the relative importance of symbols and
map elements. A map with an appropriate visual hierar-
chy is easier to interpret, and is also more attractive.
7. Construct the map with your chosen software
application. Place, modify, and arrange map elements 1.3 Contrast
according to your sketch map. Print rough drafts that
will allow you to reevaluate and refine the evolving Contrast refers to visual differences between map fea-
map. Color Plate 1 illustrates the end result of this tures that allow us to distinguish one feature from
procedure, based on the sketch map of ethnolinguis- another. Contrast adds interest to a map by providing
tic families (Figure 4). graphical variety. It can be used to differentiate features,
or to imply their relative importance. Several techniques
1.2 Visual Hierarchy can be used to create contrast, including manipulation of
the visual variables: spacing, size, perspective height, ori-
Visual hierarchy refers to the graphical representation entation, shape, arrangement, and all aspects of color.
of the intellectual hierarchy described earlier, in which The map in Figure 6A lacks appropriate contrast in
symbols and map elements are ranked according to their four respects: type size, lightness and size of thematic
relative importance. When implementing the visual hier- symbols (circles), size of lines (line width), and differ-
archy, thematic symbols are graphically emphasized ence between the mapped area and the background.
while base information is deemphasized. Similarly, more Type on this map is insufficiently differentiated by
important map elements such as the title and legend are increments of 0.5 point. The title (10 pt.), legend head-
graphically emphasized, and less important elements ing (9.5 pt.), legend definitions (9 pt.), data source (8.5
such as the bar scale and data source are deemphasized. pt.), and bar scale type (8 pt.) are almost impossible
An effective visual hierarchy attracts the map user’s to differentiate by size. This lack of contrast con-
eyes to the most important aspects of the map first, and tributes to a monotonous design in which the relative
to less important aspects later. importance of map elements is unclear at best, and
Visual hierarchy is implemented by applying contrast misleading at worst. The data source in this map
to map features, as described in section 1.3. The visual appears almost as large as the title, even though it is
weight of map features refers to the relative amount of far less important. A lack of contrast between the
attention they attract, and can be manipulated to lightness of the circles, together with a lack of con-
emphasize or deemphasize features. The map in Figure trast in circle size, contributes to a dull design that
5A reflects an inverted (incorrect) visual hierarchy makes it difficult to differentiate between classes. The
based on the general intellectual hierarchy listed ear- lines in Figure 6A also lack contrast; the frame line,
lier. Thematic symbols (i.e., symbols representing ele- circle outlines, county boundaries, bar scale, and
mentary, junior high, and high schools) have been leader lines in the legend are all 1 point wide. Again,
243
17:35:53.
Cartographic Design
A Public Schools in Lawrence
70
Elementary
59 Junior
High
High
Source: SBC
Lawrence
40
Yellow Pages,
2008.
59
UNIVERSITY OF
KANSAS
10
UNIVERSITY
HASKELL
0 1 2
Miles
B
Public Schools in Lawrence
70
Elementary
59
Junior
High
High
40
Source: SBC Lawrence

Yellow Pages, 2008.
59
UNIVERSITY OF
KANSAS
10
FIGURE 5 (A) An inverted (incor-

UNIVERSITY
HASKELL
rect) visual hierarchy. (B) A correctly

10
applied visual hierarchy, appropri-
0 1 2
ately reflecting the intellectual hier-
Miles
archy.
17:35:53.
Cartographic Design
A Mobile Homes B Mobile Homes

California, 1990 California, 1990
Units per County
Units per County
54 - 5,180 54 - 5,180
6,275 - 14,917 6,275 - 14,917
20,926 - 31,843 20,926 - 31,843
42,982 - 74,561
42,982 - 74,561
Source: U.S. Census
Bureau, United States Source: U.S. Census Bureau,
Census, 1990. United States Census, 1990.
0 200 0 200
Miles Miles
FIGURE 6 (A) Insufficient contrast in type size, lightness and size of thematic symbols (circles), line width, and difference
between the mapped area and the background. (B) Sufficient contrast in all respects.
this lack of contrast contributes to a monotonous the map user. Map design research has failed to
design in which the relative importance of map fea- produce guidelines for figure-ground that are guar-
tures is unclear. Bar scales and leader lines are less anteed to work in every situation (MacEachren and
important than thematic symbols or base information Mistrick 1992), but we have found that the following
(county boundaries in this map), and should appear guidelines work well in most cases: When accentuat-
so. Finally, the map in Figure 6A lacks sufficient con- ing points or lines, the figure-ground relationship is
trast between the mapped area and the background, established by making the points or lines darker than
resulting in the mapped area appearing to blend in their surroundings. In Figure 7A, a lack of contrast
with the background. results in a situation in which no features advance
The map in Figure 6B exhibits sufficient contrast in easily as figures, and none recede as ground. In
each of the four respects just described. It is easier to Figure 7B, the base information has been lightened,
interpret, is more visually stimulating, and is more allowing the point symbols and trails to emerge as
attractive. Notice the significant difference the gray figures. Screening is a term that describes the lighten-
background makes—it enhances the mapped area and ing of graphics to reduce their visual weight, as illus-
makes it seem more important than the background. trated by the administrative boundaries in
This technique creates a special type of contrast called Figure 7B, which have been reduced from black to
figure-ground that has already been introduced, but gray (20 percent black).
deserves special attention. When accentuating an area, the figure-ground rela-
tionship is established through opposite means—by
1.4 Figure-Ground making the area lighter than its surroundings. The
extent of the study area in Figure 7C is ambiguous,
Figure-ground was described previously as a Gestalt and is only suggested by the “STUDY AREA” label.
principle of perceptual organization. It refers to In Figure 7D, the study area is strongly emphasized
methods of accentuating certain chosen objects over and clearly defined due to its relative lightness. The
others by making the chosen objects appear closer to figure-ground relationship accentuating Western
245
17:35:53.
Cartographic Design
A B
i l i l
Tra Tra
er er
ne ne
io io
P
P
Trai l Trai l
nd nd
Ov e r l a Ov e r l a
C D
S T U D Y S T U D Y
FIGURE 7 (A) Ambiguous figure-
A R E A A R E A ground relationship. (B) Darker points
and trails emerge as figures; lighter
administrative boundaries recede as
ground. (C) Ambiguous figure-ground
relationship. (D) Lighter study area
emerges as figure.
Europe in Figure 8A is similar, but is actually a spe- figures (although less strongly than in previous exam-
cial case, one in which land is the figure and water is ples) because they appear to stand in front of the gratic-
the ground: the classic land–water contrast problem. ule, which acts as ground—an effect known as interpo-
Making an area lighter than its surroundings effec- sition. Stylized effects such as the vignette (boundary
tively emphasizes the area, but is not always appro- shading) surrounding the British Isles in Figure 9B can
priate. For example, in situations when the mapped also enhance the figure-ground relationship, but should
area is dense with areal thematic symbols and color is be used with caution because they can become visually
limited to shades of gray (Figure 8B), it would be dominant.
inappropriate to apply a shade of gray to the sur-
rounding area. This is because the areal thematic sym- 1.5 Balance
bols provide enough contrast between the mapped
area and its surroundings, and because the application Balance refers to the organization of map elements and
of gray to the surrounding area might cause it to be empty space that results in visual harmony and equilib-
confused with the thematic symbols (which are also rium. The map elements in a well-designed map tend to
gray). Notice in Figure 8C that there are actually complement one another, whereas those in a poorly
three levels, or layers, of information represented. The designed map appear to compete for space, resulting in
circles act as figures in relation to the countries visual disharmony. Before attempting to achieve bal-
(which act as ground)—the circles are dominant, and ance, you need to identify the initial available space—
appear closer to the map user. The countries, although the area the map will occupy—as defined by the frame
subordinate to the circles, act as figures in relation to line, and as illustrated in Figure 10A.
the water (which acts as ground)—the countries Once the initial available space is defined, the place-
appear to be farther away than the circles, yet closer ment of larger map elements such as the mapped area
than the water. and inset can be considered (the inset is excluded from
Figure 9 represents two alternative methods of estab- this example). The mapped area should be as large as
lishing a figure-ground relationship that accentuates possible within the available space, while leaving ample
areas. In Figure 9A, the British Isles are established as room for the remaining map elements. The mapped area
17:35:53.
Cartographic Design
A B C
FIGURE 8 (A) Land–water contrast as a special case of the figure-ground relationship. (B) Situation in which the application of
gray to the surrounding area (water) would be inappropriate. (C) Three levels of information established through figure-ground
relationships.
should also be visually centered within the available scale, and north arrow—and should be placed next.
space if possible. The concept of visually centering Visually center the legend within a larger area of avail-
objects has been discussed but warrants repetition: able space (Figure 10C), reevaluate the remaining avail-
Visually centering objects is accomplished through visual able space, and then place the smaller map elements
approximation, not precise measurement; what matters such as the data source and bar scale, as illustrated in
most is that an object looks centered within the available Figure 10D. These smaller map elements should also be
space, as Africa does in Figure 10B. With the mapped visually centered within the available spaces. Notice in
area placed, the title (which is intermediate in size) can Figure 10D that available space remains, even after all
be positioned at top center (Figure 10B). Notice that the map elements have been placed. It is not necessary to fill
title appears to be visually centered (vertically) within all available spaces, but an effort should be made to
the space above the mapped area, as well as being hori- occupy the larger areas with well-placed map elements.
zontally centered within the frame line. The addition of each map element alters the preexist-
After the mapped area and title have been placed, ing balance and available space. Map elements will prob-
the evolving map needs to be reevaluated for available ably need to be rearranged several times to achieve good
space (Figure 10B); locations must be identified for the balance. Useful questions for the cartographer to ask are
remaining map elements, based on the guidelines for whether a map looks left-heavy, right-heavy, top-heavy,
each map element. Like the title, the legend is normally bottom-heavy, or whether certain areas appear cramped
intermediate in size—larger than the data source, bar or barren. Top-heavy designs are of particular concern
because they tend to make people uncomfortable;
humans are intrinsically aware of gravity, and tend to feel
more comfortable with objects that are closer to the
A B ground, or the bottom of a page (Arntson 2003). If the
answer to any of the previous questions is yes, then the
map elements should be rearranged with the goal of visu-
ally centering them within the available spaces. Con-
sideration should also be given to the different visual
weights of the individual map elements. Don’t place too
many “heavy” objects in the same area, but, rather, try to
balance heavier objects with lighter ones.
Instead of allowing map elements to crowd one
another, try to use them to balance one another. For
example, the bar scale in Color Plate 2A crowds the leg-
FIGURE 9 Figure-ground established using (A) a graticule end, competing for its space, whereas the bar scale in
and (B) a vignette. Color Plate 2B (along with the data source) helps to
247
17:35:53.
Cartographic Design
A B
Population Density
AVA I L A B L E
S PA C E
AVA I L A B L E
S PA C E
AVA I L A B L E
S PA C E
AVAILABLE
SPACE
C Population Density D Population Density
AVA I L A B L E AVA I L A B L E
S PA C E S PA C E
Persons per Persons per

Sq. Kilometer Sq. Kilometer
0 - 10 0 - 10
11 - 33 11 - 33
34 - 58 34 - 58
59 - 101 59 - 101
102 - 723 102 - 723
AVAILABLE AVAILABLE Source: ESRI Data &

0 2,000
Maps, 2008.
SPACE SPACE Kilometers
FIGURE 10 Establishment of balance by visually centering map elements within available spaces. (A) The
frame line defines the initial available space. (B) The mapped area and title are placed. (C) The legend is
placed. (D) The data source and bar scale are placed.
17:35:53.
Cartographic Design
counterbalance the legend. The map in Color Plate 2A is of crime to the potential homeowner. The crime
poorly balanced in many respects, but most of the prob- hot-spots are hybrid symbols; they are part thematic
lems are rooted in the fact that the map elements were symbol and part base information. They are similar
not visually centered within the available spaces. Color to thematic symbols because they are directly related
Plate 2B represents a well-balanced design in which all to the map’s theme—crime was used as an input to
map elements are in harmony and equilibrium. Map ele- the site suitability ranking. They will be used here as
ments were placed according to the same sequence a visual reference to areas of relatively high crime
described for Figure 10. Certain individuals are intrinsi- and will provide a frame of reference for the the-
cally better than others at judging balance, but experi- matic symbols, much like base information does.
ence can improve one’s skills.
2.1 Steps 1–3 of the Map Communication Model
2 CASE STUDY: REAL ESTATE SITE
We will base this case study on the map communication
SUITABILITY MAP
model described previously (Figure 1), with an empha-
sis on step 4: Design and Construct the Map. Step 1
In this section we will apply the concepts, rules, and requires that we consider what the real-world distribu-
guidelines presented in this chapter to a real-world map tion of the phenomenon might look like. We are fairly
design problem. The goal will be to create an efficient, well acquainted with the study area, and have general
attractive map that represents the relative suitabilities impressions of which areas are most and least suitable
of residential building parcels (thematic symbols), for residential property ownership. However, the spe-
together with appropriate base information. The source cific distribution and ranking of parcels requires the
materials for this map are derived from county-level design and execution of spatial analysis within the
GIS data, and are limited in extent to an area slightly realm of a GIS. Spatial analysis of this sort is beyond
smaller than an urban zip code region. These materials the scope of this text, but is relatively common. It incor-
are described below. porates various geoprocessing operations that are
1. Thematic symbols are building parcels (polygons) linked within the construct of a graphical modeling
within two neighborhoods that are: application.
Step 2 of the map communication model requires us
a. Zoned for single-family residences
to determine the purpose of the map and its intended
b. Not within 500 feet of a freeway or freeway audience. The map user is defined as a client of a resi-
onramp/offramp dential real estate agent. The purpose of the map will be
c. Not within 250 feet of an active railway to provide the client with a graphical representation of
2. Thematic symbols have been ranked for suitability the geographic distribution of parcels, which are ranked
according to the following criteria. Higher suitabil- according to the criteria described above. The map will
ity parcels are characterized by: focus attention on the most suitable parcels and will
include related base information, but will not provide
a. Relatively low crime density
specific information on the parcels or the homes that
b. Relatively high percentage of college-educated occupy them.
residents (Bachelor’s degree or higher) Step 3 of the map communication model involves the
c. Relatively high elevation (the region has a high collection of appropriate data. It is assumed that this
potential for flooding) step has been completed as part of the GIS analysis
These three criteria have been weighted equally process. We are now ready to proceed to step 4 of the
and represent the same level of importance in the map communication model, where we will describe in
analysis. Details of the suitability analysis model are detail each procedure in the design process.
not discussed here.
2.2 Step 4 of the Map Communication Model: Design
3. Base information includes:
and Construct the Map
a. Unranked parcels
b. Freeways and ramps Procedure 1 of the design process, as outlined in
c. Railways section 1.1, requires us to consider how the map will
be reproduced. The chosen reproduction method will
d. River
influence almost every aspect of the cartographic
e. Descriptive type labels design process. We can assume that the reproduction
4. Additional symbols—crime hot-spots—will also be methods will involve digitally assisted offset lithogra-
included due to the immediate and present concern phy, such as computer-to-plate or direct-to-press
249
17:35:53.
Cartographic Design
methods. These methods result in high-resolution, a direct bearing on map elements and how they should
high-quality printed images in both gray tones and be implemented. Under this condition, we have decided
full color, and place few restrictions on the design to include a frame line and/or a neat line, the mapped
process. Be aware that maps shown in this chapter area, an inset (a locator map), a title and subtitle, a leg-
have been printed using black and gray tones, and that end, a data source, and a bar scale. An indication of ori-
a full-color version of the final map appears in the entation (north arrow) might be required depending on
Color Plate section of the text. the orientation of the mapped area. Specific decisions
Procedure 2 of the design process involves the selec- related to map elements and typography will be made
tion of a scale and map projection that is appropriate beginning in section 2.3.
for the map’s theme. The scale of our map will be Procedure 5 in the design process requires us to
strictly dictated by two factors: our need to represent establish an intellectual hierarchy. We will use the
two neighborhoods and our need to do so within the hierarchy presented in section 1.1 because in our expe-
confines of roughly two-thirds of a printed page in this rience, it has proved to be effective for most simple the-
text. Experimentation has resulted in an optimal scale matic maps. This hierarchy is repeated below (from
of roughly 1:20,000. Because of the very small geo- most to least important).
graphic extent of this region (in comparison with the
extent of the entire Earth), our choice of map projec- • Thematic symbols and type labels that are directly
tion is virtually inconsequential. Distortions resulting related to the theme
from the map projection process will be insignificant. • Title, subtitle, and legend
For reference, the data used in the GIS analysis phase • Base information such as boundaries, roads, place-
of this project are stored in the State Plane coordinate names, and so on
system, incorporating the Lambert Conformal Conic • Scale and north arrow
projection. This projection correctly represents the • Data source and notes
shapes of small areas, but distorts the relative sizes of • Frame and neat lines
areas. In an area as small as two neighborhoods, it is In a subsequent procedure, we will establish the visual
assumed that this distortion will be of no consequence hierarchy by graphically expressing the intellectual
to the map user. hierarchy.
Procedure 3 in the design process dictates that we Procedure 6 in the design process involves the cre-
choose appropriate data classification and symboliza- ation of one or more sketch maps that can help an
tion methods. The classification method will be quan- evolving design take shape. The creation of sketch maps
tile. This approach will appropriately represent the can be especially useful when the map designer is not
quantitative “low-to-high” suitability ranking that was completely comfortable with the software application
incorporated in the GIS analysis, and will ensure that that will be used to create the final map. A map can be
the map has roughly equal numbers of parcels in each drawn by hand without any of the distractions or prob-
class. Data standardization will not be performed on the lems associated with computers and software applica-
raw suitability values; we intend instead to represent a tions. Sketch maps become less valuable when the map
simple ranking of raw suitability values. Because our designer is completely comfortable with the software.
thematic symbols are polygons that represent the varia- Under that condition, the designer can use the software
tion in magnitude of a single attribute, a choropleth- as a sketch pad of sorts, quickly experimenting with var-
style symbolization approach will be taken. Each ious design options. A hybrid approach can also be
suitable polygon will be symbolized with a color or gray taken, in which the designer hand-sketches on top of
tone that gets darker in relation to increased suitability. computer-generated rough drafts. We will create both
The map user should be able to scan the map and dif- hand-drawn and hybrid sketch maps.
ferentiate between various levels of suitability based Procedure 7 in the design process is where the final
solely on the polygon symbolization; specific magni- map will be created using a software application.
tudes will be defined by the legend. While graphic arts applications offer the greatest
Procedure 4 in the design process involves the selec- amount of control over graphic elements and type, we
tion of appropriate map elements, consideration of how estimate that most maps are currently produced using
they should be implemented, and decisions related to GIS applications. As a result, we have chosen to create
typography. While it is important to make preliminary this map using ESRI ArcMap, and we will subse-
decisions in regard to these topics, the map designer quently export the map to Adobe Illustrator for “fin-
typically needs to progress further along in the design ishing” and preparation for print reproduction.
process in order to make final decisions. For example, Procedure 7 also includes the creation of rough drafts
the process of creating a sketch map (Procedure 6) representing the evolving map design. Rough drafts
often presents problems and reveals solutions that have can act as a surrogate for hand-sketched maps or, as
17:35:53.
Cartographic Design
described previously, can be used in conjunction with various levels of suitability—choropleth-style area sym-
hand-sketching. Whichever approach is taken, sketch bols. Base information includes unranked parcels, free-
maps and rough drafts allow the designer to evaluate ways and ramps, railways, a river, street names, and
the current state of the map, and to explore alternative place-names.
approaches to the design problem. Thematic symbols will appear only in the neighbor-
hoods of interest; it is this component of the mapped
2.3 Selection and Implementation of Map Elements area that will be sized and positioned according to the
and Typography guidelines stated above. Base information will extend
beyond the neighborhoods of interest, and will act as a
At this point we return to Procedure 4 of the design geographic frame of reference for the thematic sym-
process in order to select appropriate map elements, and bols. As a consequence, the thematic symbols will exist
to consider how each element should be implemented. completely within the frame line (without touching it),
We will also make decisions related to typography. whereas the base information will extend as far as the
Justifications for including these map elements, together frame line, and will be cropped by it, as illustrated by
with implementation details, are provided below. the sketch map in Figure 11. In this case, the frame line
also acts as a neat line, cropping or limiting the extent
1. Frame line and neat line of the base information. We will continue to refer to this
2. Mapped area line as the frame line. In comparison with a floating
3. Inset mapped area, which is removed from its geographic
context, this cropped mapped area will leave relatively
4. Title and subtitle
little available space for the remaining map elements.
5. Legend Placement of subsequent map elements will require
6. Data source careful consideration, together with the use of masks to
7. Scale cover the underlying base information. The size and
position of the mapped area will likely need to be
8. Orientation
adjusted simultaneously with the placement of addi-
tional map elements.
2.4 Frame Line and Neat Line The intellectual hierarchy dictates that thematic sym-
bols appear to be most important when represented on
The first map element placed will be the frame line. this map—they should have the greatest visual weight
It is included because it can help focus the map user’s in the visual hierarchy. Accordingly, the parcels will be
attention on what is within it. The frame line also estab-
lishes the initial available space, within which all other
map elements will be placed. This frame line will also
act as a neat line, cropping the mapped area, as
described below. A solid thin black line (0.5 point) will
serve these purposes without attracting undue atten-
tion. The size (width and height) of the frame line is
restricted in part by the maximum printable width of
this text (7 inches).
2.5 Mapped Area
The mapped area is then placed within the frame line. It

will be sized as large as possible without cramping the
frame line, and will leave ample room for the remaining
map elements. It will be visually centered (both hori-
zontally and vertically) within the frame line. At this
point, a distinction needs to be made between the two
components of the mapped area: thematic symbols and
base information. As you recall, the mapped area
always consists of thematic symbols, and sometimes FIGURE 11 A sketch map representing thematic symbols
includes base information. In this case, thematic sym- centered within the frame line, and base information cropped
bols are the residential building parcels representing by it.
251
17:35:53.
Cartographic Design
dominantly symbolized by black outlines and gray tones accentuate areas. It was stated that in order to accen-
or by a part-spectral color scheme. Higher suitabilities tuate an area, the area needs to be lighter than its sur-
will correspond with darker symbols, lower suitabilites roundings. It was also stated that this approach is not
with lighter symbols. Base information will be subdued always appropriate, especially when the mapped area
through screening, providing the map user with a frame is “dense with areal thematic symbols,” as is the case
of reference without challenging the thematic symbols here. Because the building parcels are represented by
for attention. Before adding thematic symbols, we will gray tones or colors that get progressively darker, they
focus on the appropriate symbolization of base infor- naturally emerge as figure when superimposed on a
mation. It is often useful to symbolize base information lighter background.
first, as it helps to set the context for thematic symbols. At this point, it will be useful to evaluate the avail-
The various layers of base information will need to able space and consider where the remaining map ele-
be symbolized so that each component is easily identi- ments might be placed. Figure 13 represents the
fiable. We will accomplish this by applying contrast to mapped area with the thematic symbols visually cen-
each component—not necessarily to imply relative tered within the frame line. Available space appears
importance, but to differentiate one component from both above and below the thematic symbols, with
another. Figure 12A illustrates the unranked parcels, slightly more existing below. We will now identify
freeways and ramps, railways, and river (type labels will potential locations for the remaining map elements:
be added later) in the absence of contrast; the visual inset, title/subtitle, legend, data source, scale, and orien-
weight of these symbols is excessive, making it difficult tation. We will hand-sketch each map element on top of
to differentiate the various components. Figure 12B a computer-generated rough draft.
illustrates the base information after contrast has been
applied, primarily through the use of screening (light- 2.6 Inset
ening) and the manipulation of line thickness and line
style. Appropriate symbolization allows us to differen- The geographic extent of the mapped area is quite
tiate the individual components of base information, small—smaller than a single zip code region—and
and prepares the map for the addition of thematic sym- might not be recognized without the use of an inset (a
bols. locator map). This inset will represent the primary
Figure 13 illustrates the thematic symbols domi- mapped area at a much smaller scale within the con-
nantly superimposed on the base information. text of the city boundary and major highways/free-
Compare Figure 13 with the sketch map in ways. Because an inset can serve various purposes, it
Figure 11. In both cases, the thematic symbols are visu- does not appear in the intellectual hierarchy listed
ally centered within the frame line, and the base infor- previously. In this case, the inset is for reference pur-
mation is cropped by the frame line. Also notice in poses only, and will be treated much like base
Figure 13 that the crime hot-spots (irregular polygons information. Accordingly, it will seem to have an
with horizontal line patterns) have been superimposed intermediate level of importance in the visual hierar-
on the thematic symbols and base information. chy. The inset will contain a minimal amount of type
As described earlier, the crime hot-spots are hybrid such as the name of the city and “Area of Detail,” but
symbols: part thematic symbol and part base informa- a discussion of type characteristics will be withheld
tion. The final component of base information— until the title is placed. The inset will be placed in a
descriptive type labels—will be added toward the end relatively large portion of available space, as dictated
of the design process. primarily by the frame line and thematic symbols.
The figure-ground relationship has been estab- Examination of Figure 13 reveals that the largest por-
lished as follows: thematic symbols act as “figure,” tion of available space appears in the lower left. This
appear closer to the map user than the base informa- location might be appropriate for the inset, provided
tion, and are interpreted as being more important; that a mask is placed over the underlying base infor-
base information acts as “ground,” appears farther mation. Figure 14 shows the inset sketched on top of
from the map user, and is interpreted as being less a computer-generated rough draft (a hybrid sketch
important. The hybrid crime hot-spots are, technically map). The finished inset is shown in Figure 15A. It
speaking, closer to the map user because they appear contains just enough information to allow the map
to rest on top of the underlying symbols, but they user to locate the area of detail within the larger geo-
attract only slightly more attention than the base graphic region. Be aware that the inset and all subse-
information. In section 1.4, separate guidelines were quent map elements are shown at their actual sizes—
provided for implementing figure-ground contrast to they need to be small in order to fit within the
accentuate points and lines, and, alternatively, to confines of our map.
17:35:53.
Cartographic Design
FIGURE 12 (A) Base information

in the absence of contrast. (B) Base
information differentiated by means
of contrast.
253
17:35:53.
Cartographic Design
FIGURE 13 Thematic symbols

superimposed on base information;
inclusion of crime hot-spot symbols.
FIGURE 14 A hybrid sketch map

representing possible locations for
map elements and type.
17:35:53.
Cartographic Design
2.8 Legend
A B
City of Sacramento The legend will define the suitability ranking of
Ranking thematic symbols by means of shaded rectangles and
5
80 Least Suitable definitions ranging from “Least Suitable” to “Most
Mildly Suitable Suitable.” The rectangles will be connected in order to
Moderately Suitable emphasize that a gradation of suitabilities is being rep-
Area
Are a
of Highly Suitable resented (as opposed to unique qualitative categories).
50
Detail Most Suitable
The legend heading is a good place to provide informa-
Unranked Parcels Shown in White
99
tion such as the unit of measure, type of enumeration
5
Crime Hot Spot unit (parcels in this case), and data-collection date.
Because this information will be provided in the title
and subtitle, the legend heading will simply be
“Ranking.” It is assumed that the map user will be able
to identify and use the legend without the inclusion of
FIGURE 15 (A) The inset designed for use as a locator map.
the word “Legend” in the heading.
(B) The legend defining parcel suitability ranking and crime
While thematic symbols are virtually always defined
hot-spots.
in the legend, base information is either included or omit-
ted, depending on various factors. The thematic symbols
are represented as residential building parcels, and are
identified as such in the title. Because of this, the
2.7 Title and Subtitle unranked parcels used as base information could proba-
bly be omitted from the legend. We could assume that
The title will succinctly reflect the map’s theme as the map user will intuitively identify the white polygons
“Suitability of Residential Parcels.” The subtitle will as being unranked parcels because they look like parcels,
identify the geographic region and date. (It is assumed but lack thematic symbology. However, in order to pre-
that the map user will not otherwise recognize the vent any ambiguity, we have chosen to identify the
region.) Assuming that we mask the base information, unranked parcels in the legend with the label “Unranked
there appears to be sufficient space for the title and sub- Parcels Shown in White.” Freeways, railways, and the
title at top center—we will use this space accordingly river are assumed to be self-explanatory and will also be
(Figure 14). The title and subtitle will initially be placed omitted from the legend. Crime hot-spots—areas of rela-
at top center but, as with all map elements, can subse- tively high crime—are not self-explanatory, and will need
quently be moved if necessary. to be defined in the legend. Because the crime hot-spots
At this point, we are prompted to make our first are irregular in shape, they will be identified in the leg-
typographical decisions. We have selected a simple sans end by an irregular polygon. For consistency, all type
serif type family (Helvetica Neue) over fancier, ornate characteristics in the legend will be identical to those
families. The decision to use a sans serif family instead used in the title, subtitle, and inset, with the exception of
of a serifed family is somewhat arbitrary. The use of sans type size. The legend heading will be visibly smaller than
serif type families often results in cleaner, more mod- the subtitle, and the legend definitions will be smaller
ern-looking maps, but sans serif families have no proven than the legend heading.
advantage over serifed families. The type style for the Like the inset and title, the legend will be placed in a
title and subtitle will be normal (roman). Careful choice relatively large portion of available space, as dictated
of type size will avoid the need to use a bold style, and primarily by the frame line, thematic symbols, and inset
italics will be reserved for the identification of the pub- (Figure 14). A mask will be used to cover the underly-
lication in the data source and for hydrographic fea- ing base information. The finished legend is shown in
tures. The title will be the largest type on the map, Figure 15B.
followed by the subtitle and legend heading. To increase
readability, letter spacing and word spacing will be 2.9 Data Source
slightly increased.
Remembering that top-heavy designs tend to make The data source will indicate the three sources of the the-
people uncomfortable (section 1.5), we will attempt to matic data. The heading “Data Sources:” will be included
place the remaining map elements toward the bottom to avoid confusion with map authorship, and publication
of the page, in the available space below the mapped names will be italicized. Because the data source appears
area. We will also mask the underlying base information toward the bottom of the intellectual hierarchy, it will be
where necessary. visually deemphasized through the use of a very small
255
17:35:53.
Cartographic Design
type size: 5 points. It is assumed that the average reader ished bar scale is shown with the data source in Figure
of this text will be able to discern type of this size from a 16A.
distance of roughly 12 inches. Individual lines of type will
be horizontally centered (center justified) to establish a 2.11 Orientation
self-balancing block of type. While the most appropriate
location for the data source is directly below the legend, The inclusion of a north arrow must be carefully con-
limitations of available space dictate that the data source sidered. This map is oriented to geographic north, and is
be placed elsewhere, most likely in a portion of available not intended for navigational, surveying, or orienteer-
space that can accommodate the wide but “short” dimen- ing purposes. Thus a north arrow is not warranted under
sions of the data source. Examination of Figure 14 these conditions. However, because of the irregular ori-
reveals a wide, short portion of available space in the entation of streets and freeways (many are oriented to
extreme lower right. This space might also accommodate magnetic north, roughly 17° E), we have decided to
the bar scale—a map element with similar dimensions. include a subtle north arrow in order to clarify an
With the exception of type size and the use of italics for ambiguous orientation. The north arrow will include an
publications, type characteristics will be identical to those “N” for north and a five-pointed star representing
used in the title, subtitle, inset, and legend. The finished Polaris, the North Star. It will be placed in a tall, thin
data source is shown in Figure 16A, together with the bar portion of available space, as seen in the right-hand
scale. margin in Figure 14. The finished north arrow is shown
in Figure 16B.
2.10 Scale
2.12 Final Procedures
A bar scale will be included because it will allow the
map user to assess neighborhood characteristics, such as One of the final procedures in the design process, while
block lengths and proximity to crime hot-spots. The not officially presented in the list in section 1.1, is the
scale will begin with zero (no extension scale), and the placement of descriptive type labels and other type, such
maximum distance will be 0.5 mile—a value that is as explanatory notes. In this case, descriptive type labels
round and easy to work with. A maximum value of 1 will identify base information, primarily streets and
would also be suitable, but the resulting bar scale would place-names. These labels—also referred to as annota-
be prohibitively wide. Three intermediate tics and a cen- tion—are for reference purposes only, and should not
ter value of 0.25 will be included. Like the data source, attract undue attention. They will be designed and
the bar scale will be designed to be subtle—it should placed according to the typographical rules and guide-
not attract attention. A skeletal style incorporating a lines. Labels for linear features (roads) will be set in title
very thin line width (0.25 point) is more appropriate case; areas will be labeled in all uppercase with exagger-
than bulkier checkerboard styles. Again, like the data ated letter and word spacing; italic style will be applied
source, the bar scale needs to be placed in a portion of to the hydrographic feature (the river); and all descrip-
available space that can accommodate its wide but short tive type labels will be oriented to reflect the features
dimensions. Both the data source and the bar scale can they represent. Masks and halos will be used to cover
probably fit in the available space in the lower right the underlying mapped area. Explanatory notes will also
(Figure 14). Placement of the data source and bar scale be included in order to provide basic information
in this location will also leave a portion of available related to the suitability ranking. The relatively small
space just to the left. This space can be used for notes scale of this map prevents the inclusion of a large num-
related to the site suitability ranking criteria. The fin- ber of street labels—a larger scale version would allow
for more.
In the process of placing map elements in order from
largest to smallest, and visually centering each in appro-
A B priate portions of available space, the map should
N already reflect a general sense of balance. After all map
0 0.25 0.5 Mile elements and type have been placed, however, the over-
Data Sources: Census 2000 Summary File 3, U.S. Census Bureau;
all balance of the map needs to be reconsidered. Map
2006 Crime Report Database, City of Sacramento Police Department;
County of Sacramento GIS.
elements might need to be repositioned so that they
complement rather than crowd one another, and a sense
of equilibrium between map elements and empty space
needs to be attained. This can be achieved through the
FIGURE 16 (A) The data source and bar scale. (B) The north careful examination of rough drafts and thoughtful edit-
arrow. ing. A second set of eyes can also be helpful. At this
17:35:53.
Cartographic Design
stage in the design process, the map designer often loses can have on a map. In our opinion, the map in Color
a degree of objectivity, and can benefit from the Plate 3 is easier to interpret and significantly more
comments and opinions of others. attractive. Fortunately, and due to recent advances in
Figure 17 shows our map with all map elements print repoduction methods, color maps have become
placed, and with descriptive type labels and notes far more common in publications such as daily news-
added. Compare this map with the hybrid sketch map papers.
in Figure 14. The map in Figure 17 should look very Now that we have completed the seven procedures
similar to the sketch map, which acted as a template of the map design process, we return to step 5 of the
during the design process. Color Plate 3 represents map communication model. This step involves partici-
the map in color using a sequential scheme for the pation of the map’s intended audience (the client of the
thematic symbols, cyan for the river, and various residential real estate agent) in order to evaluate its
other splashes of color. A comparison of Color Plate 3 effectiveness. Feedback solicited from the map user can
with Figure 17 reveals the dramatic impact that color be incorporated into the final map design.
Suitability of Residential Parcels

Land Park and Curtis Park Neighborhoods, 2007
50
Bro
adw
ay
2nd
Ave
.
99
d.
24th St.
Blv
ide
r.
ers
D
rk
Riv
Pa
d
Lan
Franklin Blvd.
5 C.K.
M
HIG cCLATC
H SC
CUR
H
HOO Y
Blvd
L
SA
lvlvdd. .
TIS
CR
ortrt BB
RIV
AM
reeeppo
PAR
ER
EN
Fre
N
T
CURTIS
K V
O
LAN
D
PAR
K
IL
City of Sacramento
LAG
E (pro
5
PARK
80
Ranking
pose
Least Suitable
)d
Area
Are a Mildly Suitable Locational Criteria:
of 1. Zoned for single-family residence
Detail
50 Moderately Suitable
2. Not within 500 feet of freeway or ramp
Highly Suitable 3. Not within 250 feet of active railroad
Most Suitable 0 0.25 0.5 Mile
5 99 Ranking Criteria (equally weighted):
Unranked Parcels Shown in White 1. Lower crime density
2. Higher proportion of college graduates 2006 Crime Report Database, City of Sacramento Police Department;
Crime Hot Spot 3. Higher elevation (reduced flood risk) County of Sacramento GIS.
FIGURE 17 The final real estate site suitability map. See Color Plate 3 for the color version.
257
17:35:53.
Cartographic Design
Completion of this map required several iterations software application based on your sketch maps, and
of editing, fine-tuning, and critiquing based on rough print rough drafts that will allow you to reevaluate the
drafts. The map is a relatively efficient and attractive evolving map.
communicator of geographic information, but it repre- Visual hierarchy was described as being the graphi-
sents only one possible solution to the problem at hand. cal representation of the intellectual hierarchy.
As we stated earlier in this chapter, we believe that it is Contrast is applied to map features in order to manip-
important to build a foundation of cartographic design ulate their visual weights—the relative amount of
skills by following the specific rules and guidelines attention they attract. Contrast refers to the visual dif-
presented here, and then to consider alternative ferences between map features that allow us to tell one
approaches. Always be prepared to explain or defend feature from another, and that sometimes allow certain
your design decisions. map features to appear to be more or less important.
We learned that figure-ground is a special application
of contrast that allows certain map features to be
SUMMARY emphasized by appearing closer to the map user.
Balance was described as the organization of map ele-
In this chapter, cartographic design was introduced as a ments and empty space, resulting in a map that appears
holistic process, both mental and physical, that results in to be in a state of visual harmony and equilibrium.
the creation of efficient maps that clearly communicate Good balance can be obtained, in part, through the
geographic information. Cartographic design is strongly establishment, identification, and use of available
guided by the needs of the map user and the purpose of space. Map elements are placed in order from largest
the map. It is also guided by rules, guidelines, and con- to smallest, and are visually centered within appropri-
ventions, but there is often no optimal result of the ate areas of available space.
design process; the cartographer thus crafts an appro- Finally, the concepts, rules, and guidelines set forth
priate solution from among many possibilities. in this chapter were employed to design a thematic
Cartographic design has been guided by the results of map representing the suitability of residential build-
map design research, and is strongly influenced by ing parcels. This case study was organized according
graphic design and Gestalt principles of perceptual to the map communication model and the seven pro-
organization. The design process was distilled into seven cedures of the cartographic design process. It included
procedures, which are incorporated into the map com- consideration of how the map will be reproduced;
munication model: (1) determine how the map will be selection of an appropriate scale and map projection;
reproduced; (2) select an appropriate scale and projec- determination of appropriate data classification and
tion; (3) decide on data classification and symbolization symbolization methods; selection and appropriate
methods; (4) select appropriate map elements and implementation of map elements; establishment of an
determine how to implement them; (5) establish an intellectual hierarchy; creation of sketch maps; and
intellectual hierarchy that represents the relative creation of the map using software applications. The
importance of symbols and map elements; (6) create final map effectively communicated its theme by
rough sketch maps that represent the evolving map means of an appropriate visual hierarchy, contrast, fig-
design; and (7) construct the map in your chosen ure-ground, and balance.
FURTHER READING
Arntson, A. E. (2003) Graphic Design Basics. 4th ed. Fort Forrest, D. (1999a) “Developing rules for map design: A func-
Worth, TX: Harcourt Brace College Publishers. tional specification for a cartographic design expert system.”
An introduction to principles of graphic design.
A detailed description of how expert systems might be used to
Brewer, C. A. (2005) Designing Better Maps: A Guide for GIS
design maps.
Users: Redlands, CA: ESRI Press.
Krygier, J. and Wood, D. (2005) Making Maps: A Visual Guide
A practical guide to cartographic design and typography for GIS
to Map Design for GIS. New York: The Guilford Press.
users.
An informal guide to cartographic design for GIS users.
Dent, B. D. (1999) Cartography: Thematic Map Design. 5th ed.
Boston: McGraw-Hill. MacEachren, A. M. (1995) How Maps Work: Representation,
Visualization, and Design. New York: Guilford.
design and typography.
17:35:53.
Cartographic Design
Includes comprehensive discussions of map design research and Robinson, A. H., Morrison, J. L., Muehrcke, P. C., Kimerling,
Gestalt principles of perceptual organization. A. J., and Guptill, S. C. (1995) Elements of Cartography. 6th
Monmonier, M. S. (1993) Mapping It Out: Expository ed. New York: Wiley.
Cartography for the Humanities and Social Sciences. Chicago: A classic cartography textbook with chapters on cartographic
University of Chicago Press. design and typography.
Introduces practical aspects of cartographic design for noncartog- Tufte, E. R. (1990) Envisioning Information. Cheshire, CT:
raphers. Graphics Press.
Montello, D. R. (2002) “Cognitive map-design research in the Focuses on the visual display of information, with many cartographic
twentieth century.” Cartography and Geographic Information examples.
Systems 29, no. 3:283–304. Wood, C. H., and Keller, C. P., eds. (1996) Cartographic Design:
An overview of the origins, rise, fall, and rebirth of cognitive map Theoretical and Practical Perspectives. Chichester, England:
design research. Wiley.
Robinson, A. H. (1952) The Look of Maps. Madison, WI: Includes essays focusing on the modern role and character of car-
University of Wisconsin Press. tographic design.
The book that gave birth to map design research.
GLOSSARY
balance: the organization of map elements and empty space map design research: research focusing on which mapping
that results in visual harmony and equilibrium. techniques are most effective and why.
cartographic design: a partly mental, partly physical process rough drafts: preliminary printouts of a map that are used to
in which maps are conceived and created. evaluate an evolving design.
contrast: the visual differences between map features that screening: the lightening of graphics to reduce visual weight;
allow for differentiation of features, and the implication of this is accomplished by reducing the amount of ink or toner
relative importance. applied to the print medium.
figure-ground: methods of accentuating one object more sketch map: a rough, generalized hand drawing that repre-
than another, based on the perception that one object stands sents a developing idea of what a final map will look like.
in front of another, and appears to be closer to the map user. visual hierarchy: the graphical representation of an intellec-
Gestalt principles: descriptions of the manner in which tual hierarchy.
humans see the individual components of a graphical image, visual weight: the relative amount of attention that a map
and organize the components into an integrated whole. feature attracts.
intellectual hierarchy: a ranking of symbols and map
elements according to their relative importance.
REFERENCES
Arntson, A. E. (2003) Graphic Design Basics (4th ed.). Fort Monmonier, M. (1993) Mapping It Out: Expository
Worth, TX: Harcourt Brace College Publishers. Cartography for the Humanities and Social Sciences.
Dent, B. D. (1999) Cartography: Thematic Map Design (5th Chicago: University of Chicago Press.
ed.). Boston: McGraw-Hill. Montello, D. R. (2002) “Cognitive map-design research in the
Jenks, G. F. (1973) “Visual integration in thematic mapping: twentieth century: Theoretical and empirical approaches.”
Fact or fiction?” International Yearbook of Cartography Cartography and Geographic Information Science 29, no.
13:27–35. 3:283–304.
MacEachren, A. M. (1995) How Maps Work: Representation, Robinson, A. H. (1952) The Look of Maps. Madison:
Visualization, and Design. New York: Guilford. University of Wisconsin Press.
MacEachren, A. M., and Mistrick, T. A. (1992) “The role of Tufte, E. R. (1990) Envisioning Information. Cheshire, CT:
brightness differences in figure-ground: Is darker figure?” Graphics Press.
The Cartographic Journal 29, no. 2:91–100.
259
17:35:53.
17:35:53.
Map Reproduction
261
11:53:31.
Map Reproduction
OVERVIEW
This chapter presents topics related to print and nonprint produced for each base color on a map, and proofs are
map reproduction, and introduces map dissemination created for use in map editing. Portable document formats
with regard to nonprint reproduction. In section 2, we are preferred for the delivery of digital maps to the service
encourage you to plan ahead for map reproduction to bureau. Proofing methods range from low-end techniques
avoid problems associated with specific map reproduction such as on-screen display to high-end techniques such as
methods. In section 3, we introduce map editing—a criti- separation-based proofs, which are created from color
cal activity that can save you time and money during map separations. The offset lithographic printing press trans-
reproduction. fers images from the printing plate to the print medium via
Section 4 describes the role of raster image processing a series of rolling cylinders. Registration refers to
in print reproduction. The process of printing a digital the proper alignment of colors produced by an offset
map is distilled to four steps: The map is created in press; trapping is used to correct for improper registra-
application software, converted into page description tion. Computer-to-plate and direct-to-press printing
data, interpreted by a raster image processor, and methods streamline the offset printing process by
printed on a printing device. Section 5 describes halftone eliminating the film negative. These methods can be faster
and stochastic screening, methods to create tints from and less expensive than traditional separation-based offset
base colors. lithography, but they eliminate the possibility of creating
We introduce aspects of color printing in section 6. separation-based proofs.
Process colors are mixed on the page by combining cyan, In section 8, we describe computer-based nonprint
magenta, yellow, and black (CMYK colors). Spot colors reproduction and dissemination. Nonprint reproduction
are premixed and thus allow for exact color matches involves file duplication performed by computers.
based on printed color swatches. High-fidelity process Computer-based maps can be disseminated on various
colors allow for increased color variety, vibrancy, and media and via local area networks, although it is diffi-
color-matching capabilities. Continuous tone printing cult to anticipate how a map will appear on different
allows for smooth color transitions without the use of systems. The Internet and World Wide Web offer a wide
screening. Color management systems identify and range of map dissemination options. FTP can be used to
correct for differences in color that are introduced by distribute digital maps in a rudimentary fashion over the
electronic devices. Internet, and HTTP and HTML can be employed for
Offset lithography, identified in section 7 as the domi- Web-based dissemination. Interactive map animations
nant method for high-volume print reproduction, is char- can be created in Flash format and can be viewed in
acterized by excellent print quality, high speed, and vol- Web browsers, or used as stand-alone applications. The
ume discounts. Prepress is described as a phase of Java and JavaScript programming languages allow for
high-volume reproduction centered on the service bureau, the creation of custom mapping applications, and
in which preparations are made for offset printing. Film sophisticated map dissemination environments can be
negatives (color separations) and printing plates are created through the use of map servers.
11:53:31.
Map Reproduction
1 REPRODUCTION VERSUS DISSEMINATION answer to this question influences almost every

aspect of the reproduction process.
Map reproduction involves the printing of a map 2. What is your budget? Different reproduction meth-
(print reproduction), or the electronic duplication of a ods vary greatly in cost.
map in digital form (nonprint reproduction). Map 3. When is your deadline? Make sure to allow enough
dissemination refers to the distribution of reproduced time for reproduction, considering the speed of the
maps in physical or electronic form. The term dis- method you have chosen. Consider that problems
semination is also used to describe the combination of often arise that delay the reproduction of a map.
electronic duplication and electronic distribution, in 4. Will the map be printed, or displayed on a televi-
which maps are duplicated and distributed simultane- sion or computer screen? Significant differences
ously. For example, when a map is downloaded from between print and display methods require you to
the Internet, it is reproduced and disseminated at the consider aspects such as size, resolution, and the
same time, whereas when maps are printed, reproduc- color model to be used.
tion and dissemination are distinctly different processes
occurring at different times. Emphasis is given to print 5. If the map is to be printed, what material will be
reproduction in this chapter because of the breadth and used? Paper? Cloth? Plastic? Different materials
depth of the topic; nonprint reproduction is simple in and different grades of material absorb ink and
comparison. Nonprint reproduction and electronic dis- toner differently.
semination are discussed toward the end of the chapter 6. If the map is to be displayed, what will the file
(section 8). size(s) be? If the map is to be disseminated via the
Internet, the user’s connection speed and computer
processing power need to be considered.
2 PLANNING AHEAD 7. Will the map be limited to black, white, and gray
tones, or will it be color? How many colors will be
Reproduction methods can be quite specific—each pro- used? Printers and display devices have strict
vides certain benefits but also imposes strict limitations. limitations regarding color, and the cost of print
The reproduction method you choose will influence reproduction generally increases when more col-
almost every aspect of the cartographic design process. ors are used.
If you design a map without a specific reproduction 8. What size will the map be? Printers and display
method in mind, you could very well end up with a map devices have strict limitations on size.
that cannot be practically or economically reproduced.
9. How many copies will be required? Certain repro-
For example, if you design a color map using the RGB
duction methods offer a significant decrease in the
color model and then attempt to print to a CMYK
cost per unit as the number of copies increases;
device, you will encounter problems with color accuracy
other methods don’t.
that will take time (and probably money) to resolve.
Another example involves the reproduction of folded 10. Will the map be folded? What will the fold
maps. Many large-format maps are designed to be pattern be? Folding mechanisms are limited to
folded in specific patterns so that, for example, the title certain patterns.
and cover art appear on the front and the legend 11. What level of print or display quality is acceptable?
appears on the back when the map is folded. If you Different reproduction methods provide various
design a map with a special fold pattern in mind without levels of quality.
ensuring that an appropriate folding mechanism is avail- 12. Will you copyright the map? Will the map infringe
able, you will be forced to redesign your map. If you hire on an existing copyright?
a second party to reproduce a map, choose carefully, and
make sure to establish a good relationship with this
party before making major design decisions. 3 MAP EDITING
The following is a list of questions that need to be
answered in the early stages of the design process. If you
A map must be carefully edited before it is reproduced
take the time to answer these sorts of questions care-
(Figure 1). Map editing is the critical evaluation and
fully, the map reproduction process will be less likely to
correction of every aspect of a map; it begins the first
present problems and roadblocks that will prevent you
time the cartographer views the map in its early stages,
from successfully completing a mapping project.
and culminates with a final edit that occurs just before
1. Who is the intended audience, and what is the reproduction begins. Editing could easily be treated as
purpose of the map? As in the design process, the an aspect of cartographic design, but it is presented in
263
11:53:31.
Map Reproduction
4 RASTER IMAGE PROCESSING FOR

PRINT REPRODUCTION
Virtually all modern print reproduction methods

involve raster image processing: the conversion of a dig-
ital map into a raster image that can be processed
directly by a raster-based printing device.* The resulting
raster image is representative of the raster data model
(a general category of computer data consisting of
rows and columns of square pixels), as illustrated in
Figure 2A. In contrast to raster is the vector data model,
which is composed of precisely defined points, together
with the lines that connect them. The vector data model
closely resembles what is drawn by hand, and allows the
representation of discrete points, lines, and polygons, as
illustrated in Figure 2B. High-quality type is also repre-
sented using the vector data model. The majority of the-
matic maps are created using the vector data model
because of its ability to represent clearly defined fea-
FIGURE 1 A Forest Visitor Map in the process of being tures such as wells, roads, administrative boundaries,
edited. Carefully executed corrections, additions, and dele- and so on. The raster data model is commonly used
tions save time and money during reproduction. when representing continuous surfaces such as terrain,
or when representing the results of raster-based GIS
this chapter because of its critical role in map reproduc- analysis or satellite image classification. Many maps
tion. Because reproduction methods consume time and incorporate both raster and vector data models.
money, you simply can’t afford to reproduce a map that Regardless of the data models used in the creation of a
contains errors. The following is a list of general ques- map, the map is normally converted into a raster image
tions that should be addressed when editing: for printing.
1. Map design: Does the design appropriately serve
the map user based on the map’s intended use? 4.1 Printing the Digital Map
Does it communicate the map’s information in the
most efficient manner, simply and clearly? Several steps are involved in printing a digital map, as
illustrated in Figure 3. First, the cartographer creates the
2. Completeness: Are any features, map elements, or map in digital form using application software.A descrip-
type labels missing? tion of the wide range of application software capable of
3. Accuracy: Are features, map elements, and type
labels correctly placed? Are words spelled cor-
*Certain devices, such as pen plotters, do not require that data be
rectly? Are numeric values correct?
converted into a raster image for printing, but these devices are less
Overfamiliarity and fatigue are problems faced by car- widely used today.
tographers who edit their own maps. The cartographer
spends so much time working on a map that he or she
loses a certain degree of objectivity, and often misses A B
Raster Vector
errors that are obvious to others. One solution to this
problem is to have someone else (or a group of people) Points
edit the map. Other tips for editing include the
Pixel
following:
Lines
1. Edit with “fresh eyes,” at the beginning of a work
session.
Polygons
2. Edit large maps in sections.
3. View maps upside down or sideways—this forces
you to see maps in new ways, possibly illuminating
Park Type
errors.
4. If possible, edit several days after completing a map. FIGURE 2 (A) Raster and (B) vector data models.
11:53:31.
Map Reproduction
Application Software
Produces digital map file
(vector and/or raster data models)
Printer Driver
Converts digital map file into page description data
(in a particular page description language)
Raster Image Processor (RIP)

Interprets page description data and
produces raster image FIGURE 4 An EFI Fiery® print controller and raster image
processor (RIP). Photography: Stan Musilek. Used by permis-
sion © Electronics for Imaging, Inc. 2002.
Printing Device
Processes raster image and prints the map cations that operate inside computers, such as ESRI’s
ArcPress. Many less expensive printing devices are
incompatible with PDLs and don’t possess a RIP. These
devices depend on RIP software inside the computer to
FIGURE 3 General steps involved in printing a digital map.
create the raster image, which is subsequently sent to
the printer.
producing maps is beyond the scope of this text, but gen-
The RIP produces raster images of a specific reso-
eral categories include graphic design, GIS, and remote
lution that normally coincides with the maximum res-
sensing applications; examples include FreeHand,
olution supported by the printing device that will be
ArcGIS, and Imagine, respectively. The digital map file
used. Resolution refers to the number of pixels per
produced by the application software is converted via
inch, or dots per inch (dpi), that a device can print.
the printer driver into page description data that consist
After the RIP has created the raster image, the image
of a set of printing instructions, describing every graphi-
is processed by the printing device, which in turn cre-
cal and textual component of the map. The page descrip-
ates the printed map.
tion data are written in a page description language
(PDL) such as Adobe’s PostScript, or Hewlett Packard’s
Printer Control Language (PCL). PDLs are device inde- 5 SCREENING FOR PRINT REPRODUCTION
pendent to a certain extent, allowing the user to print
from, or to, almost any device equipped to interpret a Screening is a technique used in most print reproduc-
particular PDL. PostScript has proved to be more device tion methods in which colors are made to appear lighter
independent and better suited to high-end printing than by reducing the amount of ink or toner* applied to the
other PDLs, making it the de jure standard page descrip- print medium (e.g., paper). Screening is also referred to
tion language; PostScript is the only PDL to be recog- as dithering, but the term screening is more commonly
nized by the International Organization for used in the printing industry. Screening is used to create
Standardization (ISO) (Adobe Systems Incorporated tints of a particular base color, and to represent contin-
1997). uous tone surfaces. A tint is a lighter version of a base
After the page description data have been created by color: Gray is a tint of the base color black, and pink is
the printer driver, they are interpreted by a raster image a tint of the base color red. A continuous tone surface is
processor (RIP). A RIP is software, or a combination of composed of tints that continuously vary in lightness
software and hardware, that interprets page description within a given area, as illustrated by the terrain beneath
data and converts them into a raster image that can be the “REGIONAL PARK” label in Figure 5.
processed directly by a printing device. For example, a A standard (monochromatic) laser printer is only
PostScript-compatible RIP interprets PostScript page capable of printing black because it uses black toner.
description data, and converts them into a raster image Through the application of screening, however, the
for printing. RIPs exist inside printing devices such
as laser printers, inside dedicated computing devices * Toner is powder, normally black, that is used to form images in laser
called print controllers (Figure 4), or as software appli- printers and copy machines.
265
11:53:31.
Map Reproduction
A B
FIGURE 6 (A) Halftone screening, produced by equally

FIGURE 5 Black is the only color used in this map; the illu-
spaced dots of variable size. (B) Stochastic screening, pro-
sion of gray tones is produced through screening.
duced by pseudorandomly spaced dots of uniform size.
same laser printer is capable of producing what appear

to be gray tones even though black is the only color A B
used (Figure 5). Screening can produce what appear to
be lighter versions of any base color, not just black. In Halftone dots Halftone dots
the screening process, ink or toner of a particular
color, say, dark green, is applied to the print medium
in a pattern of individual dots, allowing the color of
the underlying print medium (normally white) to show
through. If the pattern is fine enough, the human eye
is capable of combining the color of the individual
dark green dots with the white of the underlying print
medium. This allows the map user to perceive light
green instead of dark green; the pattern itself typically
goes unnoticed. Most color printing devices work on
this principle; each color that the device can produce
can be lightened through screening, allowing for the
creation of far more colors than the number of inks or
toners would imply.
5.1 Halftone and Stochastic Screening

A single halftone dot A single halftone dot
Several methods of screening have been developed, but composed of relatively composed of relatively
small pixels large pixels
two have emerged as the most widely used: halftone and
stochastic.* Halftone screening is used in almost all
print reproduction methods, with the exception of ink- FIGURE 7 Halftone dots composed of pixels: (A) output
jet printing, which normally employs stochastic screen- from a higher resolution device; (B) output from a lower
ing. Halftone screening involves the application of ink resolution device.
or toner in a pattern of equally spaced dots of variable
size, as illustrated in Figure 6A. The size of each dot
determines the degree of lightness that is achieved; the (Figure 7). As a result, halftone dots are always coarser
smaller the dot, the lighter the result, because more of than the maximum resolution of the printing device.
the underlying print medium will show through. Smaller pixels (compare Figures 7A and 7B) allow for
Halftone dots are composed of pixels produced by a the creation of finer, better defined halftone dots, and
RIP, with the pixel size normally corresponding with the allow a larger number of tints to be derived from a base
maximum resolution of the printing device to be used color.†
* Certain screening methods represent a combination of halftone and †The number of tints that can be derived from a base color can be
stochastic approaches (Hyatt 2000). calculated using this formula: (printer resolution/screen frequency)2+1.
11:53:31.
Map Reproduction
sity of dots within a given area determines the degree of

10% 30% 50% 70% 90%
lightness that is achieved; dots spaced farther apart pro-
duce lighter results. Stochastic screening is also referred
to as frequency modulation (FM) screening because
variations of lightness are achieved through the alter-
ation of the spacing, or frequency, of each dot. The term
A stochastic refers to the pseudorandom approach used to
place dots within a given area.
Halftone screening has been practiced and refined
for about 200 years, whereas stochastic screening is a rel-
atively new method. While the halftone method is well
Circle Line defined, reliable, and relatively forgiving, the stochastic
method is still being refined, and is relatively unforgiv-
ing of inaccuracies in the reproduction process. Very
B small dots are difficult to print correctly, requiring that
the entire reproduction process be tightly controlled.
The stochastic method, however, offers several improve-
ments over halftone screening. For example, it can result
in cleaner images with greater contrast because individ-
1 Inch
1 Inch
20 LPI 50 LPI ual dots can be placed closer together, and because there
is less overprinting of inks when colors are mixed on the
page (Romano 1997). It can also eliminate the possibil-
C ity of undesirable moiré patterns (described later) that
are inherent in halftone screening. Although stochastic
screening has been predicted to eclipse halftone screen-
ing, this has not yet occurred—both methods are still
widely used.
45 90 5.2 Halftone Screening Parameters
D Halftone screening is controlled by four parameters:

tint percentage, cell type, screen frequency, and screen
FIGURE 8 Halftone screening parameters: (A) tint angle. The tint percentage controls the degree to which
percentage; (B) cell type; (C) screen frequency, measured in the appearance of an ink or toner is lightened, as illus-
lines per inch (lpi); (D) screen angle, measured in degrees, trated in Figure 8A. Tint is specified as a percentage,
counterclockwise from horizontal. with lower values resulting in lighter colors. The cell
type refers to the shape of each individual mark of the
halftone pattern; the circle (dot) is the most widely used
Halftone screening is sometimes referred to as
shape, but the line and other shapes are also used
amplitude modulation (AM) screening because varia-
(Figure 8B).
tions of lightness are achieved through the alteration of
Screen frequency, or screen ruling, refers to the spac-
the size, or amplitude, of each dot. Halftone dots
ing of halftone dots within a given area, or, more specif-
increase in size until a 50 percent tint is surpassed, at
ically, the spacing of the lines that the dots are arranged
which point all color is removed from the dots and is
in (Figure 8C). The screen frequency parameter con-
applied to the area surrounding the dots, essentially cre-
trols how coarse or fine the halftone pattern is. Lower
ating white dots on a color background (70 percent and
frequencies produce a coarse halftone pattern in which
90 percent in Figure 8A). As tints greater than 50 per-
individual dots are clearly visible, which is typically
cent are specified, the background area expands, even-
employed only when a lower resolution printing device
tually filling the areas formerly occupied by dots.
is used. Higher frequencies produce a fine halftone pat-
Stochastic screening involves the application of ink or
tern in which it is more difficult to discern individual
toner in a pattern of very small, pseudorandomly* spaced
dots. This pattern is easier for the human eye to inter-
dots of uniform size, as illustrated in Figure 6B. The den-
pret as representing a solid, but lighter, color. Screen
frequency is measured in lines per inch (lpi). Lpi is
*Computing devices are incapable of generating truly random related to, but should not be confused with, dpi (dots
numbers. per inch). Dpi refers to the maximum resolution of a
267
11:53:31.
Map Reproduction
printing device, independent of the lpi setting of a associated with halftone screening do not apply to
halftone screen. Halftones with higher lpi settings con- stochastic screens. The cell type is always a very small
sist of smaller dots that are most accurately produced dot, usually representing the smallest pixel size a par-
by high-resolution printing devices. ticular printing device can produce. Because these
The screen angle parameter controls the angle at dots are placed pseudorandomly within a given area,
which the lines of halftone dots are oriented, as illus- screen frequency and screen angle become meaning-
trated in Figure 8D. Screen angle becomes important less. The irregular spacing of dots eliminates the
when multiple halftone patterns are printed in the same potential problem of moiré patterns inherent in
area. For example, imagine one halftone pattern that halftone screening.
represents a 30 percent tint of black, and another that
represents a 30 percent tint of cyan. To mix the two
6 ASPECTS OF COLOR PRINTING
colors, both halftone patterns need to be printed in
the same area, allowing black and cyan to “mix” on the
page. If both screen angles are identical (and if all other 6.1 Process Colors
halftone parameters are equal), the black halftone dots
will cover and block out the cyan dots. To prevent this, In contrast with CRTs, which generate color through
the screen angle of the black halftone would need to be the addition of red, green, and blue light, most printing
offset from the cyan halftone, preventing the direct devices generate color using inks and toners based on
overlapping of dots, as illustrated in Color Plate 1. The the subtractive primary colors: cyan (C), magenta (M),
concept of mixing colors on the page is described fur- and yellow (Y). In theory, a mixture of pure C, M, and Y
ther in section 6.1. Screen angles need to be precisely (without screening) will result in pure black. In reality,
specified in order to prevent the creation of a moiré this mixture produces a “muddy” dark brown that lacks
pattern (Figure 9), which is an unwanted print artifact crisp detail, and gray tones produced by mixing C, M,
resulting from the interplay of dots in overlying and Y are not pure. As a result, black (K) is normally
halftones (Adams and Dolin 2002). Most software used in conjunction with C, M, and Y, creating CMYK,
applications and printing devices have been pro- or the process colors. Most color printing devices mix
grammed to use screen angles that prevent the creation the process colors on the page by applying them, in
of moiré patterns. sequence, to the same area. Inks and toners based on
the process colors are semiopaque, or translucent,
allowing them to combine on the page; new colors are
5.3 Stochastic Screening Parameters created where the process colors overlap. This, together
with screening techniques, allows for the creation of a
Stochastic screening is typically controlled by one wide variety of colors and is referred to as four-color
parameter only: tint percentage. As with halftone process printing. When process colors are mixed on the
screening, the tint percentage controls the degree to page, tints of each base color are represented by
which the appearance of an ink or toner is lightened. halftone patterns, each with a unique screen angle to
Tint is specified in percentage, with lower values avoid moiré patterns. This results in a special pattern of
resulting in lighter colors. The additional parameters inks called a rosette (Color Plate 2). The human eye
interprets fine rosettes (i.e., ones that result from
halftones with high screen frequencies) as representing
a solid color. Four-color process printing is an efficient
method of producing many colors from four base
colors. However, the actual number of colors that are
possible is fewer than the number possible using the
additive primary colors (RGB). Specifically, vibrant
colors such as orange, red, green, and blue are difficult
to reproduce using the process colors (Viggiano and
Hoagland 1998).
6.2 Spot Colors
An alternative to mixing process colors on the page is to

use solid colors, or spot colors. Spot colors take the form
FIGURE 9 Moiré pattern produced from the incorrect spec- of opaque inks that are premixed before they reach the
ification of halftone screen angles. printing device. As with process colors, tints can be
11:53:31.
Map Reproduction
but include two or three additional colors. This allows

for dramatic increases in the variety and vibrancy of
colors, and also allows for accuracy in color matching
that rivals the Pantone Matching System.
Various approaches to high-fidelity color have been
taken. Early attempts included the incorporation of red,
green, and blue inks with the process colors. This
approach is still practiced but requires a total of seven
colors, and has been largely eclipsed by six-color
approaches, which have been developed by a wide vari-
ety of companies including Epson, Roland, Hewlett
Packard, Lexmark, and Pantone. Most of these systems
incorporate black; modified versions of C, M, and Y; and
two additional colors, usually orange and green,
although some systems include light cyan and light
magenta. Six-color systems produce colors with more
variety, vibrancy, and contrast than traditional process-
color systems, but are usually more expensive.
Pantone has emerged as the de facto standard
in high-fidelity color with Hexachrome, a six-color
system that incorporates black; more saturated
versions of C, M, and Y; and orange and green. The
Hexachrome system is capable of accurately reproduc-
FIGURE 10 On-screen selector for Pantone spot colors ing over 90 percent of the spot colors in the Pantone
within Macromedia FreeHand (reproduced in gray tones). Matching System, as opposed to only about 40 percent
Notice the advisory regarding accuracy of on-screen colors. in the traditional process-color systems (Pantone
Used by permission from Macromedia, Inc. Incorporated 2002). Special software is required to
work with Hexachrome, but it is either incorporated
into existing software or can be accessed through
created from spot colors through screening. Exact color the use of plug-ins—software modules that work in
matches are easier to achieve using spot colors because conjunction with software applications.
they do not rely on the printing device for mixing—a sit-
uation that often results in color inconsistencies from 6.4 Continuous Tone Color Printing
device to device. The Pantone Matching System (PMS) is
the most universally accepted spot-color system, but Some printing devices are capable of producing
TOYO and other systems are also popular. Spot colors continuous tone output: full-color prints that are
can be selected on-screen, as illustrated in Figure 10, or achieved without the use of screening techniques. Tints
(more appropriately) from printed color swatches. Spot are created through subtle variations in the volume or
inks are mixed according to specific formulas, such as density of ink or toner applied to the page. The term
those found in the Pantone Color Formula Guide contone refers either to true continuous tone printing
(Adams and Dolin 2002). A disadvantage of using spot or to a set of hybrid techniques that incorporate a
colors is the cost of reproduction when a map has many limited set of continuous tone methods with screening,
colors. Unlike the process colors, which can produce a producing results that approach true continuous tone
wide variety of colors through mixing, spot colors output. Until recently, contone techniques were best
require separate inks to be used for each color, increas- suited to printing images of continuous tone such as
ing the cost of reproduction. photographs, as opposed to maps, which are composed
primarily of discontinuous tones and fine detail;
6.3 High-Fidelity Process Colors their use in mapmaking was limited primarily to the
creation of intermediate-quality proofs, as described in
Newer methods of color printing have been developed section 7.3. Recent improvements in the resolution and
in response to the disadvantages of process and spot “apparent resolution” of continuous tone devices, how-
colors, specifically the limited variety and vibrancy of ever, have made contone techniques more suitable for
process colors and the expense of spot colors when they cartographic applications—especially those involving
are used in large numbers. High-fidelity process colors continuous tone surfaces, such as terrain. But the tech-
are based on the traditional process colors (CMYK), nology remains relatively expensive.
269
11:53:31.
Map Reproduction
6.5 Color Management Systems Solution, which facilitates the color management process
at the operating system level (Adams and Dolin 2002).
A common problem confronting cartographers is Apple’s ColorSync (Macintosh) and Microsoft’s ICM2
matching the colors on a graphic display with those (Windows) are examples of ICC Color Management
produced by a printing device. Without deliberate inter- Solutions.
vention, on-screen colors will rarely, if ever, truly match
colors that are printed. This is partly due to differences 7 HIGH-VOLUME PRINT REPRODUCTION
between the color models employed by graphic displays
and the color models employed by most printing devices
(RGB vs. CMYK). It is also partly due to differences When large numbers of maps are required (e.g., more
between software applications and display and printing than 200), issues of cost and time become critical.
technologies. Each device in the workflow, including Methods for low-volume print reproduction such as
scanners, graphic displays, and printers, introduces sub- laser and ink-jet printing become too costly or time-
tle variations in color. Even devices that are supposedly consuming when reproducing large numbers of maps,
identical—two identical printers, for example—can especially when full-color output is required. Instead of
produce different results due to variations in calibration, providing an in-depth discussion of low-volume meth-
operating conditions, and so on. All of these factors ods, we will focus on high-volume print reproduction.
contribute to differing color gamuts (ranges of colors While low-volume reproduction methods are plentiful
produced) among devices. Color management systems and varied, high-volume reproduction is dominated by
(CMSs) provide consistency and predictability of color a single method: offset lithography. Lithography is a
by identifying differences in gamuts and correcting the printing process in which ink is made to stick only
variations in color introduced by each device (Agfa to certain areas of a printing surface (through chemical
Corporation 1997). means), and is subsequently transferred to a print
CMSs are centered on the color profile, which is an medium (e.g., paper). Offset lithography is a form
electronic file that describes the manner in which a of lithography in which ink is transferred to an
particular device introduces color variations. Before a intermediate printing surface before being transferred
profile is created, the device—a color printer, for to the print medium. Virtually all mass-produced maps
example—needs to be warmed up to normal operating are the result of offset lithography, which is character-
temperature and calibrated so that it operates within ized by excellent print quality, high printing speed, and
the manufacturer’s specifications. A sample page is a significant decrease in the cost per unit as the num-
printed, which consists of a series of strictly defined col- ber of copies increases—that is, the last copy costs sig-
ors. Colors on the printed sample page are measured nificantly less than the first. The cost-per-unit decrease
using a colorimeter or spectrophotometer (both devices occurs because the majority of expenses are incurred
precisely measure color). The measured values are then in preparing for printing; the cost of paper and ink for
sent to the CMS, which creates the device’s color profile additional copies is minimal in comparison with the
by comparing the colors on the sample page with the costs of prepress and press setup, activities that occur
colors as they were originally defined. Once color before printing begins.
profiles have been created for every device in the work-
flow, the CMS is able to correct for variations in color, 7.1 The Prepress Phase
resulting in printed colors that are as similar as possible
to those seen on the graphic display, captured by the The prepress phase of high-volume map reproduction
scanner, and so on. It is impossible to match all colors in consists of various technologies and procedures that
every situation, however, because the color gamuts of make offset lithographic printing possible. The press
electronic devices can be substantially different. in “prepress” refers to the offset lithographic printing
Early CMSs were proprietary and worked only with a press (Figure 11), which is further described in section
particular manufacturer’s equipment. Early systems were 7.4. The abbreviated terms offset press and offset
also designed to work at the application level, as opposed printing are often used in regard to this device. Many
to the operating system level, which is the trend today. aspects of prepress revolve around the service bureau,
Color management has become partly standardized with a business that specializes in the creation of products
the formation of the International Color Consortium such as film negatives, printing plates, and proofs
(ICC). The ICC has developed a standard for a vendor- and provides print-related services. The service
neutral, cross-platform color profile called the ICC color bureau often plays a key role in the reproduction
profile.The ICC has also established architectures for the process, intermediate between the cartographer and
Color Management Module (CMM), which actually the press operator (the person who operates the
performs color corrections, and the Color Management offset press).
11:53:31.
Map Reproduction
A
Map Map
Film negative Printing plate

Digital
(positive)
map file
B
Cyan Cyan
Magenta Magenta
Yellow Yellow
Digital Black Black
map file Film negatives Printing plates
(color separations) (positives)
FIGURE 12 (A) Film negative and printing plate for a map

FIGURE 11 A Heidelberg four-color offset lithographic
employing one base color. (B) Film negatives and printing
printing press. Used by permission from Heidelberg USA.
plates for a full-color map to be printed using process col-
ors.
The film negative is a representation of a map, or one
color of a map, on clear plastic film. It is composed of One printing plate is created for each film negative that
black, gray tones produced through screening, and clear is produced, as illustrated in Figure 12. After it is
areas. It represents the opposite of what will ultimately mounted on the offset press, the printing plate is the
be printed: Black on film results in white when printed; first representation of the digital map to come into con-
a 70 percent tint on film results in a 30 percent tint when tact with ink, which is transferred to the print medium
printed. The film negative plays two important roles in shortly afterward. Plates are produced by service
prepress: It is used in the creation of printing plates, and bureaus or by press operators, depending on the capa-
it is used in the creation of high-quality proofs. The bilities of each.
service bureau creates the film negative from a digital The proof is a representation of what the final, repro-
map file. The file is processed according to the steps duced map will look like. It is an essential component of
described earlier and illustrated in Figure 3. The film the prepress phase, and is used in conjunction with edit-
negative is printed on a device called an imagesetter, ing to ensure that the map will be reproduced just as
which is essentially a very high-resolution, large-format the cartographer intends. Cruder, less expensive proofs
laser printer that prints on clear film. A typical resolu- (rough drafts) are created repeatedly during the map
tion of 2,400 dpi allows for the creation of extremely design process, but higher-quality (and more expensive)
well-defined linework and fine halftone screen frequen- proofing methods are required in the prepress phase.
cies. A single film negative is produced for each base Proofs are created using various technologies and are
color that a map consists of. A map consisting of one produced by individuals, service bureaus, and press
base color, say, black, requires only one film negative operators. Proofing methods are described in detail in
(Figure 12A); a map that will be reproduced using the section 7.3.
four-color process printing method requires four film
negatives—one each for cyan, magenta, yellow, and 7.2 File Formats for Prepress
black (Figure 12B). Each negative produced for a mul-
ticolor map is called a color separation, because it rep- A digital map can be delivered to the service bureau in a
resents just one of the base colors (process or spot) of variety of formats. For example, it can be delivered in the
which a map is composed. native format of the application software that created it
The printing plate is a sheet of aluminum (or poly- (an Illustrator file, an ArcMap file, etc.), but this can be a
ester) that is ultimately mounted on a roller on an off- risky option for several reasons. First, the service bureau
set lithographic printing press. It receives a positive, might not have the exact software application or fonts
latent (invisible) image from a film negative through that you used. Second, related data such as linked images
photographic means. The photosensitive surface of the or GIS data sets might not transfer well to the service
printing plate is photographically etched when exposed bureau’s computer, making it difficult or impossible for
to light that passes through an overlying film negative. the service bureau to properly open the file. Finally,
271
11:53:31.
Map Reproduction
application files are easily editable, making it possible for to embed related data and fonts, PDF files also have
the service bureau to edit your file, either intentionally or the ability to embed, or encapsulate, features such as
accidentally. Delivering a digital map in the application’s hyperlinks, movies, and keywords for searching and
native file format can yield good results, but only if the indexing. Zooming and panning capabilities for
service bureau can assure you that these issues will not on-screen viewing are also provided. PDF files can be
pose problems. viewed using Adobe Reader (a free download), and
Another option is to deliver the digital map as a file can be created from virtually any application via
that contains page description data (e.g., a PostScript Adobe’s Acrobat and Acrobat Distiller. Acrobat com-
file). Page description files can be written by choosing presses and optimizes digital maps for printing, Web
the Print to File option when printing your map. This display, and so on. Although the PDF format was orig-
can be a practical option because related data and fonts inally intended for on-screen display, it is quickly
get embedded into page description data, and because becoming the format of choice for the delivery of maps
the service bureau has the ability to download page and other documents to service bureaus for high-end
description data directly to RIPs and printing devices printing (Cardin et al. 2001).
without the software application that was used to create Portable document files are created by converting
the map. Downsides to this approach include the facts digital maps from their native file format into EPS or
that page description files can be very large, they need PDF files. This is normally accomplished either by
to contain specific information regarding the printing selecting the Export command within the application
device that will be used (information you might not that was used to create the map, or by “printing” the
have), and their contents can be viewed only by a hand- map to a PDF file using Acrobat. Be aware that on
ful of software applications. occasion, certain graphical elements of your map (e.g.,
A more attractive option is to deliver the digital map custom line patterns) will be altered or lost during the
in a portable document format such as Encapsulated conversion process. If you choose to deliver your digital
PostScript (EPS) or Portable Document Format map to the service bureau in a portable document
(PDF). An EPS file is a subset of the PostScript page format, it is especially important that you proofread
description language that allows digital maps and other your map based on output from the EPS or PDF file so
documents to be transported between software appli- that you can identify any changes resulting from the file
cations and between different types of computer. It conversion.
consists of PostScript code for high-resolution printing
and an optional low-resolution raster image for on- 7.3 Proofing Methods
screen display. Because they are written in PostScript,
EPS files can be very large, but they don’t require the Proofing methods range from low-cost, low-quality
specific printing device information required by page techniques to high-cost, high-quality techniques, as
description files. EPS files also have the ability to illustrated in Figure 13. The proofing method that most
embed related data and fonts. A PDF file is similar to closely resembles the ultimate reproduction method is
an EPS file in that it is related to PostScript but is usually the most reliable. Proofs are useful in both
“smarter” and more efficient. In addition to being able low-volume and high-volume reproduction, but the
Digital Proofs Separation-Based Proofs
On-screen
display Monochrome Color
Overlay Laminate Press check
composite composite
Lowest Cost and Quality Highest
FIGURE 13 Proofing methods for print reproduction. The most reliable method is the one that most closely resembles the
intended reproduction method.
11:53:31.
Map Reproduction
emphasis here is on proofing for high-volume repro- limitation can be compensated for by printing larger
duction using an offset press. In print reproduction, maps in sections, or tiles, that can be assembled using
the correction of errors during editing is generally scissors and tape.
cheapest when working with crude proofing methods, Color composites represent a middle ground
and becomes increasingly expensive with more sophis- between low- and high-quality proofing methods and
ticated proofing methods. are also classified as digital proofs. Various printing
On-screen display (or soft proofing) is the least devices can produce color composites including color
expensive proofing method (assuming a computer was laser and ink-jet printers, although ink-jet devices are
used to create the map); it involves viewing a digital probably the most widely used. Ink-jet printers range
map on a graphic display. Maps are typically displayed from inexpensive letter-sized units to expensive, large-
using application software or document-viewing soft- format devices developed primarily for producing color
ware such as Adobe Reader, which displays PDF files. composite proofs. Inexpensive ink-jet printers cannot
On-screen display is the primary proofing method for reproduce fine detail as well as a monochromatic laser
nonprint reproduction but is the crudest of methods for printer, but they normally have the ability to represent
print reproduction. On-screen display allows you to color. However, the color accuracy of inexpensive ink-
perform basic editing and spell-checking, but should not jet printers is not good enough to warrant using them
be entirely trusted when editing for print reproduction for critically evaluating colors. High-end, large-format
for three reasons: ink-jet printers, such as the Iris line of proofing devices,
allow large maps to be printed on one sheet, and, when
1. Because computer screens are often smaller than used in conjunction with a CMS, can produce colors that
the map being edited, the entire map cannot be are far more accurate than those produced by inexpen-
seen at full size. sive, small-format ink-jet printers. Color laser printers
2. Because computer screen resolution is far lower produce high-quality images but are usually limited to
than the resolution of printing devices, fine detail page sizes of 11 17 or smaller. As with monochro-
will not be visible on-screen unless the cartogra- matic composites, color composites for larger maps can
pher zooms in on a portion of the map. be produced on smaller-format devices, printed in sec-
3. On-screen colors rarely, if ever, truly match colors tions, and assembled using scissors and tape.
that are printed. CMSs can improve the accuracy of The highest quality, most expensive proofs available
on-screen colors, but accurate color proofing is still (before the offset press is run) are created from color
beyond the scope of on-screen display. separations. Separation-based proofs are created from
color separations (film negatives), one for each base
Despite the drawbacks of on-screen proofing, the color. Because they are produced from the same film
PDF file format provides various features that negatives that will be used to create printing plates,
enhance on-screen proofing. When viewed in Adobe separation-based proofs provide a level of quality
Reader, PDF files can be zoomed and panned, unattainable with lower priced options. They are of
searched, spell-checked, and commented on. Adobe such high quality that they can act as contract proofs:
Reader also provides proofing simulation tools that contractually binding documents. If the cartographer
allow the user to see what the map might look like is satisfied with a separation-based proof, and if the
when printed on paper. Because of these features, print operator assures the cartographer that the final,
PDF files are often accepted as contract proofs printed map will look very similar to the proof, then
(described later) in noncritical printing situations the proof can act as part of the contract between the
(WhatTheyThink.com 2002). two parties.
Creation of monochromatic composites is the second There are two categories of separation-based proofs:
least expensive proofing method. This involves printing overlay and laminate. Overlay proofs consist of a stack
a map in black, white, and gray tones, typically on a laser of transparent film sheets that are bound on one edge
printer. Monochromatic composites are classified as like a book. Examples include DuPont’s Cromacheck
digital proofs, because they are created without film and Imation’s Color Key products. Each film sheet con-
negatives. The term composite refers to the fact that all tains a positive image of one base color and its tints,
color information is applied to the page at virtually the taken directly from its corresponding color separation.
same time, and, in this case, is represented as gray tones. The sheets are registered to overlay correctly, and when
Monochromatic proofs created by laser printers are suf- viewed together, the colors in each sheet combine to
ficiently detailed for general editing tasks, but color produce an image that is very similar to what the printed
cannot be discerned, making monochromatic proofing map will look like. Individual sheets can be lifted, allow-
unsuitable for color editing. Laser prints are typically ing the cartographer to isolate problems with particular
limited to page sizes of 11 17 or smaller, but this color separations (Color Plate 3). Laminate proofs also
273
11:53:31.
Map Reproduction
consist of overlying transparent film sheets, but the indi-

vidual sheets are melded into a single sheet (individual Water Ink
sheets cannot be lifted), which is often mounted on a
backing material similar to the intended print medium. Plate
Cylinder
Examples include DuPont’s Cromalin and Imation’s
Matchprint products. Laminate proofs are considered to Printing
be superior to overlays in image quality and are more Plate
Blanket
expensive (Agfa Corporation 1999). Although separa- Cylinder
tion-based proofs represent the most reliable prepress
proofs, color composites (digital proofs) are improving
rapidly and will likely emerge as practical alternatives to
Impression
separation-based proofs. Cylinder
The ultimate (and most expensive) proof is referred
to as the press check. At the press check, printing plates
are mounted on the offset press, the press is inked and
calibrated, and sample prints are made. If the cartogra- FIGURE 14 A printing unit (simplified) of an offset litho-
pher is satisfied with the samples, the press run can graphic printing press.
begin. As the printing press runs, the press operator
periodically compares the prints with the approved ink and water to the printing plate. For simplicity, they
samples and makes adjustments to the press if neces- are represented in Figure 14 as one cylinder for water
sary. Cartographic errors discovered during the press and one for ink.
check will be extremely expensive to correct, because Offset presses are also categorized as being either
prepress and press setup activities will need to be sheet-fed or web-fed. Sheet-fed presses use precut print
repeated. Because of this, the press check should be media (individual sheets), and are typically capable of
viewed as an opportunity to identify issues that the producing between 10,000 and 12,000 prints per hour.
press operator can address, rather than an opportunity Web-fed presses use uncut print media stored on a roll,
to identify cartographic errors. For example, the press and are up to four times faster, producing almost 50,000
operator is capable of producing slight variations in the prints per hour (Fleming 2002). Web-fed presses
lightness of particular colors by altering the density of commonly have integrated folding mechanisms and are
ink that is applied to the page without increasing the used for printing newspapers, magazines, and so forth.
cost of the print job. Both categories of offset press are blazingly fast in
comparison with methods for low-volume reproduc-
7.4 Offset Lithographic Printing tion, such as ink-jet printing. Most offset presses print
on one side of the print medium at a time, although
Once the prepress phase is complete, the offset litho- some, called perfectors, are capable of printing on both
graphic printing process can begin. Offset presses are sides simultaneously.
mechanical printing devices that incorporate aspects As stated earlier, the printing plate is the first
of photographic and digital technologies. Presses are representation of a digital map to come into contact
categorized, in part, according to the number of base with ink. The ink used in offset printing is oil-based—a
colors they can produce. The simplest presses are lim- key characteristic in determining the manner in which
ited to one base color and its tints, whereas others are ink adheres to the printing plate. After a latent image is
capable of printing 10 or more base colors and their transferred to a printing plate during prepress, the plate
tints. Tints are typically created through halftone is chemically treated in a process that converts image
screening. The offset press contains one or more areas (areas that have been photographically etched)
printing units (Figure 14) composed of many cylinders into oleophilic (oil-liking) surfaces, and nonimage areas
that ultimately transfer ink to the print medium. Each into oleophobic (oil-fearing) surfaces. As the offset
printing unit is capable of printing one base color. press runs, water and ink are applied to the printing
There are three primary cylinders in each printing plate by the series of smaller cylinders near the top of
unit: the plate cylinder, to which a printing plate is the printing unit; ink adheres only to the image areas
mounted; the blanket cylinder, which receives an and is washed away from nonimage areas.
image from a printing plate and subsequently transfers Once ink is applied to the printing plate, the image
it to the print medium; and the impression cylinder, on the plate could, in theory, be transferred directly to
which helps move the print medium through the press. the print medium. If this were to occur, however, the
There are also several smaller cylinders that come into resulting image would be reversed, as illustrated by the
contact with the plate cylinder; these are used to apply image on the blanket cylinder in Figure 15. To print a
11:53:31.
Map Reproduction
Cyan
Map image Normal Magenta
Blanket Map Image
Reversed Yellow
Cylinder
Black
Normal
FIGURE 15 A blanket cylinder transferring the map image

from the printing plate to the print medium. Full-color
normal, nonreversed image, it is necessary to transfer

the image from the printing plate onto the blanket FIGURE 16 Multicolor printing achieved through the use of
cylinder, which subsequently transfers the image onto multiple printing units. In this example, process colors are
the print medium. The offset in “offset lithography” mixed on the page, resulting in full-color output.
refers to this process—the printing plate is offset from
the print medium by the blanket cylinder. The rubber
coating on the blanket cylinder allows it to perform two
additional duties. First, it acts as a buffer between the minimize the effects of misregistration. It involves the
easily worn printing plate and the print medium. manipulation of ink placement to improve the appear-
Second, it applies ink to porous print media more effec- ance of areas where inks overlap incorrectly, or where
tively than the rigid printing plate would be able to. As gaps occur (i.e., where inks should print, but don’t). The
ink is transferred to the blanket cylinder and then to trapping process is becoming largely automated
the print medium, it is forced to spread out due to pres- through the use of special software employed by
sures exerted between the cylinders. This effect is service bureaus, relieving the cartographer of what
termed dot gain. It tends to deteriorate fine detail and used to be a tedious and often difficult procedure
make tints slightly darker, as halftone dots are increased (Agfa Corporation 1999).
in size. Dot gain is sometimes compensated for by
reducing the size of halftone dots when film negatives 7.5 Computer-to-Plate and Direct-to-Press
and printing plates are produced (Romano 1997).
The procedure just described accounts for the appli- The procedures described in the previous section
cation of a single base color to the print medium. represent the current standard for offset lithographic
Multicolor print jobs are produced in the same manner, printing. However, newer methods are increasing in
except that multiple printing units are employed, one popularity, deemphasizing or abandoning photographic
for each base color, as illustrated in Figure 16. As the technology in favor of digital technology. One such
print medium passes through the press, it receives a dif- method is termed computer-to-plate (CTP), also known
ferent color ink from each successive printing unit. The as direct-to-plate. This method is similar to traditional
printing units are spaced at a particular distance from offset lithography, except that the film negative compo-
one another to allow each successive color to print on nent is bypassed entirely; instead, the printing plate is
the same region of the print medium. This allows all digitally imaged by a platesetter, directly from a digital
base colors to overlay correctly and mix on the page. map file (Figure 17A). The printing plate is normally
Registration refers to the alignment of colors in coated with a photosensitive material and is photo-
multicolor printing. When using the process colors, for graphically etched when selectively exposed to laser
example, a map feature composed of a mixture of cyan, light within the platesetter, although some platesetters
magenta, and yellow should have crisp, well-defined use thermal technology for imaging. CTP methods offer
edges, indicating that all three inks have been applied lower costs, shorter prepress times, and fewer registra-
to the exact same area of the page. In reality, misalign- tion problems. They can also improve print quality
ment often prevents the perfect placement of colors, because one generation of image transfer is removed
and misregistration occurs. Misregistration can result from the printing process. A primary disadvantage of
from poorly produced film negatives or printing plates, CTP printing is related to proofing. High-quality,
or from an imprecise or poorly calibrated printing separation-based proofs are created directly from film
press. Trapping refers to a series of techniques used to negatives; without negatives, alternative, lower-quality
275
11:53:31.
Map Reproduction
tools used to create electronic maps or of mapping

A applications themselves, but rather, is a general survey
Cyan Cyan of methods for reproducing and disseminating maps in
Magenta Magenta digital form.
Yellow Yellow
Digital Black Black 8.1 Computer-Based
map file Film negatives Printing plates
The reproduction of maps in digital form is simple in
comparison with print reproduction: It consists primar-
B ily of digital file duplication, as performed by a com-
puter’s operating system. Reproduced digital maps can
Cyan be stored on media such as the CD, DVD, and USB
Magenta “flash” memory stick—media that are also commonly
Yellow used for the dissemination of electronic maps. The local
Digital Black area network (LAN) offers an efficient method for
Printing plates
map file Film negatives mounted on electronic map dissemination within a limited area, such
printing press as within a business. Static digital maps can be dissemi-
nated in a wide variety of file formats, including the
native format of the application software that created
FIGURE 17 (A) Computer-to-plate printing. (B) Direct-to-
the map, a page description file, a portable document
press printing. (Note that in both cases, the film negative is
format such as EPS or PDF, or in one of many raster
bypassed.)
file formats. Raster formats commonly used include
forms of proofing must be employed. Another disad- JPEG, PNG, TIFF, and BMP, although almost any raster
vantage is the initial cost of CTP equipment. format can be used if the map user has software that
Direct-to-press methods take CTP one step further. will display it. Dynamic maps can be disseminated as
Direct-to-press methods are based on either offset executable files that encapsulate a map that was created
lithographic technology (direct imaging presses) or in a programming language such as Visual Basic, or that
color laser printing technology (digital toner-based was created in an application such as Macromedia*
printing), and are currently tailored primarily to lower Director and saved as a Projector file (Figure 18). The
volume, fast-turnaround reproduction needs. A sepa- Projector is an executable file that can include
rate platesetter is not involved. With offset-oriented animation, audio, interactivity, and other features.
devices, the printing plate is imaged while it is An important consideration in computer-based
mounted on the printing press (Figure 17B); if color dissemination is that the characteristics of graphic dis-
laser printer technology is involved, the device acts plays vary, and thus the cartographer cannot anticipate
much like a color laser printer. Direct-to-press meth- exactly what a map will look like on the map user’s
ods further streamline the prepress phase of reproduc- computer. Display color and resolution can differ from
tion, and virtually eliminate registration problems system to system. The problem of color accuracy will be
because the plates are imaged after they are mounted partly overcome as more people employ CMSs, but dif-
on the press. Initial equipment costs are high, and ferences in resolution will likely remain an obstacle to
proofing remains a problem with offset-oriented the consistent representation of maps on computers.
devices. However, laser-oriented devices allow the
printing plate to be reimaged on the fly, allowing inex- 8.2 Internet- and Web-Based
pensive, last-minute changes to be made (Adams and
Dolin 2002). The overall print quality of laser-based The Internet plays an important role in map dissemi-
devices is very good, but doesn’t match the quality of nation because it is capable of distributing static and
offset-based devices (Agfa Corporation 2000). interactive maps to large numbers of users almost
instantaneously. Whereas the reproduction of digital
maps is relatively simple, electronic map dissemination
8 NONPRINT REPRODUCTION AND
can be quite complex; a comprehensive discussion is
DISSEMINATION
beyond the scope of this text. However, the following is
an overview of some of the technologies, protocols, and
In this section, we focus on methods for duplicating and
distributing static and dynamic maps that will not be
printed. This is not an in-depth discussion of software * Macromedia, Inc., is now owned by Adobe Systems Incorporated.
11:53:31.
Map Reproduction
FIGURE 18 Screen shot of an inter-

active dynamic map that was dissemi-
nated as a Projector executable file.
file formats that make Internet and Web mapping (text and related images). At its core, HTML is capable
possible. The Internet allows for the transfer of digital of displaying static maps in only two raster formats
maps via basic protocols such as File Transfer Protocol (GIF and JPEG), and interactivity is limited to the use
(FTP). Using FTP software, such as WS_FTP, the of hyperlinks that, when selected, open a different
cartographer can copy a digital map to a file server—a image or direct the user to a different Web location
computer connected to the Internet and dedicated to (Kraak and Brown 2001).
the sharing of data and other software—where it can To increase the functionality of Web browsers
be accessed by a user, who downloads it to his or her beyond what is possible with HTML, specialized soft-
computer using FTP software. This is an example of ware applications called plug-ins are used that allow
simultaneous reproduction and dissemination, which Web browsers to interpret additional file formats. For
was mentioned at the beginning of this chapter; the example, you can disseminate a static digital map stored
file is duplicated and distributed at the same time. in PDF format from a Web site by placing the PDF file
Digital maps are often compressed to reduce their file on a file server and providing a hyperlink to the map on
size, allowing for faster dissemination. Several file your Web page. Assuming that the map user has the
formats, including JPEG and PDF, are automatically Adobe Reader plug-in installed, the user can simply
compressed when they are created. Other file formats click on the hyperlink, causing the map to be repro-
need to be compressed using a dedicated compression duced, transferred, and displayed in his or her Web
utility, such as WinZip. browser. A dynamic map can be disseminated in the
The Internet allows for basic map dissemination, but same way, except that a different file format and plug-in
the World Wide Web offers far greater flexibility. The will be needed. For example, a dynamic map can be cre-
World Wide Web is a subset of the Internet that is based ated in Adobe Flash format and stored on a file server.
on the Hypertext Transfer Protocol (HTTP)—a set of Flash is a Web-oriented authoring environment that can
methods defining the manner in which documents are be used to create interactive animations, graphics, and
sent and retrieved. The primary language (and file video. Dynamic maps in Flash format can be viewed in
format) for the Web is the Hypertext Markup Language Web browsers, or used as stand-alone applications. A
(HTML), which is a relatively simple, text-based pro- user can download a dynamic map in Flash format and
gramming language that is used to define the content display it in a Web browser, as long as the Flash Player
and appearance of Web pages. Web browsers such as (plug-in) is installed on the user’s computer.
Firefox and Internet Explorer are capable of interpret- TypeBrewer is an example of a dynamic mapping appli-
ing HTML documents and displaying their contents cation that is disseminated in Flash format.
277
11:53:31.
Map Reproduction
The functionality of the Web can be further The most sophisticated Web-based map dissemination
expanded to disseminate maps through the use of environments are composed of a combination of the
Java and JavaScript. Java is a platform-independent, technologies just described (Kraak and Brown 2001). For
object-oriented programming language that allows example, ESRI’s ArcIMS (Internet Map Server) incor-
the user to interact with maps and mapping applica- porates Java components and applets such as the Java
tions via the Web. To use Java applications, or applets, Viewer, a plug-in-based map browser that can be cus-
the Web browser must have access to the Java tomized using JavaScript. ArcIMS is designed to allow
Runtime Environment (JRE), which interprets Java the user to interact with data stored locally, or on remote
code. Executable code and related data are down- servers via CGI. Dedicated map servers such as this allow
loaded from a Web site (the server), stored on the the user to create, query, and manipulate maps, and to
user’s computer (the client), and processed locally by view and interact with the results, all within a Web
the user’s central processing unit (CPU). This browser (Environmental Systems Research Institute
arrangement allows Java applets to perform quickly Incorporated 2002). The National Map is an example of
once they have been downloaded (Kähkönon et al. an ArcIMS-based map dissemination environment.
1999). JavaScript is similar to Java but is simpler and ESRI’s ArcGIS Server is a newer, and more advanced,
less robust. Unlike Java, which is interpreted by the Web-based map dissemination environment. Examples
JRE, JavaScript can be embedded directly into of sophisticated map dissemination environments can be
HTML code, and interpreted and displayed by the found on the Geography Network (http://www.geogra-
Web browser itself without the need for the JRE. The phynetwork.com), a Web site where maps and mapping
Common Gateway Interface (CGI) is an additional applications can be “published” (Figure 19). MapServer
protocol that allows users to interact with mapping is an example of a popular “open source” Web-based
software and cartographic databases located on map dissemination environment developed at the
remote computers. University of Minnesota (http://mapserver.gis.umn.edu/).
FIGURE 19 The Geography

Network Web site, where digital
maps and mapping applications can
be published. Source: ESRI.
11:53:31.
Map Reproduction
MapServer is freely distributed, and provides the user (color separations), each of which represents one base
with the ability to obtain and modify the underlying color; printing plates that are mounted on the offset
source code. press; and proofs that facilitate map editing. Prepress
It is important to note that HTML is being replaced is centered on the service bureau. We learned that
gradually by the eXtensible Markup Language (XML)— portable document formats such as EPS and PDF are
an approach to programming Web content that is better preferred file formats for the delivery of digital maps
structured, more flexible, and more robust than HTML. to the service bureau.
NASA’s World Wind and Google Earth both employ Proofing methods were described on a scale ranging
XML to define the contents of map layers. The Scalable from low-end to high-end. The on-screen method is the
Vector Graphics (SVG) file format, which is based on cheapest and crudest, followed by monochromatic
XML, is an open, object-oriented standard for Web- composites normally produced on laser printers; color
based vector graphics. SVG is the most promising new composites produced on color laser printers, ink-jet
standard for high-quality static and dynamic vector- printers, and so on; separation-based proofs created
based maps on the Web (Neumann and Winter 2001). from color separations; and the press check—the high-
Google Maps has adopted the SVG format, evidence of est quality and most expensive proof.
SVG’s growing popularity. We learned that offset lithographic printing presses
contain printing units composed of rolling cylinders, and
SUMMARY that images are transferred from the printing plate to
the blanket cylinder and then to the print medium. Dot
gain was described as the spreading of ink due to pres-
In this chapter, we have presented topics related to
sures between cylinders on the press. Registration was
print and nonprint map reproduction and electronic
described as the alignment of base colors on the print
map dissemination. We have encouraged you to plan
medium, and trapping was identified as a series of meth-
ahead for map reproduction, and have stressed the
ods for correcting misregistration. Computer-to-plate
importance of map editing, both of which can help you
was described as a streamlined approach to offset
avoid difficulties when reproducing a map.
printing in which the film negative is eliminated; direct-
Raster image processing was described as being cen-
to-press is similar, but the printing plates are imaged
tral to printing digital maps, with the printing process dis-
after they are mounted on the press.
tilled to four steps: creation of the digital map using
Nonprint reproduction was introduced as being simple
application software, conversion into page description
in comparison with print reproduction. It involves digital
data, interpretation by a RIP and creation of a raster
file duplication on a computer. Computer-based map
image, and printing of the image on a printing device.
dissemination can be accomplished via several types of
Halftone and stochastic screening methods were pre-
storage media and LANs. Internet- and Web-based map
sented as ways to create tints from base colors. Aspects of
dissemination were identified as examples of simultane-
color printing were described, including process colors,
ous map reproduction and dissemination. Simple meth-
which mix on the page; spot colors, which are premixed;
ods of dissemination involve FTP, HTTP, and HTML.
high-fidelity process colors, which increase color variety
We learned that plug-ins such as Flash expand the variety
and vibrancy; continuous tone printing, which allows
of file formats that can be used online, and that program-
smooth color transitions without screening; and color
ming languages such as Java and JavaScript allow the
management systems, which identify and correct for
creation and dissemination of digital maps. These tech-
variations in color introduced by electronic devices.
nologies, together with map servers, allow sophisticated
Offset lithography was identified as the dominant
map dissemination environments to be published on the
method for high-volume print reproduction, charac-
Web. XML and SVG were identified as promising new
terized by high quality, high speed, and volume dis-
standards for Web-based map dissemination.
counts. Prepress involves the creation of film negatives
FURTHER READING
Adams, J. M., and Dolin, P. A. (2002) Printing Technology. Agfa Corporation. (1997) The Secrets of Color Management.
5th ed. Albany, NY: Delmar, Thompson Learning. Agfa-Gevaert N.V., Septestraat 27, B-2640 Mortsel, Belgium.
A comprehensive overview of printing technologies, including A primer on color management systems.
chapters on prepress, offset lithography, and digital printing. Agfa Corporation. (1999) From Design to Distribution in the
Adobe Systems Incorporated. (1997) The Adobe PostScript Digital Age. Agfa-Gevaert N.V., Septestraat 27, B-2640
Printing Primer. San Jose, CA: Adobe Systems Incorporated. Mortsel, Belgium.
An introduction to the PostScript page description language. A primer on the digital publishing process.
279
11:53:31.
Map Reproduction
Agfa Corporation. (2000) An Introduction to Digital Color Kraak, M.-J., and Brown, A., eds. (2001) Web Cartography:
Printing. Agfa-Gevaert N.V., Septestraat 27, B-2640 Mortsel, Developments and Prospects. London: Taylor & Francis.
Belgium. Includes chapters related to Web-based map dissemination.
A primer on digital printing workflows.
Loppnow, M. (2005) “Pre-press and technical support techni-
Brown, A., and Feringa, W. (2003) Colour Basics for GIS cian.” Cartographic Perspectives, no. 50: 47–50.
Users. Harlow, England: Prentice Hall.
Includes discussion of the prepress phase of high-volume map
A handbook for the use of color in GIS. reproduction.
Burnett, M. (2004) “Working with your printer.” Cartographic Lowe, J. W. (2005) “Web GIS gets flashy.” Geospatial Solutions
Perspectives, no. 49: 64–66. 15, no. 5: 33–36.
A description of issues related to printing maps. A description of Macromedia Flash as applied to Web-based map
Bruno, M. H., ed. (2000) Pocket Pal: A Graphic Arts Production dissemination.
Handbook. Memphis, TN: International Paper Company. Monmonier, M. S. (1993) Mapping It Out: Expository
A popular guide to graphic arts production. Cartography for the Humanities and Social Sciences. Chicago:
Cardin, J., Castellanos, A., and Romano, F. (2001) PDF University of Chicago Press.
Printing and Publishing: The Next Revolution After Chapter 5 includes a discussion of copyright issues related to map
Gutenberg. Agfa Corporation. Agfa-Gevaert N.V., Septestraat reproduction.
27, B-2640 Mortsel, Belgium. Neumann, A., and Winter, A. M. (2001) “Time for SVG—
An in-depth description of the Portable Document Format and Towards high quality interactive web-maps.” Proceedings, 20th
how it fits into modern, digital reproduction workflows. International Cartographic Conference, International
Cartwright, W., and Stevenson, J. (2000) “A toolbox for publish- Cartographic Association, Beijing, China, CD-ROM.
ing maps on the World Wide Web.” Cartography 29, no. 2:83–95. Describes the capabilities and shortcomings of SVG in the context
A description of tools for Web-based map dissemination. of Web-based mapping technologies.
Fleming, P. D. (2002) http://www.wmich.edu/~ppse/Offset/. Pavlicko, P., and Peterson, M. P. (2005) “Large-scale topo-
graphic web maps using scalable vector graphics.” Cartographic
A comprehensive, Web-based overview of offset lithography.
Perspectives, no. 50: 34–46.
Gartner, G., ed. (2002) Maps and the Internet 2002. Vienna,
Austria: Geowissenschaftliche Mitteilungen, Band 60, TU Describes the use of Scalable Vector Graphics in representing
Web-based maps.
Wien, Institute of Cartography and Geomedia Technique.
Peterson, M. P., ed. (2003) Maps and the Internet. Amsterdam:
Papers presented at the 2002 Annual Meeting of the International
Cartographic Association Commission on Maps and the Internet. Elsevier.
Hyatt, J. (2000) “Stochastic screening in cartographic applica- An overview of Internet-based cartography.
tions.” Cartouche 40:14–19.
An overview of screening methods and their use in cartography.
GLOSSARY
application software: software programs designed for computer-to-plate: a variation of traditional offset lithogra-
specific purposes, such as graphic design, GIS, and remote phy in which the film negative component is bypassed
sensing. entirely; the printing plate is digitally imaged by a platesetter
color composite: a digital color proof created on a color directly from a digital map file.
laser, ink-jet, thermal-wax transfer, dye sublimation, or other continuous tone: full-color printed output achieved without
printer. the use of screening techniques.
color laser printer: a type of laser printer that incorporates digital proof: a proof created without film negatives.
several colors, normally the process colors, and mixes them on digital television: a relatively new standard for high-defini-
the page. tion television (HDTV) that provides higher resolution
color lookup tables: used in association with a frame buffer images and better sound quality.
to make rapid color changes in a graphics display. digital-to-analog converter (DAC): a device within a CRT
color management system: a software application that iden- that converts a digital value to an analog voltage value that is
tifies differences in gamuts among devices and corrects the applied to an electron gun.
variations in color introduced by each device. digital versatile disc (DVD): a high-capacity digital medium
color profile: an electronic file that describes the manner in for storing and playing moving images and sound.
which a particular device introduces color variations. direct-to-press: a variation of traditional offset lithography
color separation: a film negative that represents just one of in which the film negative and traditional printing plate
the base colors that a map is composed of. components are bypassed entirely.
11:53:31.
Map Reproduction
dithering: a process in which colors are created by presum- JavaScript: a platform-independent, object-oriented pro-
ing that the reader will perceptually merge different colors gramming language similar to Java, but simpler and less
displayed in adjacent pixels (on a graphics display) or dots robust.
(on a printer). laminate proof: a separation-based proof consisting of over-
dot gain: phenomenon in which ink is forced to spread out lying transparent film sheets that are melded into a single
due to pressures exerted between the cylinders of an offset sheet.
lithographic printing press. local area network (LAN): a series of computers within a
Encapsulated PostScript (EPS): a subset of the PostScript limited geographic area that are connected, allowing for the
page description language that allows digital maps and other sharing of data and resources.
documents to be transported between software applications, map dissemination: the distribution of reproduced maps in
and between different types of computers. physical or electronic form.
eXtensible Markup Language (XML): an approach to pro- map editing: the critical evaluation and correction of every
gramming Web content that is better structured, more flexi- aspect of a map.
ble, and more robust than HTML.
map reproduction: the printing of a map on paper or similar
File Transfer Protocol (FTP): a set of methods defining the medium (print reproduction), or the electronic duplication of
manner in which electronic files and documents are sent and a map in digital form (nonprint reproduction).
retrieved via the Internet.
map server: a computer connected to the Internet that
film negative: a representation of a map, or one color of a allows the user to access and interact with digital maps using
map, on clear plastic film; composed of black and gray tones, Web browsers.
and clear areas, and represents the opposite of what ultimately
monochromatic composite: a digital proof created by print-
will be printed.
ing a map in black, white, and gray tones, typically on a laser
Flash: a Web-oriented authoring environment that can be printer.
used to create interactive animations, graphics, and video.
monochromatic laser printer: a “black-and-white” printing
four-color process printing: the combination of the process device that utilizes a focused light source, a metal drum, toner,
colors (CMYK) with screening techniques, allowing for a very and a heat source.
wide range of colors in print reproduction.
offset lithographic printing press: a mechanical printing
frame buffer: an area of memory that stores a digital repre- device that provides excellent print quality, high printing
sentation of colors appearing on the screen. speed, and a significant decrease in the cost per unit as the
gamut: a range of colors produced by a computer or a print- number of copies increases.
ing or display device. offset lithography: a form of lithography in which ink is
halftone screening: the application of ink or toner in a pat- transferred to an intermediate printing surface before being
tern of equally spaced cells of variable size to create tints from transferred to the print medium.
a base color. on-screen display: the least expensive proofing method;
high-fidelity process colors: colors based on the traditional involves the viewing of a digital map on a graphic display; also
process colors (CMYK), but including two or three additional known as soft proofing.
colors that are mixed on the page via a printing device. overlay proof: a separation-based proof consisting of a stack
hypertext markup language (HTML): a relatively simple, of transparent film sheets that are bound on one edge like a
text-based programming language that is used to define the book.
content and appearance of Web pages. page description data: digital data consisting of a set of
hypertext transfer protocol (HTTP): a set of methods defin- printing instructions that describe every graphical and textual
ing the manner in which electronic files and documents are component of a map.
sent and retrieved via the World Wide Web. page description language: a particular data structure used
imagesetter: a very high-resolution, large-format laser to describe page description data.
printer that produces film negatives. platesetter: a device that produces a latent image on a print-
ink-jet printer: a printing device (normally, full-color) that ing plate through photographic means.
incorporates nozzles, or “jets,” that squirt ink onto the print Portable Document Format (PDF): a file format that allows
medium. digital maps and other documents to be transported between
International Color Consortium (ICC): an entity that has software applications and between different types of comput-
developed a standard for a vendor-neutral, cross-platform ers, and is capable of embedding features such as hyperlinks,
color profile, and architectures for color management. movies, and keywords.
Internet: a global network of computers that permits shar- PostScript: the de jure standard page description language,
ing of data and resources. produced by Adobe Systems, Inc.
Java: a platform-independent, object-oriented programming prepress: the phase of high-volume map reproduction
language that allows users to interact with maps and mapping consisting of various technologies and procedures that make
applications via the World Wide Web. offset lithographic printing possible.
281
11:53:31.
Map Reproduction
press check: an inspection of the quality of print produced rosette: a pattern of halftone cells (consisting of the process
by an offset lithographic printing press, shortly before a press colors) that results when moiré patterns are avoided through
run begins. the appropriate specification of halftone screen angles.
print controller: a stand-alone raster image processor (RIP) Scalable Vector Graphics (SVG): a file format based on
that is often used to allow a color copy machine to be used as XML that is an open, object-oriented standard for Web-based
a printing device. vector graphics.
Printer Control Language (PCL): a standard page descrip- screen angle: the angle at which lines of halftone cells are
tion language produced by Hewlett Packard, Inc. oriented.
printer driver: software that converts digital map data from screen frequency: the spacing of halftone cells within a given
the native format of the application software into page area, expressed in lines per inch (lpi).
description data. separation-based proof: a proof created from color separa-
printing plate: a sheet of aluminum (or polyester) contain- tions (film negatives), one for each base color.
ing a latent image, which is mounted on a roller in an offset service bureau: a business that specializes in the creation of
lithographic printing press. products such as film negatives, printing plates, and proofs,
printing unit: an organization of cylinders in an offset litho- and provides print-related services.
graphic printing press that transfers one base color of a map spot colors: colors that are mixed before they reach the
to the print medium. printing device.
process colors: the subtractive primary colors (CMY), plus stochastic screening: the application of ink or toner, in a
black (K); these colors are mixed on the page via a printing pattern of very small, pseudorandomly spaced dots of uniform
device. size, to create tints from a base color.
proof: a preliminary representation of what a final repro- subtractive primary colors: the colors (cyan, magenta, and
duced map will look like. yellow) associated with printing inks that can be mixed to
raster data model: a general approach for representing geo- create a wide range of colors.
graphic features in digital form that uses rows and columns of tint: a lighter version of a base color.
square grid cells, or pixels. tint percentage: the degree to which the appearance of an
raster image processing: the conversion of a digital map into ink or toner is lightened when employing screening tech-
a raster image that can be processed directly by a raster-based niques.
printing device. trapping: a series of techniques used to minimize the effects
raster image processor (RIP): software, or a combination of of misregistration in multicolor printing.
software and hardware, that interprets page description data vector data model: a general approach for representing
and converts them into a raster image that can be processed geographic features in digital form that uses discrete points,
directly by a printing device. lines, and polygons.
registration: the alignment of base colors in multicolor World Wide Web: a system of Internet file servers that sup-
printing. port the hypertext transfer protocol.
REFERENCES
Adams, J. M., and Dolin, P. A. (2002) Printing Technology Environmental Systems Research Institute Incorporated.
(5th ed.). Albany, NY: Delmar. (2002) Geographic Information Systems for Java, an ESRI
Adobe Systems Incorporated. (1997) The Adobe PostScript White Paper. Redlands, CA: Author.
Printing Primer. San Jose, CA: Author. Fleming, P. D. (2002) “Offset lithography.” www.wmich.
Agfa Corporation. (1997) The Secrets of Color Management. edu/~ppse/Offset/.
Agfa-Gevaert N.V., Septestraat 27, B-2640 Mortsel, Hyatt, J. (2000) “Stochastic screening in cartographic applica-
Belgium. tions.” Cartouche 40:14–19.
Agfa Corporation. (1999) From Design to Distribution in the Kähkönon, J., Lehto, L., Kilpeläinen, T., and Sarjakoski, T.
Digital Age. Agfa-Gevaert N.V., Septestraat 27, B-2640 (1999) “Interactive visualization of geographical objects on
Mortsel, Belgium. the Internet.” International Journal of Geographical
Agfa Corporation. (2000) An Introduction to Digital Color Information Science 13, no. 4:429–428.
Printing. Agfa-Gevaert N.V., Septestraat 27, B-2640 Kraak, M.-J., and Brown, A. (eds.). (2001) Web Cartography:
Mortsel, Belgium. Developments and Prospects. London: Taylor & Francis.
Cardin, J., Castellanos, A., and Romano, F. J. (2001) PDF Neumann, A., and Winter, A. M. (2001) “Time for SVG—
Printing and Publishing: The Next Revolution After Towards high quality interactive web-maps.”
Gutenberg. Mortsel, Belgium: Agfa Corporation. Proceedings, 20th International Cartographic Conference,
11:53:31.
Map Reproduction
International Cartographic Association, Beijing, China, Viggiano, J. A., and Hoagland, W. J. (1998) “Colorant selec-
CD-ROM. tion for six-color lithographic printing.” IST/SID 1998
Pantone Incorporated. (2002) Designer Hexachrome Primer. Color Imaging Conference, pp. 112–115.
Pantone Incorporated, New Jersey. WhatTheyThink.com. (2002) “Proofing options: Survey exam-
Romano, F. J. (ed.). (1997) Delmar’s Dictionary of Digital ines current stats and future trends.” members.whatthey-
Printing and Publishing. Albany, NY: Delmar. think.com/allsearch/article.cfm?id = 6396.
283
11:53:31.
11:53:31.
Choropleth Mapping
OVERVIEW
This chapter covers the choropleth map, which is University to develop a broad set of color schemes for
arguably the most commonly used (and abused) method choropleth maps.
of thematic mapping. Section 1 considers the nature of Section 4 describes various approaches for specifying
data appropriate for choropleth mapping. Ideally, choro- particular colors of a scheme. For example, you might
pleth maps should be used for phenomena that have spa- decide that varying lightnesses of blue are appropriate
tial variation that coincides with the boundaries of enu- for a unipolar data set, but how should you select the
meration units. In practice, this seldom occurs, and so shades of blue making up the scheme? We will consider
data values depicted for enumeration units must be approaches for soft-copy maps, black-and-white printed
viewed as “typical,” as opposed to being uniform maps, and colored printed maps. We will find that the
throughout enumeration units. In this chapter we con- Web site for ColorBrewer is especially useful for provid-
sider some methods for standardizing data; for example, ing detailed color specifications.
areas of enumeration units can be accounted for indi- In section 5, we consider some issues pertaining to
rectly by taking the ratio of two raw totals that do not designing legends for choropleth maps. One key is to
involve area, such as a ratio of males to females. utilize readers’ prior cognitive framework. For instance,
Section 2 considers whether data for a choropleth map in designing a vertical legend, it makes sense to display
should be classed and, if so, which classification method high values at the top of the legend, using the notion that
should be used. Classed maps are commonly used people associate “up” with “higher.” On the other hand,
because they are considered simpler than unclassed maps, to enhance the overall map design, a horizontally
and thus we initially make the assumption that you wish arranged legend might be appropriate.
to create a classed map. We touch on classification meth- In section 6, we return to the issue of classed versus
ods in this chapter. unclassed maps. We argue that an unclassed map is
Section 3 considers numerous factors involved in select- appropriate when you wish to illustrate numerical
ing appropriate color schemes for choropleth maps. Some relationships among data values, or when you wish to
key factors include the kind of data (bipolar, polar, or bal- explore data (where classed and unclassed maps are both
anced); the ease of assigning names to colors (when utiliz- alternatives). We also consider various experimental
ing multiple hues, users should be able to easily assign studies of classed versus unclassed maps. Limitations of
names to the hues); ease of interpretation for the color such studies include a lack of direct comparison between
vision–impaired; and avoiding problems of simultaneous perceived regions on classed and unclassed maps, a fail-
contrast. These and other factors have been utilized by ure to measure the time required to process the map, and
Cynthia Brewer and her colleagues at Penn State viewing maps in contrived experimental situations.
285
11:54:08.
Choropleth Mapping
1 SELECTING APPROPRIATE DATA acres of wheat harvested, we might divide by acres of

wheat planted, with the resulting ratio providing a mea-
Ideally, the choropleth technique is most appropriate sure of success of the wheat crop (Figure 1A). As an
for a phenomenon that is uniformly distributed within alternative, we might divide acres of wheat harvested by
each enumeration unit, changing only at enumeration acres harvested for all major crops, producing a map
unit boundaries. For instance, state sales tax rates are illustrating the relative importance of wheat to the agri-
appropriate because they are constant throughout cultural economy of each county.
each state, changing only at state boundaries. Because A second approach for standardizing data is to create
this ideal is seldom achieved in the real world, how- a density measure by dividing a raw total not involving
ever, you should be cautious in using the choropleth area by either the areas of enumeration units or some
map. If you wish to focus on “typical” values for enu- area-based raw total. For example, if the number of
meration units, you can utilize the choropleth map, bushels of wheat produced is divided by the area of the
but you should realize that error might be present in county (in acres), then the result is bushels of wheat per
the resulting map. Instead of using the choropleth acre. This approach might not be meaningful, however, if
map because you have data readily available for enu- the raw-total data occur only within a portion of the enu-
meration units, you should consider whether some of meration unit. For example, with the wheat data, it makes
the other symbolization methods discussed in this text more sense to divide the bushels produced by the acres
might be more appropriate. of wheat harvested, yielding bushels of wheat produced
Of particular concern in choropleth mapping is the per acre of wheat (Figure 1B). A third approach for stan-
size and shape of the enumeration units. Ideally, the dardizing data is to compute the ratio of two raw totals,
method works best and is most accurate when there is neither of which involves area. For example, we might
not significant variation in the size and shape of the divide the value of wheat harvested (in dollars) by the
units. The problem can be understood by considering value of all crops harvested (in dollars). The resulting
the county-level map of population density. Imagine proportion would indicate the relative value of wheat in
trying to compute a measure of population density for each county. Although area is not included in the for-
one of the large counties in California. Because people mula, this approach indirectly standardizes for area
are unlikely to be uniformly spread throughout the because larger areas tend to have larger values for both
county, the shade shown is a poor representation for the attributes.
entire county. In contrast, a very small county, such as When computing the ratio of two raw totals not
one of the five counties of New York City, would be well involving area, it is common to express the result as a
represented because the population density is relatively rate. Although we are all familiar with rates, we often
uniform. The net result is that there is considerable vari- aren’t aware of how they are computed. To illustrate,
ation across the choropleth map in terms of our ability consider how cancer death rates are computed for
to properly represent population density. Also note that counties. First, we establish a simple proportional rela-
the largest counties provide the biggest visual impact, tionship as follows:
but also have potentially the largest error of represen- Cancer deaths for county Number of cancer deaths
tation. Thus, the choropleth technique is more appro- Population for county
=
100,000 people
priately used when enumeration units are similar in
size, such as for the counties of Iowa, remembering that We then solve for the number of cancer deaths by
ideally, a phenomenon should be relatively uniform rewriting the equation as follows:
throughout enumeration units.*
Number of Cancer deaths for county
Another important issue in selecting data for choro- = * 100,000
pleth maps is ensuring that raw-total data have been cancer deaths Population for county
adjusted to account for varying sizes of enumeration The resulting number of cancer deaths is termed the
units. We consider some approaches for standardizing cancer death rate. More generally, the formula for rates is:
data. One approach is to divide an area-based raw total
by some other area-based raw total. For example, for Units
Magnitude for category of interest
Rate = * of the
Maximum possible magnitude
* See Boscoe and Pickle (2003) for an overview of criteria that ideally
rate
should be considered in selecting enumeration units for choropleth A fourth standardization approach is to compute a
mapping. summary numerical measure (e.g., mean or standard
11:54:08.
Choropleth Mapping
A B
Wheat Harvested in Kansas, 1993 Wheat Harvested in Kansas, 1993
Ratio of Harvested to Planted Acres Yield in Bushels per Acre
53.3 to 75.0 78.5 to 85.8 87.2 to 92.1 92.4 to 96.1 96.5 to 99.4 16.2 to 24.1 25.1 to 30.1 30.6 to 36.2 37.1 to 41.6 42.6 to 52.4
FIGURE 1 Examples of approaches for standardizing data. (A) Dividing acres of wheat harvested by acres of wheat planted;
both are area-based raw totals and thus proportions (or percentages) result. (B) Dividing bushels of wheat harvested by acres of
wheat harvested; the result is a density measure (bushels per acre). (Data source: Kansas Agricultural Statistics 1994.)
deviation) for each enumeration unit. For example, we responsibility as the mapmaker to think carefully about
could compute the average size of farms in each county which standardization procedure is most appropriate,
by dividing the acreage of all farms by the number of given your data and the message that you wish to com-
farms. Note that this approach accounts for a larger municate.
acreage in a larger county by dividing by a greater num-
ber of farms.
In general, note that all of the standardization 2 DATA CLASSIFICATION
approaches discussed thus far involve ratios. Thus, one
might suggest that simply computing ratios is the key One important decision with respect to data classifica-
to standardization. A simple example illustrates that tion is whether to class the data. A potential advantage
this is not the case. Imagine that for each county, you of an unclassed map is that it can more accurately por-
computed the mean number of acres of wheat har- tray the data; for example, on a gray-tone map, the
vested over a 10-year period. Clearly, the values would intensity of a shade can be related directly to the mag-
be ratios (the numerator would be the sum of acreages nitude of individual data values. In contrast, a potential
over 10 years, and the denominator would be 10), but advantage of a classed map is that the limited number
the data would not be standardized because the of categories (on a five-class map, there will be five cat-
denominator in each case would be 10. No adjustment egories) makes the map easy to process. In deciding
would have been made to account for the fact that whether to class your data, you must also consider
larger counties tend to have larger numbers of acres whether you wish to present a map to others or to
harvested each year. explore a data set on your own. When presenting a map
It is important to recognize that the various stan- to others, you might be limited to showing one map (as
dardization approaches lead to quite different maps of in a printed publication), whereas when exploring data
the phenomenon being investigated. Figure 1A (illus- on your own, you can visualize the data in many ways
trating the percentage of planted wheat actually har- (both classed and unclassed maps can be used). For the
vested) reveals a random pattern. The map of wheat bulk of this chapter, we assume that you wish to create
yield (Figure 1B) produces yet another pattern, with a map with a limited number of categories, and that you
progressively higher values. Clearly, different methods wish to present the map to others, making the classed
of standardization produce different views of the spa- map an obvious choice. In section 6 we return to the
tial pattern of wheat harvested in Kansas. It is your issue of classed versus unclassed maps, and consider
287
11:54:08.
Choropleth Mapping
some research that has been done on this topic. We TABLE 1 Percentage of the adult population,
should also point out that the bulk of choropleth maps in U.S. states, that did not graduate from high
produced today are classed, even though unclassed school (based on 1990 data)
maps can be created easily. Finally, some mapping pack- State % Not Graduating
ages do not provide an option for mapping unclassed
Alabama 33.1
data, thus making your decision quite straightforward. Arizona 21.3
Once you have decided that classification is appropri- Arkansas 33.7
ate, you need to consider the range of classification California 23.8
methods. Remember that before selecting a method, you Colorado 15.6
need to consider the kind of data you have collected Connecticut 20.8
Delaware 22.5
(bipolar, balanced, or unipolar) and what level of round-
Florida 25.6
ing you wish. For our purposes, we assume that we wish to Georgia 29.1
map the percentage of the adult population in U.S. states Idaho 20.3
that did not graduate from high school (Table 1).This data Illinois 23.8
set is unipolar, and so we chose not to divide the data into Indiana 24.4
two groups, as we would likely do with bipolar data. Iowa 19.9
Kansas 18.7
Because the data are reported to the nearest tenth of a
Kentucky 35.4
percent and because we felt that readers of this text would Louisiana 31.7
be comfortable with this level of precision, we chose to Maine 21.2
retain it for the classification. Maryland 21.6
Next, you need to consider the actual methods of Massachusetts 20.0
Michigan 23.2
classification, such as equal intervals, quantiles,
Minnesota 17.6
mean–standard deviation, and optimal. Because we Mississippi 35.7
were creating a map for this text and knew that readers Missouri 26.1
would be familiar with the optimal method, we chose to Montana 19.0
use that method. If we were creating the map for a daily Nebraska 18.2
newspaper, we might select a different method; for Nevada 21.2
New Hampshire 17.8
instance, if we wished to focus on percentiles in the data
New Jersey 23.3
(say, the top or bottom 20 percent), then we would use New Mexico 24.9
quantiles. The last decision is to select an appropriate New York 25.2
number of classes. An advantage of the optimal method North Carolina 30.0
is that it can assist in selecting an appropriate number North Dakota 23.3
of classes. Although we considered plotting the number Ohio 24.3
Oklahoma 25.4
of classes against GADF to determine the number of Oregon 18.5
classes, we chose five classes because we wished to map Pennsylvania 25.3
this data set with both color and gray-tone maps, and we Rhode Island 28.0
felt that the gray-tone maps would be most easily inter- South Carolina 31.7
preted if no more than five classes of data were shown. South Dakota 22.9
Tennessee 32.9
Texas 27.9
Utah 14.9
3 FACTORS FOR SELECTING A COLOR SCHEME
Vermont 19.2
Virginia 24.8
In this section, we consider numerous factors that can Washington 16.2
be utilized to select a color scheme, focusing largely on West Virginia 34.0
Wisconsin 21.4
the recent work of Cynthia Brewer and her colleagues: Wyoming 17.0
Brewer (1994a; 1996), Brewer et al. (1997), Brewer
(2001), and Harrower and Brewer (2003).
should be represented by sequential steps in lightness;
3.1 Kind of Data Brewer termed this a sequential scheme. The most obvi-
ous example would be tones of gray, although in general,
In her early work, Brewer (1994a) suggested that the kind a sequential scheme can be achieved by holding hue and
of data (unipolar, bipolar, balanced) should play an saturation constant, and varying lightness (e.g., progress-
important role in selecting a color scheme. For unipolar ing from a light orange to a dark orange). Concurring with
data, she recommended that sequential steps in the data earlier cartographers’ recommendations (e.g., Robinson
11:54:08.
Choropleth Mapping
et al. 1984), Brewer advocated using light colors for low sequential schemes to emphasize the high end of the
data values and using dark colors for high data values, data, or, alternatively, a diverging scheme to emphasize
respectively. She argued that if the opposite is done, as is the midpoint of the balanced data.
sometimes the case on graphic displays, a clear legend is Brewer’s (1994a) early work was based on “personal
essential. These recommendations are supported by stud- experience, cartographic convention and the writings
ies by McGranaghan (1989; 1996). and graphics of others” (124). In more recent work,
Although Brewer argued that lightness differences Brewer and her colleagues (1997) performed experi-
should predominate for sequential schemes, she ments with people to determine the effectiveness of
stressed that visual contrast could be enhanced if satu- various color schemes; in particular, they applied
ration differences also are used. This can be seen in diverging, sequential, and spectral schemes to unipolar
Color Plate 1, in which a pure lightness scheme is com- data. The logic of applying a diverging scheme to unipo-
pared with a combined lightness–saturation scheme. In lar data is that if a quantiles classification is used, the
Color Plate 1A only lightness changes—from a light to median will fall in the middle class and serve as a logical
a dark green; in Color Plate 1B lightness again changes dividing point in the data.* A spectral scheme is based
from light to dark green, but saturation also changes, on the electromagnetic spectrum (red, orange, yellow,
increasing from the first to the third class and then green, blue, indigo, and violet). Traditionally, cartog-
decreasing for the latter two classes. Although in this raphers opposed using spectral schemes because yellow
case the increase and decrease in saturation do not cor- is inherently a light color, and thus seems out of place if
respond logically to the continual increase in the data, shown in the middle of a spectral scheme. Brewer and
Brewer (1994a, 137) indicated this is acceptable “if high her colleagues, however, argued that a satisfactory
saturation colors do not overemphasize unimportant scheme could be developed by progressing from dark
categories.” blue to bright yellow to dark red. Interestingly, Brewer
Brewer suggested that hue differences could also be and her colleagues found that the diverging, sequential,
used for sequential schemes, but that such differences and spectral schemes yielded very similar visualizations
should be subordinate to lightness differences. Color of data, in part because they took great care in
Plate 2A illustrates the middle five classes of a yellow developing these schemes. Color Plate 4 illustrates
to green to purple scheme that she advocated. Brewer some of their schemes applied to the high school
did not recommend a greater range of hues, although education data.
she noted that “sequential schemes may be con- More recently, Harrower and Brewer (2003) developed
structed that use the entire color circle” (as in Color the program ColorBrewer (http://www.ColorBrewer. org)
Plate 2B). In our opinion, the latter implies qualitative to assist those who know little about map design in select-
differences and therefore should not be used for ing appropriate color schemes. ColorBrewer provides
numerical data. three basic schemes: sequential, diverging, and qualitative.
For bipolar data, Brewer (1994a, 139) recommended Sequential schemes “are suited to ordered data that
a diverging scheme, in which two hues diverge from a progress from low to high,” whereas diverging schemes
common light hue or neutral gray. An example con- “put equal emphasis on mid-range critical values and
verging on a light hue is a dark green–greenish extremes at both ends of the data range.” Qualitative
yellow–yellow–orange–dark orange scheme, and an schemes are not relevant to this chapter because they are
example converging on a neutral is a dark red–light intended for qualitative (or nominal) data. A total of 18
red–gray–light blue–dark blue scheme (Color Plate 3 sequential and 9 diverging schemes are possible, with
illustrates the latter). Brewer argued that the greatest sequential schemes ranging from 3 to 9 classes, and diverg-
flexibility is achieved if diverging schemes are thought ing schemes ranging from 3 to 11 classes. Using a hypo-
of as two sequential schemes running end to end, with thetical spatial pattern, ColorBrewer allows users to see
the lightest tone toward the middle. whether individual colors can be easily discriminated on a
Brewer discussed balanced data in association with map (see Color Plate 5). The program also allows users to
bivariate color schemes because balanced data techni- see how the colors will look when overlaid with other
cally involve two attributes. We consider balanced data information (e.g., a road network), and provides an indi-
here because the associated legend appears only slightly cation of whether the color scheme will work in various
more complicated than a legend for a univariate map situations (e.g., projected on a screen or photocopied; see
(e.g., on a five-class map, percent English might be the lower left of Color Plate 5). We consider ColorBrewer
labeled above the legend boxes and percent French again in section 4 when we deal with the details of color
below the legend boxes). Brewer indicated that although specification.
balanced schemes can be created by overlaying two
sequential schemes—she mixed magenta and cyan to * For
the optimal classification of the high school education data, the
create a purple—she recommended using standard median also happens to fall in the middle class.
289
11:54:08.
Choropleth Mapping
3.2 Color Naming 3.3 Color Vision Impairment
In addition to working with kind of data, Brewer and her The various forms of color vision impairment can be
colleagues have utilized several other factors in develop- represented in the CIE color model by a series of
ing their color schemes. One important factor for diverg- confusion lines, or lines along which colors are confused
ing schemes is selecting colors that cannot be confused with one another. Colors running along the confusion
with one another, and thus are readily “named” as distinct lines will be confused by these groups, whereas colors
colors. For instance, if red and pink were assigned to oppo- running roughly perpendicular to the lines should be dis-
site ends of a diverging scheme, a user might have diffi- tinguishable.
culty associating a particular color with the proper end of Using CIE diagrams like the one shown in Figure 2,
the diverging scheme. In contrast, blue and red would not Olson and Brewer (1997) developed sets of confusing
be confused, as blue colors look quite different from red and accommodating color schemes. As an example,
colors and thus are readily named as distinct colors. The Figure 3 portrays their results for a diverging color
second column of Table 2 indicates which color pairs are scheme, representing gains or losses in the number of
easily named (an “OK” appears in the column). manufacturing jobs. For the confusable rendition, gains
are represented by a yellow-green to dark green scheme,
whereas losses are represented by a yellow-orange to red
TABLE 2 Color pairs appropriate for diverging color scheme. Note that these colors all fall along the same
schemes set of confusion lines in the CIE diagram. In contrast,
Confusions the accommodating rendition ranges from light blue
to dark blue and light orange-red to red, and is oriented
Color Color Color Vision Simultaneous
Pair Naming Impairment Contrast
Pink-Red * * OK
Pink-Orange * * OK
Pink-Brown * * OK
Pink-Yellow OK * OK
Pink-Green OK * * 0.8
Pink-Blue OK * OK
Pink-Purple * * OK 0.7
Pink-Gray * * OK
Red-Orange * * OK
Red-Brown * * OK 0.6
Red-Yellow OK * OK
Red-Green OK * * 0.5
Red-Blue OK OK OK
Red-Purple * OK OK y 0.4
Red-Gray OK * OK
Orange-Brown * * OK
0.3
Orange-Yellow * * OK
Orange-Green OK * OK
Orange-Blue OK OK OK 0.2
Orange-Purple OK OK OK
Orange-Gray OK * OK 0.1
Brown-Yellow * * OK
Brown-Green * * OK
Brown-Blue OK OK *
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Brown-Purple * OK OK
Brown-Gray * * OK x
Yellow-Green * * OK
Yellow-Blue OK OK * Protanopes (red-impaired)
Yellow-Purple OK OK OK Deuteranopes (green-impaired)
Yellow-Gray * OK OK
Green-Blue * * OK
Green-Purple OK * * FIGURE 2 Confusion lines for protanopes and deuteranopes
Green-Gray * * OK
drawn on an Yxy 1931 CIE diagram. Colors of similar lightness
Blue-Purple * * OK
Blue-Gray * OK OK
will be difficult to discriminate if they are placed along the same
Purple-Gray * * OK confusion line. (After Olson and Brewer 1997, p. 108.)
Source: Based on Brewer 1996, p. 81.
11:54:08.
Choropleth Mapping
Change in the
Number of Jobs Confusing Rendition Accommodating Rendition
by County Color name x y Y R G B Color name x y Y R G B
Gain > 1000 dark green 0.175 0.756 13 0 145 0 dark blue 0.153 0.064 9 50 50 250
Gain 500 - 999 green 0.268 0.658 30 115 185 0 blue 0.205 0.164 29 150 150 250
Gain 1 - 499 yellow-green 0.390 0.554 61 210 215 0 light blue 0.255 0.261 65 210 210 250
Loss 1 - 499 yellow-orange 0.480 0.484 59 255 190 0 light orange-red 0.413 0.360 47 255 180 180
Loss 500 - 999 orange 0.603 0.385 31 255 110 0 orange-red 0.519 0.380 28 230 130 130
Loss > 1000 red 0.646 0.353 13 190 60 0 red 0.658 0.342 13 200 0 0
No Data gray 0.328 0.411 19 150 150 150 dark gray (not measurable) 90 90 90
0.8 Gain > 1000 0.8
Gain 500 - 999
Gain 1 - 499
0.6 0.6
Loss 1 - 499
No Data
Loss 1 - 499 Loss 500 - 999
y 0.4 y 0.4
Loss 500 - 999
Loss > 1000 Loss > 1000
0.2 0.2 Gain 1 - 499
Gain 500 - 999
Gain > 1000

0 0
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8
x x
FIGURE 3 Confusing and accommodating diverging color schemes for those with red–green vision impairment. Note that colors for
the confusing scheme are arranged along similar confusion lines, whereas colors for the accommodating scheme cross the confusion
lines. The size of the circles reflects the relative magnitude of the CIE Y values. (After Olson, J. M. and Brewer, C. A. (1997). “An
evaluation of color selections to accommodate map users with color-vision impairments.” Annals, Association of American
Geographers, 87, no.1, p. 131. Courtesy of Blackwell Publishing.)
perpendicular to the confusion lines. Not surprisingly, classes (e.g., a dark red could be distinguished from
Olson and Brewer found that color-impaired readers moderate orange and light yellow). Standard spectral
interpreted accommodating schemes more easily. schemes (that progress through red, orange, yellow,
The third column of Table 2 indicates (with an green, and blue) are inappropriate for the color-
“OK”) which diverging color pairs are acceptable for impaired, but careful modification of the standard
the color vision–impaired. If you are showing a scheme can yield suitable colors—for instance,
diverging scheme in a printed publication that will be Brewer (1997) recommended a red, orange, yellow,
widely disseminated, you should consider using these blue-green, blue, and purple-blue as one of several
color pairs. Alternatively, if the maps are to be shown options.
in an interactive environment, then you might wish to If you would like to see how a color scheme you
provide an option for viewers to indicate whether have created will appear to the color vision–impaired,
they are color vision–impaired. Those without impair- we encourage you to experiment with the Vischeck
ment would then be able to take advantage of the Web site (http://www.vischeck.com/vischeck/). There
broader set of easily named colors shown in column 2 you will specify the type of color vision impairment
of the table. that you wish to simulate (e.g., protanope or deutera-
Although the multiple hues present in some nope) and provide a file containing a map that you
sequential color schemes could, in theory, create prob- have created. The site will then show you what the
lems for the color vision–impaired, lightness differ- map will look like to the individual with the specified
ences normally enable proper differentiation of color vision impairment.
291
11:54:08.
Choropleth Mapping
3.4 Simultaneous Contrast 3.6 Color Associations
Simultaneous contrast causes the color of an area to The fact that certain colors are often associated with
shift toward the opponent color of a surrounding color particular phenomena might allow you to pick a logical
(e.g., a gray tone surrounded by green will appear red- color scheme (e.g., in the United States, people associate
dish). Fortunately, most color pairs are not affected by the color green with money). You should be aware, how-
simultaneous contrast (“OK” predominates in the ever, that these associations are cognitive and cultural in
fourth column of Table 2). Unfortunately, note that nature, and thus could change over time and be incon-
many of those color pairs affected by simultaneous consistent among map users. A good illustration of the
trast coincide with easily named color pairs. The fact potential for change over time is the use of blue and red
that a color pair is unacceptable from the standpoint of for cool and warm temperatures. In an early study, Cuff
simultaneous contrast does not necessarily prevent it (1973) concluded that these associations were not effec-
from being used on a particular map. Rather, you need tive, but a later study by Bemis and Bates (1989) found
to examine map patterns to see whether situations arise the associations to be effective. Bemis and Bates sug-
in which simultaneous contrast is likely to be a problem gested that the different results might be a function of
(the program ColorBrewer facilitates this process). the more common use of this scheme since the time of
Cuff’s study and thus users’ greater familiarity with it.
3.5 Map Use Tasks
3.7 Aesthetics
Map use tasks refers to whether a map will be used to
present specific or general information, and whether The aesthetics of a color scheme is an important consider-
the user will acquire this information while looking at ation, regardless of how effective that scheme might be
the map or by recalling it from memory. To examine the otherwise. As an example, consider the study of Slocum
role that color schemes play in map use tasks, Janet and Egbert (1993) in which participants compared tradi-
Mersey (1990) had map readers interpret a spatial dis- tional static maps (in which the map is viewed all at once)
tribution using a variety of color schemes. In Color with sequenced maps (in which the map is built while the
Plate 6 we use the high school graduation data to illus- reader views it).Their study was split into two major parts:
trate the various color schemes that Mersey tested a formal one in which participants viewed one of the two
(whereas she utilized another unipolar data set). For types of maps (static or sequenced) and performed various
specific acquisition tasks, Mersey found that unordered map use tasks, and an informal one in which participants
hues (Color Plate 6A) worked best. This is not viewed both types of maps and were asked to comment on
surprising, given that shades for ordered schemes are them. Slocum and Egbert used a yellow–orange–red
difficult to discriminate and that this discrimination is hue–lightness scheme similar to Mersey’s in the formal
complicated by simultaneous contrast. For general portion, and a lightness-based blue scheme in the informal
acquisition tasks, Mersey found that ordered schemes portion. Although the major purpose of the informal
(Color Plates 6E and 6F) performed better, thus sup- portion was to have participants compare the method of
porting the traditional thinking on using shades of a sin- presentation (static or sequenced), they often commented
gle hue. Mersey focused just on general tasks in her dis- on the color schemes, indicating their preference for the
cussion of results for memory, finding that blue scheme over the supposedly more appropriate
lightness-based schemes outperformed hue-based yellow–orange–red scheme.
schemes, with the hue–lightness scheme (Color Plate Interestingly, studies of color preference outside the
6D) being the best overall. discipline of cartography have also found the color blue
Although Mersey’s study illustrates that certain map appealing. For example, using people’s rating of Munsell
use tasks are more effectively accomplished with partic- color chips, Guilford and Smith (1959) found that colors
ular color schemes, you should realize that it is often were preferred in the order blue, green, purple, red, and
difficult to predict how a map will be used. It is probably yellow. More recently, McManus et al. (1981) found that
more appropriate to utilize color schemes that permit a people’s preferences were more variable than Guilford
broad range of map use tasks. In Mersey’s case, she and Smith implied but that blue and yellow were still the
found that the hue–lightness scheme worked best, scor- most and least preferred, respectively. One limitation of
ing highest on 4 of the 10 tasks tested, and a close sec- such studies is that they tend to focus on particular
ond or third on the remaining tasks. In more recent groups of people. For example, McManus et al. studied
research, Brewer and her colleagues (1997) found that a only “undergraduate members of the University of
broad range of carefully selected color schemes work Cambridge” (653). What might the results have been if
well for a wide variety of map use tasks. the study had been done in other areas of the world?
11:54:08.
Choropleth Mapping
Because color preference varies among individuals, and academic journals often require authors to pay for color
because there are likely to be differences among cul- reproduction, which can cost $500 or more per page.
tures, a key point to remember is that the color schemes If you are employed in a cartographic production
you find attractive might not be attractive to others. laboratory, you generally do not design a map completely
on your own, but rather, you must respond to client
3.8 Age of the Intended Audience requirements or desires. Although you might suggest an
ideal, or optimal, set of colors, clients might also reject
The age (and presumably experience) of the intended these because they find other colors more pleasing or
audience is an important consideration in selecting col- have traditionally used other colors. For example, the
ors. For example, young children would probably not be director of a cartographic production laboratory told us
familiar with the blue–red scheme commonly used for “the school district made us make a boundary map for
representing temperature data. Research by Trifonoff attendance areas in hot pink and blue to ‘match their old
(1994; 1995) suggests that young children have a strong map.’” Although mapmakers can try to dissuade clients
preference for color over black-and-white maps: In from choosing such schemes, they must bear in mind that
comparing four methods of symbolization (gray tones, it’s the clients who are keeping the laboratory in business.
proportional circles, proportional circles coded redun-
dantly with gray tones, and shades of red), children 4 DETAILS OF COLOR SPECIFICATION
overwhelmingly chose the colored symbolization
(shades of red) as most desirable. Although this finding
In the previous section we considered numerous factors
might be true only for young children, Brewer and her
involved in selecting color schemes. Our emphasis was
colleagues (1997) found their participants (college stu-
on the general nature of color schemes, as opposed to
dents) had a strong preference for color maps, and a
detailed color specifications. In this section, we consider
black-and-white scheme was decidedly less preferred.
such details. Because we need to create color schemes
on a variety of soft-copy (on-screen) and hard-copy
3.9 Presentation versus Data Exploration (printing) devices, we have split this section into three
major parts: methods for soft-copy display, methods for
Maps can be used in two basic ways: to present informa- black-and-white printed maps, and methods for colored
tion to others (for communication purposes) and to printed maps.
explore data. In selecting color schemes, we have made
the assumption that the former is our goal. If exploration 4.1 Methods for Soft-Copy Display
is instead our goal, then the previously recommended
schemes can still be used, but a greater variety of It is logical to discuss methods for soft-copy display first
schemes also might be useful, even those that might be because we normally design our maps on-screen.
considered “inappropriate” schemes. For example, John Methods for soft-copy display include (1) color ramp-
Berton (1990, 112) argued: ing and (2) the ColorBrewer software.
The continuous blending of color is not conducive to high-
Color Ramping
lighting artifacts in the interior of the data set. While it is
usually critical that a palette not “double-back” on itself . . . A simple and traditional approach for soft-copy color
certain kinds of banding effects which violate the traditional specification is color ramping, in which users select from
order of the color wheel can provide excellent markers for a color palette endpoints for a desired scheme, and the
transitional areas in the data. computer automatically interpolates values between the
endpoints on the basis of RGB values. As an example,
imagine that a user selects endpoints of white (RBG val-
3.10 Economic Limitations and Client Requirements ues of 255, 255, 255) and black (RBG values of 0, 0, 0),
and wishes to produce a five-class map. Simple interpo-
In an ideal world, there would be no economic limitations, lation between these endpoints yields RGB values of:
and thus the cartographer would have sole responsibility 191, 191, 191; 128, 128, 128; and 64, 64, 64. The problem
for selecting colors. In the real world, of course, this usu- with this approach is that a simple arithmetic increase in
ally is not the case. One obvious economic limitation is a color gun does not correspond to an arithmetic
the expense of color reproduction in book or journal increase in perceived lightness, and thus the resulting
form: Although color might communicate information shades will not appear equally spaced. Although inter-
more effectively than black and white, it might not be polation algorithms might be modified to account for
feasible from an economic standpoint. As an example, this discrepancy, they generally are not thus modified.
As a result, color ramping must be used with caution.
293
11:54:08.
Choropleth Mapping
ColorBrewer
100
As we discussed in the preceding section, ColorBrewer
provides a broad set of color schemes appropriate for
choropleth maps. Brewer developed these schemes
using “her knowledge of the relationships between
s
en
Perceived Blackness
CMYK color mixture and perceptually ordered color
ev
ll
St
spaces, such as Munsell” (Harrower and Brewer 2003, se
30). Because a typical user would not have Brewer’s un
ar
M
ne
knowledge, he or she presumably would have difficulty 50
Li
creating these schemes from scratch, but this problem is
irrelevant because ColorBrewer provides the specifica-
tions for color schemes in a variety of formats, including
CMYK, RGB, hexadecimal, Lab, and ArcView 3.x.
4.2 Methods for Black-and-White Printed Maps

0
Because of the high cost traditionally associated with 0 50 100
producing colored printed maps, considerable research Percent Area Inked
was undertaken on selecting appropriate areal symbols
for black-and-white choropleth maps (see Kimerling
FIGURE 4 Munsell and Stevens’ gray curves for converting
1985 for a comparison of various studies). Such research
desired perceived blackness to percent area inked. The curves
resulted in gray curves or gray scales, which express the
are intended for fine-toned black-and-white areal symbols.
relationship between percent area inked and perceived
(After Kimerling 1985, 137.)
blackness. For smooth shades, or those in which the pat-
tern of marks making up the shade is not apparent,* the
Munsell curve is generally considered appropriate. In
examining this curve (Figure 4), note that at the light Step 3. Interpolate the intermediate perceived black-
end, a small difference in percent area inked equates to ness values. This involves simply adding the contrast
a large difference in perceived blackness, and at the value derived in step 2 to each perceived blackness
dark end, a large difference in percent area inked value, beginning with the lowest. For perceived black-
equates to a small difference in perceived blackness. ness values of 12 and 100, and a contrast value of 22, we
The Munsell curve can be utilized in a four-step process: have: 12 + 22 = 34; 34 + 22 = 56; 56 + 22 = 78;
and 78 + 22 = 100.
Step 1. Pick the smallest and largest perceived black- Step 4. Determine the percent area inked correspond-
ness that you desire. Let’s assume that you select values ing to each perceived blackness value. Using Figure 4,
of 12 and 100 as initial choices. (A value of 12 equates this can be accomplished by drawing a horizontal line
to 6 percent area inked and thus will produce a gray from the perceived blackness value to the Munsell
tone that can be seen as a figure against the white back- curve, and from this point drawing a vertical line to the
ground typically used on printed maps. A value of 100 proper percent area inked value. For perceived black-
will produce a solid black.) ness values of 12, 34, 56, 78, and 100, you should find
Step 2. Determine the contrast between each pair of percent area inked values of approximately 6, 20, 41, 67,
perceived blackness values, assuming the values are and 100.
equally spaced from one another. To accomplish this, Although the Munsell curve is the most widely
divide the range of perceived blackness values by the accepted gray curve for smooth shades, the Stevens
number of classes minus 1. Assuming five classes, and curve (Figure 4) has been argued as more appropriate
the preceding perceived blackness values, the result is: for unclassed maps (Kimerling 1985, 141; MacEachren
PBmax - PBmin 100 - 12 88 1994a, 105). Note that the Stevens curve is displaced
= = = 22 above the Munsell curve; thus, for light gray tones, a
NC - 1 4 4
small difference in percent area inked leads to a very
where PBmax and PBmin are the perceived maximum large difference in perceived blackness, whereas for dark
and minimum blackness, and NC is the number of classes. tones, a very large difference in percent area inked is
required to achieve a small difference in perceived
* Castner and Robinson (1969) found that shades having 75 or more blackness.
marks per inch were perceived as smooth gray tones (the pattern of To understand why the Munsell and Stevens curves
marks was not apparent). might be appropriate for classed and unclassed maps,
11:54:08.
Choropleth Mapping
respectively, it is necessary to consider the methods by Although these equations might appear complex,
which these curves were constructed. The Munsell curve they can be incorporated into a spreadsheet in the fol-
was constructed using partitioning, in which a user lowing three-step procedure:
places a set of areal shades between white and black Step 1. Determine the proportion of the data range
such that the resulting shades appear equally spaced. represented by each data value. To accomplish this, we
For example, in a study that essentially duplicated the subtract the minimum of the data from each raw data
Munsell curve, Kimerling (1975) had people place seven value, and divide the result by the range of the data:
gray tones (chosen from a set of 37) between white and
Xi - Xmin
black. It can be argued that the tones resulting from Zi =
such an approach will have maximum contrast with one Xmax - Xmin
another and thus be appropriate for a classed choro- where Xi is a raw data value, Xmin is the minimum of
pleth map. the raw data, Xmax is the maximum of the raw data, and
In contrast, the Stevens curve was created using Zi is the proportion of the data range. The resulting Z
magnitude estimation, in which a user estimates the values will range from 0 to 1.
lightness or darkness of one shade relative to another. Step 2. Convert the values calculated in step 1 to per-
Stevens and Galanter (1957) described the approach as ceived blackness. This involves computing
follows:
PBi = Zi (PBmax - PBmin ) + PBmin
Typically a particular gray was shown to [the observer] and where Zi , PBmin, and PBmax are defined as before, and
he was told to call it by some number. It was then removed PBi is the perceived blackness for an individual data
and the stimuli were presented twice each in irregular value. (To achieve the 6 to 100 percent area inked used
order. [The observer] was told to assign numbers propor-
in the classed example, you would use PBmin and
tional to the apparent lightness.
PBmax values of 13 and 100 for Munsell, and 22 and
100.8 for Stevens.)*
Some cartographers argue that this process is similar to
Step 3. Insert the perceived blackness values into
how readers compare individual areas on an unclassed
whichever equation (Munsell or Stevens) you wish to
map. Map readers do not, of course, use a number sys-
use. Figure 5 provides a comparison of the effect of the
tem, but they might conceive of ratios of gray tones (that
Munsell and Stevens curves for the unclassed high
a given gray tone is, say, five times as dark as another
school education data set. Note that when the entire
gray tone). It seems more likely, however, that readers
distribution is considered, the Munsell curve produces a
consider only ordinal relations (that is, one tone is con-
slightly darker map. On a more detailed level, the
sidered darker or lighter than another or “much” darker
Munsell curve enables greater discrimination for low
or lighter than another). If this is the case, the Stevens
values, whereas the Stevens curve permits greater dis-
curve appears inappropriate for maps.
crimination for high values.
One problem with applying either the Munsell or
Stevens curve to unclassed maps is that the four-step
process just described is tedious with a large number of Practical Application of Gray Curves
classes. To solve this problem, we digitized the curves You should be aware that if you specify a particular per-
and used curvilinear regression to fit the digitized data cent area inked using software, there is no guarantee
(Davis 2002, 207–214). The results for Munsell and that hardware will actually produce the same percent
Stevens, respectively, are: area inked. (This should be obvious to anyone who has
used laser printers for word processing: Once a toner
PA = .287029 + .329157PB + .008753PB2
cartridge has been used heavily, words will not appear
-6.24654 * 10-7PB4 + 4.16010 as solid black.) The only real solution to this problem is
* 10 - 11 PB6 to analyze available hardware carefully to ensure that
percent-area-inked specifications match software speci-
PA = .465370 + .013841PB2 - 1.26098
fications, and that black marks are actually produced.
* 10 - 4PB3 + 8.38227 * 10 - 11PB6
Even if one has taken a great deal of care in produc-
ing maps within one’s own shop, consideration must
where PB represents a desired perceived blackness and
also be given to what will happen if the map is repro-
PA represents the percent area ink necessary to achieve
duced (e.g., using a fax machine, a photocopy machine,
that level of blackness. The equations produce an excel-
lent fit in the case of Munsell (the maximum error
between predicted and digitized values for percent area * The value of 13 does not match the value of 12 used earlier because
inked is 0.3 percent) and a reasonably good fit in the the equation is an approximation of the curve shown in the graph and
case of Stevens (the maximum error is 3.4 percent). because the graph must be visually interpreted.
295
11:54:08.
Choropleth Mapping
A B
Adults Not Graduating from High School, 1990 Adults Not Graduating from High School, 1990
(Munsell) (Stevens)
Percentage of Population Percentage of Population
14.9 20.1 25.3 30.5 35.7 14.9 20.1 25.3 30.5 35.7
FIGURE 5 Unclassed maps illustrating the effect of the (A) Munsell and (B) Stevens gray curves.
or offset lithography). Most of us are familiar with how This differentiates areas falling in the highest class but
degraded an image becomes as a result of using fax or weakens the boundary between other classes, producing
copy machines. In the case of offset lithography, the a less distinctive “figure.” A third solution is to decrease
basic problem is that individual marks making up an the percent area inked only for boundaries between
areal symbol might be larger on the printed map (i.e., enumeration units falling in the highest class; Figure 6D
than the ones suggested by numerical specifications in a shows the result of doing this with an 80 percent-area-
digital file) because of “overinking or too much pres- inked shade. We thought that the latter approach was
sure between the paper and blanket cylinder” effective in differentiating areas in the highest class, and
(Monmonier 1980, 25). Although this problem might be so chose it for maps in this text.
handled by applying a correction factor similar to that
described earlier for gray scales, it is more realistic to 4.3 Methods for Colored Printed Maps
find a high-quality printer, have a good working rela-
tionship with that printer, and thus try to avoid (or at For colored printed maps, we recommend using the
least limit) the occurrence of this problem. CMYK values presented in ColorBrewer, as Brewer
and her colleagues (2003) took considerable care to
Handling Solid Black Areal Symbols ensure that the colors specified in ColorBrewer would
look appropriate not only on-screen, but also in print.
One of the issues involved in using black-and-white
To see what the color schemes are apt to look like, you
areal symbols is that when a solid black (100 percent
should examine Brewer et al.’s (2003) paper, where a
area inked) is used for the highest class, boundaries of
high-quality printed version of the color schemes
contiguous enumeration units in the highest class disap-
appears.
pear (Figure 6A). One solution to this problem is to
reduce the percent area inked for the highest class (and,
of course, to change the percent area inked for interme- 5 LEGEND DESIGN
diate classes to maintain maximum contrast); for
instance, in Figure 6B we utilized an 80 percent area We consider some finer points of legend design that
inked for the highest class. The problem with this solu- pertain to choropleth maps. Although the design of the
tion is that the contrast between classes is reduced, thus legend might seem a minor point, a map intended for
making it more difficult to determine which class an presentation could fail to communicate if the legend is
enumeration unit is a member of. A second solution is to poorly designed. One issue in legend design is whether
shade all boundaries with a gray tone that appears to be a horizontal or vertical legend should be used (Figure
midway between the two darkest shades (Figure 6C). 7). An argument for a horizontal legend is that its ori-
11:54:08.
Choropleth Mapping
Handling Solid Black Areal Symbols

A B
C D
FIGURE 6 Alternatives for handling solid black areal symbols: (A) solid black is used for the highest class and for all boundaries,
(B) the percent area inked for the highest class is reduced (and intermediate shades are adjusted to maintain maximum contrast),
(C) boundaries are depicted with a shade perceived to be midway between the shades for the two highest classes, and (D) a gray
tone is used only for boundaries between contiguous enumeration units in the highest class.
entation matches the traditional number line; as such, bottom. The decision to use a horizontal or vertical
values should increase from left to right (Figure 7A legend likely will depend on available map space; for
would be appropriate, whereas Figure 7B would be instance, we chose a horizontal design for most choro-
inappropriate). For a vertical legend, high values can be pleth maps in this text because the design seemed to fit
shown either at the top or bottom (Figures 7C and 7D). well with the geographic shapes and column widths we
The logic of showing high values at the top is that peo- were dealing with.
ple associate “up” with “higher”—think of climbing a Numeric values should be placed either at the bot-
mountain: High elevations are at the top. One problem, tom of legend boxes (for a horizontal legend) or to the
however, with having high values at the top is that we right of boxes (for a vertical legend); thus, the design in
normally read from left to right and from top to bot- Figure 8A should be avoided. The logic is that we read
tom; as such, it might seem awkward to find values from left to right and from top to bottom; we normally
increasing from left to right, but decreasing from top to see an areal symbol on the map and want to know the
297
11:54:08.
Choropleth Mapping
A A B C
34
27 - 34 24 - 34
1-2 7 - 12 15 - 19 27 - 34 19
15 - 19 14 - 23
12
B 7 - 12 5 - 13
2
1-2 1-4
1
27 - 34 15 - 19 7 - 12 1-2
FIGURE 9 Methods for indicating class limits: (A) the actual

C D
range of data falling in each class is shown, (B) class limits are
27 - 34 1-2
expanded to eliminate gaps, (C) only the minimum and maxi-
15 - 19 7 - 12 mum data values and the upper limit of each class are shown.
7 - 12 15 - 19
1-2 27 - 34
show the range of data actually occurring in each class
on the map. An advantage of the last approach is
FIGURE 7 Some horizontal and vertical legends. (A), (C), that the reader will have fewer numbers to work with;
and (D) are appropriate, whereas (B) is inappropriate on the other hand, there might be confusion concern-
because the legend limits decrease from left to right (unlike ing the bounds of each class (e.g., in Figure 9C, which
on a number line). class is “19” in?). A hyphen can be used to separate
numeric values, as we have done with the figures here.
If negative values are included in the legend, then the
A B word “to” is frequently used instead to separate
27 - 34
27 - 34
values. Because some of the data sets in this text
15 - 19 include negative values, we generally have used “to”
15 - 19
for consistency.
7 - 12 7 - 12 It might also be desirable to integrate a graphical
1-2
1-2
display with the legend. For instance, Figure 10 shows
two approaches for integrating a dispersion graph.
In Figure 10A, the legend boxes are sized to
FIGURE 8 Inappropriate legend designs: (A) the numeric reflect the range of the data in each class, whereas in
values for classes precede the legend boxes, (B) the legend Figure 10B, the legend boxes are equal in size. A prob-
boxes are not contiguous. lem with the former approach is that the size of the
small legend box for the first class might make it diffi-
associated range of values. Legend boxes should be cult to match the shade with a shade on the map.
placed directly next to one another, as opposed to being
separated by spaces (avoid Figure 8B). The logic is that 6 CLASSED VERSUS UNCLASSED MAPPING
enumeration units are contiguous to one another, so
legend boxes also should be contiguous. Prior to the early 1970s, classed choropleth maps were
Several approaches are possible for specifying class the norm.† Cartographers argued that classed maps were
limits, including: (1) indicating the range of data actu- essential because of the limited ability of the human eye
ally falling in each class, which produces numeric gaps to discriminate shades for areal symbols; also, practically
between classes (e.g., the gap between “1–2” and speaking, classed maps were the only option because of
“7–12” in Figure 9A); (2) eliminating these gaps by the time and effort required to produce unclassed maps
expanding classes (Figure 9B), and (3) indicating the using traditional photomechanical procedures. In 1973,
minimum and maximum data values and the upper the latter constraint was eliminated when Waldo Tobler
limit of each class (Figure 9C). The advantage of the introduced a method for creating unclassed maps using a
first approach is that the reader will know precisely line plotter. Today, unclassed maps can be created using a
the values falling in each class. * The latter two wide variety of hardware devices.
approaches avoid the problem of gaps, but they do not
† Interestingly, the first choropleth map (produced by Charles Dupin

* Notethat this would not be true on an equal-interval map, where the in 1827) was unclassed, but after the 1830s classed maps were much
legend intervals normally do not match the range of data in each class. more common (Robinson 1982, 199).
11:54:08.
Choropleth Mapping
A B
0 10 20 30 40 0 10 20 30 40
1-2 7 - 12 15 - 19 27 - 34 1-2 7 - 12 15 - 19 27 - 34
FIGURE 10 How a graphical display can be integrated with the legend: (A) legend boxes are scaled to reflect the range of data
in each class, (B) legend boxes are constant.
The development of unclassed mapping led to a hotly of portraying the actual data relations, a disadvantage is
contested debate on its merits and demerits (Dobson that for skewed distributions, the ordinal relations in much
1973; 1980a; 1980b; Muller 1980a; Peterson 1980), and of the data might be hidden. For example, in the case of
numerous experimental studies (Peterson 1979; Muller foreign-owned agricultural land, it is difficult to determine
1979; 1980b; MacEachren 1982b; Gilmartin and Shelton that states bordering the Great Lakes have lower values
1989; Mersey 1990; Mak and Coulson 1991). Although than Deep South states. When the data are classed, these
the results of the experimental studies can be helpful in differences become more obvious.
selecting between classed and unclassed maps, we argue
that two criteria should be considered first: (1) whether 6.2 Presentation versus Data Exploration
the cartographer wishes to maintain the numerical rela-
tions among data values, and (2) whether the map is When a map is intended for presentation, normally only
intended for presentation or exploration. In this section, one view of the data can be shown, so the mapmaker must
we consider these two criteria, and then appraise the make a choice between classed and unclassed maps. If,
results of some experimental studies. however, the intention is to explore data, then numerous
options are possible. One is to compare a variety of classi-
6.1 Maintaining Numerical Data Relations fication approaches visually; for example, Color Plate 7,
taken from the program ExploreMap, compares four
To illustrate the notion of maintaining numerical data methods of classification. With the appropriate software,
relations, consider the classed and unclassed maps for such a comparison might also include an unclassed map.
foreign-owned agricultural land and high school gradu- Another option in data exploration is to apply unclassed
ation shown in Figure 11. The optimal method was used shading only to a portion of a data set. For example, for
to create the classed maps because the optimal method the agricultural data, Maine could be assigned a unique
does the best job of minimizing classification error, and symbol, with the remainder of the data displayed using
so, in theory, will produce classed maps that are most the unclassed method (Figure 13). Note how this
similar to the unclassed maps. Shades for the classed approach makes it easier to contrast the pattern of gray
maps were selected using a conventional maximum- tones in states bordering the Great Lakes with the pattern
contrast approach in which tones are perceptually of gray tones in the Deep South.
equally spaced from one another. The approach we have taken in applying unclassed
In contrast, shades on the unclassed maps were made shading to a portion of a data set is similar to Brassel
directly proportional to the values falling in each enumer- and Utano’s (1979) quasi-continuous tone method, in
ation unit, thus maintaining the numerical relations among which the lightest and darkest shades are reserved for
the data (see section 4.2 for the equations). Clearly, the classes of outlying low and high data values and the
maps within each pair look different, with the difference remaining shades are assigned in an unclassed manner
most distinct for the agricultural maps. The difference for to intermediate data. Rather than use the darkest shade
the agricultural maps is a function of the severe skew in for the outlier (Maine), we used a unique symbol so that
the data (examine Figure 12 and note the concentration of it would be clearly distinguished from other values. If
data on the left, with Maine a notable outlier at 1). The color were used, it might be useful to portray outlying
unclassed agricultural map is a spatial expression of this values with gray tones and to apply a continuous color
skew, as we see numerous low-value gray tones and one scheme (e.g., shades of red) to the remaining data. In an
black tone (Maine). In contrast, on the classed map we see interactive environment, it is, of course, possible to make
several different tones of gray, with Maine not appearing such changes dynamically, focusing on any subset of the
quite so distinct. Although unclassed maps do a better job data desired.
299
11:54:08.
Choropleth Mapping
Classed Versus Unclassed Maps

A B
Foreign-Owned Agricultural Land, 1992 Foreign-Owned Agricultural Land, 1992
Percentage of
Privately Owned Land
Percentage of Privately Owned Land
0.0 to 0.4 0.6 to 1.2 1.5 to 3.1 5.2 14.1

0 3.5 7.1 10.6 14.1
C D
Percentage of Population
14.9 to 19.2 19.9 to 22.5 22.9 to 26.1 27.9 to 30.0 31.7 to 35.7
14.9 20.1 25.3 30.5 35.7
FIGURE 11 Optimally classed (A and C) and unclassed (B and D) maps of two attributes: foreign-owned agricultural land and
high school education.
6.3 Summarizing the Results of Experimental Studies by Peterson (1979) and Muller (1980b) also
Studies dealt with classed and unclassed maps, but these studies
are not considered here because they dealt with map
This section summarizes the results of some experimen- comparison. One problem in summarizing the studies is
tal studies that might assist the cartographer in deter- that two of them (MacEachren’s and Mersey’s) did not
mining whether classed or unclassed maps are appro- use unclassed maps, but rather varied the number of
priate. The summary is based on studies by Muller classes (from 3 to 11 and 3 to 9, respectively); thus, in
(1979), MacEachren (1982b), Gilmartin and Shelton the following discussion it is useful conceptually to sub-
(1989), Mersey (1990), and Mak and Coulson (1991). stitute “maps with few classes” for classed maps and
11:54:08.
Choropleth Mapping
0 5 10 15
Percent of Privately Owned Agricultural Land That Is Foreign-Owned
10 15 20 25 30 35 40
Percent of Adults Not Graduating from High School
FIGURE 12 Dispersion graphs for the attributes shown in Figure 11.
is mathematical, not perceptual. Visually matching a

Foreign-Owned Agricultural Land shade on an unclassed map with a shade in the legend is
difficult because there are so many shades and because
their appearance is affected by simultaneous contrast.
Note, however, that for some data sets, unclassed maps
can be useful for visualizing ordinal relations. For exam-
ple, consider ascertaining the ordinal relation of states on
the unclassed education map shown in Figure 11:
Montana appears to have a higher value than Wyoming,
which in turn is higher than Colorado. On the classed
map, such ordering is impossible to determine when the
data fall in the same class (as for these three states).
Percentage of Although ordinal relations can be obtained from
Privately Owned Land unclassed maps, the task can be difficult if enumeration
units are far apart (and thus likely appear in different
contexts), or if a distribution is highly skewed (as with
the foreign-owned agricultural land data). It also must be
0 1.7 3.5 5.2 14.1
realized that the ability to acquire specific information
will depend on the number of classes. For example, on a
FIGURE 13 An unclassed map in which continuous-tone
two-class map the error resulting from classification
shading is applied to a portion of the data set (in this case,
clearly will be greater than the error resulting from incor-
Maine is separated from the remaining data and represented
rect estimation on an unclassed map. Although experi-
by a cross-hatched shade). This is sometimes referred to as
mental studies do not indicate an ideal number of classes,
quasi-continuous tone mapping.
it appears that the error resulting from classification will
be unacceptably large on maps with four or fewer classes.
“maps with many classes” for unclassed maps. The For the recall of specific information, results are
results of the studies can be summarized conveniently inconclusive. MacEachren found that maps with fewer
under the types of information that readers acquire and classes are more effective, whereas Mersey found that
recall from maps: specific and general. although a three-class map is most effective, five- and
seven-class maps do not elicit as accurate a response as
Specific Information a nine-class map.
For the acquisition of specific information, classed maps
generally are more effective. This result might seem sur- General Information
prising because unclassed maps are usually touted as For acquisition of general information, studies gener-
being more accurate (because data are not grouped into ally have revealed no significant difference for classed
classes). The high accuracy of unclassed maps, however, and unclassed maps. Only Mak and Coulson found any
301
11:54:08.
Choropleth Mapping
significant difference, and this occurred only when about classed and unclassed maps using qualitative
individual regions were analyzed; they concluded that methods. Kevin Spradlin (2000) accomplished this by
“for more complex classed maps (six to eight classes) having individuals utilize classed and unclassed maps in
classification may have a distinct advantage over role-playing scenarios—the individuals then rated the
unclassed maps” (121). For recall of general informa- mapping techniques using word pairs (e.g., “easy to use”
tion, MacEachren found no significant relationship vs. “difficult to use”) and participated in open-ended
between the number of classes and subjects’ ability to interviews in which they provided their thoughts and
recall data, and Mersey found that a greater number of opinions on the mapping techniques. Generally, Spradlin
classes decreased recall effectiveness. Based on his found that classed maps were preferred, although he
findings, MacEachren suggested that classification noted that this might have been a function of some of
might be unnecessary for general tasks. In contrast, the tasks performed in the role-playing scenarios.
Mersey noted:
A large number of map symbols, even on a map with regu- SUMMARY

lar surface trends, may create a notion in the mind of the
user that the distribution is more complex than in fact it is. Important issues involved in choropleth mapping
Using many categories to provide the user with more include making certain that the choropleth map is a
detailed specific information may serve as “noise” when
suitable option for a particular data set, deciding
the same user attempts to reconstruct the thematic distrib-
whether or not a data set should be classed, selecting a
ution from memory. (p. 125)
method of classification, choosing a color scheme, and
designing a readable legend. You should use the choro-
pleth map only when you wish the reader to associate
Discussion “typical” values with enumeration units. If the spatial
One problem with interpreting the results of these variation of the underlying phenomenon does not coin-
studies is that usually there was no direct comparison cide with enumeration unit boundaries, and you wish
between classed and unclassed maps. For example, Mak readers to visualize the phenomenon correctly, then
and Coulson asked subjects to divide each map into another mapping technique (e.g., the isopleth or dot
five ordinal-level regions (ranging from “low” to map) is more appropriate. Ideally, the choropleth map
“high”). Rather than comparing the resulting regions should be used only when the size and shape of enu-
directly on classed and unclassed maps, they analyzed meration units are similar; if the units differ greatly in
the consistency for each type of map (classed and size and/or shape, then the resulting pattern should be
unclassed). As a result, it is possible that the consis- treated with caution. There are numerous methods for
tency was similar for both but that the locations of standardizing data to account for differing sizes of enu-
regions differed. meration units, including dividing two area-based raw
Another problem is that only one of the studies totals (e.g., dividing the number divorced by the
(Gilmartin and Shelton) measured processing time. number married), calculating a rate (e.g., the number of
Measuring processing time is important because suicides per 100,000 people), and calculating a summary
classed and unclassed maps could provide the same numerical measure (e.g., the median value of homes).
information, but one map could take less time to Now that unclassed maps can be constructed using
process. (The more rapidly processed map clearly modern computer hardware, the cartographer is faced
would be more desirable.) If classed maps are with the question of whether a classed or unclassed
processed more rapidly, as traditional classification map should be created. In this chapter, we saw that two
proponents such as Dobson argue, and if the informa- criteria can be used to assist in selecting between
tion acquired (or recalled) is identical, an argument can classed and unclassed maps. The first is whether or not
be made for classing the data. If, however, the informa- you wish to portray correct numerical relationships
tion acquired or recalled varies for classed and among the data. If the intent is to maintain correct data
unclassed maps, faster processing is a moot point: In relations, then unclassed maps are appropriate because
this case, the unclassed map might be argued for on the they provide a spatial expression of numerical relations
grounds that it provides a more correct portrayal of the in the data. A second criterion is whether the intention
spatial distribution. is to present or explore data. For presentation pur-
The studies that we have considered focused on a poses, it generally is possible to show only one map of
quantitative assessment of how effectively users could a distribution, and thus one must make a choice
acquire or recall information from classed and unclassed between classed and unclassed maps. If, however, the
maps. A different approach is to assess users’ attitudes intent is to explore data, then several visualizations of
11:54:08.
Choropleth Mapping
the data are possible: a classed map, an unclassed map, Numerous factors should be considered in selecting a
or a focus on a subset of the data. color scheme, including the kind of data (a diverging
Experimental studies for evaluating the effectiveness scheme is clearly appropriate for bipolar data, but can
of classed and unclassed maps have revealed mixed also be used for unipolar and balanced data), the ease of
results. For example, classed maps have been more naming colors (this is important for creating an accept-
effective for helping people acquire specific informa- able diverging scheme), ease of interpretation for the
tion, but neither technique has performed consistently color vision–impaired, and avoiding problems of simulta-
better in terms of helping people acquire general infor- neous contrast. The Web site for ColorBrewer provides a
mation. The experimental studies have been limited by useful approach for specifying particular colors in a
a failure to directly compare classed and unclassed scheme. For black-and-white printed maps, we have seen
maps, by not considering the time necessary for readers how gray curves can be used to select a set of gray tones
to process the map, and by not fully considering users’ that appear to be equally spaced from one another.
thoughts and opinions; thus, it appears that more Legends for choropleth maps should be designed so
research is needed to fully compare the effectiveness of that readers can readily interpret the maps. Either a
classed and unclassed maps. horizontal or vertical legend can be used, depending on
If you decide that a classed map is appropriate, you available map space. Numeric values should progress
first should consider the kind of data you have col- in a logical fashion (for instance, in a horizontal legend,
lected: Bipolar data often have a logical dividing point values should increase from left to right, as on a num-
that can be used to initially split the data into two ber line). The legend should be constructed with the
classes. You should also consider the level of precision presumption that readers will enter the legend to
you desire in your final map before classing the data. determine the numeric value for an areal symbol—
Choosing an appropriate method of data classification thus, in a vertical legend, areal symbols should appear
is a function of several factors, such as whether the to the left of numeric values (i.e., we read from left to
method considers the distribution of the data along the right and thus should encounter the symbols first).
number line, the ease of understanding the legend, and Furthermore, legend boxes should be placed contigu-
whether the method can assist in selecting the number ous to one another because on the map, enumeration
of classes. units are found contiguous to one another.
FURTHER READING
Brewer, C. A. (1994a) “Color use guidelines for mapping and Dobson, M. W. (1980a) “Perception of continuously shaded
visualization.” In Visualization in Modern Cartography, ed. by maps.” Annals, Association of American Geographers 70,
A. M. MacEachren and D. R. F. Taylor, pp. 123–147. Oxford: no. 1:106–107.
Pergamon. An example of some arguments against unclassed maps; counterar-
Discusses color schemes appropriate for various kinds of data guments for unclassed maps can be found in Muller (1979; 1980a).
(unipolar, bipolar, and balanced). Eastman, J. R. (1986) “Opponent process theory and syntax
Brewer, C. A. (1997) “Spectral schemes: Controversial color for qualitative relationships in quantitative series.” American
use on maps.” Cartography and Geographic Information Cartographer 13, no. 4:324–333.
Science 24, no. 4:203–220. Introduces the notion that opponent process theory might be
Presents recommendations for using spectral color schemes. useful in selecting color schemes, and presents an experiment
designed to evaluate the theory.
Cromley, R. G. (1995) “Classed versus unclassed choropleth
maps: A question of how many classes.” Cartographica 32, Kennedy, S. (1994) “Unclassed choropleth maps revisited:
no. 4:15–27. Some guidelines for the construction of unclassed and classed
choropleth maps.” Cartographica 31, no. 1:16–25.
Describes a method for optimally classifying data by treating clas-
sification as an integer programming problem; argues that this Summarizes the debate concerning classed and unclassed choro-
approach can assist in choosing an appropriate number of classes. pleth maps, and provides some guidelines for constructing both
kinds of maps.
Cromley, R. G. (2005) “An alternative to maximum contrast
symbolization for classed choropleth mapping.” The Kimerling, A. J. (1985) “The comparison of equal-value gray
Cartographic Journal 42, no. 2:137–144. scales.” American Cartographer 12, no. 2:132–142.
Compares various gray curves that have been developed.
Raises the issue of whether maximum-contrast symbolization is
appropriate for choropleth mapping, and notes that classed Mak, K., and Coulson, M. R. C. (1991) “Map-user response
equal-interval maps appear similar to unclassed maps. to computer-generated choropleth maps: Comparative
303
11:54:08.
Choropleth Mapping
experiments in classification and symbolization.” Peterson, M. P. (1992) “Creating unclassed choropleth maps
Cartography and Geographic Information Systems 18, with PostScript.” Cartographic Perspectives no. 12:4–6.
no. 2:109–124. Describes an approach based on PostScript that permits 9,999 dif-
An example of a study that compared readers’ ability to interpret ferent gray tones to be created for unclassed choropleth maps.
classed and unclassed maps. This is in contrast to programs such as Freehand that permit only
99 gray tones. Maps in this book were created using Freehand.
Mersey, J. E. (1990) “Colour and thematic map design: The
role of colour scheme and map complexity in choropleth map Spradlin, K. L. (2000) An Evaluation of User Attitudes Toward
communication.” Cartographica 27, no. 3:1–157. Classed and Unclassed Choropleth Maps. Unpublished M.A.
thesis, University of Kansas, Lawrence, KS.
A study of the role of map use tasks in selecting color schemes.
A qualitative study of users’ thoughts about classed and unclassed
Olson, J. M., and Brewer, C. A. (1997) “An evaluation of color
maps.
selections to accommodate map users with color-vision
impairments.” Annals, Association of American Geographers Tobler, W. R. (1973) “Choropleth maps without class inter-
87, no. 1:103–134. vals?” Geographical Analysis 5, no. 3:262–265.
An experimental study of color schemes for the color- The paper that initiated the debate concerning classed and
vision-impaired; includes RGB and Yxy values for suitable schemes. unclassed maps.
GLOSSARY
color copy machine: a device intended for image duplica- Munsell curve: a grayscale commonly used for smooth
tion, but often adapted for use as a printing device. (untextured) gray tones.
color ramping: a method for specifying colors on a graphics partitioning: a method for constructing grayscales in which a
display in which the user selects two endpoints, and the com- user places a set of areal shades between white and black such
puter automatically interpolates intermediate colors. that the resulting shades will be visually equally spaced.
diverging scheme: a sequence of colors for a choropleth or quasi-continuous tone method: a method for unclassed
isarithmic map in which two hues diverge from a common light mapping in which only a portion of the data is shown using
hue or a neutral gray, as in a dark red–light red–gray–light a smooth gradation of tones; very low and very high data
blue–dark blue scheme. values are represented by the lightest and darkest tones,
gray curves (or grayscales): a graph (or equation) express- respectively.
ing the relation between printed area inked and perceived sequential scheme: a sequence of colors for a choropleth or
blackness. isarithmic map in which colors are characterized by a gradual
magnitude estimation: (1) a method for constructing change in lightness, as in varying lightnesses of a blue hue.
grayscales in which a user estimates the lightness or darkness spectral scheme: a sequence of colors for a choropleth or
of one shade relative to another; (2) a method for determining isarithmic map in which colors span the visual portion of the
the perceived size of proportional symbols in which a value is electromagnetic spectrum (sometimes referred to as a ROYG-
assigned to one symbol on the basis of the value assigned to BIV scheme).
another symbol. Stevens curve: a grayscale that is sometimes used for smooth
mathematical scaling: sizing proportional symbols in direct (untextured) gray tones; some cartographers have recom-
proportion to the data; thus, if a data value is 40 times another, mended that it be used for unclassed maps.
the area (or volume) of the representational symbol will be 40
times as large as another representational symbol.
REFERENCES
Anselin, L., Syabri, I., and Kho, Y. (2006) “GeoDa: An intro- Berton, J. A. J. (1990) “Strategies for scientific visualization:
duction to spatial data analysis.” Geographical Analysis 38, Analysis and comparison of current techniques.”
no. 1:5–22. Extracting Meaning from Complex Data: Processing,
Anthamatten, P., and Ziegler, S. S. (2006) “Teaching geogra- Display, Interaction. Proceedings, SPIE, Volume 1259,
phy with 3-D visualization technology.” Journal of Santa Clara, CA, pp. 110–121.
Geography 105:231–237. Bemis, D., and Bates, K. (1989) “Color on temperature
Ariel, A., and Berger, N. A. (2006) Plotting the Globe: Stories maps.” Unpublished manuscript, Department of
of Meridians, Parallels, and the International Date Line. Geography, Pennsylvania State University, University
Westport, CT: Praeger Publishers. Park, PA.
11:54:08.
Choropleth Mapping
Boscoe, F. P., and Pickle, L. W. (2003) “Choosing geographic MacEachren, A. M. (1982b) “The role of complexity and
units for choropleth rate maps, with an emphasis on public symbolization method in thematic map effectiveness.”
health applications.” Cartography and Geographic Annals of the Association of American Geographers 72, no.
Information Science 30, no. 3:237–248. 4:495–513.
Brewer, C. A., MacEachren, A. M., Pickle, L. W., and MacEachren, A. M. (1994a) Some Truth with Maps: A Primer
Herrmann, D. (1997) “Mapping mortality: Evaluating color on Symbolization & Design. Washington, DC: Association
schemes for choropleth maps.” Annals of the Association of American Geographers.
of American Geographers 87, no. 3:411–438. Mak, K., and Coulson, M. R. C. (1991) “Map-user response
Brewer, C. A. (1994a) “Color use guidelines for mapping and to computer-generated choropleth maps: Comparative
visualization.” In Visualization in Modern Cartography, ed. experiments in classification and symbolization.”
by A. M. MacEachren and D. R. F. Taylor, pp. 123–147. Cartography and Geographic Information Systems 18, no.
Oxford, England: Pergamon. 2:109–124.
Brewer, C. A. (1996) “Guidelines for selecting colors for McGranaghan, M. (1989) “Ordering choropleth maps
diverging schemes on maps.” The Cartographic Journal 33, symbols: The effect of background.” The American
no. 2:79–86. Cartographer 16, no. 4:279–285.
Brewer, C. A. (1997) “Spectral schemes: Controversial color McGranaghan, M. (1996) “An experiment with choropleth
use on maps.” Cartography and Geographic Information maps on a monochrome LCD panel.” In Cartographic
Science 24, no. 4:203–220. Design: Theoretical and Practical Perspectives, ed. by C. H.
Brewer, C. A. (2001) “Reflections on mapping Census 2000.” Wood and C. P. Keller, pp. 177–190. Chichester, England:
Cartography and Geographic Information Science 28, no. Wiley.
4:213–235. Mersey, J. E. (1990) “Colour and thematic map design: The
Brewer, C. A., Hatchard, G. W., and Harrower, M. A. (2003) role of colour scheme and map complexity in choropleth
“ColorBrewer in print: A catalog of color schemes for map communication.” Cartographica 27, no. 3:1–157.
maps.” Cartography and Geographic Information Science Monmonier, M. S. (1980) “The hopeless pursuit of purifica-
30, no. 1:5–32. tion in cartographic communication: A comparison of
Castner, H. W., and Robinson, A. H. (1969) Dot Area graphic-arts and perceptual distortions of graytone.”
Symbols in Cartography: The Influence of Pattern on Their Cartographica 17, no. 1:24–39.
Perception. Washington, DC: American Congress on Muller, J.-C. (1979) “Perception of continuously shaded
Surveying and Mapping. maps.” Annals of the Association of American Geographers
Cuff, D. J. (1973) “Colour on temperature maps.” The 69, no. 2:240–249.
Cartographic Journal 10, no. 1:17–21. Muller, J.-C. (1980a) “Perception of continuously shaded
Dobson, M. W. (1973) “Choropleth maps without class maps: Comment in reply.” Annals of the Association of
intervals?: A comment.” Geographical Analysis 5, no. American Geographers 70, no. 1:107–108.
4:358–360. Muller, J.-C. (1980b) “Visual comparison of continuously
Dobson, M. W. (1980a) “Perception of continuously shaded shaded maps.” Cartographica 17, no. 1:40–51.
maps.” Annals of the Association of American Geographers Olson, J. M., and Brewer, C. A. (1997) “An evaluation of
70, no. 1:106–107. color selections to accommodate map users with color-
Dobson, M. W. (1980b) “Unclassed choropleth maps: A vision impairments.” Annals of the Association of
comment.” The American Cartographer 7, no. 1:78–80. American Geographers 87, no. 1:103–134.
Gilmartin, P., and Shelton, E. (1989) “Choropleth maps on Peterson, M. P. (1979) “An evaluation of unclassed crossed-
high resolution CRTs/The effect of number of classes and line choropleth mapping.” The American Cartographer 6,
hue on communication.” Cartographica 26, no. 2:40–52. no. 1:21–37.
Guilford, J. P., and Smith, P. C. (1959) “A system of color- Peterson, M. P. (1980) “Unclassed choropleth maps: A reply.”
preferences.” The American Journal of Psychology 72, The American Cartographer 7, no. 1:80–81.
no. 4:487–502. Robinson, A. H. (1982) Early Thematic Mapping in the History
Harrower, M., and Brewer, C. A. (2003) “ColorBrewer.org: of Cartography. Chicago: University of Chicago Press.
An online tool for selecting colour schemes for maps.” The Robinson, A. H., Sale, R. D., Morrison, J. L., and Muehrcke,
Cartographic Journal 40, no. 1:27–37. P. C. (1984) Elements of Cartography (5th ed.). New York:
Kansas Agricultural Statistics. (1994) Kansas Farm Facts. Wiley.
Topeka: Kansas State Board of Agriculture. Slocum, T. A., and Egbert, S. L. (1993) “Knowledge acquisi-
Kimerling, A. J. (1975) “A cartographic study of equal value tion from choropleth maps.” Cartography and Geographic
gray scales for use with screened gray areas.” The Information Systems 20, no. 2:83–95.
American Cartographer 2, no. 2:119–127.
Spradlin, K. L. (2000) “An evaluation of user attitudes toward
Kimerling, A. J. (1985) “The comparison of equal-value gray classed and unclassed choropleth maps.” Unpublished MA
scales.” The American Cartographer 12, no. 2:132–142. thesis, University of Kansas, Lawrence, KS.
305
11:54:08.
Choropleth Mapping
Stevens, S. S., and Galanter, E. H. (1957) “Ratio scales and Trifonoff, K. M. (1995) “Going beyond location: Thematic
category scales for a dozen perceptual continua.” Journal maps in the early elementary grades.” Journal of
of Experimental Psychology 54, no. 6:377–411. Geography 94, no. 2:368–374.
Trifonoff, K. M. (1994) “Using thematic maps in the early
elementary grades.” Unpublished PhD dissertation,
University of Kansas, Lawrence, KS.
11:54:08.
Isarithmic Mapping
OVERVIEW
Isarithmic maps (the most common form being the the edges of those triangles. Inverse-distance lays an
contour map) depict smooth, continuous phenomena, such equally spaced grid of points on top of control points,
as rainfall, barometric pressure, depth to bedrock, and, of estimates values at each grid point as a function of their
course, the Earth’s topography. After the choropleth map, distance from control points, and then interpolates
the isarithmic map is probably the most widely used the- between grid points. Kriging considers the spatial auto-
matic mapping technique, and is certainly one of the oldest, correlation in the data, both between a grid point and
dating to the eighteenth century. Section 1 considers the surrounding control points, and among control points.
kind of data appropriate for isarithmic mapping. As with Although the mathematical basis underlying kriging is
the proportional symbol map, two kinds of point data are complex, learning about it is useful because the resulting
used in isarithmic mapping: true and conceptual. True maps are often more accurate than those produced by
point data can actually be measured at a point location other interpolation methods (e.g., kriging is sometimes
(e.g., temperatures recorded at weather stations). In con- said to produce an optimal interpolation).
trast, conceptual point data are collected over either an In section 4, we consider six criteria that can be uti-
area or a volume (e.g., murder rates for census tracts), but lized to select an appropriate interpolation method for
the data are considered located at points (e.g., the centroid true point data: (1) correctness of estimated data at con-
of a census tract) for the purpose of symbolization. The trol points (i.e., does the method honor the raw data?),
terms isometric map and isopleth map are used to (2) correctness of estimated data at noncontrol points
describe maps resulting from using true and conceptual (i.e., how well does the method predict unknown
point data, respectively. Like the choropleth map, the iso- points?), (3) ability to handle discontinuities (e.g., geo-
pleth map requires that standardized data be used logic faults), (4) execution time, (5) time spent selecting
to account for the area over which conceptual data are interpolation parameters, and (6) ease of understanding.
collected. Although automated interpolation methods can save
True or conceptual points at which data are collected time (by avoiding manual interpolation), they have limi-
are termed control points. A fundamental problem in tations; in section 5, we’ll look at some of these limita-
isarithmic mapping is interpolating unknown values tions (e.g., “jagged lines” and “spurious details”).
between known values of irregularly spaced control Section 6 covers pycnophylactic interpolation, a
points. Section 2 considers manual interpolation, or technique appropriate for conceptual point data.
interpolation by “eye.” Although manual interpolation Pycnophylactic interpolation begins by raising each
is uncommon today, it often serves as a yardstick against enumeration unit to a height proportional to the value of
which automated interpolation methods are compared. its associated control point. This 3-D surface is then
Section 3 discusses several common automated interpo- gradually smoothed, keeping the volume within each
lation methods for true point data: triangulation, enumeration unit constant. This is accomplished using a
inverse-distance, and kriging. Triangulation fits a set of cell-based smoothing process analogous to generalizing
triangles to control points and then interpolates along procedures used in image processing.
307
11:54:39.
Isarithmic Mapping
In the last section of the chapter, we consider some appropriately mapped by the dot or dasymetric
issues associated with symbolizing data resulting from method.
these interpolation approaches. Basic symbolization As with the choropleth map, the isopleth map
techniques include contour lines, hypsometric tints requires that data be standardized to account for the
(shading the areas between contour lines), continuous- area over which conceptual point data are collected.
tone maps (an unclassed isarithmic map), and fishnet Thus, it would make no sense to map the raw number of
symbolization (a netlike structure that simulates the 3-D people in counties, because a larger county would tend
character of a smooth, continuous surface). An intrigu- to have a higher population total. Methods of standard-
ing possibility is to view a spectral color scheme with ization for isopleth maps are identical to those we have
special glasses that enhance the color stereoscopic effect discussed for choropleth maps.
(in which long-wavelength colors appear nearer than
short-wavelength colors).
2 MANUAL INTERPOLATION
1 SELECTING APPROPRIATE DATA A key problem in isarithmic mapping is that data are
available only at irregularly spaced control points
In considering the kind of data appropriate for isarithmic whereas the phenomenon is presumed to exist through-
mapping, it is important to remember that the underlying out the region of interest. Therefore, it is necessary to
phenomenon is presumed to be continuous and smooth. determine the values of intermediate points through the
Thus, the phenomenon is presumed to exist throughout process of interpolation. Manual interpolation, or inter-
the geographic region of interest and to change gradually polation by “eye,” was the only approach available prior
between individual point locations (as opposed to to the development of computer-based methods, which
abruptly). For instance, think of the mean yearly snowfall did not become popular until the 1970s. Although not
for the state of Michigan. There is potentially a value for commonly practiced today, it is important that you be
every location, and the change between locations is rela- aware of manual interpolation because it is frequently
tively gradual. Phenomena that are largely continuous used as a yardstick against which automated methods
and smooth but that have some discontinuities can be are compared. Manual interpolation is accomplished by
handled with the isarithmic approach, but you must spec- mentally connecting neighboring control points with
ify the discontinuities and have software available that straight lines, and then linearly interpolating along these
will handle them. For instance, a geologist might be inter- lines to create contours (or isolines) of equal value. For
ested in creating an isarithmic map showing the height of instance, in Figure 1 you can see how a contour line of 30
the Dakota sandstone in Kansas based on a series of would be drawn through a set of four control points.
sample wells. If any faults in the bedrock were known, Linear interpolation refers to the idea that the contour
these would have to be specified; otherwise, a smooth line is positioned proportionally between the control
transition between elevation values would be presumed
along the fault zone.
Data for isarithmic mapping are sampled from an
underlying phenomenon at point locations termed 34
control points. As with the proportional symbol map,
two forms of data can be collected at control points: 31
4
true and conceptual. With true point data, values are 1

actually measured at a point location; for instance, 30
weather data, such as snowfall and insolation, are mea- 30
sured at individual weather stations. In contrast, 4
conceptual point data are collected over an area (or
4
volume), but are presumed to occur at point locations;

for instance, we might collect population data for coun-
26
ties, and presume that the population is located at the 4
centroid of counties. Maps produced using true and
30
conceptual point data are termed isometric and 6
36
isopleth maps, respectively. You should be especially
cautious when working with conceptual point data, as
the assumption of continuity might not be met. For
instance, population-related data might be character- FIGURE 1 Manual interpolation involves a linear interpola-
ized by sharp discontinuities that would be more tion between control points.
11:54:39.
Isarithmic Mapping
points (e.g., in Figure 1 the contour line of 30 is 4 units of triangles, given that “simply entering the [control]
from 26 and 1 unit from 31, and so is positioned 4/5 of points in a different sequence could result in contour lines
the distance between 26 and 31). with a conspicuously different appearance.” Fortunately,
A strict definition of linear interpolation produces the this is no longer a problem, as Delaunay triangles provide
angular contour line shown in Figure 1. Because real- a unique solution regardless of the order in which con-
world phenomena generally do not exhibit such angular- trol points are specified. Delaunay triangles are closely
ity, contour lines are smoothed by considering the trend associated with Thiessen polygons, which are formed by
of control point values within the region.This is illustrated drawing boundaries between control points such that all
in Figure 3A for a larger data set. In Figure 3A, also note hypothetical points within a polygon are closer to that
that when neighboring control points are connected, a set polygon’s control point than to any other control point.
of triangles is formed. These triangles are also utilized in For example, in Figure 3B, hypothetical point S is closer
triangulation, one of several automated interpolation to control point C than to any other control point, so point
methods that we consider. When control points are posi- S is part of the Thiessen polygon associated with control
tioned near the corners of a square, it might not be clear point C. Delaunay triangles are created by connecting
where to draw the contours. For instance, in Figure 2A control points of neighboring Thiessen polygons. For
and 2B we see a situation in which the area in the middle example, in Figure 3B, triangle ABC is formed because
of the four control points can be either below or above the Thiessen polygons associated with control points A, B,
the 25-contour line. This problem is handled by averaging and C are all neighbors of one another.
the four control points to create an additional control Once Delaunay triangles are formed, contour lines
point (Figure 2C). are created by interpolating along the edges of triangles
in a fashion similar to manual interpolation. Delaunay
3 AUTOMATED INTERPOLATION triangles are desirable for this purpose because the
FOR TRUE POINT DATA longest side of any triangle is minimized, and thus the
distance over which interpolation must take place is
minimized. As with manual interpolation, a strict linear
In this section, we consider three automated interpola- interpolation along triangle edges leads to angular con-
tion methods appropriate for true point data: triangula- tour lines, as shown in Figure 3C. Smoothing of these
tion, inverse-distance, and kriging. lines is obviously necessary if the underlying phenome-
non is to be properly represented. Although details of
3.1 Triangulation the smoothing procedure are beyond the scope of this
text, the process is the 3-D equivalent of splining, which
Triangulation is logical to address first because it emu- is illustrated later for the inverse-distance method
lates how contour maps are constructed manually. A key (Davis, 2002, 378); the result is shown in Figure 3D.*
step in triangulation is connecting neighboring points to
form a set of triangles that are analogous to those
employed in manual contouring. John Davis (2002, 375) * Interested readers with a mathematical background should consult
indicated that one of the challenges for those developing McCullagh 1988, 757–761 for details on smoothing methods for both
the triangulation approach was determining a “best” set triangulation and inverse-distance methods.
A 17
B 17
C 17
25 25 25
25 25
25
27 Below 25 32 27 Above 25 32 27 22.5 32
25 25 25
25 25 25
14 14 14
FIGURE 2 When the control points to be manually interpolated form a square, a fifth control point is created by averaging the
four original control points.
309
11:54:39.
Isarithmic Mapping
A 26 30 33 35
B B D
25 38 F
22 40 A
41 G
27 36 C E
S
25
39 J
32
I
30 44
30 H M
42 L
40
34 35 K
C 30 35
D 30 35
25 25
40 40
25 25
30 30
40 40
35 35
FIGURE 3 Hypothetical data showing (A) manual contouring, (B) Thiessen polygons and Delaunay triangles, (C) simple linear
interpolation along Delaunay triangle edges, and (D) smoothed interpolation along Delaunay triangle edges.
To assist in comparing triangulation with the inverse- the grid points to actually create a contour line. Because a
distance and kriging approaches, Figure 4 portrays con- grid is used, the inverse-distance method is sometimes
tour maps of all three methods. These maps are based on termed gridding. There is no analogy to this method in
precipitation values for 376 weather stations within manual contouring—a grid is used because “it is much
South Carolina, North Carolina, and Georgia. More easier to draw contour lines through an array of regularly
specifically, the data cover a four-day period associated spaced grid nodes than it is to draw them through the
with Hurricane Hugo in 1989. Although only South irregular pattern of the original points” (Davis 2002, 380).
Carolina is shown in Figure 4, data for adjacent states The term inverse-distance is used because control
were included so that the contour map would be accu- points are weighted as an inverse function of their
rate along the state boundary of South Carolina. distance from grid points: Control points near a grid point
are weighted more than control points far away. The basic
formula for estimating a value at a grid point is:
3.2 Inverse-Distance
n
k
Inverse-distance interpolation involves three steps: a Zi /di
N = i=1
(1) laying a grid on top of the control points, (2) estimating Z n
values at each grid point (or grid node) as a function of k
a 1 / di
distance to control points, and (3) interpolating between i=1
11:54:39.
Isarithmic Mapping
Interpolation Methods
Hurricane Hugo Precipitation, 1989
A Triangulation B Inverse Distance - All Data
2.0
2.
0
2.0
3.0 3.0
2.0 3.
0
0
3.
3.
3.0 0
3.0
4.0
3.0 5.0 4.0
4. 5.0
0 6.0
5.0
9. .0
0 7
0
8.
0
6.
Contour Lines Represent Contour Lines Represent

Inches of Rainfall Inches of Rainfall
C Inverse Distance - Quadrant D Kriging

3.0
2.0
3.0
1.0 1.0
3.0
2.0 2.0
2.0 3.0 2.0
3.
0
0 3.0
3.
3.0
4.0 4.0
3.0
5.0 5.0
6.0
6.0
5.0 0
7.0 7. 8.0
0
6.
Contour Lines Represent Contour Lines Represent

Inches of Rainfall Inches of Rainfall
FIGURE 4 Contour maps of precipitation data associated with Hurricane Hugo: (A) triangulation, (B) inverse-distance using all
data, (C) inverse-distance using a quadrant approach, and (D) kriging. (Data source: Southeast Regional Climate Center.)
where Z N = estimated value at the grid point (Z /d 1 ) + (Z 2 /d 12 ) + (Z3 /d 13 ) + (Z4 /d 14 )

ZN = 1 1
Zi = data values at control points (1/d 11 ) + (1 /d 12 ) + (1 /d 13 ) + (1 /d 14 )
di = Euclidean distances from each control point
to a grid point (40>2.24) + (60>1.00) + (50>1.00) + (40>1.41)
k = power to which distance is raised =
(1>2.24) + (1>1.00) + (1>1.00) + (1>1.41)
n = number of control points used to estimate
a grid point = 49.5
To illustrate, Figure 5 depicts a portion of a hypo- Normally, however, k is set to 2, which cancels the square
thetical grid laid on top of four control points (n = 4), root computation in distance calculations, and thus saves
along with calculations for the distances from each con- computer time (Isaaks and Srivastava 1989, 259).
trol point to the central grid point. If we assume k = 1 With real-world data, the number of grid and con-
(for simplicity in manual computation), then the central trol points obviously will be greater than this simple
grid point value is: hypothetical case. A larger number of control points
311
11:54:39.
Isarithmic Mapping
eight nearest) is used to make an estimate. Quadrant

and Octant strategies also require that control points
Y
fall within an ellipse, but the ellipse is divided into four
and eight sectors, respectively, with a specified number
of control points used within each sector. Surfer also
includes options for the minimum number of points that
6
must fall within the ellipse and the maximum number
Z1 = 40
of empty sectors permitted.
Z2 = 60 To better understand these strategies, the Simple,
4 Quadrant, and Octant strategies are illustrated with
Z3 = 50 Z4 = 40 hypothetical data in Figure 6. For the sake of compari-
son, assume that each strategy requires a total of eight
2 points to make an estimate. Note that the Simple strat-
egy takes the nearest eight points regardless of the
direction of the points; the Quadrant strategy takes the
nearest two points within each of four sectors; and the
0 X Octant strategy takes the nearest point in each of eight
0 2 4 6
sectors. For these hypothetical data, it appears that
either the Quadrant or Octant strategy is preferable to
dig = √(X i – Xg)2 + (Yi – Yg)2
the Simple strategy because the latter does not use any
d1g = √(2 – 4)2 + (5 – 4)2 = 2.24 control points southwest of the grid point.
To further illustrate the effect of the strategies, consider
d2g = √(5 – 4)2 + (4 – 4)2 = 1.00 the contour maps of the Hurricane Hugo data using the
All Data and Quadrant strategies (Figures 4B and 4C).
d3g = √(4 – 4)2 + (3 – 4)2 = 1.00
For the Quadrant strategy, we specified that two control
d4g = √(5 – 4)2 + (3 – 4)2 = 1.41 points be used in each sector, permitted a grid point to be
calculated using as few as six control points, and allowed a
maximum of one sector to have no control points. We
FIGURE 5 Computation of distances for inverse-distance chose these parameters, in part, to demonstrate how the
contouring. Four control points are overlaid with a hypotheti- All Data and Quadrant strategies can lead to quite differ-
cal grid; distance calculations are shown from each control ent-looking maps. Because precipitation data were not
point to the central grid point. available for the ocean surface, the latter two parameters
also ensured that grid points would be calculated near the
land–ocean interface.
leads one to wonder which control points should be For the Quadrant strategy, we also specified differing
used to estimate each grid point. One approach is to ellipse radii (2° latitude and 2.4° longitude) because the
use all control points; in this case, n in the formula for control points were specified in degrees latitude and
ZN is simply the number of control points. This approach longitude, and at this latitude, 1° of latitude is not equal
seems unrealistic, however, because control points far to 1° of longitude.† One could also modify the radii to
away from a grid point are unlikely to be correlated account for the possibility that the phenomenon being
with that grid point. This sort of thinking, and the need mapped is more highly related (autocorrelated) along
to minimize computations, led software developers to one axis than another. For example, the user’s guide for
create various strategies for selecting an appropriate Surfer (Golden Software 2002, 109) suggests that for
subset of control points. temperature data in the upper Midwest of the United
To illustrate the range of strategies, we consider four States, a longer axis in the east–west direction might be
that are available within an early version of Surfer, a appropriate because temperatures are more similar in
common software package for creating isarithmic maps: that direction. For the Hugo precipitation data, one
All Data, Simple, Quadrant, and Octant.* The All Data might consider orienting the ellipse along the direction
strategy uses all control points in the calculation of each of the storm track, assuming that it was constant.
grid point. The Simple strategy requires that control Comparing the All Data and Quadrant maps in
points fall within an ellipse (typically, a circle) of fixed Figure 4, we see that they have similar overall patterns,
size; normally, only a subset of these points (say, the
*These options are possible in the present version of Surfer, but not † Differingellipse radii would not be necessary if the control points
using a single command. were specified in terms of an equidistant projection.
11:54:39.
Isarithmic Mapping
Simple Quadrant Octant
FIGURE 6 Three search strategies used within Surfer, a common contouring package. The cross in the middle represents a grid
point location to be estimated, and the large circle represents the search radius within which control points must be located. Actual
control points selected in each case are shown in black.
but that the All Data map exhibits more gradual changes; for control points: An average of values cannot be lesser
for example, if you compare the north central portion of or greater than any input values.*
the maps, you will note that the distance between the 2.0
and 3.0 contour lines is greater on the All Data map. This 3.3 Kriging
result is not surprising, given that all control points are
included in grid calculations, and thus we would expect The term kriging comes from Daniel Krige, who devel-
estimated values for grid points to be closer to the aver- oped the method for geological mining applications
age of all control point values. It is also interesting to note (Cressie 1990). Kriging is similar to inverse-distance
that the Quadrant map is very similar to the triangulated interpolation in that a grid is overlaid on control points,
map. One wonders whether this would be true with other and values are then estimated at each grid point as a
data sets; we will deal with this question in more detail in function of distance to control points. Rather than simply
section 4. considering distances to control points independently of
In discussing inverse-distance interpolation, we
have focused on grid point calculations because of the * One solution to this problem is to utilize trend surfaces (Davis 2002,
considerable number of options involved. Remember, 384–385).

however, that after grid point calculation, an interpo-
lation must also take place between grid points. As
with triangulation, a simple linear interpolation leads 38.5 36
to angular contour lines; angularity can be avoided
either by increasing the number of grid points (which
increases computations and memory requirements)
or by smoothing the contours. Smoothing is accom- 40 Unsmoothed
40
plished by splining, which involves fitting a mathe- Contour

matical function to points defining a contour line
(Davis 2002, 228–234); this concept is illustrated Smooth
graphically in Figure 7. Spline
One problem with inverse-distance interpolation is
that it cannot account for trends in the data. For example, 46 41
a visual examination of z values for control points might
suggest a value for a grid point higher than any of the
control points, but the inverse-distance method cannot FIGURE 7 Fitting a smooth curve (a spline) to an angular
produce such a value.This problem exists because the for- contour line. (After Sampson 1978, 50; courtesy of Kansas
mula for inverse-distance is a weighted average of z values Geological Survey.)
313
11:54:39.
Isarithmic Mapping
one another, however, kriging utilizes the spatial

autocorrelation in the data, both between the grid point
A
Y
and the surrounding control points and among the con-
trol points themselves. Understanding how spatial auto-
correlation is used in kriging requires an understanding 20
of semivariance and the semivariogram.
Z1 Z2 Z3 Z4 Z5
10
3.4 Semivariance and the Semivariogram 20 30 35 40 45
Consider the simplified graphic shown in Figure 8A, 0 X
0 10 20 30 40 50
which portrays attribute data for five hypothetical,
equally spaced control points. Normally, of course, con-
trol points are not equally spaced, but it is easier to B
understand semivariance if initially we assume that they h
are. Semivariance is defined as: 1 2 3 4
2
(Z1 – Z1 + h) 100 225 400 625
n-h 2
2 (Z2 – Z2 + h) 25 100 225
a (Z i -Z i+ h )
i=1 2
gh = (Z3 – Z3 + h) 25 100
2(n - h) 2
where Zi = values of the attribute at control points (Z4 – Z4 + h) 25
n–h
h = multiple of the distance between control 2
Σ(Zi – Zi + h) 175 425 625 625
points i =1
n = number of sample points 2(n – h) 8 6 4 2
In Figure 8A, the distance between control points is γh 21.88 70.83 156.25 312.50
10; h can take on values of 1, 2, 3, and 4 (that is, dis-
tances of 10, 20, 30, and 40, respectively, are possible);
and n is 5. Note that the largest possible multiple is 4 C
because the maximum possible distance between the 400
control points is 40.
To illustrate computations for semivariance, we can
insert the data from Figure 8A into the preceding for-
300
mula for h = 1.
γh
5-1
2
a (Zi -Zi+1) 200
i=1
g1 =
2(5-1)
100
(Z1 - Z2)2 + (Z2 - Z3)2
+(Z3 - Z4)2 + (Z4 - Z5)2
=
8 0
0 10 20 30 40
(20 - 30)2 + (30 - 35)2
h=1 h=2 h=3 h=4
+(35 - 40)2 + (40 - 45)2
= Distance between control points
8
= 21.88
FIGURE 8 Computation of semivariance using equally
spaced, hypothetical control points: (A) the equally spaced
A summary of these calculations for all values of h is
points, (B) semivariance computations, and (C) semivariogram.
shown in Figure 8B. You should compute the column
for h = 2 to ensure that you understand the method of
computation. be allowed in both distance and direction between
Real-world computation of semivariance requires two control points. For example, consider the control points
modifications to this approach. First, because control shown in Figure 9. If we assume the distance between
points are not normally equally spaced, a tolerance must control points is 5 meters, note that no control point
11:54:39.
Isarithmic Mapping
1m
1m
Sill
20º
A 5m East
γ
20º
Control Points
FIGURE 9 Determining which control points will be used in h Range

semivariance computations. A fixed distance yields no points
directly 5 meters east of point A. A tolerance of 1 meter and
20°, however, yields four control points. (From APPLIED FIGURE 10 Idealized semivariogram illustrating a flattening
GEOSTATISTICS by E. H. Isaaks and R. Mohan Srivastava. in the semivariance values. The distance at which this occurs is
Copyright © 1989 by Oxford University Press, Inc. Used by known as the range and the associated semivariance value is
permission of Oxford University Press, Inc.) the sill.
occurs exactly 5 meters east of point A. However, if we fitting a curve (or model) to the set of points compris-
permit a distance tolerance of 1 meter and an angular ing the semivariogram, a process known as modeling the
tolerance of 20°, then four control points are east of point semivariogram. Once an equation for the model is
A. The second modification is to calculate semivariance determined, a value for distance can be inserted into
in a variety of directions; thus, in addition to computing the equation, and a value for semivariance computed.
along the x axis (east–west direction), computations The simplest model is a straight-line (linear) one; other
should be made in north–south, northwest–southeast, common models include the spherical and exponential
and northeast–southwest directions. For now, we assume (Figure 11).
that such computations are combined to create a single
semivariance value. Kriging Computations
The semivariogram is a graphical expression of how
semivariance changes as h increases. For example, There are two major forms of kriging: ordinary and uni-
Figure 8C is a semivariogram for the hypothetical data versal. We focus on ordinary kriging, in which the mean
shown in Figure 8A. Clearly, as h increases (i.e., as con- of the data is assumed to be constant throughout geo-
trol points become more distant), the semivariance also graphic space (i.e., there is no trend or drift in the
increases. This basic feature is characteristic of most data).* To simplify the discussion, we assume that only
data and should not surprise us: We expect nearby geo- three control points are used to estimate a grid point;
graphical data to be more similar than distant geo- later we will relax this constraint.
graphical data. The behavior of semivariograms with In a fashion similar to inverse-distance, kriging uses a
larger data sets is characterized by the idealized curve weighted average to compute a value at a grid point.
shown in Figure 10. Note that the semivariance For three control points, the equation is
increases as it did in Figure 8C, but eventually reaches a ZN = w1 Z1 + w2 Z2 + w3 Z3
plateau. The value for semivariance at which this occurs
is known as the sill, and the distance at which it occurs where ZN = estimated value at a grid point
is known as the range. The plateau normally indicates Z1, Z2, and Z3 = data values at the control points
that the data are no longer similar to nearby values, but w1, w2, and w3 = weights associated with each control
rather that the semivariance has approached the vari- point
ance in the entire data set. The wi weights are analogous to the 1/d values used in
When using the semivariogram in kriging, it is neces- inverse-distance computations; the formula appears
sary to make an estimate of the semivariance for some
arbitrary distance between points. For example, in * See Davis (2002, 428–437) for a discussion of the more compli-
Figure 8C we might wish to estimate the semivariance cated method of universal kriging, which does account for the trend
for a distance of 26. This is normally accomplished by in the data.
315
11:54:39.
Isarithmic Mapping
the control points themselves (used in calculating

g(hij )); contrast this approach with the inverse-distance
method, which considers only distances between the
grid point and the control points.
To illustrate the nature of these equations, assume
that we are given three hypothetical control point
values and wish to estimate a value for a grid point
(Figure 12). Figure 12B lists the X and Y coordinates
for all points, and Figure 12C lists the distances between
γ pairs of points. Also assume that Figure 12D is an
appropriate linear model for a semivariogram associ-
ated with a larger data set from which the three points
were taken: The model (g = 10h) indicates that the
Linear semivariance is simply 10 times the distance (when h is
5, g is 50).
Spherical
Determining a semivariance for use in the simulta-
Exponential neous equations requires plugging distance values into
the model. For example, for control points 1 and 2, the
h distance is 3.16, and the semivariance is 31.6.
Repeating this process for each pair of points pro-
FIGURE 11 Models commonly used to summarize the points duces the semivariance values shown in Figure 12E.
comprising the semivariogram: linear, spherical, and exponen- Inserting these values into the simultaneous equations,
tial. (After Olea 1994, 31.) we have
w10.00 + w231.6 + w322.4 = 22.4
simpler for kriging because the weights are constrained w131.6 + w20.00 + w322.4 = 10.0
to sum to 1.0. w122.4 + w222.4 + w30.00 = 14.1
In kriging, the weights (wi ) are chosen to minimize w1 + w2 + w3 = 1.0
the difference between the estimated value at a grid
Davis (1986, 385–388) provided details of how these
point and the true (or actual) value at that grid point.
equations can be solved using matrix algebra. We did so
This is analogous to the situation in regression analysis
using the statistical package SPSS, and found the
in which we minimize the difference between the esti-
following values for w : w1 = 0.15, w2 = 0.55, and
mated value of a dependent attribute and its true value.
w3 = 0.30. Inserting these and the values for the con-
In kriging, minimization is achieved by solving for the
trol points into the basic weighted-average formula for
wi in the following simultaneous equations:
kriging, we have
w1g(h11) + w2g(h12) + w3g(h13) = g(h1g)
w1g(h12) + w2g(h22) + w3g(h23) = g(h2g) ZN = 0.15(50) + 0.55(40) + 0.30(25)
w1g(h13) + w2g(h23) + w3g(h33) = g(h3g) = 37
w1 + w2 + w3 = 1.0 Using kriging in real-world situations (e.g., with the
where g(hij ) = the semivariance associated with the dis- Hurricane Hugo data) differs from this simplified
tance between control points i and j (e.g., g(h12 ) = the example in two important respects: (1) more than three
semivariance for control points 1 and 2), and g(hig ) is control points normally are used to estimate each grid
the semivariance associated with the distance between point, and (2) more than one semivariogram might be
the ith control point and a grid point (e.g., g(h1g ) = the appropriate. For the number of control points, a cartog-
semivariance for control 1 and a grid point) (Davis rapher normally specifies a search strategy similar to
1986, 385).* Note that these equations consider not only that used for the inverse-distance method. For example,
the distance from control points to the grid point (used Edward Isaaks and Mohan Srivastava (1989, Chapter
in calculating g(hig )), but also the distances between 14) suggested using three points in each quadrant, for a
total of 12 points overall, and a search radius slightly
larger than the average spacing between control points.
*
For the Hurricane Hugo data, we followed Isaaks and
We reference Davis’s (1986) second edition here because we feel it
is more appropriate for the introductory student. Please be aware, Srivastava’s recommendations (using the same ellipse
however, that more recent information on kriging is available in radii as for the Quadrant map). Additionally, we
Davis’s (2002) third edition. permitted a grid point to be calculated using as few as
11:54:39.
Isarithmic Mapping
six control points, and a maximum of one sector to have

no control points; remember that these parameters
were necessary to have grid points estimated along the
land–water boundary. The resulting map (Figure 4D)
appears very similar to both the Quadrant and
Triangulated maps. Again, one wonders whether the
similarity among these maps is a function of this
particular data set, or whether these various interpola-
tion methods yield similar results for a wide variety
of data.
More than one semivariogram might be appropriate
for a particular data set because the nature of spatial
autocorrelation can vary as a function of the direction
of semivariance calculation (e.g., east–west vs.
north–south directions), as in the upper Midwest tem-
perature example discussed when considering the
inverse-distance method.* For the Hurricane Hugo pre-
cipitation data, two semivariograms might have been
useful if we had wished to consider the direction of hur-
ricane movement. For example, if over the four-day
period of rainfall the hurricane had moved in a constant
direction, we might have used one semivariogram par-
allel to the movement and one perpendicular to it.
Because we did not have data regarding hurricane
movement, we chose a single semivariogram: a linear
model based on a visual examination of a sample semi-
variogram.
Although kriging is admittedly more complex than
other methods of interpolation, it can produce a more
accurate map; as such, it is often said to produce an
optimal interpolation. It must be stressed that this is
true only if one has properly specified the semivari-
ogram(s) and associated semivariogram models. This
is important to recognize because software for kriging
might provide simplified defaults (e.g., a single semi-
variogram and linear model) that will not produce an
optimal kriged map. Another advantage of kriging is
that it provides a measure of the error associated with
each estimate, known as the standard error of the esti-
mate. This error measure can be used to establish a
confidence interval for the true value at each grid
location, assuming a normal distribution of errors. For
example, if the kriged estimate at a grid point is 2.3
inches of rainfall and the corresponding standard
error is 0.5 inch, then a 95 percent confidence interval
for the true value at that grid point would be
2.3 ; 2(0.5). This is equivalent to saying we are
FIGURE 12 Semivariance computations for a grid 95 percent certain that the true value is within the
point and three control points: (A) graph of the grid range 1.3 to 3.3.†
point and control points; (B) X and Y coordinates
for the points; (C) distances between points; (D) lin-
* It might also be necessary to fit different models (equations) to each
ear model of a semivariogram for a data set from
which the points were presumed to have been semivariogram (Isaaks and Srivastava 1989).
taken; (E) semivariances associated with distances †For a more sophisticated approach to providing a measure of error
between points. (After Davis 1986, 387.) associated with kriging, see Chainey and Stuart 1998.
317
11:54:39.
Isarithmic Mapping
4 CRITERIA FOR SELECTING AN INTERPOLATION 4.2 Correctness of Estimated Data

METHOD FOR TRUE POINT DATA at Noncontrol Points
So far in the chapter, we have discussed three methods Two basic approaches have been used for evaluating
for interpolating true point data: triangulation, inverse- correctness of data at noncontrol points: cross-validation
distance, and kriging. Given the similarity of the maps and data splitting. Cross-validation involves removing
resulting from applying these three methods to the a control point from the data to be interpolated, using
Hurricane Hugo data, how does one decide which other control points to estimate a value at the location of
method should be used? This section considers six crite- the removed point, and then computing the residual, the
ria that can be considered in selecting an interpolation difference between the known and estimated control
method: (1) correctness of estimated data at control point values; this process is repeated for each control
points (i.e., does the method honor the raw data?), point in turn. If cross-validation is done for a variety of
(2) correctness of estimated data at noncontrol interpolation methods, the resulting sets of residuals can
points (i.e., how well does the method predict unknown be compared. In data splitting, control points are split
points?), (3) ability to handle discontinuities (e.g., geo- into two groups, one to create the contour map, and one
logic faults), (4) execution time, (5) time spent selecting to evaluate it. Residuals are computed for each of the
interpolation parameters, and (6) ease of understanding. control points used in the evaluation stage. One problem
For the present discussion, we assume that the goal is to with data splitting is that it is impractical with small data
create an isarithmic map depicting values of a smooth, sets because it makes more sense to use as many control
continuous phenomenon. If the intention is to map vari- points as possible to create the contour map. Ideally, you
ous derivative measures of elevation (e.g., slope and should use either cross-validation or data splitting to
drainage), then triangulation is the natural choice evaluate the accuracy of any contour maps that you cre-
(Clarke 1995, 144–148). ate. Because this can be time-consuming, it is useful to
consider the results of those who have already done this
sort of analysis. The results of two such studies are sum-
4.1 Correctness of Estimated Data at Control Points marized next.*
Isaaks and Srivastava (1989, 249–277, 313–321,
One advantage often attributed to triangulation is that 338–349) evaluated triangulation, inverse-distance, krig-
estimated values for control points will be identical to ing, and other methods for a set of digital elevation val-
the raw values originally measured at those control ues in the Walker Lake area near the California–Nevada
points; this is known as honoring the control point data. border. One important conclusion of their study was that
For example, for the triangulated map shown in the “‘best’ [interpolation method] depends on the yard-
Figure 3D, the contour line for 30 passes directly stick we choose” (272). For example, triangulation pro-
through control point H, which has a raw value of 30. duced a smaller standard deviation of residuals than an
In contrast, the inverse-distance and kriging methods inverse-distance method (using all control points within
do not, in general, honor control points. An exception a specified search ellipse), but the inverse-distance
would be when control and grid points coincide: In this method minimized the size of the largest residual.
case, the distance between control point and grid point Overall, their study revealed that kriging is best
will be zero, and the value for the grid point can simply (318–321), although it was only slightly more effective
be made equivalent to the control point value. than an inverse-distance approach in which four control
Although honoring control point data is worthwhile, points were required in each quadrant surrounding a
using it as the key criterion for evaluating interpolation grid point.
is risky for two reasons. First, the process presumes that Franky Declercq (1996) evaluated triangulation,
data are measured without error; it is common knowl- inverse-distance, and kriging for two types of data: one
edge that physical recording devices (e.g., a rain gauge) relatively smooth and one more abrupt in nature.
are imperfect, and their readings are also subject to Smooth data consisted of mean annual hours of sun-
human error. Rather than exactly honoring control shine for the European Union, and abrupt data
points, it makes sense that such estimates should be consisted of soil erodability (K values) in northern
within a small tolerance of the raw data. A second Belgium. One of Declercq’s major conclusions was that
problem with honoring control points is that the there is little difference between inverse-distance and
process does not necessarily guarantee correctness at kriging methods. He argued that the number of control
noncontrol point locations. For example, in an evalua- points used to make an estimate is more important than
tion of the inverse-distance method, Davis (1975)
found that options providing good estimates at control *For other studies that have compared interpolation methods, see
points did poorly at noncontrol points. Mulugeta (1996).
11:54:39.
Isarithmic Mapping
the general interpolation method: He recommended using cross-validation or data splitting. This time can
few control points (four to eight) for smooth data, and be eliminated by using the evaluations of others, such
many (16 to 24) for more abrupt data. He also recom- as those described previously, although more studies
mended using either a quadrant- or an octant-sector are necessary to cover the broad range of data types
approach. Lastly, he recommended not using triangula- that geoscientists deal with. Ease of understanding
tion approaches because they produce “highly inaccu- refers to the ease of comprehending the conceptual
rate values and erratic images . . . in poorly informed basis underlying the interpolation method. Clearly,
regions or with abruptly changing data” (p. 143). kriging scores poorly on this criterion, as the kriging
literature (both published articles and software user
4.3 Handling Discontinuities guides) is often complex.
In addition to honoring the data, another advantage of 5 LIMITATIONS OF AUTOMATED INTERPOLATION

triangulation is its ability to handle discontinuities, such APPROACHES
as geologic faults or cliffs associated with a topographic
surface. McCullagh (1988, 763) indicated that “If the
locations of these special lines are entered as a logically Although automated interpolation is desirable because
connected set of points, the triangulation process . . . it eliminates the effort involved in manual interpola-
will automatically relate them to the rest of the data.” tion, a study by Gebeyehu Mulugeta (1996) illustrated
He further indicated that although grids can handle dis- some potential limitations of automated interpolation.
continuities, recognizing a fault in gridded data is Mulugeta compared automated and manual interpola-
almost impossible (p. 766). tion of two data sets: precipitation totals for a single
storm event (June 28, 1982) in a five-county area of
Michigan, and bedrock-surface altitudes for a township
4.4 Execution Time in Michigan. These data were automatically interpo-
lated via kriging and manually interpolated by experts
The amount of computer time it takes to create a con- (climatologists and meteorologists for the precipitation
tour map can be an important consideration for large data and geomorphologists for the bedrock-surface alti-
data sets. In general, a simple linear interpolated trian- tude data). A quantitative analysis using data splitting
gulation is faster than either inverse-distance or kriging revealed that the computer-interpolated precipitation
because of the numerous computations that must be map was significantly more correct than the manually
done at individual grid points with the latter methods. interpolated one, and that there was no significant dif-
Kriging is especially time-consuming because of the ference between computer and manually interpolated
simultaneous equations that must be solved ( n + 1 bedrock-surface maps. Together these results suggested
simultaneous equations must be solved for each grid that the computer-interpolated maps were at least as
point, where n is the number of control points). good as, if not better than, the manual ones.
Smoothed triangulation methods, however, can take Comments elicited from experts, however, raised seri-
substantially longer than either inverse-distance or ous questions about the appropriateness of automated
kriging methods (McCullagh 1988). interpolation. Criticisms noted by the experts can be
understood by comparing the computer-interpolated
4.5 Time Spent Selecting Parameters map of the storm event with a couple of the manually
and Ease of Understanding interpolated maps created by the experts (Figure 13).
The most significant problem noted by experts was that
Time spent selecting parameters refers to how long it “isolated peaks and depressions [on the computer-inter-
takes to make decisions such as the appropriate power polated map were] overly smoothed and too circular.”
for distance weighting, the number of control points to (Consider A1–A5 in Figure 13A.) The experts would
use in estimating a grid point, which model to use for have preferred “elongated bands . . . characteristic . . .
the semivariogram, and which smoothing method to of storm-precipitation surfaces” (p. 335). A second prob-
use for triangulation. One can avoid spending a large lem was that the computer-interpolated map portrayed
amount of time by simply taking program defaults, but “features . . . not warranted by the data points” (335).
such an approach is risky. Kriging is the most difficult An example was the peak at B4 in the middle of the
in this regard because of the need to consider whether computer-interpolated map, which “many experts
one or more semivariograms are appropriate and what showed . . . as a saddle between two elongated 2[-inch]
sort of model should be used to fit each semivariogram. isarithms [Figure 13B] or enclosed . . . in a 1.5[-inch]
The time spent selecting parameters will obviously isarithm [Figure 13C]” (335). Other problems noted by
become longer if one wishes to evaluate correctness experts on the computer-interpolated map included
319
11:54:39.
Isarithmic Mapping
“jagged lines” and “spurious details” (D1–D4 and F1–F4

Computer-Interpolated vs. Manually in Figure 13A).
Interpolated Precipitation Maps Mulugeta indicated that, ideally, mapmakers should
edit automated contour maps to alleviate such prob-
A Computer-Interpolated lems. Unfortunately, if the mapmaker does not have
F1
expertise in the domain being mapped, this task will be
C3
difficult; for example, those unfamiliar with storm pre-
D2
cipitation surfaces would not likely see the need for
D3 D1 B3 “elongated bands.” Other problems, such as “jagged
D4
lines” and “spurious details,” are capable of being
A1 B4
II I I
edited by a person without domain expertise, and thus
III
III
A3
II I III IIIIIIIIIII
III
II
1.0
III II I
III I
A2 B2 we encourage you to correct such problems within
II I
A4
II I
I
I I I I I I I I I I II
1.5
II I I
B1
C1 common design programs such as Illustrator.
IIII
II I
IIIII
IIII
III
III
IIII
III A5
III III I
III
III
5
III C2
II I
1.
I I IIIII
II II
III III V.I. = 0.5”

6 TOBLER’S PYCNOPHYLACTIC APPROACH:
II I
F2 F3 F4
0
AN INTERPOLATION METHOD FOR

1.
15 Miles
CONCEPTUAL POINT DATA
B Manually Interpolated Although the interpolation methods discussed in

section 3 were developed to handle true point data, they
C3 also have been commonly used for conceptual point
data (and thus for isopleth mapping). We accomplished
0
1.
this by assigning the percentage of wheat harvested to

the centroids of each county, which served as control
1.0
points for interpolation.

1.5
Waldo Tobler’s (1979) pycnophylactic (or volume-

1.0
1.0 preserving) method is a more sophisticated approach to
1.0 V.I. = 0.5”
handling conceptual point data.To visualize this method,
consider the standardized data associated with enumer-
15 Miles
ation units as a clay model in which each enumeration
unit is raised to a height proportional to the data. The
objective of the pycnophylactic method is to “sculpt this
C Manually Interpolated surface until it is perfectly smooth, but without allowing
any clay to move from one [enumeration unit] to
another and without removing or adding any clay”
1.0 (520).* Relating this concept to the other interpolation
approaches we have considered, we can think of volume
preservation as a form of honoring the data associated
with each enumeration unit.
To illustrate the pycnophylactic method in more
1.5
detail, we use a simplified algorithm developed by
1.0 V.I. = 0.5” Nina Lam (1983, 148–149). This algorithm illustrates
1.0
15 Miles the pycnophylactic character of Tobler’s method, but
not the smoothness constraints that Tobler specified;
for the latter, we refer you to his original paper. To
FIGURE 13 A comparison of (A) a computer-interpolated illustrate Lam’s algorithm, presume that we are given
contour map and (B – C) manually interpolated contour maps. raw counts (RCi ) for the three hypothetical enumera-
(After Mulugeta, G. (1996) “Manual and automated interpolation units shown in Figure 14A. These raw counts
tion of climatic and geomorphic statistical surfaces: An evalu-
ation.” Annals, Association of American Geographers, 86, no. 2, *See http://www.ncgia.ucsb.edu/pubs/gdp/pop/pycno.html for a visual
p. 335. Courtesy of Blackwell Publishing.) representation of pycnophylactic interpolation.
11:54:39.
Isarithmic Mapping
FIGURE 14 Basic steps of pycnophylactic (volume-preserving) contouring: (A) three hypothetical enumeration units; (B) square
cells overlaid on the enumeration units; (C) initial density values for each cell (computed by dividing the raw count for each enu-
meration unit by the number of cells in that unit); (D) smoothed cell values (achieved by averaging neighboring cells); (E)
smoothed values adjusted so that the sum within an enumeration unit equals the original total sum for that enumeration unit (the
volume is preserved). (After Lam 1983, 148–149.)
might be the number of people or the acres of wheat within that unit; this can be determined by a
harvested in each enumeration unit. In step 1 of Lam’s so-called point-in-polygon test (Clarke 1995, 207–209).
algorithm, a set of square cells is overlaid on top of the Note that four cells fall within enumeration unit E1 .
enumeration units, and it is determined which cells fall In step 2 of the algorithm, a raw count for each cell is
in each enumeration unit (Figure 14B). A cell is con- determined by dividing the raw count for each enu-
sidered part of an enumeration unit if its center falls meration unit by the number of cells in that unit; for

(pycnophylactic) (kriging)
30 30 10 10
20 30
20
20
20
30
10
30
30
20
30
30
10
10
40
30 10
50 40 50
20
20
20 60
70
Contour Lines Represent Pecentage of Land Area Contour Lines Represent Pecentage of Land Area
FIGURE 15 Comparison of interpolation methods for data collected for enumeration units: (A) the pycnophylactic approach,
which expressly deals with the fact that the data were collected from enumeration units; (B) the kriging approach, which treats the
areal data as conceptual point locations.
321
11:54:39.
Isarithmic Mapping
example, cells within enumeration unit 1 receive a similar characteristics in surrounding enumeration
value of 25/4, or 6.25 (Figure 14C). Note that this step units, as when contouring wheat harvest data for the
essentially standardizes the data by computing a den- state of Kansas.
sity measure. To contrast pycnophylactic interpolation with meth-
Steps 3 to 5 of Lam’s algorithm are executed in an ods intended for true point data, Figure 15 portrays con-
iterative fashion. The steps are as follows: tour maps of the percentage of wheat harvested in
Step 3. Each cell is computed as the average of its counties of Kansas, using both the pycnophylactic and
nondiagonal neighbors. For example, cell (2,2) becomes kriging methods. Although the two maps appear simi-
lar, there are some notable differences. One is that the
6.25 + 6.25 + 7.00 + 5.00
= 6.13 magnitude of the highest contour on the pycnophylactic
4 map is 70, whereas the highest contour on the kriged
The results for all cells for the first iteration are shown map is only 50. A higher-valued contour appears on the
in Figure 14D. This step accomplishes the smoothing pycnophylactic map because of its volume-preserving
portion of the algorithm. character. To understand this, imagine each of the coun-
Step 4. The cell counts within each enumeration unit ties in the form of the clay model described by Tobler
at the end of step 3 (Figure 14D) are added to obtain a earlier. In such a model, you would find that Sumner
total smoothed count value, SCi , for each enumeration County is highest, with a raw standardized data value of
unit. For example, the total for enumeration unit 1 is 58.5 percent. To show a smooth transition to lower-
valued surrounding counties, Sumner’s edges must be
SC1 = 6.25 + 6.50 + 5.83 + 6.13 = 24.71
beveled off, but to retain the same volume, its center
The results for enumeration units 2 and 3 are 32.40 and portion must be built up, thus resulting in a value above
39.39, respectively. 58.5 in its interior. In contrast, the kriging method we
Step 5. All cell values are multiplied by the ratio used cannot produce a value higher than any control
RCi / SCi . For cell (2,2), the result is point because each grid point is a weighted average of
surrounding control points.*
6.13 * (RC1/SC1 ) = 6.13 * (25/24.71) = 6.20
Although the pycnophylactic method is arguably
The results for all cells for the first iteration are shown more appropriate for isopleth mapping than are point-
in Figure 14E. based interpolation methods, it must be emphasized that
Note that if the counts in each cell of an enumera- the method should be used only with continuous phe-
tion unit in Figure 14E were added, the resulting sum nomena. If the phenomena are not continuous, nothing
would be equal to the original raw count for that unit; is gained by using the pycnophylactic approach. Instead,
for example, for enumeration unit 1, we have it might be more appropriate to use another method,
such as the dasymetric map.
6.32 + 6.58 + 5.90 + 6.20 = 25.00
As a result, steps 4 and 5 enforce the pycnophylactic 7 SYMBOLIZATION
(volume-preserving) constraint. Remember that steps 3
to 5 are executed in an iterative fashion. For the second
The last step in isarithmic mapping is to symbolize the
iteration, the results shown in E would be placed where
interpolated data. When mapping the Earth’s topogra-
C currently is depicted in Figure 14, and steps 3 to 5
phy, a variety of specialized symbolization techniques
would be executed again. Iteration continues until there
are possible. In this chapter, we consider symbolization
is no significant difference between the raw and smooth
approaches that are appropriate for either topographic
counts for each enumeration unit or there is no signifi-
or nontopographic phenomena.
cant change in the cell values compared with the last
iteration.
7.1 Some Basic Symbolization Approaches
One issue stressed by Tobler and not dealt with in
this algorithm is how the boundary of the study area is
A range of basic symbolization approaches is shown in
handled. His computer program for pycnophylactic
Figure 16. Contour lines (Figure 16A) frequently have
interpolation, PYCNO, provides two options: one in
been used to depict smooth, continuous phenomena, par-
which zeros are presumed to occur outside the bounds
ticularly when maps were produced manually because
of the region, and one in which a constant gradient of
change is presumed across the boundary. The former
would be appropriate if the region is surrounded by
water, as when mapping population along the coasts of * Universal kriging can estimate values beyond the range of the
the United States. The latter would be appropriate data, but it would not necessarily honor the data associated with
when mapping a phenomenon that is presumed to have the enumeration unit.
11:54:39.
Isarithmic Mapping
A B
2.0 2.0
3.0
3.0
1.0 1.0
2.0 2.0
2.0 3. 2.0 3.
0 0
3.0 3.0
3.0 3.0
4.0 4.0
3.0 3.0
5.0 5.0
6.0 6.0
5.0 0 5.0 0
7. 8.0 7. 8.0
0
0
6.
6.
C
10.0
5.0
137º
0.0
D E
2.0
3.0
1.0
2.0
10.0 3.
2.0 0
9.0
3.0
8.0
7.0
3.0
6.0 4.0
3.0
5.0 5.0
4.0 6.0
5.0 0
3.0 7. 8.0
0
6.
2.0
1.0
0.0
FIGURE 16 Comparison of several methods for symbolizing smooth, continuous phenomena: (A) using only contour lines;
(B) combining contour lines and hypsometric tints; (C) a fishnet symbolization; (D) continuous tones analogous to those
used on unclassed choropleth maps; (E) continuous tones combined with contour lines.
323
11:54:39.
Isarithmic Mapping
little production effort was involved. An obvious prob- . . . slope directions between points on the surface,
lem with contours, however, is that visualizing the surface estimat[e] relative surface values, and estimat[e]
requires careful examination of the position and values absolute values at points on the surface” (p. 282). Based
of individual contour lines—that is, the surface does not on the accuracy with which people completed these
simply pop out at you. tasks, Kumler and Groop found that continuous-tone
The addition of hypsometric tints (or shaded areas) maps were most effective. Also, when asked to pick their
between contour lines (Figure 16B) enhances the abil- favorite approach, the majority of people selected one
ity to visualize a 3-D surface because light and dark of the continuous-tone maps (a full-spectral scheme).
tints can be associated with low and high values, respec- Although Kumler and Groop’s study suggests that
tively. This visualization can be enhanced further by continuous-tone maps are particularly effective, the
using one of the color schemes that we recommended study was limited because it did not include noncontinu-
for choropleth maps instead of gray tones. Note that a ous-tone color maps. Thus, an interesting follow-up to
broader range of color schemes is possible on isarithmic their study would be a comparison of continuous-tone
maps because symbols of similar value must occur adja- and noncontinuous-tone color maps.
cent to one another (i.e., because the order of symbols The 3-D look of fishnet symbolization can also be
on the map matches the order in the legend, the legend achieved by using stereo pairs and anaglyphs, both of
needs to be consulted less frequently). Hypsometric which permit stereo views (Clarke 1995, 273–274). With
tints can also be shown without contour lines, as is com- stereo pairs, two maps of the surface are viewed with a
mon on TV weather maps.* stereoscope, which you might be familiar with from
One problem with hypsometric tints is that the limited examining 3-D views of aerial photographs. With
number of tones suggests a stepped surface, rather than anaglyphs, two images are created, one in green or
the smooth one that occurs in reality. This problem can blue, and the other in red; these images are viewed with
be ameliorated by creating a continuous-tone map anaglyphic glasses, which use colored lenses to produce
(Kumler and Groop 1990), in which each point on the a 3-D view. In their book Infinite Perspectives: Two
surface is shaded with a gray tone (or color) proportional Thousand Years of Three-Dimensional Mapmaking,
to the value of the surface at that point (Figure 16D). Brian and Jeffrey Ambroziak (1999) described a spe-
This approach is analogous to unclassed choropleth map- cialized anaglyph approach—the Ambroziak Infinite
ping, in which enumeration units are shaded with a gray Perspective Projection (AIPP). In contrast to tradi-
tone proportional to the data value in that unit. One tional anaglyphs, in which the image is viewed from
problem with interpreting a continuous-tone map is that only a single vantage point, the AIPP permits viewing
it is difficult to associate numbers in the legend with par- from multiple positions. Additionally, AIPP is “self-
ticular locations, but this problem can be solved by over- scaling” (that is, vertical exaggeration increases as one
laying continuous tones with traditional contour lines moves away from the image), permits features that
(Figure 16E). both rise above and descend below the image plane,
Another approach that assists in interpreting and permits the inclusion of additional data that are
smooth, continuous phenomena is fishnet symboliza- not part of the main anaglyph image (Ambroziak and
tion (Figure 16C). With this approach, not only does the Ambroziak 1999, 86–87). Images viewed using the
surface change gradually, but we can also actually “see” AIPP produce a striking 3-D appearance—to see this,
that certain points are higher or lower than others. we encourage you to examine the images found in the
Although a fishnet symbolization is useful, it has the Ambroziaks’ book.
same disadvantages as a prism map, including the block-
age of low points by higher ones, and the fact that rota- 7.2 Color Stereoscopic Effect
tion might produce a view unfamiliar to readers who
normally see maps with north at the top. The color stereoscopic effect refers to the notion that
When choosing among the various symbolization colors from the red portion of the electromagnetic spec-
methods shown in Figure 16, it is useful to consider a trum appear slightly nearer to the viewer than colors
study by Mark Kumler and Richard Groop (1990) that from the blue portion, primarily because the lens of the
evaluated black-and-white maps similar to those shown eye refracts light as a function of wavelength (Travis
in Figure 16, plus several continuous-tone color maps. 1991, 135–139). This effect can be used as an argument
People were asked to complete the following tasks while for utilizing a spectral color scheme (red, orange, yel-
looking at the maps: “locat[e] surface extrema, interpret low, green, blue, indigo, and violet). Cartographers nor-
mally recommend that spectral schemes be used with
* For a discussion of the use of color on weather maps, see Monmonier care, because yellow (an inherently light color) appears
(1999a). in the middle of the scheme.
11:54:39.
Isarithmic Mapping
J. Ronald Eyton has discussed several approaches for used.* In particular, based on the human spatial vision
enhancing the color stereoscopic effect, and thus literature, they argued that a lightness-based scheme
overcoming the problem with yellow. In his early work, (which they termed a “luminance-varying” scheme)
Eyton (1990) focused on how spectral colors should be should be used for high-frequency data, and a satura-
generated and how they might be combined with con- tion-based scheme should be used for low-frequency
tours and hill shading to enhance the color stereoscopic data. We illustrate this notion in Color Plate 2, where the
effect. He indicated that, ideally, the subtractive CMYK images depict the amount of ozone in the upper tropos-
process should not be used to create spectral colors phere and lower stratosphere in the spring in the
because the process is not purely subtractive, as “sub- Southern Hemisphere. For all images, the amount of
tractive ink dots . . . adjacent to each other produce ozone is depicted by the height of the surface—this
desaturated colors that are formed additively” (21). As allows us to see the nature of the spatial distribution
one solution to this problem, Eyton created maps on a regardless of which color scheme is used. In Color Plate
computer graphics display and then produced hard 2A, a traditional spectral scheme is used—concurring
copies by recording the images onto film. As another with cartographers, Rogowitz and Treinish argued
solution, he printed maps using fluorescent inks, which against using such a scheme. In Color Plate 2B, a light-
produced “brilliant, intense color” (23–24). Maps result- ness scheme is used, which allows you to see the high-
ing from his use of fluorescent ink can be found in Plates frequency artifacts of atmospheric circulation. In Color
1 to 4 of his paper. Plate 2C, a saturation-based scheme captures the broad
In spite of taking these careful approaches, Eyton found depressed region (of ozone depletion) in the middle (a
little color stereoscopic effect when spectral colors were low-frequency characteristic). Finally, in Color Plate 2D,
used alone. (He found that the effect could be enhanced a combined lightness-saturation scheme captures both
with a reading magnifier, but the desired order of the high- and low-frequency features.
colors was distorted, with green appearing above yellow
and orange, instead of below them). Eyton did find, how-
ever, that when contour lines were added, users perceived SUMMARY
a distinct color stereoscopic effect. Although Eyton could
find no explanation for this enhancement in the literature, In this chapter, we have focused on isarithmic maps,
he hypothesized that the rapid change in slope associated which are used to depict smooth, continuous phenom-
with closer contours might provide depth cues. ena. Two basic forms of data are utilized for isarithmic
Eyton achieved a particularly dramatic color stereo- mapping: true point data and conceptual point data.
scopic effect when he combined spectral colors with a True point data can actually be measured at point loca-
hill-shaded display. (Hill shading is a process in which tions (e.g., the hours of sunshine received over the
terrain is shaded as a function of its orientation and course of the year), whereas conceptual point data are
slope relative to a presumed light source, typically from collected over areas (or volumes), but conceived as
the northwest.) Eyton indicated that many viewers being located at points (e.g., birthrates for census tracts
found a combination of spectral colors, contours, and within an urban area). The terms isometric map and
hill shading to be most effective for enhancing the color isopleth map are used to describe maps resulting from
stereoscopic effect. using true and conceptual point data, respectively.
In later work, Eyton (1994) described how the A basic problem in isarithmic mapping is interpolating
color stereoscopic effect could be enhanced via spe- values between known control point locations associated
cial viewing glasses. Originally developed by with either true or conceptual point data.We covered three
Steenblik (1987), these glasses are now available interpolation methods appropriate for true point data: tri-
through http://www.chromatek.com/. The glasses can angulation, inverse-distance, and kriging. Triangulation fits
produce dramatic 3-D images when spectral colors a set of triangles to control points and then interpolates
are combined with contours or hill shading. We along the edges of these triangles, in a fashion analogous to
encourage you to acquire a set of glasses and examine manual contouring. Delaunay triangles are commonly used
Color Plate 1, which was created using a spectral because the longest side of any triangle is minimized, and
scheme developed by Eyton (1990, 24–26). thus the distance over which interpolation must take place
is minimized. Triangulation is advantageous in that it hon-
7.3 Spatial Frequency and the Use of Color ors the original control points, handles discontinuities well,
and is an obvious choice for elevation data when informa-
Bernice Rogowitz and Lloyd Treinish (1996) suggested tion such as slope and drainage are also desired. A disad-
that the spatial frequency of the data (whether the data
change rapidly or slowly over space) should determine *See Lloyd Treinish’s home page (http://www.research.ibm.com/
whether a lightness- or saturation-based color scheme is people/l/lloydt/) for other related papers.
325
11:54:39.
Isarithmic Mapping
vantage of triangulation is the execution time required to splitting. In cross-validation, a control point is removed
create smoothed contours. from the data, and other control points are used to inter-
Inverse-distance interpolation lays an equally spaced polate a value at the location of the eliminated control
grid of points on top of control points, estimates values at point. If this process is repeated for each control point,
each grid point as a function of distance from control then a set of residuals can be computed for each control
points, and then interpolates between grid points. The point. In data splitting, control points are split into two
term inverse-distance comes from the fact that nearby groups, one to create the contour map and the other to
control points are weighted more than distant points in evaluate it. Residuals are computed for each of the control
calculating a grid point. Because points far away from a points used in the evaluation stage. Because the process of
grid point are apt to be uncorrelated with the grid point, evaluating residuals can be time-consuming, an alternative
it makes sense to use a limited set of control points in cal- is to use the results of past studies such as Declercq’s.
culating a grid point. Rather than simply considering the Although automated interpolation methods for true
nearest control points, it also makes sense to consider the point data clearly avoid the effort needed for manual
direction of the control points from the grid point (e.g., an interpolation, they are not without problems. Experts
octant search can be used). An advantage of inverse-dis- note that automated interpolation methods do not con-
tance is that large data sets can be efficiently interpolated. sider the nature of the underlying phenomenon; for
Kriging is similar to inverse-distance in that a grid is example, an automated interpolation of precipitation data
overlaid on control points and values are estimated at is not likely to consider the “elongated bands . . . charac-
each grid point as a function of distance to control teristic . . . of storm-precipitation surfaces.” Automated
points. Rather than considering distances to control interpolation methods also tend to produce “jagged lines”
points independently of one another, however, kriging and “spurious details” not warranted by the data. One
considers the spatial autocorrelation in the data, both solution to these problems is to graphically edit contour
between the grid point and the surrounding control maps using design software such as Illustrator.
points, and among the control points themselves. The This chapter also considered one method for handling
spatial autocorrelation in the data is evaluated by cre- conceptual point data: Tobler’s pycnophylactic interpola-
ating a semivariogram, which illustrates changes in the tion.The pycnophylactic method begins by assuming that
semivariance (variation in the data associated with each enumeration unit is raised to a height proportional
control points) with increasing distance between con- to the value of its associated control point. This 3-D sur-
trol points. Like inverse-distance, kriging calculates an face is then gradually smoothed, keeping the volume
estimate at a grid point by using a weighted average of within each individual enumeration unit constant; the
nearby control points. The weights, however, are not smoothing is accomplished using a cell-based generaliza-
simply a function of the distance of the control points tion process analogous to procedures used in image
from the grid point, but rather consider the nature of processing. Although pycnophylactic interpolation is
the semivariogram computed for the control points. appropriate for conceptual point data (and thus isopleth
An advantage of kriging is that it can produce an maps), it should be used only if the assumption of a
optimal interpolation (if the semivariograms and asso- smooth, continuous phenomenon seems reasonable.
ciated semivariogram models are properly specified). Methods for symbolizing interpolated data include
Disadvantages of kriging are the longer execution contour lines, hypsometric tints (shading the areas
times for large data sets, the time required for the between contour lines), continuous-tone maps (a form
mapmaker to select the appropriate semivariograms of unclassed map), and fishnet symbolization (a netlike
and associated models, and the difficulty of under- structure that simulates the 3-D character of a smooth,
standing the method. continuous surface). An intriguing possibility is to view a
Ideally, you should evaluate the accuracy of interpola- spectral color scheme with special glasses that enhance
tion methods by examining residuals (the difference the color stereoscopic effect (in which long-wavelength
between known and estimated control point values). This colors will appear to be nearer than short-wavelength
can be accomplished using either cross-validation or data colors).
FURTHER READING
Bucher, F. (1999) “Using extended exploratory data analysis Burrough, P. A., and McDonnell, R. A. (1998) Principles of
for the selection of an appropriate interpolation model.” In Geographical Information Systems. Oxford: Oxford University
Geographic Information Research: Trans-Atlantic Perspectives. Press.
ed. by M. Craglia and H. Onsrud, pp. 391–403. London: Taylor
& Francis. Chapters 5 and 6 cover numerous methods of interpolation.
Proposes an expert system for selecting an appropriate method of

interpolation.
11:54:39.
Isarithmic Mapping
Cressie, N. (1990) “The origins of kriging.” Mathematical Miller, E. J. (1997) “Towards a 4D GIS: Four-dimensional
Geology 22, no. 3:239–252. interpolation utilizing kriging.” In Innovations in GIS 4, ed. by
Discusses the origins of kriging within a variety of disciplines. Z. Kemp, pp. 181–197. Bristol, PA: Taylor & Francis.
Cressie, N. A. C. (1993) Statistics for Spatial Data. Rev. ed. New Our present chapter deals with interpolation methods appropri-
York: Wiley. ate for 21⁄2-D phenomena in which one z value is associated with
each x and y location. Miller’s work considers interpolation
An advanced treatment of statistical methods for handling spatial methods for true 3-D phenomena in which multiple z values are
data; includes an extensive section on kriging. associated with each x and y location. Miller also considers spa-
Davis, J. C. (2002) Statistics and Data Analysis in Geology. 3rd tial data with a temporal component. Also see Miller (1999).
ed. New York: Wiley. Mitas, L., and Mitasova, H. (1999) “Spatial interpolation.”
Triangulation, inverse-distance, and kriging methods are discussed In Geographical Information Systems, Volume 1, ed.
on pages 370–397, 254–265, and 416–443. For kriging, the less by P. A. Longley, M. F. Goodchild, D. J. MaGuire, and
statistically inclined student might wish to consult Davis’s (1986) D. W. Rhind, pp. 481–492. New York: Wiley.
second edition. A review of various methods of spatial interpolation.
DeLucia, A. A., and Hiller, D. W. (1982) “Natural legend Monmonier, M. (1999a) Air Apparent: How Meteorologists
design for thematic maps.” The Cartographic Journal 19, no. Learned to Map, Predict, and Dramatize Weather. Chicago:
1:46–52. The University of Chicago Press.
Examines a novel approach to designing legends for isarithmic
Pages 223–226 discuss the use of color on weather maps; also see
maps; for a critique of this paper, see Paslawski (1983).
Monmonier (1999b).
Isaaks, E. H., and Srivastava, R. M. (1989) Applied
Olea, R. A. (1999) Geostatistics for Engineers and Earth
Geostatistics. New York: Oxford University Press.
Scientists. Boston: Kluwer Academic.
A widely referenced text on interpolation methods; the focus is on
Provides a mathematical treatment of kriging; a full understand-
kriging.
ing requires a solid mathematical background, but the book con-
Journal of Geographic Information and Decision Analysis 2, tains many useful thoughts on kriging.
no. 2, 1998 (entire issue). Robeson, S. M. (1997) “Spherical methods for spatial interpo-
This special issue deals with advanced issues in interpolation. lation: Review and evaluation.” Cartography and Geographic
Kumler, M. P. (1994) “An intensive comparison of triangulated Information Systems 24, no. 1:3–20.
irregular networks (TINs) and digital elevation models Reviews and evaluates spherical methods for interpolation, which
(DEMs).” Cartographica 31, no. 2:1–99. account for curvature of the Earth and are essential when inter-
polating global-scale phenomena. Also see Raskin et al. (1997).
Triangulation and an equally spaced grid are two common
approaches for representing elevation (topographic) data. This Tobler, W., Deichmann, U., Gottsegen, J., and Maloy, K. (1997)
study compares the effectiveness of these two approaches. “World population in a grid of spherical quadrilaterals.”
Lam, N. S. (1983) “Spatial interpolation methods: A review.” International Journal of Population Geography 3:203–225.
Cartography and Geographic Information Systems 10, Describes the development of a subnational database of popula-
no. 2:129–149. tion for the world based on pycnophylactic interpolation.
An overview of a variety of interpolation methods, many of which Veve, T. D. (1994) An Assessment of Interpolation Techniques
are not covered in this chapter. for the Estimation of Precipitation in a Tropical Mountainous
Mackaness, W., and Steven, M. (2006) “An algorithm for local- Region. Unpublished M.A. thesis, Pennsylvania State
ized contour removal over steep terrain.” The Cartographic University, University Park, PA.
Journal 43, no. 2: 144–156. Compares several interpolation methods, including inverse-dis-
Introduces an automated approach for removing portions of tance, kriging, and co-kriging. The latter permits the inclusion of
contour lines to ensure the clarity of the resulting contour infor- ancillary attributes in the kriging process (e.g., elevation in the
mation. case of interpolating precipitation in a mountainous area).
McCullagh, M. J. (1988) “Terrain and surface modelling Webster, R., and Oliver, M. A. (2001) Geostatistics for
systems: Theory and practice.” Photogrammetric Record 12, Environmental Scientists. Chichester, England: Wiley.
no. 72:747–779. A textbook on kriging for the application-minded environmental
scientist.
Describes and compares grid-based (using inverse-distance or
kriging) and triangulation interpolation methods.
327
11:54:39.
Isarithmic Mapping
GLOSSARY
anaglyphs: two images are created, one in green or blue, and inverse-distance: a method of interpolating data for isarith-
the other in red; when viewed through special anaglyphic mic maps; a grid is overlaid on control points, and estimates of
glasses, a three-dimensional view is produced. values at grid points are an inverse function of the distance to
color stereoscopic effect: occurs when colors from the long- control points.
wavelength portion of the electromagnetic spectrum appear isarithmic map: a map in which a set of isolines (lines of
nearer to a map reader than colors from the short-wavelength equal value) are interpolated between points of known value,
portion. as on a map depicting the amount of snowfall in a region.
conceptual point data: data that are collected over an area isometric map: a form of isarithmic map in which control
(or volume) but conceived as being located at one point for points are true point locations (e.g., snowfall measured at
the purpose of symbolization (e.g., the number of microbrew- weather stations).
eries in each state). isopleth map: a form of isarithmic map in which control
continuous-tone map: a map that uses a large number of points are associated with enumeration units (e.g., assigning
hypsometric tints to create an unclassed isarithmic map. population density values for census tracts to the centers of
contour lines: these represent the intersection of horizontal those tracts and then interpolating between the centers).
planes with the three-dimensional surface of a smooth, con- kriging: a method of interpolating data for isarithmic maps
tinuous phenomenon. that considers the spatial autocorrelation in the data.
contour map: see isarithmic map. manual interpolation: in isarithmic mapping, the estimation
control points: points that are used as a basis for interpola- of unknown values by “eye.”
tion on an isarithmic map, such as weather station locations. optimal interpolation: a term sometimes applied to kriging,
cross-validation: a method for evaluating the accuracy of which attempts to minimize the difference between the esti-
interpolation for isarithmic mapping; involves removing a mated value at a grid point and the true (or actual) value at
control point from the data to be interpolated and using other that grid point.
control points to estimate a value at the location of the ordinary kriging: a form of kriging in which the mean of the
removed point. data is assumed to be constant throughout geographic space
data splitting: a method for evaluating the accuracy of inter- (i.e., there is no trend or drift in the data).
polation for isarithmic mapping; control points are split into point-in-polygon test: a test that is intended to determine
two groups, one to create the contour map, and one to evalu- whether a point falls inside or outside a polygon.
ate its accuracy. pycnophylactic interpolation: a method of interpolating
Delaunay triangles: the type of triangles commonly used in data for isopleth maps; the data associated with enumeration
triangulation, a method of interpolation for isarithmic maps; units are conceived as prisms, and the volume within those
the longest side of any triangle is minimized and thus the dis- prisms is retained in the interpolation process.
tance over which interpolation takes place is minimized. semivariance: a measure of the variability of spatial data in
fishnet map: a map in which a fishnet-like structure provides various geographic directions (e.g., we might want to deter-
a three-dimensional symbolization of a smooth, continuous mine whether temperature values are more similar to one
phenomenon. another along a north–south axis than along an east–west
fluorescent inks: specialized printing inks that produce bril- one).
liant, intense color. semivariogram: a graphical expression of semivariance; the
gridding: a term applied to the inverse-distance method of distance between points is shown on the x axis, and the semi-
interpolation in which values are estimated at equally spaced variance is shown on the y axis.
grid locations; it should be borne in mind, however, that splining: using a mathematical function to smooth a contour
kriging also uses a grid. line.
hill shading: see shaded relief. stereo pairs: a term describing when two maps of an area
honoring the control point data: a term applied if, after an are viewed with a stereoscope, which enables the reader to
interpolation is performed, there is no difference between the see a three-dimensional view.
original value of a control point and the value of that same Thiessen polygons: polygons enclosing a set of control
point on the interpolated map. points such that all arbitrary points within a polygon are
hypsometric tints: the shaded areas sometimes used closer to the control point associated with that polygon than
between contour lines on an isarithmic map (e.g., using shades to any other polygon.
of blue between contour lines to depict increasing rainfall). triangulation: (1) a method of surveying where two points
interpolation: in isarithmic mapping, the estimation of having known locations are established and the distance
unknown values between known values of irregularly spaced between them is measured; angles from this baseline are then
control points. measured to a distant point, creating a triangle, which can then
11:54:39.
Isarithmic Mapping
serve as a basis for other triangles; (2) a method of interpolat- universal kriging: a form of kriging that accounts for a trend
ing data for isarithmic maps in which original control points (or drift) in the data over geographic space.
are connected by a set of triangles.
true point data: data that can actually be measured at a
point location, such as the number of calls made at a tele-
phone booth over the course of a day.
REFERENCES
Chainey, S., and Stuart, N. (1998) “Stochastic simulation: An McCullagh, M. J. (1988) “Terrain and surface modelling
alternative interpolation technique for digital geographic systems: Theory and practice.” Photogrammetric Record
information.” In Innovations in GIS 5, ed. by S. Carver, pp. 12, no. 72:747–779.
3–24. London: Taylor & Francis. Monmonier, M. (1991a) “Cartography at Syracuse
Clark, W. A. V., and Hosking, P. L. (1986) Statistical Methods University.” Cartography and Geographic Information
for Geographers. New York: Wiley. Systems 18, no. 3:205–207.
Davis, J. C. (1986) Statistics and Data Analysis in Geology Monmonier, M. (1999a) Air Apparent: How Meteorologists
(2nd ed.). New York: Wiley. Learned to Map, Predict, and Dramatize Weather. Chicago:
Davis, J. C. (2002) Statistics and Data Analysis in Geology University of Chicago Press.
(3rd ed.). New York: Wiley. Mulugeta, G. (1996) “Manual and automated interpolation of
Declercq, F. A. N. (1996) “Interpolation methods for scat- climatic and geomorphic statistical surfaces: An evalua-
tered sample data: Accuracy, spatial patterns, processing tion.” Annals of the Association of American Geographers
time.” Cartography and Geographic Information Systems 86, no. 2:324–342.
23, no. 3:128–144. Olea, R. A. (1994) “Fundamentals of semivariogram estima-
Davis, J. C. (1975) “Contouring algorithms.” AUTO-CARTO tion, modeling, and usage.” In Stochastic Modeling and
II, Proceedings of the International Symposium on Geostatistics, ed. by J. M. Yarus and R. L. Chambers, pp.
Computer-Assisted Cartography, pp. 352–359. 27–35. Tulsa, OK: American Association of Petroleum
Geologists.
Eyton, J. R. (1990) “Color stereoscopic effect cartography.”
Cartographica 27, no. 1:20–29. Sampson, R. J. (1978) SURFACE II Graphics System.
Lawrence, KS: Kansas Geological Survey.
Eyton, J. R. (1994) “Chromostereoscopic maps.” Cartouche
Autumn/Winter Special Issue:15. Steenblik, R. A. (1987) “The chromostereoscopic process: A
novel single image stereoscopic process.” True Three-
Golden Software, Inc. (2002) Surfer 8 User’s Guide. Golden,
Dimensional Imaging Techniques and Display
CO: Author.
Technologies. Proceedings, SPIE, Volume 761, pp. 27–34.
Isaaks, E., and Srivastava, R. M. (1989) Applied Geostatistics.
Tobler, W. R. (1979) “Smooth pycnophylactic interpolation
New York: Oxford University Press.
for geographical regions.” Journal of the American
Kumler, M. P., and Groop, R. E. (1990) “Continuous-tone Statistical Association 74, no. 367:519–536.
mapping of smooth surfaces.” Cartography and
Travis, D. (1991) Effective Color Displays: Theory and
Geographic Information Systems 17, no. 4:279–289.
Practice. London: Academic.
Lam, N. S.-N. (1983) “Spatial interpolation methods: A
review.” Cartography and Geographic Information Systems
10, no. 2:129–149.
329
11:54:39.
11:54:39.
Dasymetric Mapping
OVERVIEW
The dasymetric map is an alternative to the choropleth map Section 3 considers the work of Jeremy Mennis and
when the phenomenon is not uniformly distributed within Torrin Hultgren, who have developed an automated
enumeration units. Like the choropleth map, a dasymetric dasymetric approach known as intelligent dasymetric
map displays standardized data using areal symbols, but mapping (IDM). This approach utilizes both the car-
the bounds of the symbols do not necessarily match the tographer’s domain knowledge and the relationship
bounds of enumeration units (e.g., a single enumeration between the enumeration units and the ancillary infor-
unit might have a full range of gray tones representing dif- mation. The relationship between the enumeration units
fering levels of population density). and ancillary information is evaluated using one of three
Section 1 considers the kinds of data necessary for methods: a containment method, in which enumeration
dasymetric mapping. Although you ultimately will create units are entirely contained within an ancillary class; a
a standardized map, you will need raw totals for each centroid method, in which the centroids of enumeration
enumeration unit so that you can redistribute these raw units fall within an ancillary class; and a percent cover
totals within the enumeration units. In a fashion similar method, in which the user specifies a threshold percent-
to dot mapping, redistribution is accomplished through age and enumeration units are considered whose per-
the use of ancillary information, such as a land use/land centage covered by an ancillary class equals or exceeds
cover map. For instance, if making a map of population that threshold
density, you would likely set the population density of a In Section 4, we consider the development of
water land cover category to 0, presuming that people do LandScan, a global population database that was devel-
not live on the water (although in some areas of the oped at the Oakridge National Laboratory under the
world, this would not be true!). direction of Jerry Dobson and his colleagues. LandScan
Section 2 considers the work of Cory Eicher and provides population estimates for grid cells approxi-
Cynthia Brewer, who developed some early approaches mately 1 square kilometer in area (and finer in the
for automated dasymetric mapping and contrasted the United States); such estimates are critical for determining
effectiveness of these approaches. In their simplest populations at risk in the event of natural and manmade
approach, known as the binary method, they divided disasters. In a fashion similar to Eicher and Brewer and
land use/land cover categories into two groups: habitable Mennis and Hultgren, LandScan utilizes land use/land
and uninhabitable. In a more complex limiting variable cover information, but also has incorporated additional
method based on the classic work of John K. Wright, ancillary information, such as roads, slope, populated
Eicher and Brewer created multiple habitable categories places, and nighttime lights.
and set maximum density thresholds for some categories. Although dasymetric mapping can eliminate some of
They found that the limiting variable method produced the error found in a choropleth map, the resulting map
significantly better results than the binary method. may become so detailed that it is difficult to interpret.
331
16:07:42.
Dasymetric Mapping
One solution to this problem is to generalize the dasy- proportion of area covered by the class and the likelihood
metric map. We discuss this solution in section 5 by con- that horses will occur in that class. In some cases, the
sidering an approach developed by Mitchel Langford likelihood that horses will occur within a class will be
and David Unwin. obvious (e.g., we don’t expect horses to appear in the
water), but in others it may not be so obvious (e.g., the
1 SELECTING APPROPRIATE DATA AND ANCILLARY density of horses in a forested region would be a function
INFORMATION of the percent forest cover). In the following sections, we
will learn about some different approaches that have
been used to determine the likelihood of a phenomenon
Like the choropleth map, a dasymetric map uses areal
occurring within particular ancillary classes.
symbols to represent presumed zones of uniformity, but
unlike the choropleth map, the bounds of zones need not
match enumeration unit boundaries (Figure 1). Also like 2 EICHER AND BREWER’S WORK
the choropleth map, the dasymetric map is used to display
standardized data. Traditionally, population density has
Cory Eicher and Cynthia Brewer (2001) were among the
been the most common form of standardized data used,
first to implement and compare several methods for
but a wide variety of standardized data are possible. For
dasymetric mapping in an automated context. They
instance, we might create a dasymetric map of the density
created dasymetric maps of several population-related
of cattle, of percent forest cover, or of the rate for a spe-
attributes (e.g., population density and density of homes
cific type of cancer. Although a standardized map is the
of a specified value) for a 159-county region in four
end result, raw-total data (unstandardized data) are
states—Pennsylvania, West Virginia, Maryland, and
required for each enumeration unit so that the raw totals
Virginia—and the District of Columbia. For ancillary
can be redistributed within enumeration units.
information, they utilized a land use/land cover data set
Redistribution of raw totals is accomplished through
from the U.S. Geological Survey that split the 159-county
the use of ancillary information. Typically, this involves
region into 580 zones that fell into either urban, agricul-
using a land use/land cover map or remotely sensed
tural, woodland, forested, or water categories (the bound-
imagery. For instance, assume that you have data avail-
aries of the zones are shown in Figure 2). For the
able on the distribution of horses and that these data are
discussion to follow, we assume that population density is
available only at the county level. Let’s presume further
the attribute to be mapped using the dasymetric method.
that you acquire a land use/land cover map that provides
To allot population to dasymetric zones, Eicher
considerable detail at the county level and consists of the
and Brewer compared three methods: binary, three-
classes water, urbanized areas, grassland, cropland, and
class, and limiting variable. In the binary method, the
forest. Theoretically, you should be able to use this land
land use/land cover categories were split into two
use/land cover map to distribute horses within each
groups: habitable and uninhabitable. The habitable
county.The number of horses that you assign to each land
group included urban, agricultural, and woodland cate-
use/land cover class will be a function of the relative
gories, and the uninhabitable group consisted of the
water and forested categories. One hundred percent of
the population was assigned to the habitable group, and
0 percent was assigned to the uninhabitable group. Thus,
A B if a county consisted of 1,000 people, all of these people
were assigned to the habitable group, and a population
density figure for the area associated with that group
was calculated by dividing 1,000 by the area of the
habitable categories in the county. Obviously, this
approach produced some error in the resulting map, as
the urban category was treated identically to the
agricultural and woodland categories; furthermore, the
binary approach was unrealistic because it presumed no
population for the forested category. An advantage of
the binary approach, however, was its simplicity.
In the three-class method, Eicher and Brewer
first combined the agricultural and woodland
categories into one category they termed “agricul-
tural/woodland.” They then presumed that population
FIGURE 1 A hypothetical comparison of (A) choropleth would be assigned to only three categories: urban, agri-
and (B) dasymetric mapping. cultural/woodland, and forested—hence the name
16:07:42.
Dasymetric Mapping
agricultural/woodland, and 10 for forested. Although

an improvement over the binary method, the three-
class method was limited in that no consideration was
made for the area falling in each category. Thus, a
county with very little area in the urban category would
still have 70 percent of its population allotted to that
category, which could produce an unrealistically high
population density for the associated mapped area.
In the limiting variable method, Eicher and Brewer
used an approach based on the work of John K.
Wright (1936).* In this approach, the population was
first assigned so that the density of the three habit-
able categories just mentioned was identical. Thus, if a
county consisted of 1,000 people and was 10 square
kilometers in area, and if only habitable categories
were found in the county, a density of 1,000/10 or 100
was assigned for the entire county. Next, thresholds
were set for the maximum density allowed in
each habitable category: these were 15 and 50 people
per square kilometer for the forested and agricul-
tural/woodland categories, respectively. If the overall
density figure exceeded these thresholds, then the
densities would be set to the maximum threshold, and
the remaining population would be assigned to other
categories on the map. In our example, because the
value 100 exceeds both thresholds, the excess people
FIGURE 2 The 580 ancillary zones utilized by Eicher and all would have to be assigned to the urban category.
Brewer in their study of dasymetric mapping. (After Eicher To illustrate some of the computations involved in
and Brewer 2001. First published in Cartography and the limiting variable method, we’ll consider a simple
Geographic Information Science 28(2), p. 127. Reprinted with hypothetical example in which only two categories are
permission from the American Congress on Surveying and present (urban and forested), the population to be
Mapping. Courtesy of Cory Eicher and Cynthia Brewer.) assigned is 1,000, and the area of the county is 1 square
kilometer (Figure 3A). In this case, the population den-
sity initially assigned is 1,000/1 or 1,000 people per
three-class method. In addition to using three
categories, Eicher and Brewer assigned different * Although Wright is often recognized with developing the dasymet-
percentages of population to each of the three cate- ric method, Semenov-Tian-Shansky utilized the method in the 1920s,
gories. For counties with all three categories present, and the origins of the method precede that time (also see McCleary
the percentages were as follows: 70 for urban, 20 for 1969).
A B
Urban Urban
(1,000) (15,775)
FIGURE 3 Computing population

Forested Forested density figures for regions when a
(1,000) (15) maximum threshold is assumed for a
subregion: (A) the same density
(1,000 people per square mile) is used
for both regions; (B) a maximum
value of 15 people per square mile is
presumed for the forested region. See
text for associated formula.
333
16:07:42.
Dasymetric Mapping
square kilometer. Again, assuming a forested threshold

of 15 people per square kilometer, the density for the
urban category can be calculated using the following
formula (Wright 1936):
D - Dmam
Dn =
1 - am
where a region has been divided into two areas n and m,

D is the overall density of the region (county in our case),
Dm is the estimated density of subregion m, am is the
fractional area of region m (relative to the entire region),
and Dn is the density of region n. In Figure 3B, the
forested and urban regions would be equivalent to
regions m and n, and we would compute a density of
15,775 people per square kilometer for the urban region
as follows: FIGURE 4 A choropleth map of population density for the
159-county region utilized by Eicher and Brewer. (After
1,000 - (15) (.9375)
Dn = = 15,775 Eicher and Brewer 2001. First published in Cartography and
1 - .9375 Geographic Information Science 28(2), p. 128. Reprinted with
permission from the American Congress on Surveying and
Eicher and Brewer evaluated the effectiveness of Mapping. Courtesy of Cory Eicher and Cynthia Brewer.)
the three approaches for allotting population to dasy-
metric zones by comparing the population values
computed for the investigated zones with those for
U.S. Bureau of the Census block groups composing
the zones. This constituted a reasonably precise check 3 MENNIS AND HULTGREN’S INTELLIGENT
because the 580 zones were composed of more than DASYMETRIC MAPPING (IDM)
13,000 block groups. Using this approach, Eicher and
Brewer found that the limiting variable method pro- Jeremy Mennis and Torrin Hultgren (2006) have
duced significantly better results (from a statistical developed an automated dasymetric approach, which
standpoint) than the other two methods. The results they term intelligent dasymetric mapping (IDM). This
were also visualized using maps such as those shown approach utilizes both the cartographer’s domain
in Color Plate 1. Color Plate 1A is a dasymetric map knowledge and the relationship between enumeration
of population density resulting from the binary units and ancillary information. If the cartographer
method. Obviously, such a map portrays considerably has considerable knowledge about the phenomenon
more detail than a simple choropleth map of the data being mapped, then it may be possible to specify
(Figure 4). Color Plates 1B and 1C are error maps appropriate density values for particular ancillary
associated with the population density map. These classes. If, however, the cartographer does not have
error maps were computed by comparing the esti- that knowledge, the IDM approach will determine
mated values based on dasymetric mapping with the appropriate density values as a function of the rela-
actual values for zones based on U.S. Bureau of the tionship between the enumeration units and the ancil-
Census block groups composing each zone. The lary classes.
percent error map shows error relative to the total In the IDM approach, the enumeration units are
population of a zone, and the count error map simply termed source zones, the areas defined by the ancillary
shows the difference between the dasymetric information are termed ancillary zones, and the over-
estimates and the true values. Note the linear lap of the source and ancillary zones form target zones
patterns running northeast to southwest (in this (Figure 5). As with other dasymetric mapping
ridge-and-valley physiographic region) on the per- approaches, the problem is how the raw totals for
cent error map. Eicher and Brewer indicated that source zones should be redistributed to the target
such patterns are a function of the binary method zones. If we consider a source zone s and an ancillary
assigning zero population to forested areas (which is zone z associated with ancillary class c, then our basic
unrealistic). The limiting variable method did not problem is to determine the estimated count for a
exhibit such patterns. target zone t defined by the area of overlap of s and z.
16:07:42.
Dasymetric Mapping
A B C
Ta
r ge
tZ
on
e
Source Zones Ancillary Zones Overlay of Zones

FIGURE 5 How source and ancillary zones are overlaid to create target zones in Mennis and Hultgren’s IDM method: (A) source
zones (or enumeration units); (B) ancillary zones based on land use/land cover; and (C) an overlay of source and ancillary zones
forms target zones.
The basic formula for calculating the estimated count N c = the estimated density of ancillary class c
where D
for the target zone t is as follows: ys = the count of a source zone
Nc As = the area of a source zone
AtD
J t僆s K
yN t = ys The resulting density is thus a ratio of the sum of the
a aAtD
N cb
raw totals for all of the sampled source zones divided
by the sum of the areas of all of those source zones.
where yN t = the estimated count for the target zone t The above formulas are intended to illustrate the
ys = the count of a source zone, which overlaps basic concepts involved in IDM. Mennis and Hultgren
the target zone note that in the first of the equations, the term At
At = the area of the given target zone applies only to those target zones associated with non-
N c = the estimated density of ancillary class c
D zero values (i.e., the cartographer has not specified a
associated with the target zone density of 0). Similarly, the term As in the second equa-
Note that the numerator of the formula within tion applies only to source zones with densities other
parentheses deals with the target zone for which an than 0. Mennis and Hultgren also describe a more com-
estimate is being calculated, while the denominator plex approach in which an additional set of “region
considers all target zones that overlap the source zone. zones” are specified so that a separate density estimate
The cartographer may use his or her domain can be made for each ancillary class in each region. For
knowledge to specify the value of D N c or use one of the example, you might be using census tracts for source
following three methods to compute the value: contain- zones and wish to also utilize county-level information.
ment, centroid, or percent cover. In the containment In a fashion similar to Eicher and Brewer, Mennis and
method, IDM selects all of those source zones that are Hultgren evaluated the effectiveness of their approach by
wholly contained within the individual ancillary class working with a region of 373 census tracts along the Front
associated with the target zone. In the centroid method, Range of Colorado.The 373 census tracts served as source
IDM selects those source zones that have their centroids zones, and a land use/land cover scheme available from
contained within the individual ancillary class. In the the U.S. Geological Survey was used to create the
percent cover method, IDM allows the cartographer to following ancillary categories: high-density residential,
specify a threshold percentage and then select those low-density residential, non-residential developed,
source zones whose percentage of coverage by the vegetated, and water. The attributes examined included
ancillary class equals or exceeds that threshold. Once the total population, Hispanic population, number of children
source zones have been selected, the estimated density (people under the age of 21), and number of households.
for the ancillary class may be calculated as follows: A total of 19 different dasymetric maps were created for
m m
each of the four attributes. The 19 maps contrasted a wide
a ys n a As
Nc =
D variety of options available in IDM; for instance, maps
s=1 s=1 were created of the containment, centroid, and percent
335
16:07:42.
Dasymetric Mapping
cover methods (with various threshold values for the An important concept in LandScan is distinguishing
latter), and some included preset values for certain classes between residential and ambient populations. Normally,
based on domain knowledge, while others did not include censuses determine population counts based on where
preset values.The maps also included the binary approach people reside, and thus the result is a residential
similar to that utilized by Eicher and Brewer, and a tradi- population. For instance, the 2000 U.S. census indicated
tional areal weighting approach, in which each source that there were only 55 people in the block that used to
zone contributes to the target zone in direct proportion to house the World Trade Center (Dobson et al. 2003, 267).
the percentage of the source zone’s area that is found in In the case of natural and manmade disasters, we are
the target zone. The formula for areal weighting is as instead interested in where people are apt to be located
follows: at an instant in time—this is termed the ambient popu-
n lation. In early versions of LandScan, a single ambient
ys As ¨ z
yN t = a value was determined, but more recent versions have
s=1 As computed daytime and nighttime populations for some
where yN t = the estimated count for the target zone areas of the world, reflecting the fact that people live in
ys = the count of a source zone, which overlaps one location and work in another. Another important
the target zone aspect to keep in mind when considering LandScan is
As z = the area of intersection between the source that the quality of census data varies considerably
t
and target zones throughout the world. In some developed countries,

As = the area of the source zone such as the United States, census data are available at a
n = the number of source zones with which z very detailed level, as we have seen even for individual
overlaps blocks; but for many other countries, data are available
For all 19 maps, Mennis and Hultgren analyzed the only at the province level (which is one administrative
results by comparing the estimated values for target division below the national level), and some countries
zones with actual values at the U.S. Bureau of the Census release only a national total (Dobson et al. 2000, 850).
block level. In analyzing the results, Mennis and Hultgren Let us now consider the various ancillary information
noted the importance of domain knowledge in produc- utilized in LandScan. Dobson et al. (2000) indicated that
ing a satisfactory dasymetric map: “Those methods using “Roads are especially indicative [of population] because
preset density estimates generally performed better than of their vital role in human settlements with or without
conventional and IDM methods that did not use presets, other forms of transport” (p. 851). As of 2007, LandScan
particularly for the total population variable” (p. 189). is using the National Geospatial Intelligence Agency’s
They stressed, however, that when domain knowledge is (NGA) VMAP-1 series (scale 1:250,000) to determine
not available, the sampling approaches (containment, the location of roads. Although Dobson and his
centroid, and percent cover) can perform as good or colleagues do not provide a precise formula, they indi-
better than the classic binary and areal weighting meth- cate that 1-kilometer grid cells nearer to roads are given
ods. The best results were obtained when presets and a higher probability for population. Slope is a second
sampling were combined together. form of ancillary information used. Using a digital
elevation model (DEM), it is possible to calculate a
slope for each 1-kilometer grid cell. Dobson et al. argue
4 LANDSCAN that slope is important “because most human
settlements occur on flat to gently sloping terrain”
Developed at the Oakridge National Laboratory under (p. 851), and thus a higher probability for population is
the direction of Jerry Dobson and his colleagues (2000, assigned for grid cells having a less steep slope.
2003), LandScan (http://www.ornl.gov/sci/landscan/index. A third form of ancillary information used in
html) is a global database that provides population LandScan is land cover, which Dobson et al. (2000)
estimates for grid cells approximately 1 square kilometer describe as “perhaps the best single indicator of
in area (and finer in the U.S.). Such population estimates population density,” although they note the difficulty
are critical for determining populations at risk in the of acquiring this information in politically sensitive
event of natural and manmade disasters (e.g., the release areas (p. 852). Here again, LandScan utilizes land
of radiation from a nuclear power plant, as in Chernobyl cover information available from NGA. Populated
in 1986, or the effects of a tsunami, such as the Asian places and nighttime lights are two other key pieces
Tsunami in 2004). LandScan provides an interesting of ancillary information used in LandScan. Populated
contrast to the dasymetric approaches that we have places data are available in the VMAP-1 series.
discussed so far because in addition to land use/land Nighttime lights refers to the nighttime light emanat-
cover, Dobson and his colleagues utilized a broad range ing from the Earth’s surface, and that data are
of ancillary information, including roads, slope, populated provided by the National Oceanic and Atmospheric
places, and nighttime lights. Administration’s (NOAA) National Geophysical
16:07:42.
Dasymetric Mapping
Data Center (NGDC). Dobson et al. (2003, 35) 5 LANGFORD AND UNWIN’S GENERALIZED
indicate that three different light intensities have DASYMETRIC APPROACH
been used in LandScan. A higher probability is
assigned to a 1-kilometer cell containing a populated Traditionally, population density has been mapped
place and nighttime lights than to one containing only using choropleth symbology because population data
nighttime lights. Although Dobson (personal commu- are readily available for enumeration units, and
nication, 2007) indicates that nighttime lights data computer software frequently has the choropleth map
were not used after the development of the LandScan as a major option (if not the only option). Mitchel
2003 database because of their relatively coarse Langford and David Unwin (1994, 22) noted three
resolution, there have been proposals for launching major problems with using choropleth maps for popula-
satellites that would provide high-resolution night- tion data: (1) larger enumeration units tend to have
time light data (e.g., Elvidge et al. 2007). lower population densities (and conversely, smaller
Figure 6 summarizes some of the key inputs for enumeration units tend to have higher population
LandScan and shows a resulting population map (in densities); (2) enumeration units hide the data variation
the center) for the Mediterranean region using an early that exists within them; and (3) enumeration unit
version of LandScan produced in 1998. More recent boundaries are arbitrary, and thus unlikely to be associ-
versions of LandScan utilize higher resolution data and ated with actual discontinuities in population density. In
thus will provide more detailed maps of population. response to these problems, Langford and Unwin
FIGURE 6 A resulting population map (in the center) and some of the basic forms of ancillary information that have been uti-
lized in LandScan: slope, nighttime lights, land cover, and proximity to roads (Reproduced with permission, the American Society
for Photogrammetry and Remote Sensing, Photogrammetric Engineering and Remote Sensing 66, no. 7, pp. 849–857).
337
16:07:42.
Dasymetric Mapping
developed a novel mapping method based on remote “Residential Housing” and “All Other Categories.” A
sensing, dasymetric mapping, and generalization tech- comparison of the resulting map (Figure 7B) with the
niques. First, they used Landsat Thematic Mapper traditional choropleth map (Figure 7A) makes clear the
imagery to split pixels within their study region (north- variation in population density that can exist within
ern Leicestershire, United Kingdom) into two groups: enumeration units, and also makes clear the fact that
A B
People per km2

8,011 - 10,000
6,021 - 8,010
Residential
4,031 - 6,020
2,041 - 4,030 All Other Categories
0 km 5 0 km 5
50 - 2,040
C D
People per km2 People per km2

8,321 - 10,400 6,641 - 8,300
6,241 - 8,320 4,981 - 6,640
4,161 - 6,240 3,321 - 4,980
2,081 - 4,160 0 km 5 1,661 - 3,320 0 km 5
0 - 2,080 0 - 1,660
FIGURE 7 Langford and Unwin’s approach for mapping population density: (A) a traditional choropleth map; (B) a
binary map of “residential housing” and “all other categories” derived from remote sensing (a dasymetric map was cre-
ated from this map by dividing the population for each enumeration unit by the area of residential housing falling in
that unit); (C and D) generalized versions of the dasymetric map based on circular “windows” of 0.5 and 1 km search
radius centered on each pixel. (From Langford and Unwin 1994; courtesy of The British Cartographic Society.)
16:07:42.
Dasymetric Mapping
enumeration unit boundaries generally do not coincide This “problem” can also be viewed as an advantage,
with discontinuities in population density. however, as it provides alternative views of the data (a
Next, Langford and Unwin created a dasymetric form of data exploration).
map by combining the remote sensing–based map
with the population data associated with enumeration
units. Standardized data for the dasymetric map were SUMMARY
computed by dividing the population for each
enumeration unit by the area of residential housing At several points in this text we have stressed the
falling in that unit. They noted that the resulting notion that although choropleth maps are commonly
density values varied considerably from those used, they are often inappropriate because phenomena
depicted on the choropleth map: The dasymetric map generally are not uniformly distributed within the enu-
ranged from 2,820 to 14,900 people per square meration units composing the map. Like the choropleth
kilometer, whereas the choropleth map ranged from map, the dasymetric map uses areal symbols to repre-
50 to 10,000 people per square kilometer. The dasy- sent presumed zones of uniformity, but the bounds of
metric map initially was symbolized by shading the these zones need not match enumeration unit bound-
residential housing pixels shown in Figure 7B with an aries. The dasymetric map thus provides a potential
intensity proportional to the standardized data associ- alternative to choropleth mapping.
ated with the enumeration unit within which the pixel The key to dasymetric mapping is the availability of
fell. Unfortunately, Langford and Unwin found the ancillary information at a finer resolution than the enu-
resulting dasymetric map a “poor cartographic prod- meration units for which you have attribute data avail-
uct . . . because the reader [could] see individual pixels able. Most typically, ancillary information is available as
and too much fine spatial detail” (p. 24). To improve a land use/land cover map, although you may also be
the visual quality of the dasymetric map, Langford able to utilize remotely sensed imagery for this purpose.
and Unwin next applied a generalization operator. Of course, if you have data available for very fine
This was accomplished in three steps. First, they com- enumeration units (e.g., census tracts) and you are
puted the population associated with a pixel by divid- constructing a small-scale choropleth map, ancillary
ing the population of the enumeration unit by the information and the construction of a dasymetric map
number of residential housing pixels falling in that might be unnecessary.
unit. Next they floated a circular “window” (having a We have considered some alternative approaches for
search radius of either 0.5 or 1.0 kilometer) over each automated dasymetric mapping. Eicher and Brewer
pixel and calculated the population falling within the implemented the binary method, in which they divided
window. Finally, a population density for each pixel land use/land cover categories into two groups: habitable
was computed by dividing the population for the and uninhabitable. In a more complex limiting variable
associated window by its area. When the resulting method based on the classic work of John K. Wright,
pixel values were symbolized as gray tones (Figure Eicher and Brewer created multiple habitable categories
7C), the map depicted relatively smooth changes in and set maximum density thresholds for some categories.
population density (as opposed to the irregularities of They found that the limiting variable method produced
the dasymetric map). significantly better results than the binary method. More
Although Langford and Unwin’s resulting map recently, Mennis and Hultgren have developed an
clearly provides a much more detailed depiction than a automated dasymetric approach known as intelligent
conventional choropleth map, their approach is not dasymetric mapping (IDM). This approach utilizes both
without problems. One is that remotely sensed the cartographer’s domain knowledge and the relation-
categories are not without error. Fisher and Langford ship between the enumeration units and the ancillary
(1996, 308) suggested that this is not a serious problem, information (using either the containment, centroid, or
because “errors at the pixel level . . . can be large . . . percent cover methods). Eicher and Brewer and Mennis
without impacting the accuracy of estimates of regional and Hultgren both focused on land use/land cover as a
amounts.” Another problem is that remote sensing source of ancillary information. The LandScan global
might not provide information concerning the types of population database provides an interesting contrast
housing unit (e.g., a single-family dwelling vs. a high-rise with these approaches because it uses a broad range of
apartment complex). Langford and Unwin argued that ancillary information: roads, slope, land cover, populated
this problem is also not serious if “housing type is places, and nighttime lights.
relatively uniform within the study region” (p. 24). Although the above approaches provide improve-
A third problem is that the resulting map is a function ments over traditional choropleth mapping, you should
of the search radius used (compare Figures 7C and 7D). be aware that other sophisticated approaches are avail-
339
16:07:42.
Dasymetric Mapping
able. For instance, Wu et al. (2006, 138) describe an case of the dasymetric map, greater detail is provided
approach based on multiple regression and semivari- via ancillary information. A more detailed map can
ance that “. . . provide[s] a more accurate representation also be constructed through a mathematical manipu-
of population distribution than conventional maps of lation of the data. Tobler’s pycnophylactic method is
population density by census blocks or land-use types.” an example of this approach. Tobler’s method begins
As another example, Maantay et al. (2007) have with a set of prisms representing enumeration units
developed an approach that utilizes cadastral (tax lot) (each raised to a height proportional to the data),
information. Clearly, automated techniques for dasymet- which are then sculpted to create a smooth surface.
ric mapping are evolving, and thus you may need to The decision of whether the dasymetric or pycnophy-
consult the refereed cartographic literature to find the lactic method should be used is a function of whether
most useful methods. the phenomenon is presumed to be discontinuous (in
It is important to note that the dasymetric map is which dasymetric is appropriate) or continuous (in
only one alternative to the choropleth map. In the which pycnophylactic is appropriate).
FURTHER READING
Crampton, J. W. (2004) “GIS and geographic governance: McCleary, G. F. J. (1969) The Dasymetric Method in Thematic
Reconstructing the choropleth map.” Cartographica 39, Cartography. Unpublished Ph.D. dissertation, University of
no. 1:41–53. Wisconsin, Madison, WI.
Discusses the historical development of choropleth and dasymet- Covers the historical development of the dasymetric map and
ric mapping and the role each has played and could play in politi- describes several forms of dasymetric mapping.
cal decision making. Mesev, V., ed. (2003) Remotely Sensed Cities. London: Taylor
Fisher, P. F., and Langford, M. (1996) “Modeling sensitivity to & Francis.
accuracy in classified imagery: A study of areal interpolation Several chapters cover issues related to dasymetric mapping.
by dasymetric mapping.” Professional Geographer 48,
no. 3:299–309. Monmonier, M. S., and Schnell, G. A. (1984) “Land use and
land cover data and the mapping of population density.”
Examines to what extent the accuracy of a dasymetric map is a International Yearbook of Cartography 24:115–121.
function of the error in remotely sensed imagery.
Indicates how the denominator in a ratio for choropleth mapping
Holt, J. B., Lo, C. P., and Hodler, T. W. (2004) “Dasymetric
might be adjusted to produce a more meaningful map. For exam-
estimation of population density and areal interpolation of ple, rather than dividing population by the total area of an enu-
census data.” Cartography and Geographic Information meration unit, it could be divided by the area within which people
Science 31, no. 2:103–121. are likely to live.
An application of dasymetric mapping that uses an approach sim- Wright, J. K. (1936) “A method of mapping densities of popu-
ilar to that of Langford and Unwin, as described in Section 5; lation: With Cape Cod as an example.” Geographical Review
Figure 6 of the paper illustrates the fine detail that can result from
26, no. 1:103–110.
dasymetric mapping.
A classic paper on dasymetric mapping; introduces a formula for
Maantay, J. A., Maroko, A. R., and Herrmann, C. (2007)
“density of parts” that enables densities to be calculated for
“Mapping population distribution in the urban environment:
portions of an enumeration unit based on ancillary information.
The Cadastral-based Expert Dasymetric System (CEDS).”
Cartography and Geographic Information Science 34, no. 2: Wu, S., Qiu, X., and Wang, L. (2005) “Population estimation
77–102. methods in GIS and remote sensing: A review.” GIScience &
Remote Sensing 42, no. 1:80–96.
Reviews the broad range of dasymetric methods and presents a
new approach that uses cadastral (tax lot) information. In reviewing methods for estimating population, this paper con-
siders a broad range of methods for dasymetric mapping.
Martin, D. (1996) “An assessment of surface and zonal models
of population.” International Journal of Geographical Wu, S., Qiu, X., and Wang, L. (2006) “Using semi-variance
Information Systems 10, no. 8:973–989. image texture statistics to model population densities.”
Cartography and Geographic Information Science 33,
Describes mathematical approaches for reallocating population
within enumeration units.The method presumes that population den-
no. 2:127–140.
sity peaks at the centroid of population within an enumeration unit. Describes an advanced dasymetric mapping approach that utilizes
A choropleth symbolization is used for the resulting cell-based data. multiple regression and semivariance.
16:07:42.
Dasymetric Mapping
GLOSSARY
ancillary information: information used to more accurately intelligent dasymetric mapping (IDM): a dasymetric map-
map data associated with enumeration units (e.g., when mak- ping approach that utilizes both the cartographer’s domain
ing a dot map of wheat based on county-level data, we avoid knowledge and the relationship between enumeration units
placing dots in bodies of water). and ancillary information.
dasymetric map: like choropleth maps, area symbols are
used to represent zones of uniformity, but the bounds of zones
need not match enumeration unit boundaries.
REFERENCES
Dobson, J. E., Bright, E. A., Coleman, P. R., Durfee, R. C., Maantay, J. A., Maroko, A. R., and Herrmann, C. (2007)
and Worley, B. A. (2000) “LandScan: A global population “Mapping population distribution in the urban environ-
database for estimating populations at risk.” ment: The Cadastral-based Expert Dasymetric System
Photogrammetric Engineering and Remote Sensing 66, no. (CEDS).” Cartography and Geographic Information
7:849–857. Science 34, no. 2:77–102.
Eicher, C. L., and Brewer, C. A. (2001) “Dasymetric mapping McCleary, G. F. J. (1969) “The dasymetric method in thematic
and areal interpolation: Implementation and evaluation.” cartography.” Unpublished PhD dissertation, University of
Cartography and Geographic Information Science 28, Wisconsin-Madison, Madison, WI.
no. 2:125–138. Mennis, J., and Hultgren, T. (2006) “Intelligent dasymetric
Elvidge, C. D., Cinzano, P., Pettit, D. R., Arvesen, J., Sutton, mapping and its application to areal interpolation.”
P., et al. (2007) “The Nightsat mission concept.” Cartography and Geographic Information Science 33,
International Journal of Remote Sensing 28, no. no. 3:179–194.
12:2645–2670. Wright, J. K. (1936) “A method of mapping densities of
Fisher, P. F., and Langford, M. (1996) “Modeling sensitivity to population: With Cape Cod as an example.” The
accuracy in classified imagery: A study of areal interpola- Geographical Review 26, no. 1:103–110.
tion by dasymetric mapping.” The Professional Wu, S., Qiu, X., and Wang, L. (2006) “Using semi-variance
Geographer 48, no. 3:299–309. image texture statistics to model population densities.”
Langford, M., and Unwin, D. J. (1994) “Generating and Cartography and Geographic Information Science 33,
mapping population density surfaces within a geographical no. 2:127–140.
information system.” The Cartographic Journal 31,
no. 1:21–26.
341
16:07:42.
16:07:42.
Proportional Symbol
and Dot Mapping
343
18:01:58.
Proportional Symbol
and Dot Mapping
OVERVIEW
This chapter examines the proportional symbol map ceptual scaling yield an unclassed map. In range grad-
and the dot map, two techniques that are used to rep- ing, a classed map is created by grouping data into
resent numerical data associated with point locations. classes, and by letting a single symbol size represent all
We focus on the proportional symbol map in sections data in a class.
1–6. Section 1 discusses two basic kinds of point data Section 4 considers the design of legends for propor-
that can be displayed with proportional symbols: true tional symbol maps. Basic decisions include deciding
and conceptual. True point data are actually measured how symbols should be arranged in the legend and how
at point locations; an example would be the number of many and what size symbols should be used. In the
barrels of oil produced at oil wells. Conceptual point nested-legend arrangement, smaller symbols are drawn
data are collected over areas (or volumes), but the data within larger ones, whereas in the linear-legend
are conceived as being located at points for the pur- arrangement, symbols are placed adjacent to each other
pose of symbolization; an example would be the num- in either a horizontal or a vertical orientation.
ber of oil wells in each state. Section 1 also considers Section 5 addresses the issue of proportional symbol
how data standardization can be applied to propor- overlap. Small symbols cause little or no overlap, poten-
tional symbol maps (e.g., the murder rate for cities gen- tially resulting in a map devoid of spatial pattern; in con-
erally is a more useful measure than the raw number of trast, large symbols create a cluttered map, making it dif-
murders). ficult to interpret individual symbols. In this context, this
Section 2 introduces two general kinds of propor- section considers two issues: deciding how much overlap
tional symbols: geometric and pictographic (respective there should be, and deciding how the overlap should be
examples would be circles and caricatures of people). symbolized. Section 6 describes how redundant symbols
Although geometric symbols have been more frequently can be used on proportional symbol maps. Redundant
used, pictographic symbols are now common because of symbols represent a single attribute (e.g., magnitude of
the ease of creating them with map design software. oil production) using two visual variables (e.g., size and
Circles have been the most frequently used geometric value), thus enabling users to perform specific tasks
symbol because they are visually stable, users prefer more easily and accurately. Because redundant symbols
them, and they conserve map space. are arguably less effective for portraying general
Section 3 deals with three methods of scaling (or siz- information, they are best used in a data exploration
ing) proportional symbols: mathematical, perceptual, framework.
and range grading. In mathematical scaling, symbols The proportional symbol map is sometimes inappro-
are sized in direct proportion to the data; thus, a data priate for depicting conceptual point data because the
value 20 times another is represented by an area (or distribution of the underlying phenomenon often varies
volume) 20 times as large. In perceptual scaling, a cor- within enumeration units. For instance, we might be given
rection is introduced to account for visual underesti- the total oil production for each county in a state, but
mation of larger symbols. Both mathematical and per- suspect that the production is not uniform throughout
18:01:58.
Proportional Symbol and Dot Mapping
each county. The dot map is one solution to this problem of out-of-wedlock births, but care must be taken in
when working with raw totals (the dasymetric map interpreting any spatial patterns on this map because
is a solution when a standardized map is desired). On a cities with large populations are apt to have a large
dot map, one dot represents a certain amount of some number of out-of-wedlock births. (Note the strong
phenomenon (e.g., one dot might represent a specified visual correlation between Figures 2A and 2B, the
number of barrels of oil), and dots are placed at loca- population of major U.S. cities.)
tions within enumeration units where the phenomenon is One method of standardization is to compute the
thought to occur. ratio of two raw-total attributes. In the case of the
Section 7 considers the kinds of data appropriate for dot birth data, we can compute the ratio of out-of-wedlock
maps. Dot maps utilize raw totals such as the number of births to the total number of births, obtaining the pro-
people living in each county of a state. In a fashion similar portion of out-of-wedlock births in each city (this is
to dasymetric mapping ancillary information is necessary shown in percentage form in Figure 2C). This map has
for locating dots. For instance, if making a map of popula- a markedly different appearance than the unstandard-
tion, we might avoid placing dots representing people in ized map, in large part because of a much narrower
water bodies, presuming that people do not live on the range of data: 3.3 to 22.7 percent, as opposed to 102 to
water. Section 8 considers procedures for creating a dot 11,236 births. This map also illustrates the difficulty of
map of wheat harvested in Kansas. We consider (1) how using proportional circles when the data range is rela-
ancillary information can be utilized to select regions tively narrow—here, the largest value is more than six
where dots will be placed, (2) the selection of dot size (how times the smaller, but the map does not immediately
large each dot is) and unit value (the raw total represented suggest this. In section 3.2, we discuss how formulas
by each dot), and (3) approaches for placing dots within for circle sizes might be modified to handle this prob-
regions. lem. Another solution might be to use bars rather than
circles.
One can also argue that the microbrewery and
1 SELECTING APPROPRIATE DATA FOR brewpub data should be standardized, given that the
PROPORTIONAL SYMBOL MAPS number of microbreweries and brewpubs is likely to
be greater in more populous states. In this case, a use-
Proportional symbols can be used for two forms of point ful standardization is the number of microbreweries
data: true and conceptual. True point data are actually and brewpubs per 1 million population (Figure 1B).
measured at point locations; examples include the num- Although standardized conceptual point data can be
ber of calls made from telephone booths and tempera- represented with proportional symbols, a choropleth
tures at weather stations. Although we generally use map is more commonly used (Figure 1C) because con-
isarithmic maps to display continuous phenomena such ceptual point data are associated with areas.
as temperature, we can use proportional symbols to
focus on the raw data as they are collected at point loca-
tions. Conceptual point data are collected over areas (or 2 KINDS OF PROPORTIONAL SYMBOLS
volumes), but the data are conceived as being located at
points (e.g., the centroids of enumeration units) for pur- Proportional symbols can be divided into two basic
poses of symbolization. An example is the number of groups: geometric and pictographic. Geometric symbols
microbreweries and brewpubs in each state (Figure 1A). (circles, squares, spheres, and cubes) generally do not
Some data are not easily classified as either true or con- mirror the phenomenon being mapped, whereas
ceptual. For example, data associated with cities are col- pictographic symbols (heads of wheat, caricatures of
lected over the areal extent of each city, but the data people, diagrams of barns) do. Prior to the development
normally are treated as occurring at point locations of digital mapping, geometric symbols predominated
because at typical mapping scales, cities are depicted as because templates were readily available for manually
points. An example is the number of out-of-wedlock constructing common geometric shapes. Today, digital
births to teenagers in major U.S. cities (Figure 2A). mapping has eased the development of pictographic
The concept of data standardization discussed for symbols: One can create the basic design for a symbol
choropleth maps is also applicable to proportional by hand, scan the symbol into a computer file, and then
symbol maps. As an example, consider the data for use design software to duplicate the symbol at various
out-of-wedlock births. The unstandardized map sizes. The map of beer mugs shown in Figure 3 was cre-
(Figure 2A) is useful for showing the sheer magnitude ated using this approach. Alternatively, one can find
345
18:01:58.
Mapping Conceptual Point Data

Microbreweries and Brewpubs, 1996
A Proportional Symbol B Proportional Symbol

Number of Number per One

Establishments Million Population
3 40 76 113 149 1.2 6.1 11 15.9 20.8
C Choropleth
(standardized)
FIGURE 1 Mapping conceptual point data: (A) propor-

tional circles represent the raw number of microbreweries
and brewpubs in each state; (B) proportional circles repre-
Number per One Million Population sent standardized data—the number of microbreweries and
brewpubs per 1 million people; (C) a choropleth map repre-
senting the standardized data shown in B. (Data source:
1.2 to 3.2 3.6 to 5.9 6.9 to 9.9 12.9 to 17.5 19.2 to 20.8
http://brewpubzone.com/.)
numerous pictographic symbols already in digital form among beer mugs is arguably more difficult than judg-
in clip art files. ing size relations among circles).
The ease with which readers can associate picto- Circles have been the most frequently used geomet-
graphic symbols with the phenomenon being mapped, ric symbol. Arguments traditionally offered for using
their eye-catching appeal, and their increasingly circles include the following:
greater ease of construction suggest that these sym-
bols will become more common in the future. 1. Circles are visually stable.
Pictographic symbols, however, are not without prob- 2. Users prefer circles over other geometric symbols.
lems. One problem is that when symbols overlap, 3. Circles (as opposed to, say, bars) conserve map space.
they might be more difficult to interpret than geomet-
ric symbols (compare the northeastern portions Traditionally, cartographers recommended against
of Figures 1A and 3). Another problem is that it might using 3-D geometric symbols (spheres, cubes, and
be more difficult to judge the relative sizes of irregu- prisms) because of the difficulty of both estimating their
lar pictographic symbols (e.g., judging size relations size and constructing them. Like pictographic symbols,
18:01:58.
A Out-of-Wedlock Births to Teenagers, 1992 B Population of Major U.S. Cities, 1992

(unstandardized) (unstandardized)
Number of Births Thousands of People
102 2,886 5,669 8,453 11,236 153 1,943 3,733 5,522 7,312
C Out-of-Wedlock Births to Teenagers, 1992

(standardized)
FIGURE 2 Mapping data collected over areas (cities) that

can be conceived as points: (A) raw number of out-of-wed-
lock births for U.S. cities with a population of 150,000 or
more; (B) the population of those cities (note that this map
correlates highly with A); (C) a standardized map obtained
Percentage of All Births by dividing the number of out-of-wedlock births by the total
number of births. (Data source: National Center for Health
3.3 8.2 13.0 17.9 22.7
Statistics, 1995.)
however, geometric symbols can produce attractive, Remembering from basic math that the area of a cir-
eye-catching graphics (Figure 4). Additionally, 3-D geo- cle is equal to pr 2, we can establish the relation
metric symbols can be useful for representing a large
range in data; for example, in Figure 5 the small 3-D pri2 vi
symbol is easily detected, whereas the corresponding =
two-dimensional symbol nearly disappears. prL2 vL
3 SCALING PROPORTIONAL SYMBOLS where ri = radius of the circle to be drawn

rL = radius of the largest circle on the map
vi = data value for the circle to be drawn
3.1 Mathematical Scaling
vL = data value associated with the largest circle
Mathematical scaling sizes areas (or volumes) of point Note that this formula specifies circle areas in direct
symbols in direct proportion to the data; thus, if a data proportion to corresponding data values. The rela-
value is 20 times another, the area (or volume) of a cor- tion uses the largest radius (and largest data value)
responding point symbol will be 20 times as large for one of the circles because proportional symbol
(Figure 6). We now consider some formulas for calcu- maps are often constructed by beginning with a
lating symbol sizes; we deal with circles first because of largest symbol size to minimize the effect of symbol
their common use. overlap.
347
18:01:58.
Because the values of p cancel, this equation

Microbreweries and Brewpubs reduces to
1996
2
ri vi
2 =
rL vL
Taking the square root of both sides, we have

vi 0.5
= a b
ri
rL vL
Finally, solving for ri , we compute

vi 0.5
ri = a b * rL
vL
Number of Establishments
To apply this formula, consider how a radius for the
circle representing Los Angeles was computed for
3 to 10 11 to 15 20 to 32 44 to 72 149
Figure 2A. After some experimentation, we decided
that the largest circle (New York) should have a radius
FIGURE 3 Using a pictographic symbol (beer mugs) to rep- of 0.2125 inch (0.5398 cm). To determine the radius for
resent the number of microbreweries and brewpubs in each Los Angeles, we inserted the number of out-of-wedlock
state. The data were range graded and the sizes of symbols births for Los Angeles and New York (8,507 and 11,236,
chosen by “eye.”
FIGURE 4 An eye-catching map created using 3-D geometric symbols. (After Smith 1928. First published in The Geographical
Review, 18(3), plate 4. Reprinted with permission of the American Geographical Society.)
18:01:58.
where hi is the height of a bar to be drawn and hL is the

Data 1 1,000 height of the tallest bar;
1
ri = a b
vi 冫3
Two-Dimensional SPHERES: * rL
vL
Symbol
where ri is the radius of a sphere to be drawn and rL is
the radius of the largest sphere;
Three-Dimensional 1冫
si = a b
vi 3
Symbol CUBES: * sL
vL
where si is the length of a side of a cube to be drawn

FIGURE 5 Attempts to portray a large range of data (note and sL is the length of a side of the largest cube.
that the smallest 3-D symbol is more easily seen than the In using these formulas, there are several issues that
smallest two-dimensional symbol). you need to consider. First, although these formulas are
sometimes embedded in mapping software, you might
find that you have to enter them into spreadsheets along
respectively) into the formula, along with the largest with the raw data. Second, because these formulas typi-
radius as follows: cally involve taking the square or cube root of the ratio of
data (i.e., raising the ratio to either the 1/2 or 1/3 power),
vLos Angeles 0.5
rLos Angeles = a b * rNew York mathematical scaling is sometimes referred to as square
vNew York root or cube root scaling. We term the power to which the
0.5 ratio of data is raised the exponent for symbol scaling.
rLos Angeles = a b
8,507
* .2125 = .1849 Third, it is important to recognize that these formu-
11,236
las produce unclassed maps, as differing data values are
Formulas for other geometric symbols could be depicted by differing symbol sizes. Classed, or range-
derived in a similar fashion. The results would be as graded maps, can be created by classing the data and
follows: letting a single symbol size represent a range of data
values, but unclassed proportional symbol maps are
vi 0.5
si = a b * sL
more common. This might seem surprising given the
SQUARES:
vL frequency with which classed choropleth maps are used.
The difference stems, in part, from the ease with which
where si is the length of a side of a square to be drawn unclassed proportional symbol maps could be created
and sL is the length of a side of the largest square; in manual cartography (either an ink compass or a cir-
cle template could be used to draw circles of numerous
hi = a b * hL
vi
BARS: sizes). Also, it is easier to differentiate proportional
vL symbols from one another (especially when the range
of symbols is large) than it is to differentiate shades on
a choropleth map (e.g., the range of gray tones is fixed).
Data 10 200 Fourth, you will find that some software (e.g.,
MapViewer) permits arbitrarily specifying the smallest
and largest symbol sizes, with other symbols scaled pro-
portionally between these symbols. For symbols based
Mathematical Scaling
on area, the formulas are
vi - vS
zi =
vL - vS
Perceptual Scaling
Ai = Zi (AL - AS ) + AS
where vS and vL = smallest and largest data values

FIGURE 6 Mathematical versus perceptual scaling. In math- Zi = proportion of the data range
ematical scaling, the areas of the circles are constructed associated with the data value vi
directly proportional to the data; in perceptual scaling, the AS and AL = smallest and largest areas desired
size of the larger circle is increased (using Flannery’s expo- Ai = area of a symbol associated with
nent) to account for underestimation. the data value vi
349
18:01:58.
This approach produces an unclassed map, but the symbols where c1 is a constant equal to (1/c)1/n. In Flannery’s
are not scaled proportional to the data (i.e., a data value study, the transpose was S = (1.01902)R1.1432.* To sim-
twice another does not have a symbol twice as large). An plify computations, the value of the constant (1.01902)
advantage of the approach, however, is that it can enhance in the transposed equation can be ignored because it is
the map pattern, just as do an arbitrary exponent and range close to 1.0; thus we have S = R1.1432. Because this equa-
grading, as described in the following sections. tion expresses the relation between the areas of circles,
and because circles are constructed on the basis of radii,
3.2 Perceptual Scaling we need to take the square root of both sides, producing
S 0.5 = R0.5716. Again for simplicity, the value 0.5716 has
Numerous studies have shown that the perceived size normally been rounded to 0.57. As a result, a perceptual
of proportional symbols does not correspond to their scaling formula for circles is
mathematical size; rather, people tend to underestimate
vi 0.57
the size of larger symbols. For example, in viewing the ri = a b * rL
larger mathematically scaled circle in Figure 6, most vL
people would estimate it to be less than 20 times as
large as the smaller circle. If larger symbols are under- The result of using an exponent of 0.57 in the circle scal-
estimated, it seems reasonable to assume that formulas ing formulas can be seen for the perceptually scaled cir-
for mathematical scaling might be modified (or “cor- cles in Figure 6; for most readers, the larger circle should
rected”) to account for underestimation; this process is now appear closer to 20 times larger than the smaller
known as perceptual (or psychological) scaling. circle.
The magnitude of the power function exponent
varies, depending on the symbol type. For squares,
Formulas for Perceptual Scaling
Crawford (1973) derived an exponent of 0.93 (which
To develop formulas for perceptual scaling, it is useful yields an exponent of 0.54 for symbol scaling), indi-
to consider how researchers have summarized the cating that squares are estimated better than circles.
results of experiments dealing with perceived size. The For bars, Flannery found that underestimates were
relation between actual and perceived size typically has balanced by overestimates, and thus recommended no
been stated as a power function of the form corrective formula. Although these results suggest
R = cSn that squares and bars should be used when precise
estimates are desired, we recommend using squares
where R = response (or perceived size) and bars with caution. A study by Slocum and his
S = stimulus (or actual size) colleagues (2004) revealed that squares were not
c = a constant aesthetically pleasing (when compared with several
n = an exponent geometric and pictographic symbols) and that bars
were difficult to associate with a point location.
To differentiate the exponent n from the one for sym-
Power function exponents for “drawn” 3-D
bol scaling introduced earlier, we term it the power
symbols have been appreciably lower than for two-
function exponent.
dimensional symbols, indicating severe underestima-
The power function exponent is the key to describ-
tion. For example, for spheres and cubes, Ekman and
ing the results of experiments involving perceived size.
Junge (1961) derived exponents of 0.75 and 0.74 (cor-
If size is estimated correctly, the exponent will be close
responding to exponents of 0.44 and 0.45 for symbol
to 1.0. Underestimation and overestimation are repre-
scaling). Interestingly, research by Ekman and Junge
sented by exponents appreciably below and above 1.0.
also indicated that when truly 3-D cubes were used
For example, an oft-cited study by James Flannery
(they were “made of steel with surfaces polished to a
(1971) found that for circles, R = (0.98365)S 0.8747 ;
homogeneous, dull silvery appearance,” p. 2), the
here, the exponent of 0.8747 is indicative of underesti-
exponent was 1.0. This result suggests that the manner
mation.
in which 3-D symbols are portrayed might have an
The power function equation states what response
impact on the exponent. For example, if an interactive
arises from a certain stimulus. For constructing symbols,
graphics program gives the impression of traveling
we need to know the reverse: what stimulus must be
through 3-D space, we might expect an exponent
shown to get a certain response. Therefore, the power
closer to 1.0.
function must be transposed by dividing each side by c,
and then by raising each side to the 1/n power. The
result is
* Flannery did not raise 1/k to 1/n, so his constant differed slightly
S = c1R1/n from the value reported here.
18:01:58.
Problems in Applying the Formulas study in which he found an exponent of 0.94 when
Unfortunately, there are numerous problems in apply- three legend circles were used (Chang 1977). Chang
ing formulas for perceptual scaling: (1) the value of (1980, 161) also recommended including the statement
a power function exponent can be affected by experi- that “circle areas are made proportional to quantities,”
mental factors; (2) using a single exponent might be to encourage users to make estimates based on area.
unrealistic because of the variation that exists between Another solution to the problems noted by Chang
subjects and within an individual subject’s responses; and Griffin is to train users how to read proportional
(3) the formulas might fail to account for the symbol maps. For example, Judy Olson (1975a) found
spatial context within which symbols are estimated; and that when readers were asked to make estimates of cir-
(4) experimental studies for deriving exponents have cle size and then were given feedback on the correct
dealt only with acquiring specific map information (as answers, the power function exponent was closer to 1.0.
opposed to considering memory and general informa- Olson also found, however, that only when practice was
tion). We briefly consider each of these problems. combined with perceptual scaling did the dispersion of
Kang-tsung Chang (1980) summarized the various errors also decrease, suggesting that training and per-
experimental factors that can affect a power function ceptual scaling should be used in concert.
exponent. One factor is whether subjects are asked to The importance of spatial context on developing a
complete a ratio or magnitude estimation task. Ratio esti- perceptual scaling formula can be seen by considering
mation involves comparing two symbols and indicating the Ebbinghaus illusion shown in Figure 7: The two cir-
how much larger or smaller one symbol is than the other cles in the middle of the surrounding circles are identi-
(e.g., noting that “this symbol appears to be five times cal in size, but the one surrounded by larger circles
larger than this one”); on a map, this involves comparing appears smaller. Applying this principle to a map, a
symbols without consulting the legend. Flannery’s expo- small circle surrounded by larger circles should appear
nent of 0.87 for circles was developed on this basis. smaller than it really is, whereas a large circle sur-
Magnitude estimation involves assigning a value to a sym- rounded by smaller circles should appear larger than it
bol on the basis of a value assigned to another symbol; for really is. In an experimental study with maps, Patricia
example, if a single circle is included in a map legend, val- Gilmartin (1981) found that spatial context actually had
ues for other circles on the map can be estimated by com- these effects. On this basis, one could argue for a for-
paring them with the legend circle. Using this approach, mula in which each circle is scaled to reflect its local
Chang (1977) found that the use of a moderately sized context. Gilmartin argued against this, however, indicat-
circle in the legend led to an exponent of 0.73 for circles. ing it would “have the undesirable effect of weakening
A second factor affecting the exponent is the size of the overall pattern perception” (p. 162).
the standard symbol against which other symbols are Another limitation of perceptual scaling is that exper-
compared. For example, in magnitude estimation stud- iments for deriving an appropriate exponent have focused
ies, Chang (1977) and Cox (1976) showed that the value solely on specific map information. The portrayal of gen-
of the exponent gradually increased as the size of the eral information is a more important function of maps.
standard increased. Other experimental factors noted by Thus, it seems that general tasks should be considered in
Chang that influenced the exponent included the word- developing an exponent, or at least that the effect of the
ing of the instructions to subjects, the range of stimuli
presented, and the order in which estimates were made.
T. L. C. Griffin (1985) is one of several researchers who
have noted that reporting a single exponent for an exper-
iment hides the variation among and within subjects.
Although overall Griffin found a power function expo-
nent of 0.88 for circles (nearly identical to Flannery’s
0.87), he stressed that the exponents for individuals var-
ied from approximately 0.4 to 1.3. He also noted that
“perceptual rescaling was shown to be inadequate to cor-
rect the estimates of poor judges, while seriously impair-
ing the results of those who were more consistent” (p. 35).
One solution to the problems noted by Chang and
Griffin is to apply no correction and to stress the
importance of a well-designed legend. For example,
Chang recommended including “three standards [in FIGURE 7 The Ebbinghaus illusion: The two circles in the
the legend]—small, medium, [and] large”; he based this middle of the surrounding circles are identical in size, but the
recommendation, in part, on a magnitude estimation one surrounded by larger circles appears smaller.
351
18:01:58.
A Microbreweries and Brewpubs, 1996 B Microbreweries and Brewpubs, 1996

(mathematical) (perceptual)
Number of Number of
Establishments Establishments
7 13 22 50 149 7 13 22 50 149
FIGURE 8 Effect of mathematical versus perceptual scaling on map pattern: (A) mathematically scaled map; (B) perceptually
scaled map based on Flannery. Note that perceptual scaling appears to have little effect on the overall pattern when the largest cir-
cle on both maps is the same size.
exponent on the overall look of the map should be con- acquisition (i.e., values obtained are approximately the
sidered. To illustrate this point, consider Figure 8, which square of that for acquisition). Although the exponent in
compares mathematically and perceptually scaled maps the perceptual scaling formulas might be adjusted to
of the microbrewery and brewpub data. Here the largest account for this, we suspect that most cartographers
circle size on each map has been held constant to mini- would not make this adjustment because map readers are
mize the effects of circle overlap on the two maps. not expected to remember precise, specific information. It
Although a larger range of circle sizes appears on the per- is, however, interesting to note that the lower exponent
ceptually scaled map, the patterns on the two maps are resulting from this approach would have an effect similar
similar, suggesting that perceptual scaling has little effect to that just suggested for enhancing spatial pattern.
on general information.
Rather than basing the exponent on specific circle Range-Graded Scaling
estimates, Olson (1976a, 155–156; 1978) suggested that Range grading groups raw data into classes, and repre-
an arbitrary exponent might be used to enhance the sents each class with a different-sized symbol (as an
recognition of spatial pattern.To illustrate, Figure 9 com- example, if the data are grouped into five classes, then
pares a mathematically scaled map (the exponent in the five symbol sizes are used). Three basic decisions must
circle-scaling formula is set to 0.5) with a map having an be made in range grading: the number of classes to be
arbitrary exponent (the exponent is set to 1.0) for the shown, the method of classification to be used, and the
standardized birth data. The narrow range of data on the symbol sizes to be used for each class. The first two deci-
mathematically scaled map (Figure 9A) makes it sions are standard in any classification of numerical
difficult to detect any pattern. In contrast, on the arbi- data.
trarily scaled map (Figure 9B), it is easier to detect a pat- The sizes of range-graded symbols normally are
tern; for example, note the lower percentage of out-of- selected to enhance the visual discrimination among
wedlock births in the extreme south central part of the classes. Figure 10 portrays two sets of symbols that car-
United States (in Texas). tographers have developed for this purpose. The first
A final limitation of perceptual scaling is that most of (Figure 10A) was developed by Hans Meihoefer (1969)
the relevant experiments have been based only on infor- in a visual experiment. Meihoefer indicated that people
mation acquisition, as opposed to memory for that infor- “were able generally to differentiate” (p. 112) among all
mation. Studies by psychologists (e.g., Kerst and Howard of these circles in a map environment. Arguing that
1984) have revealed that underestimation of larger sizes some of Meihoefer’s circles were difficult to discrimi-
occurs twice: once when acquiring information and once nate, Borden Dent (1999, 181–183) used his personal
again when recalling that information. As a result, the experience in designing maps to develop the set shown
exponent for memory is appreciably lower than that for in Figure 10B. For both sets, the mapmaker simply
18:01:58.
Effect of Symbol Scaling and Position on Map Pattern

Out-of-Wedlock Births to Teenagers, 1992
A Mathematically Scaled B Arbitrarily Scaled
Percentage of All Births Percentage of All Births
3.3 8.2 13.0 17.9 22.7 3.3 8.2 13.0 17.9 22.7
C Circles Moved from Areas

of Concentration
FIGURE 9 Manipulating symbol scaling and symbol position

to enhance map pattern: (A) a mathematically scaled map
(the exponent in the circle-scaling formula is set to 0.5); (B)
Percentage of All Births an arbitrarily scaled map (the exponent in the circle-scaling
formula is set to 1.0); (C) circles are moved away from con-
3.3 8.2 13.0 17.9 22.7
gested areas.
selects n adjacent circles, where n is the number of range grading has relatively little effect on the spatial
classes to be depicted on the map. pattern because the range of circle sizes on the range-
Range grading is considered advantageous because graded and mathematically scaled maps are similar. For
readers can easily discriminate symbol sizes and thus the birth data, however, the maps are dramatically dif-
readily match map and legend symbols; another advan- ferent, with range grading resulting in considerable
tage is that the contrast in circle sizes might enhance overlap.
the map pattern, in a fashion similar to the use of an Because specific range-graded sizes have been recom-
arbitrary exponent described in the preceding section. mended only for circles, individual mapmakers must
To illustrate the latter, consider Figure 11, which com- determine appropriate sizes for other symbol types. For
pares range-graded and mathematically scaled maps for example, to create the range-graded proportional-square
both the standardized out-of-wedlock birth data and map shown in Figure 12, distinctly different small and
the raw microbrewery and brewpub data.* In the case large squares were selected, and then three intermediate
of the microbrewery and brewpub data, we see that squares were specified so that their areas were evenly
spaced between the smallest and largest squares.
*In this case, the data were classed using Jenks’s optimal approach Range grading is particularly desirable for picto-
and Dent’s five smallest circles. graphic symbols because their unusual shapes often
353
18:01:58.
A B
Meihoefer Dent
Radii Radii
in. (mm) in. (mm)
0.050 (1.27) 0.025 (0.64)
0.078 (1.99)
0.045 (1.14)
0.111 (2.82)
0.065 (1.65)
0.157 (3.99)
0.110 (2.79)
0.222 (5.64)
0.170 (4.32)
0.314 (7.98)
0.245 (6.22)
0.385 (9.77)
0.380 (9.65)
0.444 (11.28)
0.510 (12.95)
0.544 (13.82)
0.628 (15.96) 0.595 (15.11)
FIGURE 10 Potential circle sizes for range grading: (A) a set developed by Meihoefer in a visual experiment (after
Meihoefer 1969, Figure 4); (B) a set developed by Dent based on practical experience (after Dent 1999, 183).
18:01:58.
Range Grading Versus Mathematical Scaling

A B
Out-of-Wedlock Births to Teenagers, 1992 Out-of-Wedlock Births to Teenagers, 1992
(range graded) (mathematically scaled)
Percentage of Percentage of
all Births all Births
3.3 to 6.9 7.2 to 9.9 10.5 to 12.9 13.0 to 16.0 16.4 to 22.7 3.3 8.2 13.0 17.9 22.7
C D
Microbreweries and Brewpubs, 1996 Microbreweries and Brewpubs, 1996
(range graded) (mathematically scaled)
Number of Number of
3 to 10 11 to 15 20 to 32 44 to 72 149 3 40 76 113 149
FIGURE 11 Range-graded and mathematically scaled maps for both the standardized birth data and the microbrewery and
brewpub data. Maps A and C are range graded based on Dent’s five smallest circles. Maps B and D are mathematically scaled.
make precise relationships between symbols awkward the value for circle T in Figure 11A is considerably
to compute (for the mapmaker) and awkward to esti- larger than the value for circle S (say, approximately
mate (for the map reader). Thus, it makes sense to select 30 times as large), but the numbers specified in the
intermediate-sized symbols by eye, as opposed to spac- legend indicate that the values differ by only a factor
ing them regularly on the basis of area or volume. This of about 4. Another disadvantage is that range grad-
was the approach taken to construct the pictographs of ing creates clear differences in circle size, and thus
beer mugs shown in Figure 3. potentially creates a pattern when there might
A disadvantage of range grading is that readers not really be a meaningful pattern. For example, for a
might misinterpret specific information if they do not data set with minimum and maximum values of 10
pay careful attention to the legend. For example, a and 11 percent, respectively, range grading would
reader failing to examine the legend might say that create obvious differences among circles, even though
355
18:01:58.
should be ordered with the largest at the top or at

Microbreweries and Brewpubs the bottom. As with choropleth maps, we might use the
1996 argument that “people associate ‘up’ with ‘higher’ and
‘higher’ with larger data values” to justify placing larger
symbols at the top. Alternatively, we could argue that
larger symbols are visually “heavier” and thus are more
stable if placed at the bottom. With range-graded maps, it
makes sense to place smaller symbols at the top so that
data values increase from left to right and from top to
bottom, matching the order in which we normally read
A B 1,000
1,000
Number of Establishments 700
700
400
400
200
200
3 to 10 11 to 15 20 to 32 44 to 72 149 50
50
FIGURE 12 A range-graded map using squares as a point

symbol. C
a 1 percent difference might be of no interest or sig- 700

1,000
400
nificance to the reader. 200
50
4 LEGEND DESIGN FOR PROPORTIONAL SYMBOL

MAPS D
There are two basic problems in designing legends for

proportional symbol maps: deciding how the symbols
should be arranged and determining which symbols 50 200 400 700 1,000
should be included.
E F
4.1 Arranging Symbols
1,000 1,000
Two basic legend arrangements are used on propor-
tional symbol maps: nested and linear. In the nested-
legend arrangement, smaller symbols are drawn
within larger symbols (Figures 13A and 13B), whereas
in the linear-legend arrangement, symbols are placed 700 700
adjacent to each other in either a horizontal or a ver-
tical orientation (Figures 13C–F). An advantage of the
nested arrangement is that it conserves map space.
Note, however, that this arrangement might make it
400 400
difficult to compare a symbol in the legend with a
symbol on the map, because symbols in the legend
(with the exception of the smallest) are covered by
other symbols. 200 200
When a linear horizontal arrangement is used, you
must decide whether the symbols should be ordered with 50 50
the smallest on the left and the largest on the right, or
vice versa. Displaying larger symbols on the right is most
desirable, given that the traditional number line pro- FIGURE 13 Various legend arrangements: (A) and (B) are
gresses from left to right. When a linear vertical arrange- nested; (C) and (D) are linear with a horizontal orientation;
ment is used, you similarly must decide whether symbols and (E) and (F) are linear with a vertical orientation.
18:01:58.
text. Ultimately, the choice of a particular linear legend be sufficient for data with a small range, it makes sense
design is likely to depend on available map space. to use more than three when the range is large. As a rule
of thumb, we suggest using as many symbols as appear
4.2 Which Symbols to Include to be easily discriminated.
With range grading, legend symbols are a function of the

5 HANDLING OVERLAP ON PROPORTIONAL
classes displayed on the map (e.g., a five-class map yields
SYMBOL MAPS
five legend symbols). For mathematical and perceptual
scaling, there are two general methods for selecting leg-
end symbols. One is to include the smallest and largest A major issue in proportional symbol mapping is decid-
symbols actually shown on the map and then to interpo- ing how large symbols should be, and consequently how
late several intermediate-sized symbols (Figure 14A). A much overlap there should be. Small symbols cause lit-
second method is to select a set of symbols that are most tle or no overlap, and thus a map potentially devoid of
representative of those appearing on the map, which spatial pattern; in contrast, large symbols create a clut-
should minimize estimation error. The latter method can tered map, making it difficult to interpret individual
be implemented by applying Jenks’s optimal classifica- symbols. This section considers two issues: deciding how
tion to the raw data and then constructing legend sym- much overlap there should be and how the overlap
bols based on the median (or mean) of each class should be symbolized.
(Figure 14B). This method might also be refined to
include circles representing the extremes in the data 5.1 How Much Overlap?
because of the difficulty of extrapolating beyond sym-
bols shown in the legend (Dobson 1974). Regardless of Unfortunately, there are no rules regarding the appropri-
which method is used, the argument can be made that ate amount of overlap. Rather, cartographers have sug-
round numbers should be used in the legend because gested subjective guidelines; for instance, Robinson et al.
they will be easier to work with and remember. Thus, for (1984) indicated that the map should appear “neither ‘too
map A of Figure 14, legend values could be rounded to full’ nor ‘too empty”’ (p. 294). Examples of improper
the nearest 100 (100, 2,900, etc.). amounts of overlap are shown in Figures 15A and 15B.
In addition to selecting one of these general methods Although most cartographers would agree that such
for mathematical and perceptual scaling, the mapmaker extreme cases should be avoided, there would be dis-
must also decide how many symbols will be shown in the agreement as to which amount of overlap between the
legend. One possibility would be to use three symbols, as extremes is more proper. (Figures 15C and 15D are two
recommended by Chang. Although three symbols might possibilities.)
A Out-of-Wedlock Births to Teenagers, 1992 B Out-of-Wedlock Births to Teenagers, 1992

(smallest and largest circles in legend) (circles representative of classes in legend)
Number of Births Number of Births
102 2,886 5,669 8,453 11,236 396 913 1,973 4,427 9,509
FIGURE 14 Approaches for selecting circles for the legend on an unclassed proportional symbol map: (A) the smallest and
largest symbols actually shown on the map are used, along with several intermediate-sized symbols; (B) the raw data are classi-
fied, and a representative value of each class is used (in this case, the median).
357
18:01:58.
Effect of Varying Circle Overlap

Microbreweries and Brewpubs, 1996
A Too Much Overlap B Not Enough Overlap
Number of
Establishments
Number of
Establishments
3 40 76 113 149 3 40 76 113 149
C Appropriate Overlap D Appropriate Overlap
Number of Number of
3 40 76 113 149 3 40 76 113 149
FIGURE 15 Effect of varying the amount of overlap: (A) too much overlap—the map appears crowded; (B) not enough over-
lap—the map appears empty; (C) and (D) are examples of maps having an appropriate amount of overlap.
The role that overlap plays is determined to some 5.2 Symbolizing Overlap
extent by whether the map is to be used for communi-
cation or for data exploration. In the case of communi- Overlap can be handled using either transparent or
cation, you probably will want to manipulate circle opaque symbols. Transparent symbols enable readers
overlap to enhance the spatial pattern. For example, to see through overlapping symbols, whereas opaque
Figure 15C might be a more appropriate choice than symbols are stacked on top of one another (see
Figure 15D if you wish readers to note the concentra- Figure 16). On black-and-white maps, the fill within
tion of microbreweries in the northeast part of the both transparent and opaque symbols can be varied
United States. In the case of data exploration, mapping from white to solid black. Of course, color can pro-
software ideally should provide an option to easily vide considerable flexibility in overlap, as different
change symbol sizes, such as the interactive option colors can be used for the boundary and interior of
described by Slocum and Yoder (1996). symbols.
18:01:58.
A B
C D
E F
FIGURE 16 Transparent (on the left) and opaque (on the right) methods of handling overlap. From top to bottom, different fills
are shown, ranging from white to gray to black. (After Griffin 1990, 23.)
Transparent symbols allow background information the background of the rest of the map; this is most appar-
(e.g., a road network) to be seen beneath the symbols ent for the gray and black symbols in Figure 16. The
(Figures 16A, 16C, and 16E), and make it easier for the opaque approach also promotes a visual hierarchy, as the
map user to determine the bounds of individual symbols, circles appear “above” other map information (e.g., in
thus simplifying the estimation of symbol size. In con- this case, above the road network). As a result, the reader
trast, opaque symbols enhance figure-ground contrast can focus on the circles and not be distracted by the back-
because the symbols tend to appear as a figure against ground information.
359
18:01:58.
These ideas are supported by the work of T. L. C. Griffin

(1990) and Richard Groop and Daniel Cole (1978). Griffin Out-of-Wedlock Births to Teenagers
found that people were about equally split in their prefer- 1992
ence for transparent and opaque symbols; transparent
symbols were liked because they provided “maximum
information,” and those favoring opaque symbols “stressed
the quality of clarity” (24).With respect to fill, Griffin found
that for transparent circles, gray and black were about
equally popular, whereas for opaque circles, black was most
popular; almost no one liked a white fill. Groop and Cole
found that transparent symbols were more accurately
estimated than opaque symbols, and that for opaque sym-
bols, there was a strong correlation between the amount of
overlap and the error of estimation.
Other solutions for handling overlap include inset Percentage of All Births
maps, which portray a congested area at an enlarged
scale; the use of a zoom function in an interactive graph- 3.3 to 6.9 7.2 to 9.9 10.5 to 12.9 13.0 to 16.0 16.4 to 22.7
ics environment; and the possibility of moving symbols
slightly away from the center of congested areas.
FIGURE 17 Use of redundant symbols (the visual variables
Zooming typically is implemented by enclosing the area
size and value) on proportional symbol maps.
to be enlarged by a rectangular box, clicking a mouse
button, and having the outlined area fill the screen.
MapTime contains such a zoom function. Figure 9C
illustrates the effect of moving symbols away from confor a data exploration environment in which multiple
gested areas. In comparing Figures 9C and 9B, we can views of a data set are considered.
see that in Figure 9C, it is easier to compare the sizes of
individual circles and the spatial pattern is potentially
more obvious. 7 SELECTING APPROPRIATE DATA
FOR DOT MAPS
6 REDUNDANT SYMBOLS
Dot maps are ideally utilized when you have collected
conceptual data (in the form of raw totals) for enu-
Redundant symbols portray a single attribute with two meration units, and wish to show that the underlying
or more visual variables. For example, on proportional phenomenon is not uniform throughout the enumera-
symbol maps, the visual variables size and value could tion units. We create a dot map by letting one dot
be used (Figure 17). The logic of using redundant sym- equal a certain amount of some phenomenon and then
bols is that it is easier to discriminate symbols on the placing dots where that phenomenon is most likely to
basis of two visual variables than on a single visual vari- occur. For instance, imagine that you have data on the
able. As a result, users are able to perform specific map number of elephants living in each country of Africa.
tasks more easily. For example, in Figure 17 it is possible You could make a proportional symbol map of such
to differentiate the large circles from the other circles, data, but this map would be misleading because each
even though there is considerable overlap (compare country would have a single symbol, suggesting no
with Figure 11A).* variation in the spatial distribution of elephants within
Although redundant symbols promote the communi- a country. A dot map would provide a more realistic
cation of specific information, they seldom are used on representation of the distribution of elephants.
printed maps, presumably because mapmakers feel that Dot maps (like dasymetric maps) use ancillary infor-
redundant symbols impede the communication of gen- mation to create a detailed map of a phenomenon. In the
eral information (i.e., it is difficult to combine symbols case of elephants in Africa, appropriate ancillary infor-
to form a region when the symbols differ in both size mation might be the location of heavily vegetated areas
and value). Redundant symbols seem more appropriate (especially forests and wooded savannas) and protected
areas (such as national parks). Ancillary information is
* For an experimental study dealing with redundancy on proportional commonly split into limiting and related attributes.
symbol maps, see Dobson (1983). A limitation of Dobson’s study was Limiting attributes place absolute limits on where dots
that he dealt only with range-graded symbols. can be placed. For example, in general it would not make
18:01:58.
sense to place dots representing population within water wheat harvested example, they are generic to most dot-
bodies. Related attributes are those that are correlated mapping problems.
with the phenomenon being mapped but do not place
absolute limits on the location of dots or zonal bound-
8.1 Determining Regions within Which Dots Should Be
aries. For example, certain types of soil in a region might
Placed
favor a particular crop, but the crop might be found on
all types of soil within the region.
In constructing the dot map of wheat harvested in
Kansas, we first considered several attributes that might
8 CREATING A DOT MAP limit the distribution of wheat, and crops in general.
Obvious limiting attributes included the location of
This section considers three issues relevant to creating water bodies (in the case of Kansas, there are a number
such a dot map: (1) determining regions within which of reservoirs), the location of urban areas (e.g., Wichita
dots should be placed, (2) selecting dot size and unit and Kansas City), and slope (i.e., tractors cannot be
value, and (3) placing dots within selected regions. used on very steep slopes). Such limiting attributes
Although we consider these issues in the context of the might be found on paper maps and entered into a GIS
12,000 Acres of Wheat
FIGURE 18 Locating wheat directly using remote sensing versus locating wheat based on cropland from a land use/land cover
map. The upper left map depicts the distribution of wheat within Finney County based on an analysis of 30-meter-resolution
Landsat imagery, whereas the lower right map portrays the same distribution (the gray-shaded county) based on a 1-kilometer res-
olution land use/land cover map. (Courtesy of Kansas Applied Remote Sensing Program, University of Kansas.)
361
18:01:58.
as layers via scanning or digitizing (or they might One obvious limitation of the Kansas land use/land
already be available in digital form). The resulting lay- cover map is that it does not distinguish the location of
ers could then be overlaid, and dots thus placed only in individual crops. This limitation might be handled by
areas not constrained by the limiting attributes. considering ancillary information that could assist in
Rather than employ an overlay approach, we used a locating wheat within the cropland areas on the land
Kansas land use/land cover map developed by Jerry use/land cover map. One potential ancillary attribute
Whistler and his colleagues (1995) because it provided would be precipitation, because it varies from a high of
a convenient synthesis of limiting attributes. The land about 35 inches in eastern Kansas to only about 15
use/land cover map, which was created from Landsat inches in western Kansas. Presuming an ideal amount of
Thematic Mapper remote sensing data, split the land- precipitation for wheat, a mathematical function could
scape into 10 classes: five urban (residential, commer- be developed that would place a higher probability on
cial-industrial, grassland, woodland, and water) and locating dots near the ideal precipitation area. Ideally,
five rural (cropland, grassland, woodland, water, and such an approach also would consider other complicat-
other). The map defined directly two of the limiting ing attributes, such as the timing of the precipitation.
attributes we had considered (urban areas and water Another potential ancillary attribute would be irri-
bodies), and included two limiting attributes that we gated cropland. The 1994 Kansas Farm Facts specifies
had not immediately thought about: woodland and the acres of wheat harvested from irrigated land, so if
grassland. Most important, however, was the cropland the location of irrigated cropland could be determined,
class, which indirectly handled the slope attribute and a percentage of dots could be placed in such areas
any other limiting and related attributes likely associ- equivalent to the percentage of wheat irrigated in that
ated with cropland. area. Remote sensing could be used for this purpose
Although a mapmaker might be tempted to place (Eckhardt et al. 1990), particularly in the case of center-
within the cropland class all dots representing wheat, pivot irrigation, which traces a distinctive circle on the
it is important to recognize that land use/land cover landscape (Astroth et al. 1990). Still another possible
maps based on remote sensing techniques are not ancillary attribute would be the distribution of soils
without error. Whistler and his colleagues (1995, 773) within each county. One might suspect that certain soils
estimated the overall accuracy of the Kansas land would be more conducive to wheat production than to
use/land cover map to be 85 to 90 percent or better, other crops. In discussing this possibility with those
but the accuracy of individual categories within partic- knowledgeable about Kansas agriculture (a farmer and
ular counties deviated considerably from this. For a county extension agent), we found that the selection
example, using data provided by Whistler et al., we of crops at the county level was more likely a function
found that only 53 percent of the cropland in Butler of farm policy and associated programs rather than of
County was estimated to be correctly classified. To soil type. For example, wheat would continue to be
handle this error, we placed some dots representing planted at a particular location because a farm program
cropland within grassland areas (the category typically specified that the same crop must be planted for finan-
confused with cropland on imagery). The number of cial support to be retained.
dots so placed was a function of the estimated per- Although ancillary information could assist in creat-
centage of incorrectly classified cropland in a county: ing a more detailed dot map, remote sensing alone
If 47 percent of the cropland in a county was estimated provides the capability to create the most precise map
to be misclassified as grassland, then 47 percent of the because of its ability to pinpoint the location of individ-
dots were placed in the grassland category within that ual crops. For example, Stephen Egbert and his col-
county. Although these percentage figures suggest that leagues (1995) used remote sensing to differentiate
large numbers of dots were affected, they were not wheat, corn, grain sorghum (milo), alfalfa, fallow land,
because there was generally little wheat harvested in shortgrass prairie, and sand-sage prairie in Finney
counties with large error values; less than 5 percent of County, Kansas, and found classification accuracies to
all dots were affected by this procedure. be as high as 99 percent for wheat. They accomplished
The map shown in the lower right of Figure 18 this by using imagery for three different time periods, as
resulted from using the dot placement procedure just opposed to the single-date imagery employed for the
discussed. Kansas land use/land cover map. The map shown in the
18:01:58.
upper left of Figure 18 is illustrative of the kind of detail in selecting dot size and unit value. To use the nomo-
they were able to obtain. Although using remote sens- graph, we must first calculate the number of dots per
ing in this fashion can provide considerable detail, the square centimeter for a sample area on our map. For
greater amount of imagery incurs a greater expense, example, imagine that we found our densest enumer-
and the necessary verification (ground truth) data is ation unit to be a county with 500,000 acres of wheat
also costly (Egbert, personal communication 1996). and an area of 6 square centimeters (at the mapped
Thus, mapmakers wishing to use remote sensing must scale). If we presume a unit value of 1 dot equals
carefully consider its cost and benefits for a particular 1,000 acres, then the number of dots required would
application. be 500 (500,000/1,000), and the number of dots per
square centimeter would be 83.3. (500/6). (The
8.2 Selecting Dot Size and Unit Value dashed vertical line in Figure 19 represents the value
of 83.3.) If we now construct a line from the origin to
Dot size (how large each dot is) and unit value a desired dot diameter, and note where this line inter-
(the count represented by each dot) are important sects the vertical dashed line, we will have an indica-
parameters in determining the look of a dot map. tion of the approximate distance between the edges
Cartographers have argued that very small dots proof dots. For instance, in Figure 19 we can see that a
duce a “sparse and insignificant” distribution, and that dot diameter of approximately 0.889 centimeter has
very large dots “give an impression of excessive den- been selected, and thus the distance between dots
sity.” Similarly, a small unit value produces a map that will be slightly greater than 0.0254 centimeter. Note
“gives an unwarranted impression of accuracy,” that this distance falls in the “zone of coalescing
whereas a large unit value results in a map that “lacks dots.” It should be noted that these calculations pre-
pattern or character” (Robinson et al. 1995, 498). sume that dots are distributed in a relatively uniform
Generally, it has been argued that dots in the densest hexagonal pattern.
area should just begin to coalesce (that is, merge with In practice, some experimentation is generally
one another). required to select an appropriate dot size and unit value,
J. Ross Mackay (1949) developed a graphical regardless of whether the nomograph is used. For our
device known as the nomograph (Figure 19) to assist wheat map, we selected several counties that had both a
1.0
0.9
.1778
43
6
.152
2
9
01
76
.12
88
.11
.0
.1
.0
AGGREGATE AREA OF DOTS IN SQUARE CENTIMETERS
0.8
DIAMETER OF DOTS IN CENTIMETERS
TERS
0.7
TIME
0.6 .025 35
CEN
4 .06
S IN
Z O N E O F FIGURE 19 The nomograph—a

0.5 C O
DOT
A L E device used for computing appropriate

S C I N
EEN
.05 G
0.4 08 dot size and unit value. A vertical line is
TW
8
D .050 drawn representing a relatively dense
E BE
O
area on the map, and a second line is
.0
ANC
0.3
76
drawn from the origin to a desired dot

2
S
DIST
.1
01
size. The distance between dots can be

6
.0381
.1
0.2
27
determined by where the two lines

.15 778
0
.1 032
24
intersect. (After Mackay 1949. First

.2
TIM ETERS
IN CEN
0.1 DOTS published in Surveying and Mapping 9,
TWEEN
CE BE
DISTAN p. 7. Reprinted with permission from
0 the American Congress on Surveying
0 15.50 31.00 46.50 62.00 77.50 93.00 108.50 124.00 139.50 155.00 170.50
DOTS PER SQUARE CENTIMETER and Mapping.)
363
18:01:58.
high percentage of cropland in wheat and a large value

for acres of wheat harvested. Selecting counties on the Dot Placement Methods
basis of percentage of cropland in wheat ensured that Soybeans in Kansas
dense areas of harvested wheat would be considered,
and having a large value for acres harvested ensured
A Uniform
that the dense area would consist of a relatively large
number of dots. Using a graphic design program
(Freehand), we placed all necessary dots within the sam-
pled counties using several dot sizes and unit values and
then subjectively evaluated the resulting maps for coa-
lescence. Once we found an acceptable dot size and unit
value, we used Freehand to place dots in all counties.
One problem with the approach that we have
described is that it fails to consider the difficulty that
map readers have in correctly estimating the number
and density of dots within subregions of a map. B Geographically Weighted
Somewhat analogous to the underestimation of propor-
tional symbol size (see section 3.2), readers also under-
estimate the number and density of dots. Judy Olson
(1977) indicated that it is possible to correct for this
underestimation, but the effect on the overall pattern is
small, and thus of questionable utility.
It is interesting to note that mapmakers can ignore
the issue of dot size and unit value when the phenome-
non is located explicitly via remote sensing (as in the
Finney County example shown in the upper left of
Figure 18); in this case, the darkened pixels tell us pre-
cisely where the phenomenon is located. One disadvan-
C Geographically Based
tage of this approach, however, is that it might be diffi-
cult to make a visual estimate of the magnitude of
production, as individual pixels merge together. Of
course, because the map is automatically generated, the
computer can be used to make a precise estimate of the
acres of wheat grown in any area (albeit with the esti-
mate a function of the ability to correctly interpret the
phenomenon via remotely sensed imagery).
8.3 Placing Dots within Regions
When dots were located manually, cartographers gener-

ally used one of three approaches for placing dots: uni- FIGURE 20 Approaches used by cartographers to place dots
form, geographically weighted, and geographically based. on a dot map: (A) uniform, (B) geographically weighted, and
For each of these approaches, it was presumed that the (C) geographically based.
mapmaker had calculated the number of dots associated
with particular enumeration units by dividing the raw clustering and large gaps between some dots. Thus, the
total for the enumeration unit by the unit value. In the result might better be termed uniform with a random
uniform approach (Figure 20A), an attempt was made to component. The end result simulated a choropleth map,
place dots associated with an enumeration unit in a rela- as the boundaries of enumeration units were sometimes
tively uniform fashion throughout the enumeration unit. apparent (Figure 20A).
Mackay (1949, 5) described this process as “placing the In the geographically weighted approach, dots were
first dot near the center [of the enumeration unit] and shifted (weighted) toward neighboring enumeration units
then each successive dot in the largest remaining space.” of higher value, as shown in Figure 20B. This approach
Perfect uniformity was not desirable because patterns in took advantage of the notion of spatial autocorrelation—
nature are not perfectly uniform. Pure randomness also that high values tend to be located near other high val-
was not desired because this could lead to unrealistic ues. This was an obvious improvement over the uniform
18:01:58.
approach because it tended to diminish the effect of

county boundaries (note that county boundaries are less
A B
apparent in Figure 20B than in Figure 20A). It essentially 45 90 108
simulated an isopleth map, based on the assumption of
smooth, continuous changes between enumeration units. 27 60 60
In the geographically based approach (Figure 20C), dots
were located based on the ancillary information that we 27 18 27
have described previously, albeit using the uniform
approach for areas within which dots were supposed to
be placed.* This, of course, was the preferred approach,
as it took advantage of geographic knowledge and pro- C D
duced the most detailed dot map.
Unfortunately, present-day software for dot mapping
generally does not include satisfactory approaches for
dot placement. Software vendors have developed dot-
mapping techniques (commonly referred to as dot-
density routines), but these techniques generally place
dots in a purely random fashion, which, as we have said,
can lead to unrealistic clusters and gaps in the dot
pattern. These routines also tend to presume that the
user wishes to place dots within enumeration units with- FIGURE 21 A depiction of Lavin’s dot-density shading
out considering ancillary information. Furthermore, approach: (A) a matrix of grid points serves as input;
there is the presumption that dots will be placed within (B) square cells are presumed to surround each grid point;
polygons (a vector-based structure), as opposed to (C) square cells are split into equally sized subcells, with the
within a cell-based (raster) structure.† To illustrate the number of subcells a function of the number of dots to be
difficulty of working with a cell-based structure, con- placed; (D) a dot is placed within each subcell. (After Lavin
sider the portion of the Kansas land use/land cover map 1986. First published in The American Cartographer 13(2),
shown in the left portion of Color Plate 1. Clearly, it is p. 145. Reprinted with permission of Stephen Lavin and the
not easy to define polygons, as remote sensing tech- American Congress on Surveying and Mapping.)
niques used to create the land use/land cover map have
produced very detailed patterns. At the end of the Step 2. Compute the proportion of dot coverage, Pi ,
following section, we suggest an automated approach for each of the square cells. The formulas are
that might handle this problem.
vi - vs
zi =
An Approach for Automatic Dot Placement vL - vs
In this section, we consider an approach for automated
dot placement that can circumvent the problems of Pi = zi (P max - Pmin ) + Pmin
clusters and gaps resulting from purely random place-
ment. The approach is based on a dot-density shading where vi = data value for a particular grid cell
technique that Stephen Lavin (1986) developed for vs and vL = smallest and largest data values in
mapping continuous phenomena (e.g., rainfall). Lavin’s the grid
approach requires four basic steps: zi = proportion of the data range associated
with data value vi
Step 1. Insert an equally spaced gridded network of
Pmin and Pmax = minimum and maximum desired
data values into the computer (Figure 21A). Such a grid
proportions of dot coverage
is normally the end result of the inverse-distance
Pi = proportion of dot coverage for data
(or gridding) approach for automated contouring.
value vi
Conceptually, each grid intersection is presumed to be
Assuming that the minimum and maximum desired pro-
surrounded by a square cell within which dots will be
portions of dot coverage are 0.08 and 0.50,‡ respectively,
placed (Figure 21B).
‡ We chose the minimum and maximum values so that we could
reproduce the dots in Lavin’s illustrations. Normally, you would
*In its most sophisticated form, the geographically based approach choose a maximum value that would cause dots in the densest
could also include a weighting component. region to coalesce. The minimum value would then be chosen
† For a description of vector and raster techniques, consult a GIS so that the density of dots properly reflected the data relations on
textbook such as Lo and Yeung (2007). the map.
365
18:01:58.
the following would be computed for the lower left grid

point value 27 in Figure 21A:
A 180º 120º 60º 0º 60º 120º
90º
27 - 18 60º 60º
zi = = 0.10
108 - 18 30º 30º
Pi = 0.10(0.50 - 0.08) + 0.08 = 0.1220)
0º 0º
Step 3. Compute the number of dots, Nd , to be placed
within a square cell. The formula is 30º 30º
Pi * Ac
Nd = 60º 60º
Ad
90º
where Ac is the area of the cell, and Ad is the area of a 120º 60º 0º 60º 120º 180º
dot. For the lower left grid point, we have

B
0.1220 * 0.1296
Nd = = 8.784 L 9
0.0018
Step 4. Partition the square cells surrounding each grid

point into n equally sized subcells, where n is as close as
possible to Nd. In the case of the lower left grid point, the
number of cells (9) matches Nd exactly (Figure 21C).
The subcells serve as plotting boundaries for each dot: A
dot is plotted randomly within each cell and cannot
extend outside the cell (Figure 21D). A map resulting
from this process is shown in Figure 22, along with a tra-
ditional isoline map. FIGURE 22 A comparison of (A) Lavin’s dot-density shad-
Applying Lavin’s dot-density shading to a dot map ing approach and (B) the traditional isoline method. Maps are
would involve the following steps: First, standardized val- based on data for land elevations, and contour lines are in
ues would have to be computed for each subregion hundreds of feet. (After Lavin 1986. First published in The
within which dots are to be placed (subregions would be American Cartographer 13(2), p. 142. Reprinted with
determined by limiting and related attributes). Stand- permission of Stephen Lavin and the American Congress on
ardization would be accomplished by dividing a raw-total Surveying and Mapping.)
value (say, acres of wheat) associated with a subregion by
the area of that subregion. Second, a grid would be laid
over the map. All grid points falling within a particular The resulting map would appear more like the
subregion would be assigned the standardized value for detailed map of Finney County shown in the upper
that subregion. Finally, steps 2 to 4 just described would left portion of Figure 18 than a traditional dot map.
be executed. A limitation, however, is that it would not exhibit
One problem with applying Lavin’s approach to dot some of the regularities of the Finney County map,
mapping is that subregions within which dots are to be such as the circular features representing center-pivot
placed are often in raster (pixel-based) form, as we irrigation.
have indicated for the Kansas land use/land cover
map (Color Plate 1). In this case, note that individual SUMMARY
subregions are frequently smaller than the dots that
would be placed within them. In this case it might make
In this chapter, we have covered a number of important
more sense to develop a plotting algorithm based on
principles involved in constructing proportional symbol
pixels (rather than dots), such as the following:
maps. We have stressed that two kinds of data can be
Determine the proportion of cropland within an mapped using proportional symbols: true point data,
enumeration unit to be symbolized as wheat. (Divide which can actually be measured at point locations, and
acres of wheat harvested by acres of cropland.) conceptual point data, which are collected over areas
For each pixel within the enumeration unit: (or volumes) but are conceived as being located at
If the pixel is cropland, points. Respective examples would be the water
Generate a random number between 0 and 1. released by wells associated with center-pivot irrigation
If the random number is … the proportion of systems and the number of fatalities due to drunk dri-
cropland in wheat, set the pixel to wheat. ving in each state. It is often useful to standardize data
18:01:58.
intended for a proportional symbol map by considering and thus can readily match map and legend symbols,
raw totals for one attribute relative to values of another and because the contrast in circle sizes might enhance
attribute. For example, we might divide the gallons of the map pattern.
water pumped at a well by the area covered by the asso- Other important issues in proportional symbol map-
ciated center-pivot irrigation system. ping include legend design, symbol overlap, and symbol
Two basic kinds of proportional symbols are possible: redundancy. Basic problems of legend design include the
geometric and pictographic. Geometric symbols (e.g., arrangement of symbols (nested or linear) and the num-
circles and spheres) generally do not mirror the phenom- ber and size of symbols. A nested arrangement conserves
enon being mapped, whereas pictographic symbols (e.g., map space, but it might make it more difficult for the
drawings of oil derricks to represent oil production) do. map user to compare a symbol in the legend with a sym-
Conventionally, geometric symbols have been used more bol on the map. In terms of circle overlap, small symbols
frequently, but pictographic symbols are becoming com- cause little or no overlap, and thus potentially create a
mon because of the ease of creating them in a digital map devoid of spatial pattern; in contrast, large symbols
environment. Three-dimensional symbols (e.g., spheres) can create a cluttered map, making it difficult to interpret
traditionally have been frowned on because of the diffi- individual symbols. Redundant symbols represent a sin-
culty readers have in estimating their size, but 3-D sym- gle attribute (say, the magnitude of production at water
bols can produce attractive, eye-catching graphics and wells) by two visual variables (e.g., size and value), thus
are useful for representing a large data range. enabling users to perform specific tasks more easily and
There are three methods for scaling (or sizing) pro- accurately. Because redundant symbols are arguably less
portional symbols: mathematical, perceptual, and effective for portraying general information, they are
range grading. In mathematical scaling, symbols are best used in a data exploration framework.
sized in direct proportion to the data; thus, a data value When you have collected conceptual data (in the
10 times another is represented by an area (or volume) form of raw totals) for enumeration units, and wish to
10 times as large as the area of the other data value. In show that the underlying phenomenon is not uniform
perceptual scaling, a correction is introduced to throughout the enumeration units, a dot map is appro-
account for visual underestimation of larger symbols; priate. The key to constructing an effective dot map is
thus, larger symbols are made bigger than would nor- the availability of ancillary information. As with the
mally be specified by mathematical scaling. For exam- dasymetric map, we found that a land use/land cover
ple, when constructing proportional circles, a Flannery map can provide the necessary ancillary information.
correction is normally applied so that larger circles Important issues in dot mapping include selecting the
appear even larger (a symbol scaling exponent of 0.57 dot size (how large each dot is) and the unit value (the
is used). We also noted a number of problems with count represented by each dot). A nomograph can be
applying such a correction (e.g., using a single expo- used for this purpose, although you will generally find
nent might be unrealistic because of the variation that some experimentation with potential values is nec-
in perception that exists among subjects). In range essary. Another issue in dot mapping is placing dots
grading, data are grouped into classes, and a single within regions specified by ancillary information.
symbol size is used to represent all data falling in a Unfortunately, mapping software often utilizes a purely
class. Range grading is considered advantageous random approach that leads to unrealistic clusters and
because readers can easily discriminate symbol sizes gaps in the dot pattern.
FURTHER READING
Brewer, C. A., and Campbell, A. J. (1998) “Beyond graduated Dobson, M. W. (1983) “Visual information processing and
circles: Varied point symbols for representing quantitative cartographic communication: The utility of redundant
data on maps.” Cartographic Perspectives, no. 29:6–25. stimulus dimensions.” In Graphic Communication and
Summarizes literature related to proportional symbols, and illus- Design in Contemporary Cartography. ed. by D. R. F. Taylor,
trates both univariate and bivariate point symbols. pp. 149–175. Chichester, England: Wiley.
Chang, K. (1980) “Circle size judgment and map design.” An experimental study dealing with redundancy on proportional
American Cartographer 7, no. 2:155–162. circle maps.
Summarizes various experimental factors that can affect a power Flannery, J. J. (1971) “The relative effectiveness of some com-
function exponent. mon graduated point symbols in the presentation of quantita-
Dent, B. D. (1999) Cartography: Thematic Map Design. 5th ed. tive data.” Canadian Cartographer 8, no. 2:96–109.
Boston: McGraw-Hill. A frequently cited study for determining an appropriate exponent
for scaling circles.
Chapter 8 covers basic principles of dot mapping. For more on
basic principles, see pages 497–502 of Robinson et al. (1995). Gilmartin, P. P. (1981) “Influences of map context on circle
367
18:01:58.
perception.” Annals, Association of American Geographers 71, Develops an appropriate set of range-graded circles.
no. 2:253–258. Olson, J. M. (1975a) “Experience and the improvement of
Examines spatial context as a factor in estimating proportional cartographic communication.” Cartographic Journal 12, no.
symbol sizes. 2:94–108.
Griffin, T. L. C. (1985) “Group and individual variations in Considers the notion that readers might be trained to improve
judgment and their relevance to the scaling of graduated cir- their estimates of symbol size.
cles.” Cartographica 22, no. 1:21–37. Olson, J. M. (1976a) “A coordinated approach to map com-
Considers the notion that power function exponents may vary as munication improvement.” American Cartographer 3, no.
a function of the individual person. 2:151–159.
Griffin, T. L. C. (1990) “The importance of visual contrast for Considers how changes in both map design and reader training
graduated circles.” Cartography 19, no. 1:21–30. might be used to enhance map communication; of particular inter-
est is the discussion of adjustments to the symbol scaling exponent
Covers various methods for handling overlap of proportional
to enhance map pattern.
circles.
Kerst, S. M., and Howard, J. H. J. (1984) “Magnitude estimates Patton, J. C., and Slocum, T. A. (1985) “Spatial pattern recall:
of perceived and remembered length and area.” Bulletin of An analysis of the aesthetic use of color.” Cartographica 22,
the Psychonomic Society 22, no. 6:517–520. no. 3:70–87.
Contemplates the notion that we should consider memory for Examines the effect of circle color on the ability to recall the
information when developing symbol scaling exponents. pattern of proportional circles.
Lindenberg, R. E. (1986) The Effect of Color on Quantitative Rice, K. W. (1989) The Influence of Verbal Labels on the
Map Symbol Estimation. Unpublished Ph.D. dissertation, Perception of Graduated Circle Map Regions. Unpublished
University of Kansas, Lawrence, KS. Ph.D. dissertation, University of Kansas, Lawrence, KS.
Examines the role of color in the adjustment of circle scaling An extensive study of the role that verbal labels play in the
exponents, and concludes that when all circles are the same color, perception of regions on proportional circle maps.
Flannery’s adjustment is appropriate. Slocum, T. A. (1983) “Predicting visual clusters on graduated
Meihoefer, H. (1969) “The utility of the circle as an effective circle maps.” American Cartographer 10, no. 1:59–72.
cartographic symbol.” Canadian Cartographer 6, no. Describes a model for predicting the visual clusters that readers
2:105–117. see on proportional circle maps.
GLOSSARY
ancillary information: information used to more accurately Flannery correction: a method of adjusting the sizes of pro-
map data associated with enumeration units (e.g., when mak- portional circles to account for the perceived underestimation
ing a dot map of wheat based on county-level data, we avoid of larger circle sizes; a symbol scaling exponent of 0.57 is used.
placing dots in bodies of water). geometric symbols: symbols that do not necessarily look like
clip art: pictures that are available in a digital format; such the phenomena being mapped, such as using squares to depict
pictures can serve as the basis for creating pictographic sym- barns of historical interest; contrast with pictographic sym-
bols. bols.
conceptual point data: data that are collected over an area inset: the map element consisting of a smaller map used in
(or volume) but conceived as being located at one point for the context of a larger map (e.g., a map that is used to show a
the purpose of symbolization (e.g., the number of microbrew- congested area in greater detail).
eries in each state). linear-legend arrangement: a form of legend design for
dot-density shading: a technique for mapping smooth, con- proportional symbol maps; symbols are placed adjacent to
tinuous phenomena (e.g., precipitation) in which closely each other in either a horizontal or a vertical orientation.
spaced dots depict high values of the phenomenon, whereas magnitude estimation: (1) a method for constructing
widely separated dots depict low values. grayscales in which a user estimates the lightness or darkness
dot map: a map in which small symbols of uniform size of one shade relative to another; (2) a method for determin-
(typically, solid circles) are used to emphasize the spatial pat- ing the perceived size of proportional symbols in which a
tern of a phenomenon (e.g., one dot might represent 1,000 value is assigned to one symbol on the basis of the value
head of cattle). assigned to another symbol.
dot size: how large dots are on a dot map. nested-legend arrangement: a form of legend design for
Ebbinghaus illusion: a phenomenon in which a circle sur- proportional symbol maps in which smaller symbols are
rounded by large circles will appear smaller than the same- drawn within larger symbols.
sized circle surrounded by small circles. nomograph: a graphical device that can assist in selecting
dot size and unit value on a dot map.
18:01:58.
opaque symbols: proportional symbols that do not permit ratio estimation: a method for determining the perceived
the reader to see base information beneath the symbols; read- size of proportional symbols; two symbols are compared, and
ers also must infer the maximum extent of each symbol. the viewer indicates how much larger or smaller one is than
perceptual scaling: a term used when proportional symbols the other (e.g., one symbol appears five times larger than the
are sized to account for the perceived underestimation of other symbol).
large symbols; thus large symbols are drawn larger than is sug- redundant symbols: used to portray a single attribute with
gested by the actual data. two or more visual variables, such as representing popula-
pictographic symbols: symbols that look like the phenom- tion for cities by both size of circle and gray tone within the
ena being mapped, such as using diagrams of barns to depict circle.
the locations of barns of historical interest; contrast with geo- transparent symbols: proportional symbols that enable
metric symbols. readers to see base information beneath them and the maxi-
power function exponent: the exponent used for the stimu- mum extent of each proportional symbol.
lus in the power function equation relating perceived size and true point data: data that can actually be measured at a
actual size; this exponent is used to summarize the results of point location, such as the number of calls made at a tele-
experiments involving perceived size of proportional symbols. phone booth over the course of a day.
proportional symbol map: a map in which point symbols are unit value: the count represented by each dot on a dot map.
scaled in proportion to the magnitude of data occurring at zoom function: in an interactive graphics environment, an
point locations, such as circles of varying sizes representing area to be focused on is enlarged, commonly by enclosing a
city populations. desired area with a rectangular box and clicking the mouse.
range-graded map: a term used to describe a classed pro-
portional symbol map.
REFERENCES
Astroth, J. H. J., Trujillo, J., and Johnson, G. E. (1990) “A 1995 ACSM/ASPRS Annual Convention & Exposition,
retrospective analysis of GIS performance: The Umatilla Technical Papers, Volume 3, Charlotte, NC, pp. 836–845.
Basin revisited.” Photogrammetric Engineering and Ekman, G., and Junge, K. (1961) “Psychophysical relations in
Remote Sensing 56, no. 3:359–363. visual perception of length, area and volume.” The
Chang, K. (1977) “Visual estimation of graduated circles.” Scandinavian Journal of Psychology 2, no. 1:1–10.
The Canadian Cartographer 14, no. 2:130–138. Eckhardt, D. W., Verdin, J. P., and Lyford, G. R. (1990)
Chang, K. (1980) “Circle size judgment and map design.” The “Automated update of an irrigated lands GIS using SPOT
American Cartographer 7, no. 2:155–162. HRV imagery.” Photogrammetric Engineering and Remote
Cox, C. W. (1976) “Anchor effects and estimation of gradu- Sensing 56, no. 11:1515–1522.
ated circles and squares.” The American Cartographer 3, Flannery, J. J. (1971) “The relative effectiveness of some
no. 1:65–74. common graduated point symbols in the presentation of
Crawford, P. V. (1973) “The perception of graduated squares quantitative data.” The Canadian Cartographer 8, no.
as cartographic symbols.” The Cartographic Journal 10, 2:96–109.
no. 2:85–88. Griffin, T. L. C. (1985) “Group and individual variations in
Cressie, N. (1990) “The origins of kriging.” Mathematical judgment and their relevance to the scaling of graduated
Geology 22, no. 3:239–252. circles.” Cartographica 22, no. 1:21–37.
Dent, B. D. (1999) Cartography: Thematic Map Design (5th Griffin, T. L. C. (1990) “The importance of visual contrast for
ed.). Boston: McGraw-Hill. graduated circles.” Cartography 19, no. 1:21–30.
Dobson, M. W. (1974) “Refining legend values for Gilmartin, P. P. (1981) “Influences of map context on circle
proportional circle maps.” The Canadian Cartographer 11, perception.” Annals of the Association of American
no. 1:45–53. Geographers 71, no. 2:253–258.
Dobson, M. W. (1983) “Visual information processing and Groop, R. E., and Cole, D. (1978) “Overlapping graduated
cartographic communication: The utility of redundant circles: Magnitude estimation and method of portrayal.”
stimulus dimensions.” In Graphic Communication and The Canadian Cartographer 15, no. 2:114–122.
Design in Contemporary Cartography: Progress in Kerst, S. M., and Howard, J. H. J. (1984) “Magnitude
Contemporary Cartography, ed. by D. R. F. Taylor, pp. estimates of perceived and remembered length and area.”
149–175. Chichester, England: Wiley. Bulletin of the Psychonomic Society 22, no. 6:517–520.
Egbert, S. L., Price, K. P., Nellis, M. D., and Lee, R.-Y. (1995) Lavin, S. (1986) “Mapping continuous geographical distribu-
“Developing a land cover modelling protocol for the high tions using dot-density shading.” The American
plains using multi-seasonal thematic mapper imagery.” Cartographer 13, no. 2:140–150.
369
18:01:58.
Lo, C. P., and Yeung, A. K. W. (2007) Concepts and Robinson, A. H., Sale, R. D., Morrison, J. L., and Muehrcke,
Techniques of Geographic Information Systems (2nd ed.). P. C. (1984) Elements of Cartography (5th ed.). New York:
Upper Saddle River, NJ: Pearson Prentice Hall. Wiley.
Mackay, J. R. (1949) “Dotting the dot map.” Surveying and Slocum, T. A., Sluter, R. S., Kessler, F. C., and Yoder, S. C.
Mapping 9, no.1:3–10. (2004) “A qualitative evaluation of MapTime, A program
Meihoefer, H.-J. (1969) “The utility of the circle as an effec- for exploring spatiotemporal point data.” Cartographica
tive cartographic symbol.” The Canadian Cartographer 6, 39, no. 3:43–68.
no. 2:105–117. Slocum, T. A., and Yoder, S. C. (1996) “Using Visual Basic to
Olson, J. M. (1975a) “Experience and the improvement of teach programming for geographers.” Journal of
cartographic communication.” The Cartographic Journal Geography 95, no. 5:194–199.
12, no. 2:94–108. Whistler, J. L., Egbert, S. L., Jakubauskas, M. E., Martinko, E.
Olson, J. M. (1976a) “A coordinated approach to map com- A., Baumgartner, D. W., and Lee, R. (1995) “The Kansas
munication improvement.” The American Cartographer 3, State Land Cover Mapping Project: Regional scale land
no. 2:151–159. use/land cover mapping using Landsat Thematic Mapper
data.” 1995 ACSM/ASPRS Annual Convention &
Olson, J. M. (1977) “Rescaling dot maps for pattern enhance-
Exposition, Volume 3 (Technical Papers), Charlotte, NC,
ment.” International Yearbook of Cartography 17:125–136.
pp. 773–785.
Olson, J. M. (1978) “Graduated circles.” The Cartographic
Yoder, S. C. (1996) “The development of software for
Journal 15, no. 2:105.
exploring temporal data associated with point locations.”
Robinson, A. H., Morrison, J. L., Muehrcke, P. C., Kimerling, Unpublished MA thesis, University of Kansas,
A. J., and Guptill, S. C. (1995) Elements of Cartography Lawrence, KS.
(6th ed.). New York: Wiley.
18:01:58.
Multivariate Mapping
OVERVIEW
Mapmakers need to display multiple attributes. For Section 2 begins by considering small multiples—the
example, a climatologist might wish to simultaneously simultaneous comparison of three or more maps. The dif-
view temperature, precipitation, barometric pressure, and ficulty of synthesizing information depicted via small
cloud cover for a geographic region. The cartographic multiples has led to numerous approaches for combining
display of such multiple phenomena is known as attributes onto one map. Techniques discussed include the
multivariate mapping; if only two attributes are dis- trivariate choropleth map (in which three choropleth
played, the process is termed bivariate mapping. maps are overlaid), the multivariate dot map (in which
Bivariate mapping is covered in section 1, and multivari- different-colored dots are used to represent multiple phe-
ate mapping involving three or more attributes is cov- nomena), and multivariate point symbols (an intriguing
ered in section 2. example is the Chernoff face, in which various facial
A fundamental issue in multivariate mapping is features are used to represent multiple attributes).
whether individual maps should be shown for each Although it is possible to combine a large number of
attribute (in which case maps are compared) or whether attributes on a single map, it is often difficult to visually
all attributes should be displayed on the same map (in interpret the resulting symbols.A solution to this problem is
which case maps are combined). Thus, sections 1 and 2 to explore multivariate data in an interactive graphics envi-
are divided into two subsections: one for map ronment. Several of the symbolization approaches that we
comparison and one for map combination. The map will discuss have this capability. For instance, SLCViewer, a
comparison section for bivariate mapping focuses on software package developed by David DiBiase and his col-
selecting an appropriate method of classification when leagues (1994a), allowed users to view data as small multi-
comparing choropleth maps. Although choropleth ples or to combine up to three attributes using proportional
maps have their limitations, they are commonly used in point symbols, weighted isolines (isolines of varying
bivariate mapping, just as they are used in univariate width), and area shading. Software that also provides
mapping. exploratory capability for multivariate phenomena
The map combination section for bivariate mapping includes Project Argus, CommonGIS, ESTAT, and IDV.
first describes how choropleth maps can be overlaid to In sections 1 and 2, we make the presumption that the
create a bivariate choropleth map. Other symbols mapmaker wishes to map the attributes directly. An alter-
considered in that section include the rectangular native approach is to perform a cluster analysis prior to
point symbol, in which the width and height of a rec- mapping the data. In cluster analysis, we group observa-
tangle are made proportional to the values of two tions based on their scores on a set of attributes; for
attributes being mapped, and the bivariate ray-glyph, instance, we might group counties in California based on
in which the angle of rays (straight-line segments) which counties have similar incomes and similar voting
pointing to the left and right of a small central circle behavior. Thus, rather than mapping the raw data, we
represent two attributes. map the resulting clusters, which is relatively simple from
371
18:02:32.
the standpoint of symbolization. Cluster analysis is the TABLE 1 Class limits for each of the classification
topic of section 3. methods used in Figure 1
Optimal
1 BIVARIATE MAPPING
Attribute
1.1 Comparing Maps Class 1 2 3
1 0–5 26–37 126–134

In this section, we consider approaches for bivariate
2 13–16 40–53 137–142
mapping in which individual maps are used to show
3 23–25 56–68 144–149
each attribute (i.e., maps are compared). We focus on
choropleth maps because they have been commonly Mean–Standard Deviation
used for map comparison.
Attribute
Choropleth Maps Class 1 2 3
As with a single choropleth map, an important consid-
1 None 26–32 126–129
eration in comparing several choropleth maps is decid- 2 0–16 37–56 131–144
ing whether the data should be classed, and, if so, which 3 23–25 63–68 147–149
method of classification should be used. To begin, we
assume that we wish to class the data, and thus we focus Nested Means
on the method of classification. Initially, we also assume Attribute
that we wish to compare two attributes for a single
point in time (say, median income and percent of the Class 1 2 3
adult population with a college education, both col-
1 0–2 26–37 126–132
lected for 2000). Later, we consider comparing maps for 2 3–5 40–45 133–137
the same attribute collected over two time periods. 3 13–16 47–53 138–142
In selecting a method of classification, it is critical to 4 23–25 56–68 147–149
consider the distribution of each attribute along the
Quantiles
number line. If the attributes have differing distribu-
tions (e.g., if one is skewed and the other is normal), Attribute
certain classification methods can lead to an inappro-
Class 1 2 3
priate visual impression of correlation between the
attributes. To illustrate, consider the hypothetical distri- 1 0–2 26–41 126–133
butions shown in Figure 1A. Attribute 1 is clearly posi- 2 3–5 42–47 134–139
3 13–25 48–68 140–149
tively skewed, whereas attributes 2 and 3 appear to have
normal distributions.* In Figure 1B, values of these
attributes have been assigned spatially so that
and normal distributions might not be surprising (com-
extremely high correlations result in each case (the cor-
pare the optimal maps for attributes 1 and 2), but note
relation coefficients, r, appear in Figure 1C).†
that there is also a lack of visual association between maps
The optimal method of classification is often recom-
of the normal distributions (attributes 2 and 3). The opti-
mended because it minimizes classification error, it seems
mal method fails to reflect the high correlations between
natural to ask whether the optimal method might also be
the attributes because it focuses on the precise distribu-
used for map comparison. If the optimal method is applied
tion of the individual attributes along the number line.
to all three attributes shown in Figure 1B, we obtain the
Classification methods that are more appropriate for
maps shown in Figure 1D. (Table 1 specifies the limits of
map comparison include mean–standard deviation,
each class.) Although these maps suggest positive associa-
nested means, quantiles, and equal areas. The
tions between each pair of attributes, they do not support
mean–standard deviation method specifies class limits
the high correlation coefficients found in Figure 1C
by repeatedly adding or subtracting the standard devia-
(remember that 1.0 is the maximum possible value for r).
tion from the mean. The results of applying the
The lack of a strong visual association between the skewed
mean–standard deviation approach to the hypothetical
* These visual assessments were confirmed by a Shapiro–Wilk test, an
distributions are shown in Figure 1D. The visual appear-
inferential test for normality (Stevens 1996, 245). ance of these maps supports the high correlation
† Kendall’s rank correlations (Burt and Barber 1996, 396–398), which between the normal distributions, but not the high cor-
are arguably more appropriate for skewed data, were 0.98 between relations between the skewed and normal distributions.
attributes 1 and 2 and 1 and 3 and 1.00 between attributes 2 and 3. These results are not surprising, given the fact that
18:02:32.
A D
Optimal
Attribute 1 Attribute 2 Attribute 3
Attribute 1
0 5 10 15 20 25
Attribute 2
20 30 40 50 60 70
Mean–Standard Deviation
Attribute 3
125 130 135 140 145 150
B
Nested Means

1 0 5 4 37 26 47 44 131 126 139 137
1 4 13 23 32 43 48 56 129 135 140 147
2 3 2 24 40 42 41 63 132 134 133 148
14 16 4 25 50 53 45 68 142 144 138 149
Quantiles
Correlation Coefficients
Attribute
Unclassed
1 2 3
1 1.00 0.92 0.94
Attribute
2 0.92 1.00 0.98
3 0.94 0.98 1.00
FIGURE 1 Using choropleth maps to compare geographic distributions: (A) three hypothetical attributes (attribute 1 is positively
skewed, whereas attributes 2 and 3 are normal); (B) maps of the raw data for the three attributes; (C) correlation coefficients (r
values) between each pair of attributes; (D) maps for differing methods of classification.
373
18:02:32.
means and standard deviations generally should be used data, we concluded that the optimal method is inappro-
only with normal distributions. priate. It should be noted, however, that if we wished to
In the nested-means approach, the mean of the data is compare the same attribute for two points in time (say,
used to divide the data into two classes—values above median income for 1990 and 2000), the data could be
and below the mean. The resulting classes can be further combined into a single data set, and the optimal method
subdivided by again computing the mean of each class, could be applied to the combined set.
and this process can be repeated until no further subdivi- One limitation of this section is that our conclusions
sion is desired.The results of using nested means to create are based on a subjective interpretation of only those
four classes are shown in Figure 1D. Like the mean–stan- attributes shown in Figure 1. To alleviate this problem,
dard deviation approach, nested means seems to do a bet- let us now consider some of the more formal studies
ter job of portraying the similarity of the normal distribu- done by cartographers. Robert Lloyd and Theodore
tions. This is to be expected, as the mean used to define Steinke (1976; 1977) found that the visual correlation of
class breaks is an appropriate measure of central ten- maps is affected by the amount of blackness on each
dency for normal distributions, but not for skewed ones. map (assuming that gray tones are used for symboliza-
Another weakness of nested means is that the number of tion); in other words, if maps A and B and C and D have
classes can be only a power of 2 (2, 4, 8, 16, etc.); this com- the same statistical correlation, maps A and B will be
plicates a comparison of this approach with the other judged more similar if their blackness levels are more
methods of classification. similar than the blackness levels of maps C and D. As a
The quantiles method of classification places an result, Lloyd and Steinke argued for using equal areas
approximately equal number of observations in each (which, as we have indicated, produces results similar to
class based on the ranks of the data. Because the classes quantiles) when comparing choropleth maps.
resulting from the quantiles method are unaffected by Judy Olson undertook two studies to determine
the magnitudes of the data, the method is arguably which classification methods appeared to preserve the
appropriate for comparing differently shaped distribu- correlation between two attributes. In Olson’s (1972b)
tions. For example, for the hypothetical data, the quan- first study, she analyzed 300 pairs of attributes derived
tiles method portrays high correlations between not only from theoretically normal distributions and found that
the normal distributions but also the skewed and normal the quantiles method was best at reflecting the correla-
distributions (Figure 1D). The equal-areas method of tion between the attributes. In her second study, Olson
classification is similar in concept to quantiles, but rather (1972a) analyzed 300 pairs of real-world attributes
than placing an equal number of observations in classes, (which were primarily not normally distributed) and
an equal portion of the map area is assigned (i.e., the found that the mean–standard deviation and nested-
desired area in each class is simply the area of the map means methods were most effective. These results con-
divided by the number of classes desired). If enumera- flict with some of our own conclusions because of
tion units are equal in size (as in the hypothetical data), Olson’s stress on broad relationships (she focused on a
the equal-areas method produces a map identical to scatterplot of the correlation coefficient and on a mea-
quantiles. sure of rank correlation for all pairs of attributes) as
Up to this point, we have focused on comparing opposed to looking at detailed graphs and maps of each
classed choropleth maps. As with univariate choropleth distribution, as was done in Figure 1.
maps, it is natural to ask whether the issue of selecting Studies by Michael Peterson (1979) and Jean-Claude
an appropriate classification method might be obviated Muller (1980b) dealt with people’s visual comparison of
simply by not classing the data. To illustrate, consider both classed and unclassed choropleth maps. These
the unclassed maps of the hypothetical data shown in studies concluded that people perceive similar correla-
Figure 1D. Here the visual impression seems similar to tions on pairs of classed and unclassed maps; as a result,
that of the mean–standard deviation and nested-means the authors raised questions about the need to class
approaches, as attributes 2 and 3 (the normal distribu- data for choropleth mapping. Although the results of
tions) appear similar to one another, whereas attribute Peterson and Muller’s studies contradict our recom-
1 (the skewed distribution) does not appear as highly mendation to use unclassed maps only when comparing
correlated with the other attributes. This result leads distributions having the same shape, we suspect that
one to the conclusion that unclassed maps should be a detailed examination of data distributions would
used for comparative purposes only when the distribu- support our recommendation.
tions have similar shapes. Strong support for our conclusion that quantiles is
At the beginning of this section, we assumed that we particularly appropriate for map comparison comes
wished to compare two attributes for a single point in from the work of Cynthia Brewer and Linda Pickle
time (say, median income and percent of the adult (2002). In comparing a wide variety of classification
population with a college education for 2000). For such methods in actual map comparison tasks, they found
18:02:32.
quantiles to be the most effective (a minimum boundary high-rate area is most problematic because it is located
error method based on Cromley’s (1996) work was a where the raw number of deaths is high.
close second). They concluded that “quantiles seems to
be one of the best methods for facilitating comparison 1.2 Combining Two Attributes on the Same Map
as well as aiding general map-reading” (p. 679).
In this section, we consider approaches for bivariate
Miscellaneous Thematic Maps mapping in which two attributes are combined on the
same map. Again, we focus on choropleth maps because
Although it is common to hold the mapping method
of their common use.
constant when comparing maps (e.g., showing two
choropleth maps or two isopleth maps), useful infor- Bivariate Choropleth Maps
mation often can be acquired by comparing two dif-
In the 1970s, the U.S. Bureau of the Census developed
ferent kinds of thematic maps. This is especially true
the bivariate choropleth map—a method for combining
when one map is used to show raw totals and another
two colored choropleth maps (see Meyer et al. 1975 for
map is used to show standardized data. For example,
a summary of the technical methods used at that time).
consider Figure 2, which compares a proportional sym-
Color Plate 1 depicts a bivariate choropleth map using
bol map of the raw number of infant mortalities in
a color scheme similar to the one used by the Bureau of
New Jersey with a choropleth map of the infant mor-
the Census, along with each of the univariate distribu-
tality rate. The proportional symbol map suggests that
tions used to create the bivariate map.* A listing of the
the “problem” of infant mortality is in the northeast-
factors considered in constructing the Bureau of Census
ern part of the state (where the largest circles are).
color schemes is given in Table 2.
This map by itself, however, is not very meaningful
Although bivariate choropleth maps were deemed a
because the pattern is likely a function of population
success from a technical standpoint, they received con-
(counties with more people are apt to have more
siderable criticism for their presumed failure to commu-
infant deaths). In contrast, the choropleth map stan-
nicate either information about individual distributions
dardizes the raw mortality data by considering the
or the correlation between them. In response to these
number of deaths relative to the number of live births,
criticisms, Judy Olson (1981) conducted an experimental
and suggests that the “problem” is found in three areas
study using color schemes similar to those used by the
of the state (represented by the shades in the highest
Bureau of the Census. In contrast to the earlier criticism,
class). Unfortunately, a high rate on the choropleth
map might not be meaningful if there are also few * Specifications for the colors used in Color Plate 1 were taken from
deaths. Only when the two maps are viewed together Olson (1981, 269). For a discussion of the logic behind the selection of
can the complete picture emerge: The northernmost these colors, see Olson (1975b).
Infant Mortality in New Jersey, 1988 FIGURE 2 A comparison of

Deaths of Persons Less Than One Year Old by County of Residence proportional symbol and
choropleth maps. The pro-
portional symbol map pro-
vides information on the raw
number of infant mortalities,
whereas the choropleth
Infant Mortality Rate map focuses on the rate of
Number of Deaths
Deaths per 1,000 infant mortality. To under-
7 Live Births
25 stand the complete picture
4.7
50 of infant mortality, both
7
maps are necessary. (From
100 9.9* Monmonier, Mapping It Out:
12 Expository Cartography for
214
15.5 the Humanities and Social
Statewide Total = 1,168 *U.S. mean = 9.9 Sciences, p. 166; © 1993 by
The University of Chicago
Press. All rights reserved.
Published 1993.)
375
18:02:32.
TABLE 2 Factors considered in developing color As an alternative to the Bureau of the Census
schemes for the U.S. Bureau of the Census bivariate approach, J. Ronald Eyton (1984a) developed a bivari-
choropleth maps ate method based on complementary colors, or colors
1. All colors must be distinguishable. that combine to produce a shade of gray.† Eyton chose
2. The transition of colors should progress smoothly in a red and cyan as the complementary colors for his maps,
visually coherent way. presumably because they produced an attractive map.
3. The individual categories of each distribution should be Using the subtractive primary colors (CMY), Eyton
visually distinguishable or coherent, and the two distribu- created red by combining magenta (M) and yellow (Y),
tions as a whole should be separable from one another.
and he created cyan (C) directly from the cyan subtrac-
4. The arrangement of the colors presented in the legend
should correspond to the arrangement of a scatter diagram. tive primary. Because overprinting colors in CMY does
5. Tones should progress from lighter to darker corresponding not produce a true gray, Eyton used black ink (K)
to a change in the numerical values from low to high. for areas in which a true gray was desired. A bivariate
6. Extreme values (legend corners) should be represented by map resulting from Eyton’s process is shown in Color
pure colors. Plate 2.‡
7. There should be coherence in the triangle of cells above
Eyton argued that most of the factors listed in
and below the main diagonals to show positive and nega-
tive residuals. Table 2 were accounted for by his complementary
8. To convey relationships, positive diagonals (lower left to method and that users appeared to understand the
upper right) and negative diagonals (upper left to lower map more easily than one based on Bureau of the
right) should have visual coherence. Census colors. A visual comparison of Color Plates 1
9. The design of the color-coding scheme should take into and 2 suggests that Eyton was correct. The Bureau of
account the difficulty in mentally sorting large numbers of
the Census color scheme implies nominal differences
colors in the legend.
10. The color scheme should relate to the data in such a way and requires careful examination of the legend,
that the map relationship reflects as closely as possible whereas Eyton’s scheme appears more logically
the statistical relationship. ordered and allows patterns on the map to be dis-
11. The crossed version of the map should be constructed as a cerned more easily. Note, for example, the ease with
direct combination of the specific sets of colors assigned to which the reddish-brown values (those above the
the two individual maps.
12. The combination of colors on the two individual maps
white–gray–black diagonal) can be found using the
should look like combinations of the specific colors Eyton scheme, compared with the corresponding val-
involved. ues using the Bureau of the Census color scheme.
13. The number of categories to be used should not exceed the In addition to promoting complementary colors,
number that can be dealt with by the reader. A 3 * 3 legend Eyton also recommended that statistical parameters
is both mechanically and visually simpler than a 4 * 4
associated with the attributes be considered in bivariate
arrangement and might actually convey more to the reader.
14. Alternatives to a rectangular arrangement to the legend mapping. Specifically, Eyton utilized the reduced major
should be considered. The rectangular form creates map axis and bivariate normality. The reduced major axis
interpretation problems and affects the message of the method fits a regression line to a set of data such that
statistical relationship. the line bisects the regression lines of Y on X (Y is
After Olson 1975b, as specified by Eyton, J. R. (1984a) “Complementary-
treated as the dependent attribute) and X on Y (X is
color two-variable maps.” Annals, Association of American Geographers, treated as the dependent attribute; see Figure 3). The
74, no. 3, p. 480. Courtesy of Blackwell Publishing. reduced major axis is thus appropriate when it is not
clear which attribute should be treated as the depen-
Olson found that bivariate maps provided “information dent attribute.
about regions of homogeneous value combinations” and A distribution is considered bivariate-normal if the Y
that users, rather than being confused by these maps, values associated with a given X value are normal and if
actually had a positive attitude about them (p. 275). the X values associated with a given Y value are also
Olson, however, stressed that a clear legend was critical normal. Generally, this condition is met if the individual
to understanding bivariate maps, that both bivariate and distributions of X and Y are normal. Given a bivariate-
individual maps should be shown,* and that an explana- normal distribution, it is possible to construct an
tory note should be included describing the types of equiprobability ellipse enclosing a specified percentage
information that could be extracted (see p. 273 of of the data. The equiprobability ellipse will be centered
Olson’s work for an example).
† Colors opposite one another on the Munsell color circle are
* The univariate maps should be shown in black and white so that the complementary.
two distributions can be readily compared. In Color Plate 1 they are ‡ Eyton used cross-hatching (narrowly spaced horizontal and vertical
shown in color so that the colors composing the bivariate map are lines) for intermediate shades, whereas we have chosen to use smooth
clear. tones.
18:02:32.
diagonal of purple-blues.) If one attribute is unipolar

3 X on Y and the other bipolar, she recommended using a
Reduced
Major Axis sequential scheme for the unipolar data and a diverging
scheme for the bipolar data. If both are bipolar, she rec-
2 Y on X
ommended using two diverging color schemes (Color
Plate 4).
Standardized Y values
A potential concern in bivariate choropleth mapping

1
is the number of classes used for each attribute. In the
examples presented thus far, we have used three classes
0 for each attribute. Our thinking here parallels item 13
X, Y
in Table 2, which states, “The number of categories to
Equiprobability be used should not exceed the number that can be dealt
-1 ellipse with by the reader.”
The approaches we have considered to this point
for bivariate choropleth mapping have all been based
-2
on smooth (untextured) colors. As an alternative, both
Laurence Carstensen (1982; 1986a; 1986b) and
-3
Stephen Lavin and J. Clark Archer (1984) created
-3 -2 -1 0 1 2 3 bivariate maps using cross-hatched shading con-
Standardized X values sisting of horizontal and vertical lines (Figure 4).
Interpretation of these maps focuses on the size and
shape of the boxes formed by the cross-hatched lines
FIGURE 3 A reduced major axis and associated equiproba-
(low values on both attributes are represented by
bility ellipse. The reduced major axis is appropriate when one
large squares, whereas high values on both attributes
does not wish to specify a dependent attribute. The equiprob-
are represented by small squares). A high positive
ability ellipse can be used to enclose a specified percentage of
correlation is represented on the map by a predomi-
the data associated with a bivariate normal distribution.
nance of squares of varying sizes within enumeration
units, and a high negative correlation is depicted by
rectangles. In the case of Figure 4, we see a fair num-
on the mean of the data and oriented in the direction of ber of rectangles, which supports the correlation of
the reduced major axis (Figure 3). Using this ellipse and r = - .71.
the means of the two attributes, Eyton created a bivari- Both Carstensen and Lavin and Archer stressed
ate map analogous to the one shown in Color that unclassed cross-hatched bivariate maps should be
Plate 3.* Eyton argued that the resulting map clearly con- used (thus the legend in Figure 4 contains no class
trasted observations near the means of the data (as breaks). This suggestion contrasts with that of Eyton
defined by the 50 percent equiprobability ellipse) with (1984a, 485–486), who found smooth-toned unclassed
extreme observations (i.e., those well above the mean on bivariate maps difficult to interpret. The reason that
both attributes, those well below the mean on both attrib- unclassed cross-hatched maps appear to be more
utes, and those high on one attribute and low on the effective is that the line-spacing attribute allows indi-
other). For example, in Color Plate 3, we can clearly see vidual attributes to be seen. Carstensen (1982) found
that the countries of Botswana, Zimbabwe, Lesotho, and that cross-hatched bivariate maps are reasonably
Kenya (those shown in cyan) have a low percent of urban effective in communicating concepts about correla-
and high female life expectancy. tion, but noted two problems: the difficulty of shading
Cynthia Brewer (1994a) argued that color schemes small enumeration units and the unpleasant appear-
for bivariate maps should be a function of whether the ance of the symbology. These problems cause this
attributes are unipolar or bipolar in nature. For two method to be used less frequently than the color meth-
unipolar data sets (e.g., percent urban and female life ods mentioned earlier.
expectancy), she recommended using either a comple-
mentary scheme or two subtractive primaries; the latter Alternative Bivariate Maps
produces a diagonal of shades of a constant hue, as The bivariate point symbol is one alternative to the
opposed to the grays resulting from complementary bivariate choropleth map. Figure 5 depicts one form of
colors. (Brewer combined magenta and cyan to create a bivariate point symbol, the rectangular point symbol,
in which the width and height of a rectangle are made
*Given the requirement of normality, we chose to map two normal proportional to each of the attributes being mapped.
distributions: percent urban and life expectancy for females. Both Stanton Wilhelm (1983) and Sean Hartnett (1987)
377
18:02:32.
Median Asking Price and Percent of

Homes Lacking Plumbing
Wisconsin, 1980
6.8
6.1
Percent Lacking Plumbing

5.3
4.6
3.8
3.1
2.4
1.6
0.9
0.1
-0.6
-1
-2
7,
17
27
37
46
56
66
76
86
48
2,
,3
,3
,2
,1
,9
,8
,7
,6
,5
26
89
64
40
17
94
70
47
23
00
6
Median Asking Price ($)
FIGURE 4 A bivariate choropleth map based on cross-hatching (the attributes are represented by horizontal and vertical lines
of varying spacing). Note that each attribute is unclassed; negative values in the legend are a function of the major axis scaling
used to fit the bivariate data. (From Carstensen 1986a, p. 36; courtesy of Laurence W. Carstensen.)
proposed this method as an alternative to cross-

Homes Heated with Wood or Utility Gas hatched symbology; Hartnett argued that examining
1980 25 rectangular point symbols is much easier than inspect-
ing the small boxes formed by cross-hatched lines and
Wood (%)
8
that the resulting map is more aesthetically pleasing.
4 Because point symbols are more readily associated
0 with point locations than with areas, bivariate point
0 20 40 82
Utility Gas (%) symbols might be particularly appropriate for mapping
25 true point data (say, the number of out-of-wedlock
births to teenagers and the number of hours of sex edu-
Wood (%)
15 cation for high school students for major cities in the

8
United States).*
4
Another form of bivariate point symbol is the
0
0 20 40 80 bivariate ray-glyph, which Daniel Carr and his col-
Utility Gas (%)
leagues (1992) used to examine the relationship between
r = -0.8098
nitrate (NO3 ) and sulfate (SO4 ) concentrations in the
eastern United States and Canada. In Figure 6, rays (the
FIGURE 5 A bivariate map in which attributes are repre-
sented by the width and height of a rectangular point symbol.
(Courtesy of Sean Hartnett.) * Ellipses would be an alternative to rectangles (MacEachren 1995, 95).
18:02:32.
2 MULTIVARIATE MAPPING INVOLVING

Nitrate and Sulfate Concentrations THREE OR MORE ATTRIBUTES
2.1 Comparing Maps
If more than two attributes are symbolized, each as a

separate map, the result is termed a small multiple
(Figure 7). Edward Tufte (1990, 33) argued, “Small mul-
tiples, whether tabular or pictorial, move to the heart of
visual reasoning—to see, distinguish, choose . . . . Their
multiplied smallness enforces local comparisons within
our eyespan, relying on the active eye to select and
Bivariate Trends
1982 to 1987
make contrasts.” Although much can be gleaned from
kg/ha Per Year small multiples, they clearly have their limitations. A
general problem is that comparing two particular points
NO3 SO4
or areas across a set of attributes can be difficult: Try
1.6 2.4
using Figure 7 to describe the nature of agriculture in
0.9 1.1
the state of Michigan. In the case of choropleth maps,
0.2 -0.1
one problem is the difficulty of discerning small enu-
-0.5 -1.4 meration units (e.g., a number of countries would dis-
-1.2 -2.6 appear in a small multiple of choropleth maps of
Africa). Problems such as these have led researchers to
develop methods for combining multiple attributes on
FIGURE 6 A bivariate map based on a ray-glyph symbol. the same map.
Rays (straight-line segments) pointing to the right and left
represent sulfate and nitrate concentrations, respectively.
(After Carr et al. 1992. First published in Cartography and
2.2 Combining Attributes on the Same Map
Geographic Information Systems 19(4), p. 234. Reprinted with
permission from the American Congress on Surveying and
Trivariate Choropleth Maps
Mapping.) The notion of overlaying two colored choropleth maps
can be extended to three choropleth maps, thus produc-
ing a trivariate choropleth map. Ideally, this approach
straight-line segments) pointing to the left represent should be used only for three attributes that add to 100
nitrate concentrations, and rays pointing to the right percent; examples include soil texture (expressed as per-
represent sulfate concentrations. High values on the cent sand, silt, and clay) and voting data for three political
attributes occur when the rays extend toward the top of parties (e.g., percent voting Republican, Democrat, and
the symbol, whereas low values occur when the rays independent). Colors can be assigned to the three attrib-
extend toward the bottom of the symbol. An advantage utes using a variety of approaches: CMY (Brewer 1994a,
of the ray-glyph is that the small symbols can be 142; Metternicht and Stott 2003); RGB (Byron 1994); and
squeezed into a relatively restricted space. It seems red, blue, and yellow primaries (Dorling 1993, 172). The
likely, however, that the patterns represented by these result of using RGB is shown in Color Plate 5.The advan-
symbols would be more difficult to interpret than those tage of using attributes that add to 100 percent is that the
for the rectangular symbol. resulting colors will be restricted to a triangular two-
A relatively common approach in bivariate mapping dimensional space. Brewer (1994a) argued that if the
is to combine proportional and choropleth symbols with attributes do not sum to 100 percent, a 3-D cube-shaped
the size of the proportional symbol used for raw-total legend will be required and the resulting map will be dif-
data and a choropleth shade within the symbol used for ficult to interpret.
standardized data. For example, this approach could be An alternative to creating a trivariate choropleth
used to depict on a single map the infant mortality data map is to use patterns (or textures) as a substitute for
presented as two maps in Figure 2. These same symbols smooth, colored tones. For instance, Victoria Interrante
could be used redundantly (that is, both size and shad- (2000) created the multivariate map of agriculture for
ing could be used to represent a single attribute). U.S. counties shown in Color Plate 6 by using texture
Obviously, a legend is critical to determining whether for two attributes and color for a third attribute. The
the symbols should be interpreted in a bivariate or a type of texture (weave vs. rocks) was used for percent
redundant fashion. change in the number of farms, and the size of the
379
18:02:32.
A B C
D E F
G H I
FIGURE 7 The small multiple: a method for multivariate mapping in which each attribute is depicted as a separate map.
Maps depict the following agricultural data for 1954 for the central portion of the United States: (A) class V farms, (B) sugar
cane, (C) cotton, (D) turkeys, (E) pasture, (F) hogs and pigs, (G) dairy, (H) expenditures for lime, and (I) residential farms.
(After Lindberg 1987; courtesy of Mark B. Lindberg.)
18:02:32.
texture (small vs. large rocks) was used for the land in colored dot maps, but unfortunately his method has not
farms. Color was used to depict the third attribute been published.
(average age of farm operators). Here we can see some A general question that can be asked of multivariate
of the difficulty of interpreting three attributes simulta- maps is how effective they are compared to a set of indi-
neously (for instance, try describing the character of vidual maps of each attribute (i.e., how would a multi-
agriculture throughout Nevada based on the three variate colored dot map compare to a set of dot maps in
attributes), and of creating a suitable legend (the legend small multiple format?). Jill Rogers and Richard Groop
shown here is a modification of Interrante’s). (1981) evaluated this question by having readers identify
regions on both a multivariate dot map (consisting of
Multivariate Dot Maps three categories) and its component univariate dot maps.
Readers were asked to identify both “homogeneous
The notion of univariate dot mapping can be extended regions” in which one category appeared to predominate
to create a multivariate dot map if a distinct shape or and “mixed regions” in which there appeared to be a mix-
color of symbol is used for each attribute to be mapped.* ture of two or three categories. Rogers and Groop found
In the case of color, George Jenks (1953b) introduced that the multivariate dot map “was slightly more effective
the notion of pointillism for multivariate dot mapping in communicating perceptions of both homogeneous and
and developed two major maps based on it (Jenks 1961; mixed regions” (p. 61).Their results, however, suggest that
1962). Pointillism was a technique used by nineteenth- the perception of mixed regions on maps with more than
century painters to create various color mixtures by three categories might not be effective.
assuming the viewer would visually combine very small
dots of selected colors. Jenks applied this principle by
letting different-colored dots represent various crops (or Multivariate Point Symbol Maps
farm products; see Color Plate 7); he argued that view- When multivariate data are depicted using a point sym-
ers could visually merge the separate colors to create bol, the result is termed a multivariate point symbol.
mixtures, thus gaining a more realistic view of the transi- These symbols are obviously appropriate for point phe-
tional nature of cropping practices often found in the nomena, but they must also be used for areal phenomena
landscape. Furthermore, Jenks noted that if the dots because of the difficulty of creating multivariate areal
were large enough, the map could provide detail regard- symbols (note the limitations of the trivariate choropleth
ing the location of individual crops in selected areas. map described previously when only three attributes are
Jenks (1953b, 5) also provided some useful suggestions shown). Two distinct forms of attributes are commonly
for dot colors: mapped using multivariate point symbols: related (or
1. Colors should remind the map reader of the crop additive) and nonrelated (or nonadditive). Related attrib-
that they represent. utes are measured in the same units and are part of a
larger whole. An example would be the percentages of
2. High-value, low-acreage crops, such as tobacco or various racial groups in a population: White, African-
truck, should be of more intense hue than the more American, Asian/Pacific Islander, Native American, and
extensive and widely grown crops. so on. Such attributes can be depicted using the familiar
3. Selected minor crops, such as peanuts or soy- pie chart, in which a circle is divided into sectors repre-
beans, which tend to change the crop character of senting the proportion of each attribute.
broader areas, should have colors of moderately Nonrelated attributes are measured in dissimilar units
high intensity. and thus are not part of a larger whole (e.g., percent urban
and median income). Multivariate point symbols used to
Although considerable information can be obtained
represent nonrelated attributes are commonly termed
from multivariate dot maps, readers have difficulty
glyphs, several of which are illustrated in Figure 8. The
determining the meaning of the areas of color mixture
multivariate ray-glyph or star (Figure 8A) is constructed
because the legend contains colors for only individual
by extending rays from an interior circle, with the lengths
crops. This problem might be solved if the map were
of the rays proportional to the values of each attribute.
viewed in an interactive data exploration environment
When originally designed by Edgar Anderson (1960), rays
in which individual map areas could be focused on and
were extended only from the top portion of the circle, but
enlarged, and individual categories of dots could be
it has become common to extend rays in all directions.† If
turned on and off so that the relative contribution of
each category could be determined. Richard Groop
(1992) developed an automated method for creating † Several terms have been used to describe this symbol. The term star
comes from Borg and Staufenbiel (1992), whereas Anderson called
*For a discussion of the use of shape on multivariate dot maps, see them metroglyphs. We use the term multivariate ray-glyph because of
Turner (1977). their similarity to the bivariate ray-glyph introduced earlier.
381
18:02:32.
A B C D E
FIGURE 8 Examples of multivariate point symbols: (A) a multivariate ray-glyph or star: the length of rays are proportional to
the values of attributes; (B) a polygonal glyph or snowflake: a polygon connects the endpoints of the rays shown in A; (C) three-
dimensional bars: the height of bars is proportional to the magnitude of attributes; (D) data jacks: the spikes of the jack are pro-
portional to the magnitude of each attribute; and (E) Chernoff faces: individual facial features (e.g., the size of the eyes) are asso-
ciated with individual attributes.
rays composing the multivariate ray-glyph are connected,

a polygonal glyph or snowflake is created (Figure 8B). Low High
Donna Cox (1990) and her colleagues at the Service Employees (%)
National Center for Supercomputing Applications at Electorate Voting (%)
the University of Illinois at Urbana-Champaign have Adult Employment (%)

Mean Housing Price (%)
developed several novel multivariate point symbols.
One of these is three-dimensional bars, in which the
height of the bars is made proportional to the magni-
tude of various attributes (Figure 8C); in Cox’s imple-
mentation, the individual bars were shown in different
colors, as opposed to using the different textures seen
in Figure 8C. Another technique is the data jack
(Ellson 1990), in which triangular spikes are extended
from a square central area and made proportional to
the magnitude of each attribute (Figure 8D). As with
bars, the spikes of jacks can be distinguished most eas-
ily if they are displayed in different colors. An advan-
tage of the 3-D structure of jacks is that they can be
viewed from arbitrary positions in 3-D space.
A particularly intriguing multivariate point symbol is
the Chernoff face (Figure 8E), in which distinct facial
features are associated with various attributes. For exam-
ple, fatness of the cheeks might represent one attribute,
whereas size of the eyes might represent another
attribute. Figure 9, taken from Daniel Dorling’s work
with cartograms, is illustrative of Chernoff faces. In this
case, the width of the face represents mean house price
(fat cheeks indicate more expensive housing); mouth
style represents the percent of adult employment (a
smile for high employment); nose size represents the per-
cent voting (a larger nose indicates a higher percent); eye FIGURE 9 A cartogram in which Chernoff faces are used to
size and position represent the percent employed in ser- display multivariate data. (Environment and Planning B:
vices (large eyes located near the nose indicate a high Planning and Design 1995, Vol. 22, pp. 269–290, Pion,
percent employed in services); and total area of the face London.)
18:02:32.
represents the number of voters in a parliamentary

constituency.
For simplicity, only four attributes are depicted by
the symbols shown in Figure 8. Several of these sym-
bols, however, can be modified to represent many more
attributes. For example, numerous rays can be added to
the star, and up to 20 different attributes can be repre-
sented by Chernoff faces (Wang and Lake 1978, 32–35).
The multivariate point symbols that we have consid-
ered thus far are often associated with points or areas in
which there is some spacing between the point symbols.
In contrast, several multivariate point symbols have
been developed that are intended to depict numerical
data that change continuously across geographic space
(and thus there may be little or no spacing between
point symbols). One example is the “stick-figure” icons
(Figure 10) used within Exvis, a system for exploratory
visualization developed at the University of Lowell,
MA. Icons within Exvis were selected on the basis of
preattentive processing, or “the ability to sense differ-
ences in shapes or patterns without having to focus
attention on the specific characteristics that make them FIGURE 11 A map created using Exvis. Five satellite images
different” (Grinstein et al. 1992, 638). Up to 15 attributes were combined to display water features in the eastern por-
could be mapped using one of these icons by varying the tion of the Great Lakes. (From Smith et al. 1991,
angles, lengths, and intensities of the limbs composing p. 197; courtesy of SPIE—The International Society for
the icon.* Grinstein et al. noted that when the icons are Optical Engineering and Stuart Smith.)
dense enough, “they form a surface texture” and that
“structures in the data are revealed as streaks, gradients,
or islands of contrasting texture” (p. 638). A geographic been combined into a single image. In this case, Stuart
example is shown in Figure 11, where five satellite Smith and his colleagues (1991) argued that “We can see
images of the eastern portion of the Great Lakes have not only Lake Erie and Lake Ontario, but also Lake
* Technically, they specified 17 attributes, but 2 of these are accounted
Huron, Georgian Bay, Lake Simcoe, and some of the
for by the x and y coordinate location of the icon.
smaller outlying lakes” (p. 197).
A second form of multivariate point symbol
intended for continuous geographic data is the “percep-
tual texture elements” (or pexels) developed by
Christopher Healey and his colleagues at North
Carolina State University (Healey and Enns 1999;
Healey 2001). Pexels are small 3-D bars (similar to
those utilized by Cox and her colleagues) that can be
varied in height, spacing, and color.† As an illustration,
consider Color Plate 8, which portrays three attributes
related to plankton in the northern Pacific Ocean
(plankton are an important food source for salmon).
The density of plankton is indicated by color, with low
to high plankton values indicated by blue, green, brown,
FIGURE 10 Some stick-figure icons (a form of multivariate
red, and purple, respectively. The speed of ocean cur-
point symbol) used in Exvis, a system for exploratory visual-
rent is indicated by the height of pexels (taller pexels
ization. Multiple attributes can be mapped using one of these
indicate a stronger current), and the magnitude of sea
icons by varying the angles, lengths, and intensities of the
surface temperature is indicated by the spacing of pex-
limbs composing the icon. (“Visualization for Knowledge
els (a tighter spacing indicates a warmer temperature).
Discovery,” G. Grinstein, J. C. J. Sieg, S. Smith, and M. G.
In examining Color Plate 8A, we see that in February,
Williams, from the International Journal of Intelligent Systems,
Vol. 7. Copyright 1992 by Wiley Publishing, Inc. All rights †Pexels can also vary in orientation, and the density can include a
reserved. Reproduced here by permission of the publisher.) regularity parameter, but we do not illustrate these here.
383
18:02:32.
the density of plankton is relatively low (we see mostly assigned to each attribute; in Figure 12, shades of cyan,
blues and greens). At this time, ocean currents are rela- green, red, and blue are presumed. Color Plate 9 illus-
tively weak in the north central Pacific and along the trates the attribute block approach in mapped form,
south coast of Alaska (low bars are found at these where four attributes associated with the water balance
spots), and most of the ocean is relatively cold (pexels in January for the United States and portions of Canada
are relatively widely spaced), although a warmer region and Mexico are depicted: potential evapotranspiration
can be seen in the south. In contrast, in June (Color (cyans), soil moisture holding capacity (greens), tem-
Plate 8B) we see a higher density of plankton (reds and perature (reds), and precipitation (blues). Miller argues
purples predominate), stronger currents seem to have that these colors are commonly used in climatology
shifted southward, and water is warmer farther north, applications. On the map we can see several key char-
although the ocean off the Alaskan and British acteristics associated with each attribute: high precipita-
Columbia coasts is still relatively cold. These sorts of tion in the northwestern United States (the dark blue
summaries come not only from looking at static images region), high soil moisture holding capacity in the upper
such as those shown here, but also from using interac- Midwest (the bright green region), higher temperatures
tive software that allows the user to view the 3-D image as we move south (the brighter red areas), and a some-
from different angles, as well as zoom in and pan what higher potential evapotranspiration in Florida,
around. It should be noted that Healey and his col- Mexico, and the Baja (the brighter cyan areas). One
leagues undertook numerous experiments to ensure problem with Miller’s approach is that the map may be
that they utilized a reasonable set of parameters for misinterpreted if the attribute blocks on the map are
pexels. not related carefully to the color scheme. For instance, a
James Miller (2007) recently developed still another bright cyan might be misinterpreted as a blue if the
form of multivariate point symbol intended for data position of cyan within the attribute block is not care-
that change continuously across geographic space. fully noted. Miller, however, argues that his technique
Miller’s approach utilizes a block of pixels to represent compares favorably with both the stick-figure and the
each attribute, with the blocks arranged in rows and pexel approaches.
columns (Figure 12). Miller terms the resulting symbol Interestingly, nongeographers developed most of the
an attribute block. In Figure 12, the complete block is multivariate point symbols we have described. Micha
10 pixels x 10 pixels, and the four individual attributes Pazner is one geographer who has experimented with
are 5 pixels * 5 pixels. Note that a color scheme is multivariate point symbols. With his colleagues at the
University of Western Ontario (Pazner and Lafreniere
1997; Zhang and Pazner 2004), Pazner has developed
an approach based on the geo-icon, in which adjacent
10 Pixels
pixels are combined in various ways to represent differ-
ent attributes. For instance, in Figure 13 pixels are com-
5 Pixels 5 Pixels bined to create various shapes that are intended to
depict the attributes associated with the ability to tra-
verse off-road terrain: (1) road proximity, (2) stream
5 Pixels
Shades Shades proximity, (3) slope steepness, (4) terrain roughness, and
of Cyan of Green (5) vegetation resistance. Note how each of the symbols
serves to signify the attribute of interest. The entire geo-
10 Pixels
icon is 20 pixels * 20 pixels in size. The values for each

attribute are depicted by changing the color within each
symbol. For instance, Color Plate 10 illustrates the
5 Pixels
Shades Shades
of Red of Blue results for the off-road travel example using a blue to
red color sequence for each symbol. In A, B, and C of
the Color Plate, we see what the results look like for
varying resolutions, ranging from a small scale (A) to a
large scale (C). Zhang and Pazner (p. 39) admit that
such geo-icon images “. . . are difficult to read, and
users find their interpretation tedious and time con-
FIGURE 12 The attribute block, a technique for multivariate suming.” Zhang and Pazner’s experience with such
mapping, where each attribute is represented by a block of pix- images, however, is that they can provide useful multi-
els; in this case, each attribute is 5 pixels * 5 pixels, and the variate information, especially when the user is fully
entire block is 10 pixels * 10 pixels (after Miller 2007, p. 60). involved in the design of the image.
18:02:32.
attribute infant mortality); note that this is a difficult

task for the combined star and snowflake symbol. A
form of combined general information would involve
examining the pattern of the size of symbols across the
state (e.g., note that there appears to be a band of rela-
1 2 tively less desirable counties running through the
southern part of the state).
5 The apparent difficulty of visualizing the pattern of
an individual attribute in Figure 14 raises the question
of whether certain multivariate symbols might be more
effective for one form of information than for another.
Nelson and Gilmartin (1996) evaluated this question
by having map readers examine four types of multi-
4 3 variate point symbols: a modified Chernoff face, a cir-
cle divided into quadrants, a cross, and boxed letters
representing attribute names (Figure 15). Each point
FIGURE 13 The geo-icon, a technique for multivariate map- symbol represented four attributes depicting quality of
ping, in which adjacent pixels are combined in various ways to life on a map consisting of nine enumeration units.
represent different attributes; in this case, pixels are combined Nelson and Gilmartin found that all symbols were
to form shapes representing attributes related to the ability to processed equally well if time to examine the map was
traverse off-road terrain: (1) road proximity, (2) stream prox- not a factor. If time was a consideration, then boxed
imity, (3) terrain roughness, (4) slope steepness, and (5) vege- letters were the most effective (i.e., resulted in the
tation resistance. The entire geo-icon is 20 pixels x 20 pixels fastest reaction time), followed by crosses, divided cir-
(after Zhang and Pazner 2004, 37). cles, and Chernoff faces. Nelson and Gilmartin stressed
that certain symbols (e.g., Chernoff faces and letters)
worked better for individual attributes, whereas other
Specific versus General Information. Two kinds of symbols (e.g., crosses and circles) worked better for
information can be acquired from univariate thematic combined attributes.
maps: specific and general. These same kinds of infor- One problem with such experimental studies is that
mation can also be acquired from multivariate the- they often are unable to consider all of the factors that
matic maps, but the information can be acquired either might affect map reader performance. In Nelson and
for individual attributes or for combinations of attrib- Gilmartin’s case, they mentioned aesthetics (e.g., that
utes. Thus, we can conceive of the kinds of information Chernoff faces might be appropriate because of their
acquired from a multivariate map as a two-by-two attention-getting quality) and the number of attributes
matrix, with specific–general on one axis and nature of mapped (they noted that other researchers found
the attributes (i.e., individual or combined) on the Chernoff faces effective when displaying a large num-
other axis. Although it is clear that this notion could be ber of attributes) as potential factors to study. Other
applied to a broad range of multivariate techniques, factors mapmakers should consider include the diffi-
Elisabeth Nelson and Patricia Gilmartin (1996) illus- culty of discriminating attributes when small point
trated this notion in the context of multivariate point symbols are used (imagine interpreting the boxed let-
symbols. As an illustration, consider Figure 14, which ter for 20 attributes); the number of enumeration units
uses combined star and snowflake symbols to depict (Nelson and Gilmartin considered only a nine-county
“quality of life” in South Carolina counties in 1992. An region); and the viewing environment (e.g., in a 3-D
example of specific information for an individual environment, data jacks presumably would be more
attribute would be the desirability associated with a effective than, say, stars).
particular attribute within a county (for instance, that
the southernmost county is highly desirable in terms of Combining Different Types of Symbols
attribute 2, median income). A combined form of spe- In combining three or more attributes on the same map,
cific information would involve comparing the size of we have focused on using the same type of symbol for a
one symbol with the size of another (for instance, not- particular application (we used an area symbol for the
ing that the southernmost symbol is larger than the one trivariate choropleth map and a point symbol for the
to its immediate northwest). An example of general multivariate point symbol map). It is also possible to
information for an individual attribute would involve combine various symbol types to display multivariate
examining the distribution of a single attribute across data. A good example is the software SLCViewer, which
the state (e.g., examining the distribution of the David DiBiase and his colleagues (1994a, 303–309) used
385
18:02:32.
Quality of Life in South Carolina, 1992
8
7 1
6 2
5 3
4
’
8
7 1
6 2
5 3
4
7 8 1
6 2 Each variable is symbolized by rays of three lengths,
5 4 3 indicating low, medium, or high desirability.
FIGURE 14 A multivariate map based on a combination of the star and snowflake symbols shown in Figure 8. (Data source:
South Carolina State Budget and Control Board 1994.)
A B C D
EIH
C
FIGURE 15 Multivariate symbols used in a study by Nelson and Gilmartin (1996): (A) a modified Chernoff face; (B) a circle
divided into quadrants; (C) a cross; and (D) boxed letters representing attribute names.
18:02:32.
FIGURE 16 Using SLCViewer to create a multivariate map by combining point, line, and area symbolization. The attributes
mean annual evaporation, precipitation, and temperature associated with a climate model are displayed as proportional circles,
weighted isolines, and choropleth shading, respectively. (From DiBiase, D., Sloan, J. L., and Paradis, T. (1994b), “Weighted Isolines:
An alternative method for depicting statistical surfaces.” The Professional Geographer 46, no. 2, p. 219. Courtesy of Blackwell
Publishing.)
to explore data produced by climate models. SLCViewer bols for each attribute are separable or integral.
permitted the analyst to view up to four climatic attrib- Separable symbols are those that can be attended to
utes as small multiples or to overlay three attributes to independently, thus allowing the map reader to focus
create a multivariate map. For the multivariate map, on individual attributes. On cartograms, size and light-
point, line, and area symbolization could be overlaid. For ness are considered separable symbols, as illustrated
example, Figure 16 shows mean annual evaporation, pre- in Figure 17A. Here we can ignore the shading of the
cipitation, and temperature displayed as proportional states and focus on their size, noting the higher num-
circles, weighted isolines, and choropleth shading, ber of tornadoes that appear to occur in the Great
respectively. Weighted isolines were created by making Plains and southern states. Alternatively, we can
the width of contour lines proportional to the data; note ignore the size of the states and focus on the shading,
that this approach does not require labeling isolines noting regions where mobile homes are common.
because wider lines are logically associated with more of Integral symbols are those that map readers tend to
the phenomenon being mapped.* The obvious advan- integrate, meaning that individual attributes cannot
tage of using different symbol types is that one symbol be attended to easily, but that it is possible to examine
type will not conflict with another; thus, each attribute the correlation between attributes. For instance, in
can be seen individually, and the attributes can be Figure 17B we would have a difficult time identifying
related to one another. the spatial pattern of one of the attributes, say, per-
centage of murders committed with handguns, but we
Separable versus Integral Symbols can see the correlation between the two types of mur-
When multivariate data are displayed on a single map, ders—for instance, that the percentage of murders
a question that sometimes arises is whether the sym- due to both handguns and knives tends to be moder-
ate to high for many states along the east coast of the
* In an experimental study, DiBiase et al. (1994b) found that weighted
United States. The notion of separable and integral
isolines were more effective than traditional labeled isolines and symbols was developed in psychology, but has
shadowed contours when the objective was to detect low and high recently been promoted and studied by a cartogra-
areas in the data. pher—Elisabeth Nelson (2000). Thus far, Nelson has
387
18:02:32.
A
Tornadoes and Mobile Homes
State Housing Classified

as Mobile Homes (%)
0.5 - 4.9 5.0 - 9.9 10.0 - 17.0
Average Number of Tornadoes

Recorded Each Year, by State
100
50
20
5
B
Murder: Handguns versus Knives, 1996
Murders Committed
with Handguns (%)
Murders Committed
Low Medium High

with Knives (%)
Low Medium High
FIGURE 17 Examples of (A) sep-

arable and (B) integral symbols.
(Courtesy of Elisabeth Nelson.)
focused on bivariate maps, as the examples here illus- mentally overlay the maps to derive various combinations
trate, but the concept could also be extended to mul- of attributes for particular counties. More useful would be
tivariate maps. a method that combines the maps and tells us which
counties are similar to one another and how they are
similar—this is the purpose of the mathematical technique
3 CLUSTER ANALYSIS of cluster analysis, which we describe in this section.
Cluster analysis techniques can be divided into hierar-
Here we consider classification methods appropriate for chical and nonhierarchical methods. Hierarchical meth-
multiple numeric attributes. As an example, consider the ods begin by assuming that each observation (county, in
data shown in Table 3, representing the value of various the case of the New York data) is a separate cluster, and
attributes for counties in New York State (excluding New then progressively combine clusters.* The process is
York City) recorded by the U.S. Bureau of the Census for hierarchical in the sense that once observations are
the 2000 Census. Our interest is in whether there are cer- combined in a cluster, they remain in that cluster
tain counties that have similar scores on one or more of throughout the clustering procedure. Nonhierarchical
these attributes. For instance, we might wonder whether
some counties have a high percentage of African
Americans, a high infant mortality rate, and a decreasing * Throughout this section we assume that we are clustering enumera-
population (a negative percent population change). We tion units, an areal phenomenon. It is possible, however, to cluster
might tackle such questions by visually comparing sepa- other forms of phenomena, such as points—thus, we have used the
rate maps of the attributes, but this would require us to generic term observation.
18:02:32.
TABLE 3 Cluster analysis data for New York State counties

% Population Change % African- Infant Mortality Rate
County (1990–2000) % Unemployed American (Deaths per 1,000 Live Births)
Albany 0.6 2.8 11.1 6.1

Allegany -1.1 6.7 0.7 5.4
Broome -5.5 3.3 3.3 8.1
Cattaraugus -0.3 6.4 1.1 9.6
Cayuga -0.4 4.5 4.0 6.4
Chautauqu -1.5 4.8 2.2 8.0
Chemung -4.3 4.8 5.8 4.9
Chenango -0.7 4.8 0.8 4.9
Clinton -7.1 5.2 3.6 15.2
Columbia 0.2 2.9 4.5 6.0
Cortland -0.7 5.9 0.9 3.6
Delaware 1.5 4.8 1.2 8.0
Dutchess 8.0 3.1 9.3 5.0
Erie -1.9 4.8 13.0 8.2
Essex 4.6 6.6 2.8 5.2
Franklin 9.9 7.6 6.6 10.7
Fulton 1.6 5.8 1.8 4.6
Genesee 0.5 4.9 2.1 14.6
Greene 7.7 5.0 5.5 4.1
Hamilton 1.9 8.2 0.4 0.0
Herkimer -2.1 5.0 0.5 8.6
Jefferson 0.7 8.2 5.8 5.2
Lewis 0.6 7.8 0.4 9.0
Livingston 3.1 4.6 3.0 8.5
Madison 0.4 4.4 1.3 4.8
Monroe 3.0 3.8 13.7 8.2
Montgomery -4.4 5.8 1.2 3.4
Nassau 3.6 2.7 10.1 4.8
Niagara -0.4 5.9 6.1 10.6
Oneida -6.1 3.8 5.7 5.9
Onondaga -2.3 3.5 9.4 6.7
Ontario 5.4 3.7 2.1 6.8
Orange 11.0 3.1 8.1 6.5
Orleans 5.6 5.3 7.3 5.4
Oswego 0.5 6.3 0.6 4.8
Otsego 2.1 4.7 1.7 5.1
Putnam 14.1 2.5 1.6 8.2
Rensselaer -1.2 3.8 4.7 10.7
Rockland 8.0 3.0 11.0 4.3
Saratoga 10.7 3.2 1.4 4.8
Schenectady -1.8 3.5 6.8 6.2
Schoharie -0.8 4.8 1.3 14.9
Schuyler 3.0 5.5 1.5 4.6
Seneca -1.0 4.9 2.3 5.4
Steuben -0.4 4.9 1.4 5.9
St. Lawrence 0.0 8.0 2.4 6.8
Suffolk 7.4 3.2 6.9 6.2
Sullivan 6.8 5.0 8.5 4.8
Continued
389
18:02:32.
TABLE 3 Continued
% Population Change % African- Infant Mortality Rate
County (1990–2000) % Unemployed American (Deaths per 1,000 Live Births)
Tioga -1.1 3.3 0.5 4.7

Tompkins 2.6 2.7 3.6 8.2
Ulster 7.5 3.3 5.4 3.6
Warren 6.9 4.0 0.6 5.8
Washington 2.9 4.1 2.9 3.3
Wayne 5.2 4.4 3.2 10.3
Westchester 5.6 3.0 14.2 4.9
Wyoming 2.2 5.8 5.5 11.3
Yates 7.9 3.7 0.6 3.0
methods presume a specified number of clusters and is appropriate because larger enumeration units are apt to
associated members, and then attempt to improve the have larger values of a countable phenomenon.* For the
classification by moving observations between clusters. New York State data, we standardized the raw number of
Because hierarchical methods are a bit easier to under- African Americans living in each county by dividing by
stand and more common, we focus only on them. the total population of each county. For our small hypo-
thetical agricultural data set, we will assume the data are
3.1 Basic Steps in Hierarchical Cluster Analysis already standardized.
A second form of standardization adjusts for differ-
To summarize the process of cluster analysis, we utilize ent units of measure on the attributes. For instance,
an eight-step process that is a modification of the six consider the case of median family income and percent
steps recommended by Charles Romesburg (1984). voting in a presidential election. Median family income
likely would be measured in thousands of dollars rang-
Step 1. Collect an Appropriate Data Set ing from, say, 20,000 to 60,000, whereas percent voting
in a presidential election is a percentage attribute rang-
The first step is to collect a data set that the mapmaker
ing from, say, 30 to 50. If such data were clustered, the
wishes to cluster. Ultimately, we will cluster the data
percentage values would have little impact on the clus-
shown in Table 3, but for illustrative purposes, we also
ter analysis; we would essentially be clustering family
will work with the small hypothetical data set for live-
income. The most common approach for standardizing
stock and crop production listed in Figure 18A and plot-
for different units of measure is to compute z scores:
ted in graphical form in Figure 18B.
—
Romesburg (p. 38) indicates that cluster analysis can Xi - X
be used for three purposes: zi =
s
• To create a scientific question. where zi is the resulting z score for the ith observation, Xi
• To create a research hypothesis that answers a scien- is a raw data value for the ith observation, X is the mean
tific question. of an attribute, and s is the standard deviation of an
• To test a research hypothesis to decide whether it attribute. Applying this formula to our hypothetical data
should be confirmed or disproved. in Figure 18A, we would find respective mean and stan-
dard deviation values of 14.80 and 9.65 for the livestock
Because we were relatively unfamiliar with the census
attribute, and a z score of (5.00 - 14.80) / 9.65 = 1.02 for
data for New York, we wanted to use it to create a
the first observation. Because we presume that our hypo-
scientific question—to explore the patterns that might
thetical data are in the same units of measure, however,
arise. Those knowledgeable about the geography of New
this standardization is not necessary and thus we will work
York might use the resulting patterns to look for other
with the unstandardized data. We chose, however, to stan-
attributes to explain the results or consider other attrib-
dardize the New York State data using z scores because
utes that might be suitable for a cluster analysis.
three of the attributes were measured in percent and the
other (infant mortality) was measured in number of
Step 2. Standardize the Data Set
deaths per 1,000 births.
The second step in cluster analysis is to standardize
the data set, if necessary. Two forms of standardization
are possible. One accounts for enumeration units of vary- *If data associated with points (as opposed to areas) are clustered, it
ing size. Just as for choropleth maps, this standardization might also be appropriate to standardize the data.
18:02:32.
A B
Y
1
20 3
Livestock Crop
Observation Production Production 15
Crop Production
2
1 5 20
2 5 15 10
3 17 19
4 20 5
5
4 5
5 27 5
0 X
0 5 10 15 20 25 30
Livestock Production
C D
Y 1-2
Observation (or cluster)
1
1 2 3 4 5 20 3
1 - - - - -
15
Crop Production
2
2 5.00 - - - -
10
3 12.04 12.65 - - -
4 21.21 18.03 14.32 - - 5

4 5
5 26.63 24.17 17.21 7.00 - 0 X

0 5 10 15 20 25 30
E F
Y 1-2
1
Observation (or cluster) 20 3
3 4 5 1-2
3 - - - - 15
Crop Production
4 14.32 - - -
10
5 17.21 7.00 - - 4-5

5
4 5
1-2 12.35 19.62 25.40 -
0 X
0 5 10 15 20 25 30
FIGURE 18 (Continued)
391
18:02:32.
G H
Y
1 1-2-3
20 3
Observation
(or cluster)
3 1-2 4-5
15
Crop Production
2
3 - - -
Observation
(or cluster)
10
1-2 12.35 - -
4-5
4-5 15.77 23.00 - 5
4 5
0 X
0 5 10 15 20 25 30
FIGURE 18 The basic cluster analysis process using the UPGMA method: (A) a hypothetical raw data set; (B) a graphical plot
of each observation; (C) the initial resemblance matrix of Euclidean distance coefficients; (D) observations (or clusters) 1 and 2
combine because they have the smallest Euclidean distance coefficient; (E–H) the clustering process continues by combining clus-
ters that have the smallest Euclidean distance coefficient.
Step 3. Compute Initial Resemblance Coefficients considering each observation as a separate cluster and
The third step of cluster analysis is to compute an initial then combining those clusters that are most similar to
set of resemblance coefficients that expresses the simi- one another, as expressed by the initial resemblance
larity of each pair of observations. Although numerous coefficients. In our case, observations (clusters) 1 and 2
coefficients are possible, we will use a Euclidean have the shortest Euclidean distance (see Figure 18C)
distance coefficient because it is commonly used and and thus combine to form a new cluster, as shown in
readily interpreted graphically. The Euclidean distance Figure 18D. For our purposes, we designate this as clus-
for two attributes is ter 1–2.
When clusters are combined, we must recompute the
dij = 2(Xi - Xj)2 + (Yi - Yj)2 resemblance coefficient between the newly formed clus-
ter and all other existing clusters. Although numerous
where dij is the Euclidean distance between the ith and methods are possible for recomputing resemblance coef-
jth observations, Xi and Xj are the values of observa- ficients, we focus on one—the unweighted pair–group
tions i and j on attribute X, and Yi and Yj are the values method using arithmetic averages (UPGMA)—because
of observations i and j on attribute Y. For instance, the Romesburg (1984) indicated that it has been commonly
Euclidean distance between the first two observations used. In UPGMA, revised resemblance coefficients are
in Figure 18A is computed by averaging coefficients between observa-
tions in the newly formed cluster and observations in
existing clusters. For the newly formed cluster 1–2, we
d12 = 2(X1 - X2)2 + (Y1 - Y2)2
would have
= 2(5 - 5)2 + (20 - 15)2 = 5.00
d3(1-2) = 1/2(d13 + d23 ) = 1/2(12.04 + 12.65) = 12.35
Note that this value corresponds to the distance
d4(1-2) = 1/2(d14 + d24 ) = 1/2(21.21 + 18.03) = 19.62
between points 1 and 2 in the graph in Figure 18B.
Figure 18C provides a complete set of Euclidean d5(1-2) = 1/2(d15 + d25 ) = 1/2(26.63 + 24.17) = 25.40
distance coefficients for all pairs of observations listed
in Figure 18A. The results of these computations appear in the last
row of the revised resemblance matrix shown in
Step 4. Cluster the Data Figure 18E. Evaluating this revised resemblance
The fourth step of cluster analysis is to actually cluster matrix, we find that observations (clusters) 4 and 5 are
the data. In hierarchical cluster analysis, we begin by grouped next to form cluster 4 –5 (see Figure 18F).
18:02:32.
Again, we compute revised resemblance coefficients combine at a Euclidean distance of 7 (compare

by averaging coefficients between observations in the Figure 19 with Figure 18C and E). Figure 20 is a den-
newly formed cluster and observations in existing clus- drogram of the results of UPGMA for the New York
ters. For clusters 3 and 4 –5, the result is State data.
d3(4-5) = 1/2(d34 + d35 ) = 1/2(14.32 + 17.21) = 15.77 Step 5. Determine an Appropriate Number of Clusters
One key aspect of cluster analysis is determining an
while for clusters 1–2 and 4–5, the result is appropriate number of clusters. As we can see in the
dendrograms, we begin with each observation as a sep-
d(1-2) (4-5) = 1/4 (d14 + d15 + d24 + d25 ) arate cluster, and ultimately combine all observations
to create a single cluster. Clearly, the latter is not our
= 1/4(21.21 + 26.63 + 18.03 + 24.17) desire as our goal is to create groups of observations,
= 23.00 with each group being relatively homogeneous and dif-
ferent from the other groups. One common approach
The resulting distances are shown in Figure 18G, where for selecting an appropriate number of clusters is to
we see that the smallest Euclidean distance is 12.35 look for breaks in the dendrogram—places where com-
(between clusters 1–2 and 3). These clusters are combining two clusters would require making a fairly large
bined in Figure 18H. Finally, we combine clusters 1–2–3 jump in the resemblance coefficient. One simple way
and 4–5 at an average Euclidean distance of 20.26 (the to do this is to cover the dendrogram with a piece of
average of the lengths of the dashed lines shown in paper and then to gradually slide the paper along the
Figure 18H). dendrogram (in our case, from right to left). If you do
The results of a clustering process normally are this for the New York State data (Figure 20), you will
summarized in the form of a dendrogram, which is a find a relatively large break for three clusters, but this
treelike structure that illustrates the resemblance is only the relatively uninteresting result that all but
coefficient values at which various clusters combined. two observations fall in a single cluster. If you continue
For instance, for the hypothetical agricultural data, to slide the paper, you will find another relatively large
Figure 19 illustrates that clusters 1 and 2 combine at break at seven clusters. Note that we have marked this
a Euclidean distance of 5, and that clusters 4 and 5 solution in Figure 20 with a line of long dashes and
Euclidean Distance
0 5 10 15 20
0
Observations (or clusters)
FIGURE 19 Dendrogram for the hypothetical data shown in Figure 18A using Euclidean distance as the
resemblance coefficient.
393
18:02:32.
Distance At Which Clusters Combine

0 5 10 15 20 25
5
FIGURE 20 Dendrogram of the
UPGMA method for the New York
State data. Numbers on the left rep-
resent cluster numbers specified by
6
7 the SPSS software used to generate
the clusters and correspond to those
shown in Table 4.
indicated the composition of clusters with lines of short Table 4 shows the mean z scores for the New York
dashes. State data for the UPGMA method. We can name a
cluster by using attributes with mean values that are
Step 6. Interpret the Clusters relatively high or low (values near ; 1.0 represent
One of the challenges of cluster analysis is interpret- moderately high or low scores, whereas values near
ing the resulting clusters. One approach is to com- ; 2.0 represent distinctly high and low scores). Thus,
pute, for each attribute, the mean z score for all cluster 1 has moderately low unemployment and a
observations falling in each cluster; for instance, high percentage of African Americans, whereas clus-
18:02:32.
TABLE 4 Mean z scores for attributes on each cluster

Cluster % Population Change % Unemployed % African-American Infant Mortality Rate
1 .02 -.86 2.19 -.08

2 -.56 -.03 -.40 -.23
3 -.40 1.95 -.50 .33
4 -.47 .31 -.17 2.07
5 1.37 -.67 .26 -.48
6 1.74 1.95 .63 1.36
7 -.04 2.35 -1.03 -2.26
ter 6 has a rapidly increasing population, high unem- Step 8. Determine Whether Clusters
ployment, and moderately high infant mortality. Are Really Appropriate
(Keep in mind that these statements are being made One should be cautious in interpreting the results of a
relative to other counties in New York State. A high cluster analysis because it is possible to cluster any data
value here might be low relative to some other area set, even a set of random numbers, for multiple attrib-
of the country.) utes. One approach used to judge the effectiveness of a
cluster analysis is the cophenetic correlation coefficient,
Step 7. Map the Resulting Clusters which measures the correlation between raw resem-
Once you have selected an appropriate set of clusters, blance coefficients (Euclidean distance values in the
you will want to map the results (Color Plate 11). In case of UPGMA) and resemblance coefficients derived
symbolizing clusters, you should bear in mind that clus- from the dendrogram (normally referred to as the
ters are likely to be qualitatively different, so symbols cophenetic coefficients). For our hypothetical data, these
used to represent the clusters should reflect this quali- coefficients are depicted in Figure 21 as the resemblance
tative difference. For the New York State data, we and cophenetic matrices, respectively. Values for the
chose to use qualitatively different hues developed by resemblance matrix are taken from Figure 18C, and
Cynthia Brewer and Mark Harrower (http://www. those for the cophenetic matrix are determined by con-
ColorBrewer.org). sidering the Euclidean distance at which observations
Resemblance Matrix Cophenetic Matrix

X Y
Observation Observation
1 2 3 4 5 1 2 3 4 5 5.00 5.00
12.04 12.35
1 - - - - - 1 - - - - -
12.65 12.35
2 5.00 - - - - 2 5.00 - - - - 21.21 20.26

Observation
Observation
18.03 20.26
3 12.04 12.65 - - - 3 12.35 12.35 - - - 14.32 20.26

21.63 20.26
4 21.21 18.03 14.32 - - 4 20.26 20.26 20.26 - - 24.17 20.26
17.21 20.26
5 26.63 24.17 17.21 7.00 - 5 20.26 20.26 20.26 7.00 - 7.00 7.00
FIGURE 21 Computation of the cophenetic correlation involves relating corresponding values in the resemblance and
cophenetic matrices.
395
18:02:32.
combine in the dendrogram shown in Figure 19. For a shade of gray); Eyton also utilized a reduced major
instance, cluster 1–2–3 does not combine with cluster axis (an appropriate regression method when it is not
4 –5 until a distance of 20.26, and so cophenetic values clear which attribute should be treated as the depen-
for observations 1 and 4, 2 and 4, and 3 and 4 are all dent attribute) and an equiprobability ellipse (an ellipse
20.26. Once the two matrices are determined, they are that encloses a specified percentage of the data in a
converted to two linear lists of X and Y values, as shown scatterplot so that values near the mean can be con-
in Figure 21. A correlation coefficient r is then computed trasted with more extreme observations). Brewer noted
between X and Y. For the matrices shown in Figure 21, r that color schemes for bivariate choropleth maps
is equal to 0.91, which suggests that the clusters are should be a function of whether the attributes are
reflecting the true relationships in the data, given that unipolar or bipolar. A cross-hatched symbology for
the maximum possible r is 1.0.* For the New York State bivariate choropleth mapping has been developed, but
data, we used the package R (http://www.r-project.org/ ) it is rarely used because of its coarse appearance.
to compute the cophenetic correlation, deriving an r Although choropleth maps have been the most fre-
value of 0.83. This value also suggests that our clustered quently used symbology for combining two attributes
results are useful, because Romesburg indicates that on the same map, some interesting bivariate point sym-
cophenetic values of 0.8 or greater are desirable (p. 27). bols have been developed, including the rectangular
point symbol (in which the width and height of a rec-
SUMMARY tangle are varied), and the bivariate ray-glyph (in which
straight-line segments point to either the right or left of
a small central circle). It appears that the rectangular
In this chapter we have covered a variety of methods
point symbol might be a suitable substitute for coarse
for multivariate mapping involving the cartographic dis-
cross-hatched symbology. An advantage of the ray-
play of two or more attributes. When only two attrib-
glyph is that it can be squeezed into a relatively
utes are displayed, the process is commonly termed
restricted space. A bivariate point symbol also can be
bivariate mapping. Multivariate mapping can be accom-
created by placing a choropleth shade within a propor-
plished through either map comparison (in which a
tional point symbol, but this technique should be used
separate map is created for each attribute) or by com-
with caution as it could be confused with redundant
bining all attributes on the same map. For bivariate map
symbology.
comparison, we stressed the importance of using an
When more than two maps are displayed simultane-
appropriate method of data classification. Although
ously, the result is termed a small multiple. Although
optimal data classification is often recommended for
small multiples can be useful for comparing the patterns
univariate mapping, it is generally inappropriate for
on multiple maps, they are difficult to interpret when
bivariate mapping because it focuses on the precise dis-
comparing subregions within each map. A common
tribution of individual data values along the number
alternative to the small multiple is the multivariate
line. (An exception would be when the same attribute is
point symbol or glyph. Examples of glyphs include the
compared for two different time periods.) The quantiles
star (in which multiple rays extend from a central
and equal-areas methods of classification are more
circle), the snowflake (in which rays of the star are con-
appropriate choices for bivariate mapping, although the
nected), three-dimensional bars (in which bars of vary-
mean–standard deviation and nested-means methods
ing height are placed alongside one another), data jacks
can be useful if the data are normally distributed.
(in which triangular spikes extend from a square central
Unclassed maps can be used for map comparison, but
area), and Chernoff faces (in which distinct facial fea-
only when the data distributions have a similar shape.
tures are used). Although a considerable number of
When two choropleth maps are overlaid, the result is
attributes can be represented by such methods, it is
termed a bivariate choropleth map, a technique devel-
questionable whether map readers can understand the
oped by the U.S. Bureau of the Census in the 1970s.
resulting symbols. Multiple attributes can also be com-
Bivariate choropleth maps have been criticized because
bined on choropleth and dot maps to create trivariate
of their presumed failure to communicate information,
choropleth maps and multivariate dot maps.
although they can be effective if a clear legend is used,
The difficulty of interpreting multivariate symbols has
both bivariate and individual maps are shown, and an
led to the development of software for exploring multi-
explanatory note is included. Eyton developed a logical
variate data. In the case of SLCViewer, an analyst could
color scheme for the bivariate choropleth map based on
view up to four climatic attributes as a small multiple or
complementary colors (colors that combine to produce
overlay three attributes to create a multivariate map. The
latter is a particularly intriguing option because propor-
* For more sophisticated approaches to evaluating cluster solutions, tional symbols, weighted isolines (in which the width of
see Aldenderfer and Blashfield (1984, 62–74).
18:02:32.
contour lines is proportional to the data), and choropleth for example, we might want to know which census
shading are all included on the same map. tracts are similar in terms of voting behavior, income,
When displaying multiple attributes on the same map, and attendance in private schools. In the eight-step
the associated symbols can be termed either separable or clustering process, we saw that there are numerous
integral. Separable symbols are those that a map reader issues that must be tackled in cluster analysis, includ-
can readily attend to independently, such as cartogram ing standardizing the data, selecting a clustering
size and lightness, whereas integral symbols are those that method (we considered one hierarchical method—
can be integrated to examine the correlation between UPGMA), selecting an appropriate number of clus-
attributes, such as with the rectangular point symbol. ters, interpreting those clusters, and symbolizing the
In this chapter we also examined cluster analysis clusters. We should caution you that we have only
methods, which are appropriate when we wish to touched the surface of cluster analysis, as entire books
combine multiple numeric attributes on a single map; are written on this topic.
FURTHER READING
Aspaas, H. R., and Lavin, S. J. (1989) “Legend designs for IEEE Transactions on Visualization and Computer Graphics 5,
unclassed, bivariate, choropleth maps.” The American no. 2:145–167.
Cartographer 16, no. 4:257–268. Describes a method for creating pexels, a form of glyph for multi-
Examines the effect of legend designs on the interpretation of variate symbolization; includes a discussion of related perceptual
unclassed bivariate choropleth maps. experiments. See also Healey (2001).
Brewer, C. A. (1994a) “Color use guidelines for mapping and Lindberg, M. B. (1987) Dot Map Similarity: Visual and
visualization.” In Visualization in Modern Cartography, ed. by Quantitative. Unpublished Ph.D. dissertation, University of
A. M. MacEachren and D. R. F. Taylor, pp. 123–147. Oxford: Kansas, Lawrence, KS.
Pergamon. Examines the issue of comparing dot maps, in both the visual and
Provides guidelines for using color in both univariate and multi- the numerical sense.
variate mapping. Maxwell, B. A. (2000) “Visualizing geographic classifications
Brewer, C. A., and Campbell, A. J. (1998) “Beyond graduated using color.” The Cartographic Journal 37, no. 2:93–99.
circles: Varied point symbols for representing quantitative Presents a method for selecting colors to depict clusters of data
data on maps.” Cartographic Perspectives, no. 29:6–25. associated with narrow coastal zones.
Presents methods for creating bivariate point symbol maps. Mersey, J. E. (1980) An Analysis of Two-Variable Choropleth
Chang, K. (1982) “Multi-component quantitative mapping.” Maps. Unpublished MS thesis, University of Wisconsin-
The Cartographic Journal 19, no. 2:95–103. Madison, Madison, WI.
Considers cluster analysis and other statistical approaches for An experimental study of the effectiveness of bivariate choro-
combining multivariate data. pleth maps.
Cox, D. J. (1990) “The art of scientific visualization.” Academic Monmonier, M. S. (1975) “Class intervals to enhance the
Computing 4, no. 6:20–22, ff. visual correlation of choroplethic maps.” The Canadian
Describes several multivariate symbolization methods that have
been used in a range of disciplines. Introduces a method for enhancing the visual correlation of
choropleth maps by modifying the boundaries of class intervals.
DiBiase, D., Reeves, C., MacEachren, A. M., Von Wyss, M.,
Also see Monmonier (1976).
Krygier, J. B., Sloan, J. L., and Detweiler, M. C. (1994a)
“Multivariate display of geographic data: Applications in Monmonier, M. S. (1977) “Regression-based scaling to facili-
earth system science.” In Visualization in Modern tate the cross-correlation of graduated circle maps.” The
Cartography, ed. by A. M. MacEachren and D. R. F. Taylor, Cartographic Journal 14, no. 2:89–98.
pp. 287–312. Oxford: Pergamon. Describes a technique for enhancing the visual correlation of pro-
An overview of various methods for displaying multivariate portional symbol maps.
spatial data. Monmonier, M. (1993) Mapping It Out: Expository
Hancock, J. R. (1993) “Multivariate regionalization: An Cartography for the Humanities and Social Sciences. Chicago,
approach using interactive statistical visualization.” AUTO- IL: University of Chicago Press.
CARTO 11 Proceedings, Minneapolis, MN, pp. 218–227. Pages 227–241 describe methods for representing geographic cor-
Introduces a method for multivariate regionalization based on relation.
interactive statistical visualization. Olson, J. M. (1981) “Spectrally encoded two-variable maps.”
Healey, C. G., and Enns, J. T. (1999) “Large datasets at a glance: Annals, Association of American Geographers 71, no.
Combining textures and colors in scientific visualization.” 2:259–276.
397
18:02:32.
An experimental study of the effectiveness of bivariate choro- Tufte, E. R. (1990) Envisioning Information. Cheshire, CT:
pleth maps. Graphics Press.
Romesburg, H. C. (1984) Cluster Analysis for Researchers. This book provides a wealth of methods for representing both
Belmont, CA: Lifetime Learning Publications. nonspatial and spatial data. Chapter 4 deals entirely with small
Provides an overview of methods for cluster analysis; also see multiples.
Aldenderfer and Blashfield (1984) and Everitt et al. (2001). Wang, P. C. C., ed. (1978) Graphical Representation of
Rowles, R. A. (1991) Regions and Regional Patterns on Multivariate Data. New York: Academic Press.
Choropleth Maps. Unpublished PhD dissertation, University A collection of chapters on the graphical representation of multi-
of Kentucky, Lexington, KY. variate data. Many of the chapters focus on Chernoff faces.
Compares regions on an optimally classified choropleth map with
those formed by a cluster analysis method.
GLOSSARY
bivariate choropleth map: the overlay of two univariate glyphs: a term applied to multivariate point symbols when
choropleth maps (e.g., one map could be shades of cyan and the attributes being mapped are in dissimilar units (i.e., not
the other shades of red). part of a larger whole).
bivariate mapping: the cartographic display of two attrib- integral symbols: the individual symbols representing sepa-
utes, such as median income and murder rate, for census tracts rate attributes cannot be attended to easily, but it is possible
within a city. to examine the correlation between the data sets (e.g., a rec-
bivariate-normal: when y values associated with a given x tangular point symbol, where the height represents one
value are normal, and the x values associated with a given y attribute and the width represents another attribute).
value are also normal. multivariate dot map: a dot map in which distinct symbols
bivariate point symbol: a point symbol used to portray two or colors are used for each attribute to be mapped (e.g.,
attributes simultaneously (e.g., representing two attributes by wheat, corn, and soybeans could respectively be represented
the width and height of ellipses). by green, blue, and red dots).
bivariate ray-glyph: used to map two attributes by extending multivariate mapping: the cartographic display of two or
straight line segments to either the right or the left of a small more attributes; for example, we might simultaneously map
central circle. ocean temperature, salinity, and current speed.
Chernoff face: distinct facial features that are associated multivariate point symbol: the depiction of three or more
with individual attributes (e.g., a broader smile depicts cities attributes using a point symbol.
with a higher per capita expenditure on public schools). multivariate ray-glyph: a term used when rays are extended
complementary colors: colors such as cyan and red that from a small interior circle, with the lengths of the rays made
combine to produce a shade of gray. proportional to values associated with each attribute being
mapped.
cophenetic correlation coefficient: in a cluster analysis,
measures the correlation between raw resemblance coeffi- nested means: a method of data classification in which the
cients (Euclidean distance values in the case of UPGMA) and mean is used to repeatedly divide the data set; only an even
resemblance coefficients derived from the dendrogram number of classes is possible.
(normally referred to as the cophenetic coefficients). pexels: a multivariate point symbol consisting of small 3-D
data jack: triangular spikes are drawn from a square central bars that can be varied in height, spacing, and color.
area and made proportional to the magnitude of each pie chart: a modification of the proportional circle in which
attribute being mapped. a portion of the circle is filled in (e.g., to represent percentage
dendrogram: a treelike structure illustrating the resem- of land area from which wheat is harvested).
blance coefficient values at which clusters in a cluster analysis polygonal glyph: a symbol formed by connecting the rays of
combine. a multivariate ray-glyph.
equal-areas: a method of data classification in which an rectangular point symbol: the width and height of a rectan-
equal portion of the map area is assigned to each class. gle are made proportional to each of two attributes being
equiprobability ellipse: if data are bivariate normal, an mapped.
ellipse can be drawn that encloses a specified percentage of resemblance coefficients: in cluster analysis, the values that
the data. express the similarity of each pair of observations (e.g., the
geo-icon: a method for multivariate mapping in which adja- Euclidean distance between two observations).
cent pixels are combined in various ways to depict multiple separable symbols: symbols representing attributes can be
attributes; for instance, the shape of each grouping of pixels attended to independently, thus allowing the map reader to
could signify a particular attribute (e.g., a vegetation attribute focus on individual data sets (e.g., cartogram size and light-
could be depicted by a treelike pixel grouping). ness are separable).
18:02:32.
small multiple: many small maps are displayed to show the unweighted pair–group method using arithmetic averages
change in an attribute over time or to compare many attrib- (UPGMA): a method for computing resemblance coeffi-
utes for the same time period. cients when clusters in a cluster analysis are combined;
snowflake: see polygonal glyph. involves averaging the coefficients between observations in
the newly formed cluster with coefficients for observations in
star: see multivariate ray-glyph. existing clusters.
three-dimensional bars: the height of three-dimensional- weighted isolines: the width of contour lines is made pro-
looking bars are made proportional to each of the attributes portional to the associated data; thus labeling of contour lines
being mapped. is not required.
trivariate choropleth map: the overlay of three univariate
choropleth maps; this approach should be used only when
attributes add up to 100 percent, such as percent voting
Republican, Democratic, and Independent.
REFERENCES
Aldenderfer, M. S., and Blashfield, R. K. (1984) Cluster Carstensen, L. W. J. (1982) “A continuous shading scheme
Analysis. Beverly Hills, CA: Sage. for two-variable mapping.” Cartographica 19, no.
Anderson, E. (1960) “A semigraphical method for the analy- 3/4:53–70.
sis of complex problems.” Technometrics 2, no. 3:387–391. Cox, D. J. (1990) “The art of scientific visualization.”
Anselin, L., Syabri, I., and Kho, Y. (2006) “GeoDa: An intro- Academic Computing 4, no. 6:20–22, 32–34, 36–38, 40.
duction to spatial data analysis.” Geographical Analysis 38, DiBiase, D., Reeves, C., MacEachren, A. M., Von Wyss, M.,
no. 1:5–22. Krygier, J. B., Sloan, J. L., and Detweiler, M. C. (1994a)
Anthamatten, P., and Ziegler, S. S. (2006) “Teaching geogra- “Multivariate display of geographic data: Applications in
phy with 3-D visualization technology.” Journal of Earth system science.” In Visualization in Modern
Geography 105:231–237. Cartography, ed. by A. M. MacEachren and D. R. F. Taylor,
pp. 287–312. Oxford, England: Pergamon.
Ariel, A., and Berger, N. A. (2006) Plotting the Globe: Stories
of Meridians, Parallels, and the DiBiase, D., Sloan, J. L. I. I., and Paradis, T. (1994b)
“Weighted isolines: An alternative method for depicting
International Date Line. Westport, CT: Praeger Publishers.
statistical surfaces.” The Professional Geographer 46,
Borg, I., and Staufenbiel, T. (1992) “Performance of no. 2:218–228.
snowflakes, suns, and factorial suns in the graphical
Dorling, D. (1993) “Map design for census mapping.” The
representation of multivariate data.” Multivariate
Cartographic Journal 30, no. 2:167–183.
Behavioral Research 27, no. 1:43–55.
Ellson, R. (1990) “Visualization at work.” Academic
Brewer, C. A., and Pickle, L. (2002) “Evaluation of methods
Computing 4, no. 6:26–28, 54–56.
for classifying epidemiological data on choropleth maps in
series.” Annals of the Association of American Everitt, B. S., Landau, S., and Leese, M. (2001) Cluster
Geographers 92, no. 4:662–681. Analysis (4th ed.). London: Arnold.
Brewer, C. A. (1994a) “Color use guidelines for mapping and Eyton, J. R. (1984a) “Complementary-color two-variable
visualization.” In Visualization in Modern Cartography, ed. maps.” Annals of the Association of American Geographers
by A. M. MacEachren and D. R. F. Taylor, pp. 123–147. 74, no. 3:477–490.
Oxford, England: Pergamon. Grinstein, G., Sieg, J. C. J., Smith, S., and Williams, M. G.
Burt, J. E., and Barber, G. M. (1996) Elementary Statistics for (1992) “Visualization for knowledge discovery.”
Geographers (2nd ed.). New York: Guilford. International Journal of Intelligent Systems
7:637–648.
Carr, D. B., Olsen, A. R., and White, D. (1992) “Hexagon
mosaic maps for display of univariate and bivariate Groop, R. E. (1992) “Dot-density crop maps of the
geographical data.” Cartography and Geographic Midwest.” Abstract, Association of American Geographers
Information Systems 19, no. 4:228–236, 271. Annual Meeting, San Diego, CA, p. 89.
Byron, J. R. (1994) “Spectral encoding of soil texture: A new Hartnett, S. (1987) “Employing rectangular point symbols in
visualization method.” GIS/LIS Proceedings, Phoenix, AZ, two-variable maps.” Abstract, Association of American
pp. 125–132. Geographers Annual Meeting, Portland, OR, p. 39.
Carstensen, L. W. (1986a) “Bivariate choropleth mapping: Healey, C. G. (2001) “Combining perception and impression-
The effects of axis scaling.” The American Cartographer ist techniques for nonphotorealistic visualization of multi-
13, no. 1:27–42. dimensional data.” SIGGRAPH 2001 Course 32:
Nonphotorealistic Rendering in Scientific Visualization (Los
Carstensen, L. W. (1986b) “Hypothesis testing using univari-
Angeles, California): 20–52. http://www.csc.ncsu.edu/fac-
ate and bivariate choropleth maps.” The American
ulty/healey/abstract/pubs.html#avi.02.
399
18:02:32.
Healey, C. G., and Enns, J. T. (1999) “Large datasets at a Olson, J. M. (1972b) “The effects of class interval systems on
glance: Combining textures and colors in scientific choropleth map correlation.” The Canadian Cartographer
visualization.” IEEE Transactions on Visualization and 9, no. 1:44–49.
Computer Graphics 5, no. 2:145–167. Olson, J. M. (1975b) “The organization of color on two-
Interrante, V. (2000) “Harnessing natural textures for variable maps.” AUTO-CARTO II, Proceedings of the
multivariate visualization.” IEEE Computer Graphics and International Symposium on Computer-Assisted
Applications 20, no. 6:6–11. Cartography, pp. 289–294, 251, 264–266.
Jenks, G. F. (1953b) “Pointillism as a cartographic technique.” Olson, J. M. (1981) “Spectrally encoded two-variable maps.”
The Professional Geographer 5, no. 5:4–6. Annals of the Association of American Geographers 71, no.
Jenks, G. F. (1961) Crop patterns in the United States: 1959. 2:259–276.
Map, U.S. Bureau of the Census. (Scale 1:5,100,000.) Pazner, M. I., and Lafreniere, M. J. (1997) “GIS icon maps.”
Jenks, G. F. (1962) Livestock and livestock products sold in 1997 ACSM/ASPRS Annual Convention & Exposition,
the United States: 1959. Map, U.S. Bureau of the Census. Volume 5 (AUTO-CARTO 13), Seattle, WA, pp. 126–135.
(Scale 1:5,100,000.) Peterson, M. P. (1979) “An evaluation of unclassed crossed-
Lavin, S., and Archer, J. C. (1984) “Computer-produced line choropleth mapping.” The American Cartographer 6,
unclassed bivariate choropleth maps.” The American no. 1:21–37.
Cartographer 11, no. 1:49–57. Rogers, J. E., and Groop, R. E. (1981) “Regional portrayal
Lindberg, M. B. (1987) “Dot map similarity: Visual and with multi-pattern color dot maps.” Cartographica 18, no.
quantitative.” Unpublished PhD dissertation, University of 4:51–64.
Kansas, Lawrence, KS. Romesburg, H. C. (1984) Cluster Analysis for Researchers.
Lloyd, R. E., and Steinke, T. R. (1976) “The decisionmaking Belmont, CA: Lifetime Learning.
process for judging the similarity of choropleth maps.” The Smith, S., Grinstein, G., and Pickett, R. (1991) “Global
American Cartographer 3, no. 2:177–184. geometric, sound, and color controls for iconographic
Lloyd, R., and Steinke, T. (1977) “Visual and statistical displays of scientific data.” Extracting Meaning from
comparison of choropleth maps.” Annals of the Association Complex Data: Processing, Display, Interaction II, SPIE—
of American Geographers 67, no. 3:429–436. The International Society for Optical Engineering, Volume
1459, Santa Clara, CA, pp. 192–206.
MacEachren, A. M. (1995) How Maps Work: Representation,
Visualization, and Design. New York: Guilford. South Carolina State Budget and Control Board. (1994)
South Carolina Statistical Abstract. Columbia, SC: Office of
Metternicht, G., and Stott, J. (2003) “Trivariate spectral
Research and Statistical Services.
encoding: A prototype system for automated selection of
colours for soil maps based on soil textural composition.” Stevens, J. (1996) Applied Multivariate Statistics for the Social
Proceedings of the 21st International Cartographic Sciences (3rd ed.). Mahwah, NJ: Lawrence Erlbaum
Conference, Durban, South Africa, pp. 2341–2353 (CD- Associates.
ROM). Tufte, E. R. (1990) Envisioning Information. Cheshire, CT:
Meyer, M. A., Broome, F. R., and Schweitzer, R. H. J. (1975) Graphics Press.
“Color statistical mapping by the U.S. Bureau of the Turner, E. J. (1977) “The use of shape as a nominal variable
Census.” The American Cartographer 2, no. 5:100–117. on multipattern dot maps.” Unpublished PhD dissertation,
Miller, J. R. (2007) “Attribute blocks: Visualizing multiple University of Washington, Seattle, WA.
continuously defined attributes.” IEEE Computer U.S. Bureau of the Census. (1970) “Population distribution,
Graphics and Applications 27, no. 3:57–69. urban and rural, in the United States: 1970.” United States
Muller, J.-C. (1980b) “Visual comparison of continuously Maps, GE-50, No. 45.
shaded maps.” Cartographica 17, no. 1:40–51. Wang, P. C. C., and Lake, L. T. G. E. (1978) “Application of
Nelson, E. S. (2000) “The impact of bivariate symbol design graphical multivariate techniques in policy sciences.” In
on task performance in a map setting.” Cartographica 37, Graphical Representation of Multivariate Data, ed. by
no. 4:61–78. P. C. C. Wang, pp. 13–58. New York: Academic.
Nelson, E. S., and Gilmartin, P. (1996) “An evaluation of Wilhelm, S. D. (1983) “Two symbols for unclassed two-vari-
multivariate quantitative point symbols for maps.” In able mapping: The bivariate box and the ratio bar.”
Cartographic Design: Theoretical and Practical Unpublished MA thesis, University of North Carolina,
Perspectives, ed. by C. H. Wood and C. P. Keller, pp. Chapel Hill, NC.
191–210. Chichester, England: Wiley. Zhang, X., and Pazner, M. (2004) “The icon imagemap
Olson, J. (1972a) “Class interval systems on maps of technique for multivariate geospatial data visualization:
observed correlated distributions.” The Canadian Approach and software system.” Cartography and
Cartographer 9, no. 2:122–131. Geographic Information Science 31, no. 1:29–41.
18:02:32.
Cartograms and Flow Maps
OVERVIEW
This chapter covers cartograms and flow maps—two focus will be on quantitative data such as the number of
thematic mapping techniques that are arguably less barrels of oil shipped from Saudi Arabia to countries
commonly used, but that provide unique visualizations around the world. There are many different types of flow
and are often the only logical choice for certain kinds of maps, ranging from distributive flow maps that depict the
data. Section 1 introduces the cartogram, a technique in movement of commodities, people, or ideas between geo-
which spatial geometry is distorted to reflect a theme; for graphic regions to network flow maps that depict flows
example, sizes of countries might be made proportional within a network (e.g., a transportation network) to
to the population of each country. An advantage of car- continuous flow maps that depict flows of continuous phe-
tograms is that small enumeration units with high nomena such as wind speed and direction.We will consider
attribute values that would be hidden on a conventional a variety of software for depicting flows, including Waldo
equivalent projection can be readily seen on a Tobler’s software for migration mapping (recently updated
cartogram. Traditionally, the digital construction of car- by David Jones), Doantam Phan and his colleagues’ soft-
tograms was fraught with two problems: the inability to ware for handling a broad range of flow mapping prob-
retain correct topological relationships between enumer- lems, and two pieces of software for handling continuous
ation units and slow execution times. Using the principle flow maps, Stephen Lavin and Randall Cerveny’s unit-
of diffusion from physics, Michael Gastner and Mark vector density mapping and Greg Turk and David Banks’
Newman have developed a novel approach for creating image-guided streamline placement.
cartograms that deals with these problems.
Another important development in cartograms has
been Daniel Dorling’s construction approach, in which 1 CARTOGRAMS
uniformly shaped symbols (normally, circles) represent
each enumeration unit. This is advantageous because When creating thematic maps, cartographers generally
extremely detailed maps can be created relatively easily avoid distorting spatial relationships. For example, an
using this simplified symbology. In Dorling’s approach, equivalent (or equal-area) projection is normally used
a circle’s size is typically a function of population, and for a dot map so that the density of dots is solely a func-
some other attribute (or attributes) are displayed within tion of the underlying phenomenon (and not the map
the circle. Dorling contends that the resulting map more projection). Sometimes, however, cartographers purpose-
properly reflects the human geography of a region; this is fully distort space based on values of an attribute; the
in contrast to traditional choropleth maps, which, he resulting maps are known as cartograms. Probably
argues, reflect the physical extent of the region. the most common cartograms used in everyday life are
Section 2 introduces flow maps, which generally utilize distance cartograms, in which real-world distances
lines of varying width to depict the movement of phenom- are distorted to reflect some attribute, such as the time
ena between geographic locations.Although flow maps can between stops on subway routes: Here cartograms
be based on either qualitative or quantitative data, our are appropriate because the time between (and order of)
401
18:02:59.
stops is more important than the actual distance between

stops. In the geographic literature, area cartograms
A
are the most common form of cartogram.* Area car-
tograms are created by scaling (sizing) enumeration
units as a function of the values of an attribute associated
with the enumeration units. For example, consider
Figure 1, which illustrates the states and territories of
Australia as they might appear (A) on an equivalent map
projection and (B) on a cartogram when scaled in direct
proportion to the population of each state or territory.
Although population is the most typical attribute por-
trayed on area cartograms, any ordinal or higher-level
attribute (either unstandardized or standardized) can be
portrayed.
Two forms of area cartogram traditionally have been
recognized: contiguous and noncontiguous. With
contiguous cartograms, an attempt is made to retain the
contiguity of enumeration units; as a result, the shape of
units must be greatly distorted (as in Figure 1B). With
noncontiguous cartograms, the shapes of enumeration B
units are retained, but gaps appear between units (as in
Figure 2). The contiguous cartogram arguably best rep-
resents a “true” map because there is an attempt to
maintain boundary relationships between enumeration
units, there are no gaps between enumeration units, and
the shape of the total study area is easier to maintain
(Dent 1999, 209). A downside of the contiguous car-
togram is that the shapes of enumeration units may
make them difficult to identify.
Traditionally, the digital construction of contiguous
cartograms was complicated by two problems: the
inability to retain correct topological relationships
between enumeration units and slow execution times.
Michael Gastner and Mark Newman (2004, 2005) have
developed a novel approach that maintains topological
relationships and is efficient (i.e., execution times are
on the order of seconds). Reflecting their physics back-
ground, Gastner and Newman (2005, 4) describe their
approach as follows:
It is a trivial observation that on a true population

FIGURE 1 Creating a cartogram: (A) states and territories of
cartogram the population is necessarily uniform: once
areas of regions have been scaled to be proportional to Australia as they might appear on an equivalent projection;
their population, by definition, population density is the (B) a contiguous-area cartogram in which states and territories
same everywhere. Thus, one way to create a cartogram . . . are scaled on the basis of 1976 population. (From Griffin 1983,
is to allow population to “flow away” from high-density p. 18; courtesy of T. L. C. Griffin.)
areas into low-density ones until the density is equalized
everywhere. This . . . brings to mind the diffusion process
of elementary physics. . . .
Figure 3A is a dot map of lung cancer cases among
Although the mathematics of their approach is males in New York State from 1993 to 1997 based on
beyond the scope of this text, Gastner and Newman 1,598 zip codes on an equivalent map projection. It can
present numerous applications illustrating its effective- be argued that this map is not particularly useful
ness. Figure 3 illustrates one of these applications. because the areas of concentration do not necessarily
represent an increased likelihood of cancer, but rather
* Dent (1999) termed these value-by-area cartograms; we have used are indicative of higher population densities (note the
the term area cartogram for simplicity. heavy concentration in the New York City area in the
18:02:59.
Number of Persons 65 Years of Age and Over, 1970
100,000 Persons 65 and Over

Source: U.S. Bureau of the Census
FIGURE 2 A noncontiguous area cartogram; the shape of enumeration units is retained by introducing gaps between units. (After
Olson, J. M. (1976b) “Noncontiguous area cartograms.” The Professional Geographer 28, no. 4, p. 372. Courtesy of Blackwell Publishing.)
southeast portion of the map). In contrast, Figure 3B is dence of lung cancer is not a function of environmental
a cartogram created using the same data—in this case, factors, but rather individual behavior (Gastner and
the zip codes have been scaled to represent the under- Newman 2005, 10).
lying population. Now we can see that the dots are rel- The Worldmapper site (http://www.sasi.group.shef.
atively uniformly distributed, suggesting that the inci- ac.uk/worldmapper/index.html) provides an extensive
A B
FIGURE 3 (A) A dot map of lung cancer cases for males in New York State from 1993 to 1997 based on 1,598 zip codes on an
equivalent map projection; (B) a cartogram of the same data in which zip codes are scaled based on population. (From Gastner,
M. T., and Newman, M. E. J. (2004) “Diffusion-based method for producing density-equalizing maps.” Proceedings of the National
Academy of Sciences 101, no. 20:7499–7504; Copyright (2004) National Academy of Sciences, U.S.A.)
403
18:02:59.
collection of maps that further illustrates the usefulness features along boundaries of enumeration units (e.g.,
of Gastner and Newman’s approach. The purpose of distinctive meanders of a river), include an inset
Worldmapper is “to communicate to the widest possi- map depicting actual areal relations of enumeration
ble audience . . . how parts of the world relate to each units (or, alternatively, the units should be labeled),
other and the implications of these relationships for and not use area cartograms if the anticipated reader-
society ” (Dorling et al. 2006, 757). This is accomplished ship is unfamiliar with the region depicted. Another
by displaying a wide variety of data sets in cartogram potential way to enhance the understanding of car-
form using Gastner and Newman’s algorithm. Colors tograms is to smoothly change (as in an animation)
for territories on these cartograms were chosen from a between a standard equivalent projection and the
rainbow scale, with the richest regions being a shade of cartogram. Alan Keahey (1999) developed an early
violet and the poorest regions being a dark red, thus version of such an algorithm, which can be viewed at
enabling one territory to be compared across maps. http://alan.keahey.org/research/papers/ica.99.html.
Since the cartogram greatly distorts regions, Dorling et
al. (2006) argue that an interactive capability may ulti- 1.1 Dorling’s Cartograms
mately need to be added to “allow users to ‘brush’ over
the cartograms and see the names of territories change Daniel Dorling (1993; 1994; 1995b; 1995c) has devel-
under their eyes . . . and morph from one projection to oped an intriguing approach for constructing car-
another to better illustrate just how unevenly peoples, tograms that begins by placing a uniformly shaped sym-
goods and livelihoods are distributed over the surface bol (typically, a circle) in the center of each
of this planet.” (p. 762). Dorling et al. argue, however, enumeration unit, with the size of the symbol a function
that even in their noninteractive form these maps have of population. Initially, symbols overlap one another
received a great deal of enthusiasm. because small enumeration units can represent rela-
Area cartograms are commonly used for their tively large populations (Figure 4A). To eliminate over-
dramatic impact; in fact, it can be argued that an area lap, an iterative procedure is executed in which symbols
cartogram should not be used if the relations between are gradually moved away from one another (Figure
values of an attribute are similar to the relations 4B–E). Wherever possible, points of contact between
between sizes of enumeration units, as the resulting map symbols should reflect points of contact between actual
will not appear distorted. It should be recognized, how- enumeration units, but sometimes it is not possible to
ever, that readers may have difficulty interpreting area meet this constraint. As a result, Dorling termed this a
cartograms because they look so different from other “noncontiguous form of cartogram.” The result of
thematic maps, in which the size relations between applying Dorling’s algorithm to the 9,289 wards of
enumeration units are similar to how they appear on a England and Wales is shown in Figure 5. Figure 5A
globe (in the case of equivalent projections, the size depicts the boundaries of wards as they would appear
relations match the globe). To ease the interpretation on a traditional equivalent projection, and Figure 5B
of cartograms, mapmakers should retain characteristic displays the resulting cartogram. (The upper right portion
A B C D E
0 4 16 64 256
FIGURE 4 Dorling’s algorithm for creating cartograms: (A) Uniformly shaped symbols (circles, in this case) are scaled on the
basis of population, and are placed in the center of enumeration units on an equivalent projection; (B–E) the circles are gradually
moved away from one another so that no two overlap. The numbers beneath each illustration represent the number of iterations
in the algorithm. (After Dorling 1995b, p. 274.)
18:02:59.
A B
FIGURE 5 Applying Dorling’s algorithm to the 9,289 wards of England and Wales: (A) boundaries as they would appear on a
traditional equivalent projection; (B) a cartogram resulting from applying Dorling’s algorithm. (From Dorling 1993, p. 171; cour-
tesy of The British Cartographic Society.)
of each figure portrays wards grouped into counties.) country is dominated by relatively low population densi-
The basic difference between the maps is that small ties (the blues and greens), whereas the cartogram
land areas with large populations are much more appar- (B) provides a detailed picture of the variations in pop-
ent on Dorling’s cartogram than they are on the tradi- ulation densities within urban areas. Dorling contends
tional map. This is particularly apparent on the maps of that his cartogram depicts the human geography of a
county boundaries: Note that London is a mere dot on region rather than focuses on its physical extent, as
the traditional map, but encompasses a substantial por- depicted on the choropleth map.
tion of the cartogram.* One obvious difference between Dorling’s car-
A major purpose of Dorling’s approach is not to map tograms and traditional cartograms is that his provide
population but, rather, to serve as a base on which other no shape information for enumeration units. As a result,
attributes can be displayed. For example, Color Plate 1 the addition of a conventional equivalent projection
portrays population density in Britain on both a would seem essential when examining Dorling’s car-
traditional equivalent projection (the result is a choro- tograms, particularly if the map reader is unfamiliar
pleth map) and on Dorling’s cartogram. Note the drastic with the region depicted. Another difference is that
difference in appearance between these two maps. The Dorling’s cartograms typically show a large number of
equivalent projection (A) suggests that most of the enumeration units, and thus considerable detail. As a
result, readers must study these maps meticulously. In
*No theme is displayed on the traditional map because just the
fact, the complexity of Dorling’s cartograms suggests
boundaries of wards appear; the dark areas represent where small that they might be examined most effectively in a data
wards bleed together. exploration environment. For example, imagine having
405
18:02:59.
both a traditional equivalent projection and a car-

togram displayed on a computer screen at the same
time. As the cursor is moved over an enumeration unit
on the cartogram, this same enumeration unit could be
highlighted and identified on the equivalent projection
(or vice versa). Alternatively, the user could zoom in to
examine subregions in greater detail on both maps.
2 FLOW MAPPING
Flow maps depict the movement of phenomena

between geographic locations, typically using lines of
varying widths (for quantitative data), although other
visual variables may be used (such as hue for qualita-
tive data). Although we will see that a wide variety of
phenomena can be portrayed on flow maps, you are
probably most familiar with those representing the flow FIGURE 6 A flow map showing the number of vehicles in a
of goods or movement of people between countries. portion of Seattle, Washington, in 2004; lines representing
individual streets are scaled in direct proportion to the num-
ber of vehicles traveling between various street intersections.
2.1 Selecting Appropriate Data
(Courtesy of the City of Seattle, Seattle Department of
Transportation.)
When considered in the broadest sense, flow maps can
be used to map either qualitative or quantitative data. individual streets are scaled in direct proportion to the
An example for qualitative data would be the depiction number of vehicles traveling between various street
of various linear features such as roads, streams, and intersections. An additional feature of this flow map is
railroads by, respectively, red, blue, and black lines (with that the total number of vehicles is embedded in each
small perpendicular black marks for the railroads). flow line.
Here the intention would be to illustrate that the types Also similar to proportional symbol maps, flow
of flows for these features differ qualitatively from one maps can be based on either unstandardized or stan-
another. Such a map might also portray quantitative dardized data. Unstandardized data are by far more
data; for instance, narrow and wide red lines might be common. For true point locations, this is not problem-
used to represent interstate and U.S. highways, respec- atic, but for conceptual point locations, the resulting
tively. In the latter case, the intention would be to flow map should be treated with caution. For instance,
express an ordinal relationship between the two types a large raw migration from China is to be expected
of roads—consequently, no numerical information because of its large population, but one might wonder
would be provided in the legend. how the rate of migration from China compares with
More commonly, quantitative flow maps depict that from Japan.
numerical (interval- or ratio-level) data.
In a fashion similar to proportional symbol mapping, 2.2 Types of Flow Maps
flows can be distinguished between either true or con-
ceptual point locations. An example for true point loca- For our purposes, it is convenient to consider five types
tions would be a map depicting traffic flows along city of flow maps. Terms for the first three of these come
streets, where the street intersections represent the true form the work of Mary Parks (1987), who conducted an
point locations. For instance, Figure 6 depicts a flow extensive survey of various types of flow maps.
map showing the number of vehicles in a portion of Distributive flow maps depict the movement of com-
Seattle, Washington, in 2004, where lines representing modities, people, or ideas between geographic regions.
18:02:59.
It is useful to distinguish two subcategories of distribu-

tive flow maps. The first typically maps the entire world,
or a large portion thereof, and attempts to depict the
actual routes of flow. Color Plate 2, illustrating the vol-
ume of French wine exported by sea in 1864, is a classic
example of this subtype, which was created by the
famous French cartographer Charles Joseph Minard. It
is interesting to note that Minard adjusted the geo-
graphic base to place more emphasis on the shipping
routes. Arthur Robinson (1982, 151) stated:
The logic is unassailable: if a flow line of a given width is to

portray ocean transport it would be ridiculous to have it
cross land, such as the margins of the Channel or the Straits
of Gibralter. . . . The solution was simply to widen the
Channel or the Strait.
Clearly, Robinson was impressed by Minard’s flow

maps, noting that Minard “. . . certainly brought that
class of cartography to a level of sophistication that has
probably not been surpassed” (Robinson 1967, 105). To
a certain extent, Robinson’s statement still applies
today, at least for flow maps depicting the transport of
people or goods around the world.
The second subcategory of distributive flow map
depicts the flow of a phenomenon within a portion of the
Earth’s land surface (e.g., within a country). In this case,
the precise route of flow is not as important as the general
direction and magnitude; thus, straight lines of varying FIGURE 7 Migration to and from California, 1965–1970.
width can be drawn between the origin and destination (After Tobler 1987. First published in the American
points. The maps of migration between states of the Cartographer 14(2), p. 160. Reprinted with permission from
United States shown in Figures 7 and 8 are examples of the American Congress on Surveying and Mapping.)
this. In section 2.4, we’ll discuss software specifically
intended for mapping this sort of migration data.
Network flow maps depict flows within networks,
such as the transportation network shown in Figure 6.
Parks (1987) noted that maps of general shipping routes
are a form of network flow map because their focus is
on the network of the shipping route. In this case, a leg-
end is often not used “. . . since the flow lines are
intended to convey the relative importance of the
routes in the network pattern rather than precise val-
ues” (Parks 1987, 54).
Parks used the term radial flow maps to describe
flow maps that have a distinctive radial pattern, as in
Color Plate 3, which shows patterns of migration to U.S.
states in 1960. Each state includes a set of radial flow
lines that conceptually illustrates both the quantity of
migrants into the state as well as the general region
from which they came. The choropleth shading within
each state represents the total of all migrants for each FIGURE 8 Net migration 1965–1970 for the 48 contiguous
state. You should consider whether the use of choro- U.S. states, with flows below the mean net migration not
pleth shading is appropriate or inappropriate in this shown. (Reprinted by permission from Geographical
instance. Another example of radial flows would be a Analysis, Vol. 13, No. 1 (Jan. 1981). Copyright 1981 by Ohio
map centered on the state of Kansas showing migration State University Press. All rights reserved.)
407
18:02:59.
from Kansas to other states as a result of the Dust Bowl information. We will also see that certain forms of flow
in the 1930s. symbolization (e.g., Lavin and Cerveny’s unit-vector
There are two types of flow maps that Parks did not method) do not lend themselves to ready numerical
consider. One is continuous flow maps, which depict the interpretation.
movement of a continuous phenomenon such as wind
or ocean currents (Color Plate 4). Such phenomena typ- 2.4 Software for Flow Mapping
ically are composed of two attributes, magnitude and
direction, that can change at any geographic location. In this section we consider various software that has
For example, at any point on the ocean surface, we can been developed for flow mapping. Our intent is not to
measure the speed of the current and the direction from provide an exhaustive discussion of software, but rather
which it flows. In section 2.4, we’ll discuss two to provide a sense of the range of software that has
approaches that have been developed for creating these been developed. One limitation we will see is that soft-
sorts of maps. ware for distributive flow maps stressing the actual flow
A final type of flow map depicts flows of telecom- routes (as in Color Plate 2) has yet to be fully devel-
munications technology, including the Internet and oped, although software by Phan and his colleagues
its associated information spaces—we’ll refer to (2005) appears to be moving in that direction.
these as telecommunications flow maps. Parks consid-
ered these a form of network map, but the abstract Mapping Migration
appearance of telecommunications maps related to the Migration data are especially logical for digital portrayal
Internet (Color Plate 5) suggests that a specialized cat- because of the large number of movements that must be
egory is appropriate for these maps. Since space does depicted. For example, for the 48 contiguous U.S. states,
not permit us to cover telecommunications flow maps there are 2,256 possible movements (assuming that all
in detail here, we recommend that you consult the fine pairs of states are considered); if we consider U.S. coun-
work of Dodge and Kitchen (2001a, 2001b; http://www. ties, there are more than 9 million possible movements,
cybergeography.org/atlas/) for more information. and this does not even consider the attribute of time!
Waldo Tobler (1987) was one of the first to develop
2.3 Issues in Designing Flow Maps software for displaying migration flows. One of the sim-
pler options in Tobler’s software was the depiction of
Borden Dent (1999, 227–228) provided an excellent one-way migration to or from a particular state by arrows
summary of “essential design decisions” in creating flow of varying widths (Figure 7). When one wishes to show
maps: the migration between all pairs of enumeration units
simultaneously, a different approach is required because
1. Flow lines are highest in intellectual [importance] of the large number of arrows that result. The key is rec-
and therefore highest in visual/graphic importance. ognizing that many migration movements are small;
2. Smaller flow lines should appear on top of larger deleting such movements will allow the map reader to
flow lines. focus on more important movements. Tobler indicated
3. Arrows are necessary if direction of flow is critical that by deleting migration values below the mean, he
to map meaning. generally was able to remove 75 percent or more of the
flow lines, while deleting less than 25 percent of the
4. Land and water contrasts are essential (if the
migrants. For example, Figure 8 illustrates this approach
mapped area contains both).
for net migration for states in the United States from
5. Projection, its center and aspect, are used to direct 1965 to 1970.
readers’ attention to the flow pattern important to Another interesting feature of Tobler’s software is
the map’s purpose the ability to route data through enumeration units
6. All information should be kept simple, including lying between the starting and ending points for migra-
flow line scaling. tion, thus reflecting the route over which people were
presumed to migrate. For example, migration data from
7. Legends should be clear and unambiguous, and
New York to California would ideally have to be routed
include units where necessary.
through Pennsylvania, Ohio, and numerous other states.
Since we have discussed similar design issues elsewhere Although the details of how Tobler achieved this are
in this text, we will not expand on items 1 to 6. With beyond the scope of this text (see Tobler 1981, 7–8 for a
respect to legend design, it is important to emphasize that summary), it is interesting that he used some of the
some flow maps do not include a legend. The logic is that same concepts implemented in his pycnophylactic
sometimes we are more interested in interpreting the method. Figure 9 illustrates the result for the migration
general spatial pattern than in deriving precise numerical data used for Figure 8.
18:02:59.
algorithm in which the geography was not distorted; these

results are shown in Figure 10B, albeit for a different data
set than in Figure 10A. Although Phan et al. chose to use
abbreviated state names to identify migration destina-
tions, it would be possible to use state boundaries
instead—in fact, for interpreting pattern information, we
feel that using state boundaries would be more useful
(at least for the undistorted version of their algorithm).
Phan et al. did not illustrate a digital map similar to
Minard’s in their paper, but on a Web site associated
with their work (http://graphics.stanford.edu/papers/
flow_map_layout/), they point to the efforts of others
who have used their algorithm as a basis for creating
FIGURE 9 Migration data depicted in Figure 8 are rerouted flow maps of the world. Clearly, Phan et al.’s approach
to pass through states between the starting and ending points; has considerable potential for enhancing the creation of
again, flows below the mean are not shown. (Reprinted by per- automated flow maps.
mission from Geographical Analysis, Vol. 13, No. 1 (Jan. 1981).
Copyright 1981 by Ohio State University Press. All rights Mapping Continuous Flows
reserved.) Lavin and Cerveny’s Unit-Vector Density Mapping.
Stephen Lavin and Randall Cerveny (1987) developed
their unit-vector density method (Figure 12) as an
In 2003, David Jones updated Tobler’s original software alternative to wind arrows, which were conventionally
to a Windows-based environment (http://www.csiss.org/ used to represent wind speed and direction (Figure 11).
clearinghouse/FlowMapper/). Although the updated soft- Lavin and Cerveny deemed wind arrows awkward be-
ware does not include all of the options described in cause (1) readers had to separate them visually to create
Tobler’s original work, it enables users to easily explore the the individual distributions of wind speed and direction,
sorts of flows that Tobler discussed. Sunita Yadav-Pauletti’s and (2) the symbols representing speed (the flaglike por-
(1996) MigMap (http://web.ku.edu/~cagis/MigMap.zip) tion) were abstract and thus time-consuming to interpret.
was another early migration mapping software package. Unit-vector density mapping involves changing the den-
Yadav-Pauletti symbolized migration data using both sity and orientation of short fixed-length line segments
choropleth and arrow symbols, and utilized small multiples called unit vectors. The density and orientation of unit
and animation to depict changes over time.Although it was vectors are a function of the magnitude and direction of
possible to create a number of interesting maps with her the vector-based flow being mapped. Typically, the line
software (Color Plate 6), a limitation was the inability to segments have an arrowhead to indicate movement, al-
input one’s own migration data. though Lavin and Cerveny argued that they could be
used effectively without arrowheads.
Phan et al.’s Method Lavin and Cerveny’s algorithm for unit-vector density
Doantam Phan and his colleagues (2005) have developed mapping is analogous to the algorithm Lavin used in his
software intended for handling a broad range of flow map- dot-density shading technique. The algorithm presumes
ping problems. A key to their approach is an algorithm that one has first created an equally spaced gridded net-
that enables flow lines to merge smoothly and to shift the work of vector-based flow values; thus, each of the crosses
underlying base map to make room for the flow lines, just in Figure 12A would have two data values—one for vec-
as Minard did on his maps. Although the details of their tor magnitude and one for vector direction. Creating such
algorithm are beyond the scope of the present text, it is data requires twice running one of the gridding
noteworthy that their algorithm is based on hierarchical approaches for interpolation, once for vector magnitude
clustering. and once for vector direction. Next, it is presumed that
Phan et al. illustrate the results of applying their algo- each grid intersection is surrounded by a square cell
rithm using the same data that Tobler utilized for his within which unit vectors will be placed (Figure 12B).
migration maps (compare Figures 7B and 10A). In exam- Each of these square cells is then divided into subcells,
ining Figure 10A, we can see that the underlying geogra- with the number of subcells a function of the magnitude
phy has been distorted in order to permit flow lines to of the data being mapped (Figure 12C). Finally, individual
merge and to minimize the likelihood that the flow lines unit vectors are placed within each subcell, with the ori-
cross one another. Since such distortion may be confus- entation corresponding to the direction of the flow being
ing, Phan et al. also experimented with a version of the mapped (Figure 12D).
409
18:02:59.
FIGURE 10 Flow maps created

using Phan et al.’s approach: (A) the
underlying geography has been dis-
torted to permit flow lines to merge
and to minimize the likelihood of
flow lines crossing one another;
(B) a version in which the under-
lying geography is not distorted.
(After Phan, D. et al. (2005) “Flow
map layout.” Proceedings of the 2005
IEEE Symposium on Information
Visualization; © 2005 IEEE.)
As an example of the unit-vector density technique, low wind speed defined by the differing densities of unit
consider the map of average January surface winds for vectors (e.g., the central high plains and portions of the
the United States shown in Figure 13. Lavin and Cerveny Rocky Mountains, respectively).
noted that this map illustrates “many of the features A less obvious but intriguing use of unit-vector
sought on more traditional” maps, but avoids the diffi- density mapping is the portrayal of topography. In this
culty of determining wind speed (p. 135). In particular, case, elevation is represented by the density of unit
they noted (1) areas where arrows appear to converge or vectors, whereas aspect (the direction the slope faces)
diverge, which are indicative of low- and high-pressure is depicted by the orientation of the unit vector; no
areas, respectively (in the extreme southwestern portion arrow symbol is necessary because topography does not
of the United States, there is an apparent area of diver- move. The density of unit vectors readily identifies the
gence); (2) the flow off the Great Lakes, which can lead to regions of highest elevation, and the orientation of
“lake effect” precipitation; and (3) the areas of high and vectors indicates significant characteristics of aspect.
18:02:59.
FIGURE 11 Wind arrows: a traditional method for

depicting wind speed and direction. (From Lavin and
Cerveny 1987, p. 132; courtesy of The British
Cartographic Society.)
A B
FIGURE 12 Lavin and Cerveny’s unit-vector den-

C D sity method for mapping vector-based flows (i.e.,
data that have both magnitude and direction): (A)
input is an equally spaced gridded network of mag-
nitudes and directions; (B) a square cell is presumed
around each grid point; (C) the cells in B are divided
into subcells, the number of which is proportional to
the magnitude of the data in each cell; (D) a unit
vector is placed within each subcell, with the orien-
tation corresponding to the direction of the data.
(From Lavin and Cerveny 1987, p. 134; courtesy of
The British Cartographic Society.)
To illustrate, Figure 14 contrasts present-day Himalayas, the Greenland ice sheet, and the Andes. In
topography with the estimated situation 18,000 years Figure 14B, these areas are again apparent, but other
ago. In Figure 14A, the density of unit vectors assists in high-elevation areas are also apparent, which reflect the
identifying major topographic features such as the fact that major ice sheets once covered much of North
411
18:02:59.
FIGURE 13 A unit-vector density map based on average January surface winds for 187 weather stations. (From Lavin and
Cerveny 1987, p. 136; courtesy of The British Cartographic Society.)
America and Europe. Also note the orientation of the streamlets—the individual curved line segments in
unit vectors in Figure 14B; particularly distinctive are Figure 15A) small amounts in random directions and
those radiating from source regions for glaciation in measuring the energy of the resulting map, where the
what is now Canada. energy was based on a blurred image of the original
map. Although the details of their measure of energy are
Turk and Banks’ Image-Guided Streamline Place- beyond the scope of this text, we can illustrate the notion
ment. Greg Turk and David Banks (1996) developed an graphically. Figure 15B is a blurred image of the raw vec-
image-guided streamline placement method as an alter- tor plot shown in Figure 15A; Turk and Banks argued
native to vector plot methods such as Lavin and Cerve- that the bright and dark locations are indicative of situ-
ny’s.Actually,Turk and Banks did not appear to be aware ations where the streamlets are too far apart or too close
of Lavin and Cerveny’s particular implementation, but together, respectively. When their algorithm was applied
instead referenced vector plotting methods in the com- (the results are shown in Figure 16A), the blurred image
puter science literature that appeared to share some of was much more uniform, suggesting that the streamlets
the characteristics of Lavin and Cerveny’s method. Turk were neither too far apart nor too close together (Figure
and Banks noted two problems with traditional vector 16B).
plotting methods: (1) “. . . the human eye often picks out In addition to preventing individual vectors from
runs of adjacent arrows and groups them together visually, crossing one another, Turk and Banks also developed
despite the fact that these groups are an artifact of the methods for merging streamlets into longer
underlying grid pattern . . .” and (2) “portions of neigh- streamlines (arguing that “each endpoint of a stream-
boring arrows come very close to one another and may line distracts from the visual flow of the image”
even overlap” (p. 454). Figure 15A is a traditional vector (p. 455)) and for mapping scalar quantities associated
map exhibiting these problems. with a vector field (wind speed would be a scalar
Turk and Banks attempted to solve the above prob- quantity associated with wind direction). Color Plate
lems by repeatedly moving individual vectors (or 4 is an example of a map illustrating wind speed and
18:02:59.
A
FIGURE 14 A comparison of unit-vector density maps for (A) present-day topography and (B) estimated topography
18,000 years before the present. (From Lavin and Cerveny 1987, p. 139; courtesy of The British Cartographic Society.)
413
18:02:59.
A A
B B
FIGURE 15 (A) An illustration of grouping and overlap FIGURE 16 (A) The result of applying Turk and Banks’ algo-
problems that may arise with traditional vector plotting meth- rithm to the data used to generate Figure 15A; (B) a blurred
ods; (B) a blurred version of A. (From Turk, G. and Banks, D. version of A. (From Turk, G. and Banks, D. (1996) “Image-
(1996) “Image-guided streamline placement.” SIGGRAPH 96 guided streamline placement.” SIGGRAPH 96 Conference
Conference Proceedings, New Orleans, Louisiana, pp. 453–460. Proceedings, New Orleans, Louisiana, pp. 453–460. © 1996
© 1996 ACM, Inc. Reprinted by permission.) ACM, Inc. Reprinted by permission.)
direction over Australia resulting from applying Turk depict a wide range of attributes, such as forest loss,
and Banks’ method. In this case, higher wind speeds nuclear power production, and wealth (see Worldmapper
are represented by larger, more opaque arrows. Not at http://sasi.group.shef.ac.uk/worldmapper/index.html
only does Turk and Banks’ method produce com- for these and other examples). Area cartograms can be
pelling images, but a study by David Laidlaw and his either contiguous or noncontiguous, depending on
colleagues (2001) also revealed that two variants of whether enumeration units are adjacent or not adjacent,
Turk and Banks’ method were deemed particularly respectively. The digital construction of contiguous car-
effective when compared to other methods for visual- tograms has been plagued by two problems: the inability
izing flows. to retain correct topological relationships between enu-
meration units and slow execution times. Fortunately, a
SUMMARY new algorithm developed by Michael Glastner and Mark
Newman has dealt with these issues. Daniel Dorling also
In this chapter, we have focused on cartograms and has developed a novel method for creating cartograms in
flow maps. Cartograms involve distorting spatial geome- which circles of varying size represent the population of
try to reflect an attribute. A popular form is the area each enumeration unit. When the resulting circles are
cartogram, in which sizes of enumeration units are scaled shaded with the attribute (or attributes) of interest,
to reflect attribute values. Although population is a tradi- Dorling contends that the resulting map more properly
tional attribute depicted on cartograms, it is possible to reflects the human geography of a region because of
18:02:59.
its focus on people; this is in contrast to traditional fixed-length line segments called unit vectors. The
equivalent projections (and the associated choropleth density of vectors is a function of the magnitude of the
map), which reflect the physical extent of a region. vector-based flow (e.g., a higher wind speed produces
Flow maps depict the movement of phenomena a higher density of vectors), whereas orientation is a
between geographic locations, typically using lines of function of the direction of flow (a north wind pro-
varying width.There are many types of flow maps, includ- duces a north–south-oriented vector). Lavin and
ing distributive flow maps, network flow maps, and Cerveny suggested that unit-vector density symbology
continuous flow maps. Waldo Tobler developed the first provides more information than traditional wind
significant piece of software for displaying migration arrow symbology because users can readily interpret
flows, showing that important migration movements both wind speed and direction. One problem with
could be highlighted by displaying only movements Lavin and Cerveny’s approach is that unit vectors
above the mean of the data. Tobler also developed an often appear to overlap one another. To avoid this
ingenious method for directing migration flows to reflect problem, Turk and Banks developed the image-guided
the route over which people were presumed to migrate. streamline placement approach.
A limitation of Tobler’s software was the lack of user Overall, we have seen that cartograms and flow maps
interaction, but an updated version of Tobler’s software often provide eye-catching graphics and are particularly
permits interaction in a Windows environment. appropriate for certain kinds of data. Thus, we would
Stephen Lavin and Randall Cerveny’s unit-vector encourage you to use these approaches even when tra-
density mapping is one solution for creating con- ditional techniques are possible, as the resulting maps
tinuous flow maps. Unit-vector density mapping are more likely to attract the reader and communicate
involves changing the density and orientation of short the message that you intended.
FURTHER READING
Breding, P. (1998) The Effect of Prior Knowledge on Eighth House, D. H., and Kocmoud, C. J. (1998) “Continuous car-
Graders’ Abilities to Understand Cartograms. Unpublished togram construction.” Proceedings, Visualization ’98, Research
M.S. thesis, University of South Carolina, Columbia, SC. Triangle Park, NC, pp. 197–204.
Describes an experiment on the effectiveness of cartograms; one Describes an algorithm for constructing a contiguous-area car-
conclusion was that contiguous cartograms result in a higher per- togram that tends to preserve shape. In contrast to the “pinching
centage of correct responses than proportional symbol maps and ballooning” found in the Gusein-Zade and Tikunov algorithm
although the cartograms took longer to process. (see preceding citation), the authors argue that their approach
Cuff, D. J., Pawling, J. W., and Blair, E. T. (1984) “Nested “preserves the distinctive shapes of [regions]” (p. 202).
value-by-area cartograms for symbolizing land use and other Jackel, C. B. (1997) “Using ArcView to create contiguous and
proportions.” Cartographica 21, no. 4:1–8. noncontiguous area cartograms.” Cartography and Geographic
Introduces a novel mapping technique in which nested cartograms Information Systems 24, no. 2:101–109.
are centered in enumeration units and scaled proportionally to Presents Avenue code for contiguous- and noncontiguous-area
the percentage of various land uses. cartograms. (Avenue is a programming language used in associa-
Dent, B. D. (1999) Cartography: Thematic Map Design. 5th ed. tion with ArcView.)
Boston: McGraw-Hill. Keim, D. A. (1998) “The Gridfit Algorithm: An efficient and
Chapters 11 and 12 provide an overview of cartograms and flow effective approach to visualizing large amounts of spatial
mapping, respectively. data.” Proceedings, Visualization ’98, Research Triangle Park,
Dorling, D. (1993) “Map design for census mapping.” The NC, pp. 181–188.
Cartographic Journal 30, no. 2:167–183. Describes an approach in which individual plotted points (e.g.,
Provides a thorough discussion of Dorling’s novel cartogram lightning strikes) are plotted near their actual location to avoid
method; for atlases comprised of cartograms, see Dorling (1995a) overplotting. The approach thus distorts point locations, as
and Dorling and Thomas (2004). opposed to the area distortion common in cartograms.
Gusein-Zade, S. M., and Tikunov, V. S. (1993) “A new technique Marble, D. F., Gou, Z., Liu, L., and Saunders, J. (1997) “Recent
for constructing continuous cartograms.” Cartography and advances in the exploratory analysis of interregional flows in
Geographic Information Systems 20, no. 3:167–173. space and time.” In Innovations in GIS 4, ed. by Z. Kemp,
pp. 75–88. Bristol, PA: Taylor & Francis.
Describes an algorithm for constructing a contiguous-area car-
togram that tends to preserve shape. The authors argue that in com- Describes data exploration software developed at Ohio State
parison to earlier algorithms, this one “obtains better resultant University for analyzing interregional flows in both space and time.
images, especially in cases where a sharp difference in the initial dis- The software had the ability to map more than one attribute simulta-
tribution of the variable exists” (p. 172). neously, which was accomplished using “projection pursuit” concepts.
415
18:02:59.
Rittschof, K. A., Stock, W. A., Kulhavy, R. W., Verdi, M. P., and Provides a history of cartograms.
Johnson, J. T. (1996) “Learning from cartograms: The effects Tobler, W. (2003) “Movement mapping.” http://www.csiss.org/
of region familiarity.” Journal of Geography 95, no. 2:50–58. clearinghouse/FlowMapper/MovementMapping.pdf.
An experimental study of the effectiveness of cartograms; also see Tobler provides some thoughts on flow mapping and includes
Griffin (1983). maps, Web links, and references; for some early work on mapping
Thompson, W., and Lavin, S. (1996) “Automatic generation of continuous flows, see Tobler (1981).
animated migration maps.” Cartographica 33, no. 2:17–28. Tobler, W. (2004) “Thirty-five years of computer cartograms.”
Summarizes methods for mapping migration and develops an ani- Annals of the Association of American Geographers 94,
mated method in which either small arrows or circles appear to no. 1:58–73.
move from one region to another. Provides a historical overview of digital methods for creating
Tikunov, V. S. (1988) “Anamorphated cartographic images: cartograms.
Historical outline and construction techniques.” Cartography
17, no. 1:1–8.
GLOSSARY
area cartogram: a map in which the areal relationships of flow map: a map used to depict the movements of phenom-
enumeration units are distorted on the basis of an attribute ena among geographic locations; generally, this is done using
(e.g., the sizes of states are made proportional to the number “flow lines” of varying thickness.
of deaths in each state due to AIDS). image-guided streamline placement: an approach for map-
cartogram: a map that purposely distorts geographic space ping continuous flows that measures the energy of the result-
based on values of a theme (e.g., making the size of countries ing map, where the energy is based on a blurred image of the
proportional to their population). original map.
continuous flow map: depicts the movement of a continuous network flow map: depicts flows within networks, such as a
phenomenon such as wind or ocean currents. transportation network.
continuous vector-based flow: a flow composed of two radial flow map: a flow map that has a distinctive radial pat-
attributes, magnitude and direction, that can change at any tern, such as a map centered on Kansas showing migration
point (e.g., at any point in the atmosphere, we can compute from Kansas to other states as a result of the Dust Bowl in the
the speed of and direction from which the wind blows). 1930s.
distance cartogram: a map in which real-world distances are telecommunications flow map: depicts flows of telecommu-
distorted to reflect some attribute (e.g., distances on a subway nications technology, including the Internet and its associated
map might be distorted to reflect travel time). information spaces.
distributive flow map: depicts the movement of commodi- unit-vector density map: a map in which the density and ori-
ties, people, or ideas between geographic regions. entation of short fixed-length segments (unit vectors) are used
equivalent (equal-area) projection: a map projection for to display the magnitude and direction, respectively, of a con-
which areas are preserved compared to the same areas on the tinuous vector-based flow, such as wind speed and direction.
Earth’s surface.
REFERENCES
Dent, B. D. (1999) Cartography: Thematic Map Design (5th Dorling, D. (1995c) “Visualizing changing social structure
ed.). Boston: McGraw-Hill. from a census.” Environment and Planning A 27:353–378.
Dorling, D. (1993) “Map design for census mapping.” The Dorling, D., Barford, A., and Newman, M. (2006)
Cartographic Journal 30, no. 2:167–183. “Worldmapper: The world as you’ve never seen it before.”
Dorling, D. (1994) “Cartograms for visualizing human IEEE Transactions on Visualization and Computer
geography.” In Visualization in Geographical Information Graphics 12, no. 5: 757–764.
Systems, ed. by H. M. Hearnshaw and D. J. Unwin, pp. Dorling, D., and Thomas, B. (2004) People and Places: A 2001
85–102, plates 7–10. Chichester, England: Wiley. Census Atlas of the UK. Bristol, UK: The Policy Press.
Dorling, D. (1995a) A New Social Atlas of Britain. Dodge, M., and Kitchin, R. (2001a) Atlas of Cyberspace.
Chichester, England: Wiley. Harlow, England: Addison-Wesley.
Dorling, D. (1995b) “The visualization of local urban change Dodge, M., and Kitchin, R. (2001b) Mapping Cyberspace.
across Britain.” Environment and Planning B: Planning London: Routledge.
and Design 22, no. 3:269–290.
18:02:59.
Gastner, M. T., and Newman, M. E. J. (2004) “Diffusion- Parks, M. J. (1987) “American flow mapping: A survey of the
based method for producing density-equalizing maps.” flow maps found in twentieth century geography text-
Proceedings of the National Academy of Sciences 101, books, including a classification of the various flow map
no. 20:7499–7504. designs.” Unpublished MA thesis, Georgia State
Gastner, M. T., and Newman, M. E. J. (2005) “Density-equal- University, Atlanta, GA.
izing map projections: Diffusion-based algorithm and Phan, D., Xiao, L., Yeh, R., Hanrahan, P., and Winograd, T.
applications.” GeoComputation 2005, Ann Arbor, MI. (2005) “Flow map layout.” Proceedings of the 2005 IEEE
http://www-personal.umich.edu/~mejn/papers/ Symposium on Information Visualization, Minneapolis,
geocomp.pdf. MN, pp. 219–224.
Griffin, T. L. C. (1983) “Recognition of areal units on http://graphics.stanford.edu/papers/flow_map_layout/.
topological cartograms.” The American Cartographer 10, Robinson, A. H. (1967) “The thematic maps of Charles
no. 1:17–29. Joseph Minard.” Imago Mundi 21: 95–108.
Keahey, T. A. (1999) “Area-normalized thematic views.” Robinson, A. H. (1982) Early Thematic Mapping in the History
Proceedings of the 19th International Cartographic of Cartography. Chicago: University of Chicago Press.
Conference, Ottawa, Canada, pp. Section 6:12–21 (CD- Tobler, W. R. (1981) “A model of geographical movement.”
ROM). Geographical Analysis 13, no. 1:1–20.
Laidlaw, D. H., Davidson, J. S., Miller, T. S., da Silva, M., Tobler, W. R. (1987) “Experiments in migration mapping by
Kirby, R. M., et al. (2001) “Quantitative comparative computer.” The American Cartographer 14, no. 2:155–163.
evaluation of 2D vector field visualization methods.”
Turk, G., and Banks, D. (1996) “Image-guided streamline
Proceedings of the Conference on Visualization ‘01, San
placement.” SIGGRAPH 96 Conference Proceedings, New
Diego, CA, pp. 143–150.
Orleans, LA, pp. 453–460.
Lavin, S. J., and Cerveny, R. S. (1987) “Unit-vector density http://doi.acm.org/10.1145/237170.237285.
mapping.” The Cartographic Journal 24, no. 2:131–141.
Yadav-Pauletti, S. (1996) “MIGMAP, a data exploration
Olson, J. M. (1976b) “Noncontiguous area cartograms.” The application for visualizing U.S. Census migration data.”
Professional Geographer 28, no. 4:371–380. Unpublished MA thesis, University of Kansas,
Lawrence, KS.
417
18:02:59.
18:02:59.
Visualizing Terrain
OVERVIEW
In this chapter we consider approaches for symbolizing a (or bird’s-eye views), and draped images. Although
particular type of smooth, continuous phenomenon—the such approaches do not provide the geometric accu-
Earth’s terrain. By “terrain,” we mean the Earth’s eleva- racy of a vertical view, their realism (and associated
tion (both above and below sea level) and associated fea- lack of abstraction) makes them an attractive alterna-
tures found on the Earth—its landscape. Specialized tech- tive. Traditionally, only a single oblique view of a land-
niques have been developed for symbolizing terrain scape was possible and so some information was of
because of its importance in everyday life and in the bur- necessity hidden. Now, however, interactive software
geoning area of GIS. permits users to create their own oblique views.
Section 1 briefly considers the kinds of data normally Another limitation was the time needed to create real-
used to represent terrain. Fundamental data consist of ele- istic oblique views, but digital design tools are now
vation values above (or below) mean sea level for indi- making it easier for cartographers to create realistic-
vidual point locations. These data are most commonly looking views.
presented in a square gridded network (a so-called digital Although maps based on an oblique view have a dis-
elevation model, or DEM). Data for depicting the Earth’s tinctive 3-D appearance, they are still created on a two-
landscape can be acquired from mapping agencies such dimensional surface (either paper or a graphics display
as the USGS. device). It can be argued that a more effective approach
Section 2 considers approaches for symbolizing terrain is a physical model that is truly 3-D (i.e., we can hold
that have focused on a vertical (or plan) view, as though the and manipulate small models in our hands). Such physi-
map reader is viewing the Earth from an airplane high cal models are the topic of section 4. Physical models
above it. Methods covered include hachures, contour-based have been around for hundreds of years, but only
methods, the physiographic method, shaded relief, and recently has technology enabled us to create such models
various morphometric methods (which focus on the struc- from digital data.
ture of the terrain such as aspect and slope). Assuming that
the topographic surface is illuminated obliquely (typically
from the northwest), these methods utilize various 1 NATURE OF THE DATA
approaches for distinguishing illuminated areas from areas
in the shadows. Because the visual effect of shaded relief Fundamental data for depicting terrain consists of
can be rather dramatic, and because it has become a popu- elevation values above (or below) mean sea level for
lar computer-based method, we cover some of the details of individual point locations. Although elevation data might
associated computational formulas. be available at irregularly spaced control points (as
In section 3, we consider approaches for symboliz- with other smooth, continuous phenomena), these data
ing terrain that have focused on an oblique view of are most commonly provided in a square gridded net-
the Earth, including block diagrams, panoramas work (a so-called digital elevation model, or DEM).
419
18:03:38.
Visualizing Terrain
Elevation data are sometimes also provided as a causes us to expect a light in a similar orientation in
triangulated irregular network (TIN). The procedures for other situations such as relief depiction. If an alternate
creating DEMs and TINs are described in GIS textbooks light direction is used, then the relief might appear to be
(e.g., Chapter 9 of Lo and Yeung 2007). DEM data can be inverted (e.g., valleys might seem to be mountains).†
acquired from various mapping agencies such as the
USGS (http://edc.usgs.gov/products/elevation/dem.html).* 2.1 Hachures
When symbolizing terrain, you will often want to
display not only the elevation of the landscape but also Hachures are constructed by drawing a series of paral-
features found on the landscape (e.g., roads, streams, lel lines perpendicular to the direction of contours. Two
and land use). There are two basic approaches for por- forms of hachures were traditionally used in manual
traying such features. One is to utilize “raw” remotely cartography (Yoeli 1985). In the first, known as slope
sensed images, such as those obtained by satellites (e.g., hachures, the width of hachures was proportional to the
Landsat); the other approach is to use maps that have steepness of the slope; thus, steeper slopes had thicker
been either derived from remotely sensed images (e.g., hachures and were darker. Because slope hachures gen-
through visual interpretation) or acquired from ground erally failed to create the impression of a third dimen-
surveys. An excellent source of such data for the United sion, shadow hachures were developed (Figure 1) in
States is the EROS Data Center (http://edc.usgs.gov/). which an oblique light source was presumed. In this
approach, the width of hachures was a function of
2 VERTICAL VIEWS whether the hachure was illuminated or in the shadow.
Although hachures normally are associated with the
bygone era of manual cartography, Imhof (1982, 229)
In this section, we consider approaches for symbolizing argued that they should not be forgotten:
terrain that traditionally emphasized a vertical (or plan)
view of the landscape.A vertical view was emphasized for Hachures alone, without contours, are more capable of
two reasons. First, there was a desire to visualize the land- depicting the terrain than is [relief] shading alone. . . .
scape of a particular region (say, the Rocky Mountains in Hachures also possess their own special, attractive graphic
Colorado) in its entirety. Given the time and cost con- style. They have a more abstract effect than [relief]
straints of creating printed maps, this generally meant that shading, and perhaps for this reason are more expressive.
a single map would be produced and that to visualize the
entire region, this map would have to be viewed vertically Additionally, Pinhas Yoeli (1985, 112) argued that
(because an oblique view would of necessity hide some hachures could be used for a structural analysis of the
features). Second, there was a desire to show features in landscape (e.g., to depict different degrees of slope, as
their correct geographic location. In this respect, cartog- in Figure 2).
raphers distinguished between maps and paintings. For The maps shown in Figures 1 and 2 were developed
instance, Eduard Imhof (1982, 79–80) stated: by Yoeli using the following rules recommended by
Imhof:
The map is not only a picture; it is primarily, a means of
1. Each hachure follows the direction of steepest
providing information [italics in original]. It has other
gradient.
responsibilities than has painting . . . . Above all, it must
include the depth of the valleys and the heights of moun- 2. Hachures are arranged in horizontal rows.
tains, not only in the visual sense but in the geometrical 3. The density of hachures is constant except for the
sense also. horizontal plane, which is blank.
4. The length of hachures corresponds to the local
To create the impression of a 3-D view, the bulk of the
horizontal distance between assumed contours.
methods in this section assume that the surface is lit
obliquely from the northwest. According to Imhof 5. Wider hachures are used on steeper slopes.
(1982, 178), a northwest light source is commonly used Imhof suggested that these rules are appropriate for
because of the preponderance of people who write with large-scale maps, but that they should be relaxed for
their right hand. When writing with the right hand, small-scale maps (i.e., scales smaller than 1:500,000).
lighting is placed above and to the left to eliminate the Based on this notion, Patrick Kennelly and Jon
possibility that a shadow will be cast over the written
material. Our experience in using a light in this situation
†It should be noted that by placing a method in this section, we do
not mean to imply that these approaches could be used only with a
* An extensive set of Web links to sources of digital spatial data for vertical view, but rather that they have been most commonly used to
the United States can be found at http://libinfo.uark.edu/gis/us.asp. create vertical views.
18:03:38.
Visualizing Terrain
of hachure lines to remove the ambiguity of which end

of the hachure is upslope or downslope (Figure 4). Their
method calculates a slope and an aspect grid from the
DEM, converts these grids to point themes, and spa-
tially joins the point themes into one theme containing
both slope and aspect information. Arrows are then
drawn in the direction of aspect (Rule 1), with thickness
being a function of slope (Rule 5). The hachure density
is constant because all points derived from the DEM
are evenly spaced (Rule 3).*
2.2 Contour-Based Methods
Several methods for portraying terrain have been

developed that involve manipulating contour lines.
Tanaka’s method (Tanaka 1950) involves varying the
width of contour lines as a function of their angular
relationship to a presumed light source; lines perpen-
dicular to the light source have the greatest width,
whereas those parallel to the light source are narrowest.
Furthermore, contour lines facing the light source are
drawn in white, whereas those in shadow are drawn in
black (Figure 5). Although this technique was used as
early as 1870 (Imhof 1982, 153), Kitiro Tanaka is
generally credited with originating it because of the
attractive maps he produced (Figure 6) and the mathe-
matical basis he provided.
Like hachures, Tanaka’s method originally was pro-
duced manually. Berthold Horn (1982) was one of the
first to create a computer-based version of Tanaka’s
method. A visual analysis of Horn’s maps suggests that
Tanaka’s method compares favorably with other com-
puter-assisted methods for representing relief (see
Horn 1982, 113–123). More recently, Patrick Kennelly
and Jon Kimerling (2001) suggested improvements to
Tanaka’s method. One of the attractive maps that
resulted from their work appears in Color Plate 1. Here
we see that they drew contours in a fashion similar to
Tanaka, but added colors between the contours “to
give the topography a somewhat realistic appearance”
(p. 117).
J. Ronald Eyton’s (1984b) illuminated contours , a
FIGURE 1 Hachures created using Yoeli’s computer-based raster (pixel-based) approach for computer-based con-
procedure. (From Yoeli 1985, 123; courtesy of The British touring, was a simplification of Tanaka’s method. Similar
Cartographic Society.) to Tanaka, Eyton brightened contours facing the light
source and darkened contours in the shadow. Eyton’s
approach, however, differed in that the width of contour
Kimerling (2000) devised a method based on regularly lines was not varied (in order to speed computer pro-
spaced point symbols derived from a DEM. The result- cessing). The image resulting from Eyton’s approach is
ing hachure maps (Figure 3) obey Rules 1, 3, and 5 but not as distinctive as Tanaka’s (compare Figures 6 and 7)
ignore Rules 2 and 4, which assume a contour frame-
work for hachure construction. Kennelly and
Kimerling’s obliquely illuminated map avoids changing * Kennelly and Kimerling used a northeastern illumination rather
the thickness or spacing of hachures by using white and than the usual northwestern illumination because they felt that “it
black hachures on a gray background. Kennelly and highlighted a few features slightly better” (Patrick Kennelly, personal
Kimerling also added arrowheads to the downslope end communication).
421
18:03:38.
Visualizing Terrain
FIGURE 2 Hachures used for a structural analysis of the landscape, in this case depicting differing degrees of slope. (From Yoeli
1985, 124; courtesy of The British Cartographic Society.)
FIGURE 3 A small-scale map of the Cascade Mountains in Washington State that was created using Kennelly and Kimerling’s
hachure method. (Courtesy of Patrick Kennelly.)
18:03:38.
Visualizing Terrain
FIGURE 4 An enlargement of a por-

tion of Figure 3 showing the character
of individual hachures. (Courtesy of
Patrick Kennelly.)
FIGURE 5 A simplified representation of

Tanaka’s method for depicting terrain. The
width of contour lines is a function of their
angular relationship with a light source;
contour lines facing the light source are drawn
in white, whereas those in the shadow are
drawn in black. (After Tanaka 1950. First
published in The Geographical Review 40(3),
p. 446. Reprinted with permission of the
American Geographical Society.)
423
18:03:38.
Visualizing Terrain
FIGURE 6 A map created using Tanaka’s method for depicting terrain. (After Tanaka 1950. First published in The Geographical
Review 40(3), p. 451. Reprinted with permission of the American Geographical Society.)
but still gives the impression of three dimensions, espe- easily recognized symbols. A portion of Raisz’s (1967)
cially when viewed from a distance. classic map Landforms of the United States is shown in
Shadowed contours is another approach for repre- Color Plate 2A, and a sampling of his standard symbols
senting terrain by modifying contour lines. In this is shown in Figure 10. Raisz argued that his physio-
approach, contour lines facing the light source are graphic map “appeals immediately to the average [per-
drawn in a normal line weight, whereas contour lines in son]. It suggests actual country and enables [the indi-
the shadow are thickened (Figure 8). This technique dif- vidual] to see the land instead of reading an abstract
fers from Tanaka’s principally in that the contours fac- location diagram. It works on the imagination” (Raisz
ing the light source are drawn in black rather than 1931, 303). To our knowledge, no one has ever auto-
in white (compare Figures 5 and 8). Figure 9 depicts a mated Raisz’s method, probably because it would
computer-assisted implementation of shadowed require extensive knowledge of geomorphology and
contours developed by Yoeli (1983). involve a subjective interpretation of the landscape.
2.3 Raisz’s Physiographic Method 2.4 Shaded Relief
Based on the work of A. K. Lobeck (1924), which we Shaded relief (also known as hill shading or chiaroscuro)
consider later, Erwin Raisz (1931) developed the has long been considered one of the most effective
physiographic method, in which the Earth’s geomor- methods for representing terrain. Swiss cartographers
phologic features are represented by a set of standard, (Imhof in particular) have been widely recognized for
18:03:38.
Visualizing Terrain
FIGURE 8 A simplified representation of shadowed

contours in which contour lines facing a light source are
drawn in black, whereas contour lines facing away from the
light source produce a black shadow. (After Yoeli 1983. First
published in The American Cartographer 10(2), p. 106.
Reprinted with permission from the American Congress on
Surveying and Mapping.)
FIGURE 7 A map created using Eyton’s illuminated
contours in which contours facing the light source are drawn
in white, whereas contours in the shadow are drawn in black;
(e.g., ArcGIS) and software explicitly intended for map-
note that the width of contour lines is not varied. (Courtesy of
ping smooth, continuous phenomena (e.g., Surfer).
J. Ronald Eyton.)
The effectiveness of automated shaded relief maps
is superbly illustrated by Gail Thelin and Richard
Pike’s (1991) relief map of the conterminous United
their superb work. Like the preceding methods, hill
States. Thelin and Pike (3) noted that their map not
shading presumes an oblique light source: Areas facing
only portrays well-known features such as the “struc-
away from the light source are shaded, and areas directly
tural control of topography in the folded Appalachian
illuminated are not shaded. Slope can also be a factor,
Mountains,” but also portrays less familiar features
with steeper slopes shaded darker. When produced man-
such as the Coteau des Prairies, a flat-iron shaped
ually, shading is accomplished using pencil and paper or
plateau in South Dakota associated with glaciation.
an airbrush. Color Plate 3 is an example of one of the
Pierce Lewis (1992) described Thelin and Pike’s map
maps that Imhof created, and Color Plate 2B is a portion
as follows:
of a map that Richard Edes Harrison (1969) produced
for the 1970 U.S. National Atlas.
Let us not mince words. The United States Geological
Numerous people have contributed to automating Survey has produced a cartographic masterpiece: a relief
shaded relief maps, with pioneering work done by Yoeli map of the coterminous United States which, in accuracy,
(1967; 1971), Brassel (1974), Hügli (1979), and Horn elegance, and drama, is the most stunning thing since
(1982). A discussion of sophisticated lighting algorithms Erwin Raisz published his classic “Map of the Landforms
that would apply to more than just shaded relief can be of the United States.” (p. 289)
found in basic computer graphics texts such as those by
Foley et al. (1996) and Rogers (1998). Today, shaded Because Raisz’s map has been so widely used by
relief maps are a standard element of GIS software geographers, it is natural to ask which is better, the
425
18:03:38.
Visualizing Terrain
contrast, the Raisz map represents one knowledgeable

geographer’s view of the landscape, with certain
features emphasized, as Raisz deemed appropriate.
Another good illustration of automated shaded relief
is Lloyd Treinish’s (1994) rendition of the Earth’s ter-
rain (Color Plate 4), which he developed using the IBM
Visualization Data Explorer (DX; http://www.
research.ibm.com/dx/). The attractiveness of this image
is apparent in the fact that a variant of the image was
chosen for the cover of the December 1992 issue of GIS
World, a popular magazine for those interested in GIS.
The image was created by combining shaded relief tech-
niques with hypsometric tints representing elevation
values above and below sea level. The shaded relief
technique was based on Gouraud shading, which inter-
polates shades between vertices on a polygon (or cell)
boundary, and thus produces a smoother appearance
than if the same shade were used throughout a polygon
(Foley et al. 1996, 734–737). The color scheme for eleva-
tion values (see the top of Color Plate 4) was developed
so that the portion above sea level had “the appearance
of a topographic map” (Treinish 1994, 666), whereas the
portion below sea level used darker shades of blue to
represent greater depth.
Although fully automated shaded relief methods
can produce highly effective renditions, they are some-
times deficient compared to their manual counterparts.
On this basis, Lorenz Hurni and his colleagues (2001)
FIGURE 9 A map created using shadowed contours. (After noted that interpretation of shaded relief maps can be
Yoeli 1983. First published in The American Cartographer enhanced through user-controlled local adjustments of
10(2), p. 109. Reprinted with permission from the American the assumed position of the light source, vertical exag-
Congress on Surveying and Mapping.) geration, brightness, elevation-dependent contrast, and
the transition between diffuse reflection and aspect-
USGS map or the Raisz map? Lewis (1992) argued that based shading. Similarly, Wendy Price (2001) described
both maps have merit: The USGS map is “more like an the limitations of software tools intended to simulate
aerial photograph” (298), allowing us to see the the traditional airbrush, and suggested that the manual
structure of the landscape as it actually exists (at least airbrush still has its uses.
within the constraints of the computer program). In Some cartographers have utilized multiple light
sources when producing shaded relief (e.g., Hobbs 1995,
1999). An intriguing extension of this idea is the work of
Patrick Kennelly and James Stewart (2006), who devel-
oped a uniform sky illumination model in which the light
source is presumed to be evenly distributed throughout
the sky. A comparison of shaded relief images created
using a single point source illumination and the uniform
sky illumination model is shown in Figure 11. For the
single point source (Figure 11A), a nontraditional north-
east light source is used at a 45° angle above the hori-
zon. In this case, intermediate slopes with aspects facing
the northeast are brightest, and areas of steep slope fac-
FIGURE 10 Examples of the standard symbols that Raisz ing the southwest are darkest. In the uniform model
used to create his physiographic diagrams. (After Raisz 1931. (Figure 11B), the brightest areas are either mountain
First published in The Geographical Review 21(2), p. 301. ridge lines or relatively large, flat areas, whereas narrow
Reprinted with permission of the American Geographical valley bottoms are darkest. Comparing the two images,
Society.) it appears that the uniform model helps us visualize sub-
18:03:38.
Visualizing Terrain
A B
FIGURE 11 A comparison of shaded relief images using (A) a single point source illumination and (B) a uniform sky illumination
model. The point source approach shows overall terrain structure and aspect, while the uniform approach shows subtle detail and
relative elevation differences. (After Kennelly and Stewart 2006. First published in Cartography and Geographic Information
Science 33(1), p. 30. Reprinted with permission from the American Congress on Surveying and Mapping.)
tle terrain differences that are not apparent with the tra- shown in Figure 12. We will presume that our objective is
ditional point source approach. For instance, note the to determine an appropriate gray tone for the central cell
detail shown in the north–south trending valley in the of the 3 * 3 window. The following steps are based on
southeastern portion of the map. The deepest under- Eyton’s (1991) formulas.*
standing of the landscape, however, appears to be
Step 1. Compute the slope of the window. The slopes
achieved when the two images are viewed together. The
along the x and y axes, slx and sly, are
traditional approach appears useful for showing overall
terrain structure and aspect, while the uniform model is
Z(2,3) - Z(2,1)
useful for showing subtle detail and relative elevation slx =
differences. 2D
Z(3,2) - Z(1,2)
Computation of Shaded Relief sly =
2D
A DEM is the basic input necessary for creating a
computer-assisted shaded relief map. Fundamental calcu- * Fora discussion of alternative approaches for computing slope and
lations involve manipulating 3 * 3 “windows” of eleva- aspect, see Chrisman (2002, 174–179). Studies of various approaches
tion values within the grid of the DEM. To illustrate this can be found in Hodgson (1998), Jones (1998), and Dunn and
concept, we will use the window and hypothetical data Hickey (1998).
427
18:03:38.
Visualizing Terrain
A B
Representative Window Hypothetical Data
Z(1,1) Z(1,2) Z(1,3) 100 104 110
Z(2,1) Z(2,2) Z(2,3) 105 110 115
Z(3,1) Z(3,2) Z(3,3) 110 114 120
D 30
FIGURE 12 (A) A representative window and (B) hypothetical data used to compute digital shaded relief.
where the Z values represent digital elevation values as an azimuth (an angle measured clockwise from a north
shown in Figure 12A, and D is the distance between base of 0°). (f) is defined as
grid points.
f = cos - 1 c d
For the sample data, the results are slx
115 - 105 slo
slx = = 0.167
60
For the hypothetical data, we have
114 - 104
sly = = 0.167
f = cos - 1 c d = 45
60 0.167
0.236
The overall slope, slo, is then defined as
To convert (f) to an azimuth (u), we use the signs
2(slx)2 + (sly)2 shown in Table 1. For the hypothetical data, we compute
u = 270° ⫹ 45° ⫽ 315°. Note that this result confirms a
For the hypothetical data, we have visual examination of the 3 * 3 window in Figure 12:
The window appears to face northwest (an azimuth of
2(0.167)2 + (0.167)2 = 0.236 315°).
Because the resulting slope is expressed as a tangent

(the ratio of rise over run), it can be converted to an TABLE 1 Conversion of local angle (f) to azimuth (u)
angular value by computing the arc tangent (tan-1) . for calculations associated with computer-based shaded
The arc tangent for the hypothetical data is relief
tan-1 0.236 = 13.3°. Thus, the hypothetical data have Sign of Slope in x(slx) Sign of Slope in y(sly) Azimuth
an overall slope of 13.3°.
Positive Positive 270° + f
Step 2. Compute the downslope direction, or azimuth,
Positive Negative 270° - f
for the 3 * 3 window. This is accomplished by calculat-
Negative Positive 90° - f
ing a local angle (f) between the overall slope and the
Negative Negative 90° + f
slope in the x direction, and then by converting this to
18:03:38.
Visualizing Terrain
Step 3. Estimate the reflectance for the central cell of elevations of 315° and 45°, respectively, then the
the 3 * 3 window, the cell associated with Z(2, 2). A sim- Lambertian reflectance for the hypothetical data is
ple but reasonably effective approach is to presume a
Lr = cos(315 - 315) cos(90 - 13.3) cos(45)
Lambertian reflector, which reflects all incident light
+ sin(91 - 13.3) sin(45)
equally in all directions. The formula for Lambertian
= 0.850
reflectance is
Step 4. Convert the Lambertian reflectance value to
Lr = cos(A f - A s) cosEf cos Es
either RGB or CMYK values appropriate for display or
+ sin Ef sinEs
printing. Presuming a graphics display with intensities
where Lr = relative radiance (scaled 0.0 - 1.0) ranging from 0 to 255 for each RGB color, RGB values
would be calculated by multiplying the Lambertian
Af = azimuth of a slope facet (0° to 360°)
reflectance by 255 (the same value would be used for
As = azimuth of the sun (0° to 360°) each color to produce a gray tone). For our hypothetical
data, the result would be 255 * 0.850 = 217 for each
Ef = elevation of a normal to the slope facet
color. Thus, the central cell of the 3 * 3 window would
(90°minus the slope magnitude in degrees) have a relatively bright reflectance value.
Es = elevation of the sun (0° to 90°)
2.5 Morphometric Techniques
A visual representation of these terms is shown in
Figure 13. If we presume solar azimuth and solar The techniques that we have described thus far are
intended to provide an overall view of terrain. Often,
however, it is useful to focus on particular morphomet-
ric (or structural) elements of terrain. Aspect and slope
Zenith are two of the most common structural elements that
have been visualized. Aspect deals with the direction
the land faces (e.g., north or south), whereas slope
Sun
considers the steepness of the land surface. For exam-
ple, a developer of a ski resort might desire north-facing
aspects (to minimize snowmelt) and a diversity of
slopes (to accommodate skiers’ varied abilities). In this
Normal section we consider two approaches that have been
developed for handling aspect and slope—the first deals
solely with aspect, whereas the second deals with both
aspect and slope. We then consider a software package,
North
LandSurf, that has been developed to visualize and ana-
Es lyze a broad range of morphometric elements.
Ef
Symbolizing Aspect: Moellering and Kimerling’s
MKS-ASPECT™ Approach
A f As Harold Moellering and Jon Kimerling (1990) developed
an approach for symbolizing aspect based on the oppo-
nent-process theory of color. Their approach is known
as MKS-ASPECT™ and is protected by two patents
A f = azimuth of slope facet (No. 5,067,098 and No. 5,283,858) held by the Ohio
State University Research Foundation in the name of
A s = azimuth of sun Harold Moellering and A. Jon Kimerling. MKS-
Ef = elevation of the normal to slope facet ASPECT treats aspect as a nominal phenomenon (e.g.,
a northwest aspect is different from a southeast aspect)
Es = elevation of sun
and assigns colors to opposing aspects based on oppos-
ing color channels; thus, red and green, and blue and
FIGURE 13 A visual representation of terms associated with yellow are placed opposite one another. Mixtures of
computing Lambertian reflection. (From Eyton, 1990, nonopposing hues can then be used for intermediate
Cartographica, p. 27. Reprinted by permission of University aspects (Figure 14).
of Toronto Press Incorporated. © University of Toronto Moellering and Kimerling argued that the luminance
Press 1990.) (or brightness) of a color should ideally be highest for a
429
18:03:38.
Visualizing Terrain
Moellering and Kimerling selected their colors

using the HLS (hue–lightness–saturation) color speci-
Light
Source fication model. The HLS model is similar to the HSV
Aspect
model except that the HLS model is a double rather
0º than a single hexcone. Initially, Moellering and
Kimerling created colors by holding the lightness and
315º 45º saturation of the HLS model constant at 50 (one-half
the maximum) and 100 (the maximum), respectively,
Y YG G
and simply varying hue. Unfortunately, because HLS is
not a perceptually based model, they found that this
approach produced colors relatively far from the ideal
curve shown in Figure 15. Only after considerable
270º RY HUE GB 90º
experimentation did they arrive at the set of eight
colors plotted in Figure 15 and shown in Table 2. A
map created using Moellering and Kimerling’s color
scheme is shown in Color Plate 5; in this case, 128 col-
R RB B
ors are used (rather than 8) to provide greater color
225º 135º fidelity.
180º Symbolizing Aspect and Slope: Brewer and Marlow’s

Approach
FIGURE 14 Using opponent-process colors to represent Cynthia Brewer and Ken Marlow (1993) attempted to
aspect. (After Moellering and Kimerling 1990, 154.) rectify Moellering and Kimerling’s difficulty in working
with HLS by employing the perceptually based HVC
northwest aspect (assuming the usual light source from model, at the same time incorporating slope in addition
the northwest) and decrease to a minimum for a southeast
aspect; because yellow is inherently a bright color, it is log-
ical to use yellow for the northwest aspect. More specifi-
cally, they argued that colors should ideally fall on a curve L
represented by the following equation (Figure 15):
1.0 White
L = 50 * [cos(315 - f) + 1]
where f is the azimuth of the aspect and L is the

relative luminance as measured by a colorimeter, a
device for measuring the brightness of a color.
Blue Cyan
240º 180º
Y
100
Relative Luminance (L)
YG Magenta 0.5 Green

300º 120º
75
RY
50
Red Yellow
25 R G GB 0º 60º
B RB B
0
135 225 315 45 135
Aspect (φ)
FIGURE 15 The relationship between aspect (as measured

H
in degrees clockwise from a north value of 0) and relative 0.0 S
luminance in Moellering and Kimerling’s method for symbol- Black
izing aspect. The curve shows the ideal relationship, while dots
represent the actual relationship. (After Moellering and FIGURE 16 The HLS color model. Note that it is a double
Kimerling 1990, 158.) hexcone, as opposed to HSV, which is a single hexcone.
18:03:38.
Visualizing Terrain
TABLE 2 HLS, Yxy, and RGB specifications for Moellering and Kimerling’s method for representing aspect*
Aspect H L S Y x y R G B
0 70 50 100 37.5 .354 .544 213 255 0

45 120 40 100 15.1 .288 .584 0 204 0
90 190 40 100 10.4 .200 .257 0 170 204
135 240 40 100 2.3 .160 .087 0 0 204
180 290 40 100 4.7 .250 .139 170 0 204
225 0 50 100 10.6 .594 .357 255 0 0
270 50 50 100 27.1 .446 .476 255 213 0
315 60 50 100 42.4 .396 .514 255 255 0
*Aspect is measured in degrees clockwise from north, which is 0°. HLS values are based on the color model shown in Figure 16. Yxy values are taken from
Moellering and Kimerling (1990, 158).
to aspect. Like Moellering and Kimerling, Brewer and the edge of the circle represent the hues selected for the
Marlow chose eight aspect hues. For each aspect, how- eight aspects. The hues do not appear to be equally
ever, they also depicted three slope classes (5 to 20 per- spaced (as one might expect, given the perceptual basis
cent, 20 to 40 percent, and 7 40 percent) as the differing of HVC) because a wide range of lightness and satura-
levels of saturation; they represented the 6 5 percent tion values are not possible for all hues; for example, it is
slope category as gray for all aspects. In total, they not possible to create a variety of lightnesses and satu-
showed 25 classes of information (8 aspects * 3 slope rations between green and blue. The saturation (or
classes, plus the 6 5 percent slope category). The colors chroma) of a slope class within an aspect is represented
selected by Brewer and Marlow for aspect and slope are by the distance of a black dot from the center of the dia-
shown graphically as HVC coordinates in Figure 17. gram (these dots are equally spaced along the lines,
Each black dot in the graph represents one of the 25 col- reflecting the perceptual basis of the HVC model). The
ors used. The eight lines extending from the interior to numerical values of distance can be determined using
the scale labeled “Chroma” in the diagram; for example,
for the blue hue, the third dot from the center of the dia-
135º Hue Angle gram appears to have a saturation of approximately 45.
110º H The numbers printed alongside each dot represent the
Northeast
Green
Nor
G re
145º H value (lightness) of each color. Note that higher light-

th
en-
85º H
ness values tend to be found facing the direction of the
N
Yell
or
Ye
th
155º H light source (the northwest), and that lightness also

ow
w
l
lo
80
es
w
65
t
95 varies with slope. Overall, Brewer and Marlow selected

77
87
67 colors based on their knowledge of the HVC color
80
75
70
space and on their desire to have a suitable range of
West
45º H 80 77 75 50 Chroma 100 lightnesses and saturations available for each hue. HVC,
Orange
72
60
65
57
Yxy, and RGB specifications for the 25 slope–aspect
70 65 50
classes developed by Brewer and Marlow are shown in
67 47
st 57
Eas
t Table 3, and a map created using their color scheme is
we 65 Blu
ut
h 35 e 245º H shown in Color Plate 6.
So
d
Re 50
5º H
So
Wood’s LandSerf
uth ple
ta
Pu
eas
en
h
Although aspect and slope are the most common

Sout
r
Mag
285º H parameters utilized in examining the morphometry of

325º H terrain, those interested in a detailed analysis (e.g.,
Slope (%) geomorphologists) utilize a richer set of parameters.
>40
For instance, two additional parameters are the
20 to 40
profile curvature, which describes the rate of change
of slope, and plan curvature, which describes the rate
5 to 20
of change of aspect. These four parameters can be
0 to 5
used to define the following features: pits, channels,
passes, ridges, peaks, and planar regions. Jo Wood has
FIGURE 17 A plot of HSV colors used by Brewer and developed an intriguing package, LandSerf ( http://
Marlow to depict aspect and slope; see the text for an expla- www.soi.city.ac.uk/~jwo/landserf/ ), that allows map-
nation. (After Brewer and Marlow 1993, 331.) makers to visualize and analyze these parameters and
431
18:03:38.
Visualizing Terrain
TABLE 3 HVC, Yxy, and RGB specifications for Brewer and Marlow’s method for representing aspect and slope* (After
Brewer and Marlow 1993. First published in Auto-Carto 11 proceedings. Reprinted with permission from the American
Congress on Surveying and Mapping.)
Aspect H V C Y x y R G B
Maximum Slopes ( 7 40 percent slope)

0 110 80 56 55.3 .339 .546 132 214 0
45 135 65 48 35.2 .252 .486 0 171 68
90 245 50 46 19.8 .180 .191 0 104 192
135 285 35 51 10.3 .240 .135 108 0 163
180 325 50 60 19.6 .350 .198 202 0 156
225 5 65 63 33.7 .442 .319 255 85 104
270 45 80 48 53.8 .421 .421 255 171 71
315 85 95 58 81.1 .395 .512 244 250 0
Moderate Slopes (20 to 40 percent slope)

0 110 77 37 50.6 .319 .451 141 196 88
45 145 67 33 37.0 .253 .400 61 171 113
90 245 57 31 25.7 .221 .238 80 120 182
135 285 47 34 17.4 .262 .209 119 71 157
180 325 57 40 25.1 .332 .240 192 77 156
225 5 67 42 34.3 .390 .316 231 111 122
270 45 77 32 47.3 .377 .384 226 166 108
315 85 87 39 63.7 .359 .445 214 219 94
Low Slopes (5 to 20 percent slope)

0 110 75 19 45.4 .305 .381 152 181 129
45 155 70 16 39.4 .270 .348 114 168 144
90 245 65 15 32.7 .260 .284 124 142 173
135 285 60 17 27.5 .280 .272 140 117 160
180 325 65 20 32.6 .311 .282 180 123 161
225 5 70 21 37.2 .339 .317 203 139 143
270 45 75 16 41.8 .328 .348 197 165 138
315 85 80 19 49.7 .321 .375 189 191 137
Near-Flat Slopes (0 to 5 percent slope)

None 0 72 0 39.0 .290 .317 161 161 161
* Aspect is measured in degrees clockwise from north, which is 0°.
associated features. Color Plate 7 illustrates some of cells, Color Plate 7D was created using 65 * 65
the capability of LandSerf. In Color Plate 7A, we see windows of 30-meter cells. Obviously, the larger win-
a shaded relief map of the Columbia River area in dow size has generalized the landscape. (The border
Washington State in which the colors depict elevation around the map appears because of the large window
using a green–brown–purple–white scheme. In Color size and the lack of information surrounding the map.)
Plate 7B, we see a slope map (with the steepest slopes We could continue the analysis by examining the char-
shown in yellow and red) overlaid on the shaded relief acteristics of the features (e.g., channels and ridges)
map. The slope map obviously shows quite different associated with this larger scale landscape. Because
information than the shaded relief map shows because space does not permit us to illustrate the full potential
the former identifies the steepest slopes as being near of LandSerf here, we encourage you to explore the
major drainage basins. In Color Plate 7C, we see some capability of this software on your own.
of the key features found in this region: Channels are
shown in blue, ridges in yellow, and planar regions in
gray. At this scale, we cannot see them, but a blowup of 3 OBLIQUE VIEWS
the region would also show peaks (in red) and passes
(in green). Finally, in Color Plate 7D, we see the effect In contrast to the preceding section, which focused on
of broadening the scale of analysis. In contrast to the vertical views, several approaches for depicting terrain
images shown in Color Plates 7A–C, which were cre- have focused on oblique views. Here we consider three:
ated by processing 3 * 3 windows of 30-meter DEM block diagrams, panoramas, and “draped images.”
18:03:38.
Visualizing Terrain
FIGURE 18 Example of a traditional block diagram. (From Figure 98, p. 69, of Lobeck, A. K. (1958) Block Diagrams. 2nd ed.
Amherst, MA: Emerson-Trussel Book Company.)
3.1 Block Diagrams layers. Color Plate 5.1 is an example of an image that
could result from this process. A major advantage of
In a fashion similar to the physiographic method automating block diagrams is that users are not restricted
described in the preceding section, block diagrams to a single view—programs such as EarthVision (devel-
(Figure 18) are useful for examining the Earth’s oped by Dynamic Graphics; http:/ /www.dgi.com) allow
geomorphologic and geologic features, but with an users to rotate the diagram to different positions and
oblique rather than a vertical view. Although block to explore the data by, for example, slicing the diagram at
diagrams have their origins in the eighteenth century different locations.
(Robinson et al. 1995), A. K. Lobeck (1958) is commonly
considered to have originated them because of his book 3.2 Panoramas and Related Oblique Views
Block Diagrams. Lobeck indicated that block diagrams
have a twofold function: “First, to present pictorially the Panoramas (also known as bird’s-eye views) provide
surface features of the ground; and second, to represent a view of the landscape that we might expect from
the underground structure” (p. 1). Obviously, a purely a painting (which is how traditional panoramas were
vertical view could not achieve this function. Lobeck created), but they also pay careful attention to geogra-
recognized that cartographers might be concerned by phy so that we can clearly recognize known fea-
the lack of accuracy in the block diagram (because cer- tures. Panoramas were especially popular in the late
tain features are purposely hidden, whereas others are nineteenth and early twentieth centuries. Tom Patterson
exaggerated), but he felt that any inaccuracies were out- (2005, 61) described this as the golden era of birds-eye
weighed by the ease of understanding: “No conventions views in the United States, noting that “As towns and
are needed to represent the topography. No explanation cities grew and became prosperous thanks to industriali-
is needed to indicate the position of the geological cross- zation, it became fashionable and a matter of civic
section. It carries its message directly to the eye and pride to advertise this newfound economic vitality
leaves a visual impression unencumbered by lengthy in the form of oblique panoramic maps.” Many of
descriptions” (p. 2). these are available from the U.S. Library of Congress
Like the physiographic method, a full automation of (http://memory.loc.gov/ammem/pmhtml/panhome.html);
methods for creating block diagrams would be difficult we have illustrated one of these for Anniston, Alabama,
because of the extensive knowledge base required and in Color Plate 8.
the subjective interpretation involved. Many parts of the One of the best-known panoramists was Heinrich
process, however, have been automated. For instance, if Berann, whose work Patterson (2000) so ably described.
sample points indicating the elevation of sedimentary Color Plate 9, a classic panorama of Denali National Park
rock layers are known, it is possible to interpolate in Alaska, is an example of the fine work of Berann.
between these points to estimate the bounds of individual Traditionally, creating panoramas such as Berann’s
433
18:03:38.
Visualizing Terrain
FIGURE 19 An image of Yello-

wstone Lake, the largest mountain
lake in North America. This shows
the nature of attractive oblique
3-D images that can be created using
modern digital tools. (Courtesy of
the National Park Service.)
required considerable artistic skill, but Patterson (1999) 4. Texturize by adding “graphical noise” to areas, and
showed how an array of digital tools can allow those with even lines. He argued that this provides “a tactile
limited artistic ability to create panoramas and other appearance that mimics nature.”
oblique views of the 3-D landscape. Figure 19 is an exam-
Color Plate 10 shows a map of Crater Lake National
ple of one of the intriguing images that Patterson created.
Park that Patterson created using such rules. Because
Patterson argued:
space does not permit us to provide details of the
approaches that Patterson used to implement these rules,
Humans naturally tend to visualize physical landscapes in
profile based on our grounded lives on the Earth’s surface, we encourage you to examine his Web site (http://www.
rather than as flat maps. . . . [Thus] an obliquely viewed 3D shadedrelief.com/), where he discusses these approaches.
landscape is . . . probably more understandable to users,
especially those with limited map reading skills. (p. 217) 3.3 Draped Images
In his more recent work, Patterson (2005) continues to In recent years, the most popular form of oblique view
promote the potential of digital panoramic views, but has been draping remotely sensed images (or other
he also stresses that such views frequently require a information, such as land use and land cover) on a 3-D
considerable amount of effort and a range of talents. map of elevation (this is essentially a fishnet map, but
Patterson (2002) has also suggested that the realism we can’t see the fishnet once the image is draped on the
inherent in oblique 3-D landscapes should be employed map of elevation); the resulting maps are termed
in maps that have a vertical view. If this is done, he draped images. The popularity of this approach is indi-
argues, “Users can comprehend . . . map information cated by the fact that several hundred software pack-
relatively effortlessly without explanation (such as text ages have been developed for 3-D visualization of
and legends)” and that more attractive maps will result, terrain (see http://www.tec.army.mil/research/products/
which is important in our “media-driven age.” Patterson TD/tvd/survey/).* As with modern forms of block dia-
suggested four rules for creating realistic topographic grams, the problem of blockage is generally handled by
maps having a vertical view: either producing an animated fly-by of the landscape or
allowing users to rotate the draped image. Depending
1. Remove lines because they are rare in nature and on the nature of the draped information, the resulting
can be visually distracting. maps can appear quite realistic; for instance, with
2. Rasterize by converting vector lines and fills to a 1-meter satellite imagery, it is possible to identify indi-
pixel form. Again, this will make the map appear vidual cars. Given the realism of draped images, a full
more natural.
3. Modulate tones rather than use flat area tones. How *We have no way of knowing how many of these software packages
often do we see a lake that is perfectly uniform? permit draping of images, but we presume that many of them do.
18:03:38.
Visualizing Terrain
discussion of them leads us into the area of virtual SUMMARY

environments (VEs).
This chapter has examined methods for symbolizing
4 PHYSICAL MODELS the Earth’s terrain and associated landscape features.
Basic data for depicting terrain are generally
In the preceding two sections, we covered approaches for provided in an equally spaced gridded format (such as
depicting terrain that provide the illusion of a third a digital elevation model, or DEM). DEMs and related
dimension even though one does not actually exist. The data representing the Earth’s landscape can be
associated maps were created either on a piece of acquired from mapping agencies such as the USGS.
paper or on a graphics display screen, both of which are We divided the methods for symbolizing terrain into
two-dimensional.A natural alternative is to create a truly vertical views, oblique views, and physical models.
3-D physical model.Although physical models have been Methods that have emphasized a vertical view include
around since the time of Alexander the Great, cartogra- hachures (drawing a series of parallel lines perpendicular
phers have used them infrequently because of the diffi- to the direction of contours), contour-based methods
culty of obtaining terrain information, the cost and time (e.g., Tanaka’s method), and shaded relief (or hill
of creating them manually, and the difficulty of disposing shading). These methods assume that the topographic
of them. Today, such limitations no longer exist, as digital surface is illuminated, typically obliquely from the north-
data are readily available, several technologies have been west. The various methods use different approaches to
developed for converting the digital data into a solid distinguish areas that appear illuminated from those that
model, and disposal is less of an issue, given the range of appear in the shadows. For example, in Tanaka’s method,
available technologies (Caldwell 2001). contour lines perpendicular to the light source have the
To illustrate the capability of physical models, we’ll greatest width, whereas those parallel to the light source
consider various models that have been created by Solid are narrowest; furthermore, contour lines facing the light
Terrain Modeling (STM; http://www.stm-usa.com/). For source are drawn in white, whereas those in shadow are
instance, STM has created 17 models for the National drawn in black. Today, vertical views of terrain are most
Geographic Explorer’s Hall in Washington, DC, with the commonly created using computer-based shaded relief,
largest (the Grand Canyon) measuring 52 * 6 feet. but recent software developed by Kennelly and
Based on DEM data, a machine cuts the basic structure Kimerling might revive interest in other methods for
of the model out of dense polyurethane foam; colorful depicting a vertical view. We also considered morphome-
satellite images (or any other image) are then printed on tric methods as a special form of vertical view that empha-
the foam model. Color Plate 11 shows two examples of sizes the structural elements of terrain. For instance,
models that have been created for Explorer’s Hall. The Moellering and Kimerling developed an approach for
Mt. Everest example (on the left) illustrates the dramatic symbolizing aspect based on the opponent-process theory
3-D character of the models, and the Hurricane Floyd of color, and Wood has developed the package LandSerf
example (on the right) illustrates the colorful images that to visualize a variety of topographic features including
can be depicted. Recently, STM has created an even pits, channels, passes, ridges, peaks, and planar regions.
larger model (40 * 72 feet) of the province of British Approaches for creating oblique views include block
Columbia for the historic Crystal Garden in Victoria, diagrams, panoramas, and draped images. Software for
Bristish Columbia (http://www.stm-usa.com/bc.htm). In creating block diagrams (e.g., EarthVision) has been
addition to showing what this region looks like from available for quite some time, whereas utilizing comput-
space, this model also accurately depicts the curvature of ers to create panoramas is a more recent phenomenon.
the Earth, resulting in viewers at different levels seeing Particularly interesting is the work of Patterson, who has
different views of the model. created attractive oblique views in the spirit of the famed
There appears to be considerable excitement about panoramist Berann. Patterson also has illustrated how the
the potential of physical models. Although one cannot realism normally depicted in oblique views can be utilized
interact with these models as one could with an image on in vertical views. Draping remotely sensed images (or
a graphics display (e.g., pointing to a location to deter- other information, such as land use and land cover) on a
mine its elevation), the ability to hold the model and 3-D map of elevation is a common approach for creating
manipulate it in one’s hands (assuming a small-scale oblique views, and hundreds of software packages have
version) appears to be a very desirable feature. An inter- been developed that provide 3-D visualizations of terrain.
esting research project would be to compare the effec- Physical models are an attractive possibility for
tiveness of physical models with 3-D visualizations depicting terrain because the models are truly 3-D.
created on graphics displays. Such studies might help us Thus, users can work with these models as they would
build more effective 3-D visualizations for graphic displays. other 3-D objects in the real world; for instance, you can
435
18:03:38.
Visualizing Terrain
hold a small physical model in your hands and easily convert digital data into solid models. A good indication
view it from different angles. Physical models have been of the excitement these models have generated is the
around since the time of Alexander the Great, but only large number of models that National Geographic has
in the last few years has technology become available to portrayed in their Explorer’s Hall in Washington, DC.
FURTHER READING
Ambroziak, B. M., Ambroziak, J. R., and Bradbury, R. (1999) Presents an algorithm for shaded relief that is appropriate for ele-
Infinite Perspectives:Two Thousand Years of Three-Dimensional vation data stored as contour lines, and illustrates how multiple
Mapmaking. New York: Princeton Architectural Press. light sources can be used to enhance the visualization of terrain;
also see Hobbs (1999).
Provides maps illustrating much of the history of symbolizing
terrain and proposes a novel approach based on anaglyphs. Hodgson, M. E. (1998) “Comparison of angles from surface
slope/aspect algorithms.” Cartography and Geographic
Bernhard, J., and Hurni, L. (2006) “Swiss-style colour relief
Information Systems 25, no. 3:173–185.
shading modulated by elevation and by exposure to illumina-
tion.” The Cartographic Journal 43, no. 3: 198–207. Compares several algorithms for computing the slope and
aspect of DEM cells; also see Jones (1998) and Dunn and
Reviews the historic development of the Swiss-style color-modu- Hickey (1998).
lated shaded relief and describes a computer-based implementation.
Horn, B. K. P. (1982) “Hill shading and the reflectance map.”
Caldwell, D. R. (2001) “Physical terrain modeling for Geo-Processing 2, no. 1:65–144.
geographic visualization.” Cartographic Perspectives,
no. 38:66–72. A dated, but thorough, review of automated approaches for
creating shaded relief.
Reviews modern approaches for creating physical models of the
Earth’s landscape and discusses several related cartographic Hurni, L., Jenny, B., Dahinden, T., and Hutzler, E. (2001)
issues. “Interactive analytical shading and cliff drawing: Advances
in digital relief presentation for topographic mountain
Cartographic Perspectives no. 42, 2002.
maps.” Proceedings, 20th International Cartographic Con-
Several papers in this special issue on “Practical Cartography” ference, International Cartographic Association, Beijing,
deal with symbolizing terrain. China, CD-ROM.
Ding, Y., and Densham, P. J. (1994) “A loosely synchronous, Describes software that allows users to interact with a shaded
parallel algorithm for hill shading digital elevation models.” relief map, and thus produce a result that simulates manual relief
Cartography and Geographic Information Systems 21, no. 1:5–14. shading; also describes an approach for digital cliff drawing.
Describes a sophisticated method for creating shaded relief that Imhof, E. (1982) Cartographic Relief Presentation. Berlin:
incorporates both shadows cast by other terrain features and Walter de Gruyter.
atmospheric scattering; the method was implemented on a parallel
computer. A classic text on methods for symbolizing terrain.
Eyton, J. R. (1991) “Rate-of-change maps.” Cartography and Patterson, T. (2002) “Getting real: Reflecting on the new look
Geographic Information Systems 18, no. 2:87–103. of National Park Service maps.” Accessed June 21, 2003. http://
www.shadedrelief.com/realism/index.html.
Includes basic calculations for creating a shaded relief map, but
also several more advanced concepts such as computing and Discusses digital approaches for creating realistic topographic
mapping the convexity and concavity of a 3-D surface. maps when a vertical view is used; much of Patterson’s work can
be found at http://www.shadedrelief.com.
Foley, J. D., van Dam, A., Feiner, S. K., and Hughes, J. F. (1996)
Computer Graphics: Principles and Practice. 2nd ed. Reading, Raper, J. (2000) Multidimensional Geographic Information
MA: Addison-Wesley. Science. London: Taylor & Francis.
The notion of trying to simulate the appearance of terrain by using An advanced treatment of methods for mapping true 3-D data.
a presumed light source is analogous to attempts to simulate the Thelin, G. P., and Pike, R. J. (1991) “Landforms of the conter-
appearance of other phenomena, such as how a room looks when minous United States—A digital shaded-relief portrayal.”
lit from a certain direction. Chapter 16 deals with simulating the Map I-2206. Washington, DC: U.S. Geological Survey. Scale
appearance of these other phenomena. 1:3,500,000.
Heller, M., and Neumann, A. (2001) “Inner-mountain cartog- Describes the methods used to create Thelin and Pike’s shaded
raphy—from surveying towards information systems.” relief map, along with an interpretation of physiographic features
Proceedings, 20th International Cartographic Conference , seen in the map; also see Pike and Thelin (1990–91; 1992).
International Cartographic Association, Beijing, China, Treinish, L. A. (1994) “Visualizations illuminate disparate
CD-ROM. data sets in the earth sciences.” Computers in Physics 8,
Describes software for visualizing caves; also see Gong et al. no. 6:664–671.
(2000).
Discusses the creation of the dramatic rendition of the Earth’s
Hobbs, F. (1995) “The rendering of relief images from digital terrain shown in Color Plate 4, along with several other
contour data.” The Cartographic Journal 32, no. 2:111–116. visualizations of spatial data.
18:03:38.
Visualizing Terrain
GLOSSARY
aspect: (1) placement of a projection’s center with respect to illuminated contours: a raster approach for depicting topog-
the Earth’s surface; common aspects include equatorial, polar, raphy in which contours facing a light source are brightened,
and oblique; (2) the direction in which a topographic slope whereas those in the shadow are darkened.
faces (e.g., a northwest aspect would face the northwest). luminance: brightness of a color as measured by a physical
bird’s-eye view: see panorama. device.
block diagram: a technique developed by A. K. Lobeck for panorama: provides a view of the landscape that we might
depicting the three-dimensional structure of geomorphologic expect from a painting (which is how traditional panoramas
and geologic features. were created), but careful attention is given to geography so
chiaroscuro: see shaded relief. that we can clearly recognize known features.
chorodot map: a combination of choropleth and dot maps; physical model: a truly 3-D model of the Earth’s landscape
the “dot” portion derives from using small squares within (i.e., we can hold and manipulate small models in our hands).
enumeration units, whereas the “choropleth” portion derives physiographic method: a method developed by Raisz for
from the shading assigned to each square. depicting topography in which the Earth’s geomorphological
colorimeter: a physical device for measuring color; typically, features are represented by standard, easily recognized symbols.
color specifications are given in the CIE system. shaded relief: a method for depicting topography; areas fac-
command-line interface: a computer interface in which the ing away from a light source are shaded, whereas areas
user types commands specifying processes to be performed. directly illuminated are not; slope might also be a factor, with
steeper slopes shaded darker.
draped image: a remotely sensed image (or other informa-
tion; e.g., land use and land cover) is draped on a 3-D map of shadowed contours: a method for depicting topography;
elevation. contour lines facing a light source are drawn in a normal line
weight, whereas those in the shadow are thickened.
Gouraud shading: a sophisticated method for creating
shaded relief; shades are interpolated between vertices of a slope: the steepness of a topographic surface.
polygon (or cell) boundary, and thus a smoother appearance Tanaka’s method: a method for depicting topography in
is produced than when the same shade is used throughout a which the width of contour lines is varied as a function of their
polygon. angular relationship with a light source; also, contour lines
hachures: a method for depicting topography in which a facing the light source are white, whereas those in shadow are
series of parallel lines is drawn perpendicular to the direction black.
of contours. uniform sky illumination model: a method for creating
hill shading: see shaded relief. shaded relief in which the light source is presumed to be
evenly distributed throughout the sky.
REFERENCES
Brewer, C. A., and Marlow, K. A. (1993) “Color representa- Foley, J. D., van Dam, A., Feiner, S. K., and Hughes, J. F.
tion of aspect and slope simultaneously.” Auto-Carto 11 (1996) Computer Graphics: Principles and Practice (2nd
Proceedings, Minneapolis, MN, pp. 328–337. ed., C ed.). Reading, MA: Addison-Wesley.
Caldwell, D. R. (2001) “Physical terrain modeling for Harrison, R. E. (1969) Shaded Relief, Map. United States
geographic visualization.” Cartographic Perspectives, no. Department of the Interior, Geological Survey. In The
38:66–72. National Atlas of the United States of America, 1970, pp.
Chrisman, N. (2002) Exploring Geographic Information 56–57. (Scale 1:7,500,000.)
Systems (2nd ed.). New York: Wiley. Hügli, H. (1979) “Vom Geländemodell zum Geländebild. Die
Dunn, M., and Hickey, R. (1998) “The effect of slope Synthese von Schattenbildern, Vermessung,
algorithms on slope estimates within a GIS.” Cartography Photogrammetrie.” Kulturtechnik 77:245–249.
27, no. 1:9–15. Hurni, L., Jenny, B., Dahinden, T., and Hutzler, E. (2001)
Eyton, J. R. (1984b) “Raster contouring.” Geo-Processing “Interactive analytical shading and cliff drawing: Advances
2:221–242. in digital relief presentation for topographic mountain
maps.” Proceedings, 20th International Cartographic
Eyton, J. R. (1990) “Color stereoscopic effect cartography.”
Conference, International Cartographic Association,
Beijing, China, CD-ROM.
Eyton, J. R. (1991) “Rate-of-change maps.” Cartography and
Hobbs, F. (1995) “The rendering of relief images from digital
contour data.” The Cartographic Journal 32, no. 2:111–116.
437
18:03:38.
Visualizing Terrain
Hobbs, K. F. (1999) “An investigation of RGB multi-band Patterson, T. (2000) “A view from on high: Heinrich Berann’s
shading for relief visualization.” International Journal of panoramas and landscape visualization techniques for the
Applied Earth Observation and Geoinformation 1, U.S. National Park Service.” Cartographic Perspectives,
no. 3–4:181–186. no. 36:38–65.
Hodgson, M. E. (1998) “Comparison of angles from surface Patterson, T. (2002) “Getting real: Reflecting on the new
slope/aspect algorithms.” Cartography and Geographic look of National Park Service maps.” http://www.shade-
Information Systems 25, no. 3:173–185. drelief.com/realism/.
Imhof, E. (1975) “Positioning names on maps.” The American Patterson, T. (2005) “Looking closer: A guide to making
Cartographer 2, no. 2:128–144. bird’s-eye views of National Park Service cultural and
Horn, B. K. P. (1982) “Hill shading and the reflectance map.” historical sites.” Cartographic Perspectives, no. 52:59–75,
Geo-Processing 2, no. 1:65–144. 96–106.
Jones, K. H. (1998) “A comparison of algorithms used to Price, W. (2001) “Relief presentation: Manual airbrushing
compute hill slope as a property of the DEM.” Computers combined with computer technology.” The Cartographic
& Geosciences 24, no. 4:315–323. Journal 38, no. 1:107–112.
Kennelly, P., and Kimerling, A. J. (2000) “Desktop hachure Raisz, E. (1931) “The physiographic method of representing
maps from digital elevation models.” Cartographic scenery on maps.” The Geographical Review 21,
Perspectives, no. 37:78–81. no. 2:297–304.
Kennelly, P., and Kimerling, A. J. (2001) “Modifications of Raisz, E. (1967) Landforms of the United States. Map. (Scale
Tanaka’s illuminated contour method.” Cartography and approximately 1:4,500,000.)
Geographic Information Science 28, no. 2:111–123. Robinson, A. H., Morrison, J. L., Muehrcke, P. C., Kimerling,
Kennelly, P. J., and Stewart, A. J. (2006) “A uniform sky A. J., and Guptill, S. C. (1995) Elements of Cartography
illumination model to enhance shading of terrain and (6th ed.). New York: Wiley.
urban areas.” Cartography and Geographic Information Rogers, D. F. (1998) Procedural Elements for Computer
Science 33, no. 1: 21–36. Graphics. Boston: McGraw-Hill.
Lewis, P. (1992) “Introducing a cartographic masterpiece: A Tanaka, K. (1950) “The relief contour method of represent-
review of the U.S. Geological Survey’s digital terrain map ing topography on maps.” The Geographical Review 40, no.
of the United States.” Annals of the Association of 3:444–456.
American Geographers 82, no. 2:289–304. Treinish, L. A. (1994) “Visualizations illuminate disparate
Lo, C. P., and Yeung, A. K. W. (2007) Concepts and data sets in the earth sciences.” Computers in Physics 8, no.
Techniques of Geographic Information Systems (2nd ed.). 6:664–671.
Upper Saddle River, NJ: Pearson Prentice Hall. Yoeli, P. (1967) “The mechanization of analytical hill
Lobeck, A. K. (1924) Block Diagrams. New York: Wiley. shading.” The Cartographic Journal 4, no. 2:82–88.
Lobeck, A. K. (1958) Block Diagrams (2nd ed.). Amherst, Yoeli, P. (1971) “An experimental electronic system for
MA: Emerson-Trussell. converting contours into hill shaded relief.” International
Moellering, H., and Kimerling, A. J. (1990) “A new digital Yearbook of Cartography 11:111–114.
slope-aspect display process.” Cartography and Yoeli, P. (1983) “Shadowed contours with computer and
Geographic Information Systems 17, no. 2:151–159. plotter.” The American Cartographer 10, no. 2:101–110.
Patterson, T. (1999) “Designing 3D landscapes.” In Multimedia Yoeli, P. (1985) “Topographic relief depiction by hachures
Cartography, ed. by W. Cartwright, M. P. Peterson, and G. with computer and plotter.” The Cartographic Journal 22,
Gartner, pp. 217–229. Berlin: Springer-Verlag. no. 2:111–124.
18:03:38.
Map Animation
OVERVIEW
In this chapter, we focus on animated maps, or maps as a “movie.” Although animations shown in a computer
characterized by continuous change while the map is environment often allow users to start and stop them,
viewed. You are probably most familiar with animated jump to an arbitrary frame, and play them forward and
maps from watching the weather segment of the nightly backward, we still classify them as having “little” interac-
news or The Weather Channel, but we will see that tion. In contrast, some animations permit substantial
virtually any form of thematic symbology can be ani- interaction, such as allowing the user to specify subsets of
mated (albeit some more effectively than others). information to be displayed, linking the map with graph-
Although animation commonly is associated with ical representations, and allowing the user to interact with
recent technological advances, in section 1 we will see the display to acquire information.
that the earliest map animations were produced in the Normally, animations are displayed in 2-D space, whereas
1930s, and that cartographers were discussing the the third dimension (the z axis) is used to display something
potential of animation as early as the 1950s. that is three-dimensional (e.g., a mountain or a cloud).
In section 2, we consider visual variables for anima- Section 4 considers the possibility of instead using the z axis
tion and ways of categorizing animations. Visual vari- to depict time. This approach provides an alternative to ani-
ables refer to the perceived differences in map symbols mation that may allow us to detect information that cannot
that are used to represent spatial data; examples include be seen or that is unlikely to be seen in an animation. Since
spacing, size, orientation, and shape. Animated maps can symbols in 3-D space frequently block one another, this
utilize these same visual variables, but they also use oth- approach requires that users interact with the display.
ers, including duration, rate of change, order, display As we discuss the animation examples, one issue that
date, frequency, and synchronization. Animations can we raise is the difficulty of understanding many anima-
be split into categories according to whether the animations. We highlight this issue in section 5 by briefly con-
tion emphasizes (1) a change in either position or sidering some of the studies that have been done to
attribute, (2) the location of some phenomenon, or (3) evaluate the effectiveness of animation. Although those
the spatial distribution of an attribute. who have developed animations often seem enamored
Section 3 covers numerous examples of map anima- by them, studies to date do not clearly indicate the
tions, focusing on those that have been discussed in the effectiveness of animation. We suspect that the effec-
refereed academic literature. For convenience, we split tiveness of animation ultimately will depend on numer-
these examples into two groups: those having “little or no ous factors, including the symbology used, the level of
interaction” and those “characterized by substantial inter- interaction that the user is permitted to have with the
action.” Animations shown on television or in a movie animation, whether the user has expertise in the
theater represent the extreme case of “no interaction,” as domain being mapped, and the user’s past experience
the user has no control and thus must view the animation with animation.
439
11:52:56.
Map Animation
In recent years, numerous animations have been created 2 VISUAL VARIABLES AND CATEGORIES
that are not discussed in the refereed literature—we pro- OF ANIMATION
vide links to these on the home page for this text (www.
prenhall.com/slocum). Animations have also been devel- In 1991, David DiBiase and his colleagues created a
oped in association with data exploration software and for videotape in which they formulated visual variables
depicting data uncertainty. unique to animation and categorized various types of
animation. This videotape was supplemented by an arti-
1 EARLY DEVELOPMENTS cle (DiBiase et al. 1992), and the visual variables were
later expanded on by Alan MacEachren (1995).
Although map animation is often viewed as a recent
development, the notion has been around since the 2.1 Visual Variables for Animation
1930s, and was promoted by cartographers as early as
the 1950s, most notably by Norman Thrower (1959; In the videotape, DiBiase and his colleagues described
1961), who clearly was enamored by its potential: how the visual variables for static maps could be
applied to animated maps. For example, they animated
By the use of animated cartography we are able to create a choropleth map of New Mexico population from 1930
the impression of continuous change and thereby to 1980 to show how the visual variable lightness (rep-
approach the ideal in historical geography, where phe- resented by gray tones) could represent changes over
nomena appear “as dynamic rather than static entities.” time, and they transformed the state of Wisconsin from
Distributions which seem to be particularly well suited its real-world shape to that of a cow to illustrate how
to animation include the spread of populations, the the visual variable shape could be used in animation.
development of lines of transportation, the removal of
More important, however, they showed how animation
forests, changing political boundaries, the expansion of
requires additional visual variables beyond those used
urban areas, and seasonal climatic patterns. (1959, 10)
for static maps, namely, duration, rate of change, and
Because computers were only beginning to be developed order.
in the 1950s, animations at that time were created using In its simplest form, duration is defined as the length of
hand drawings (“cels”) and a camera to record each time that a frame of an animation is displayed. More gen-
frame of the animation. erally, a group of frames can be considered a scene (in
Early computer-based animations were exemplified which there is no change between frames), and we can
by Waldo Tobler’s (1970) 3-D portrayal of a popula- consider the duration of that scene.* Because duration is
tion growth model for the city of Detroit, Michigan, measured in quantitative units (time), it can be logically
and Hal Moellering’s (1976) representation of the used to represent quantitative data; for example, DiBiase
spatiotemporal pattern of traffic accidents. Mark and his colleagues animated electoral college votes for
Harrower and his colleagues (2000, 281) argued that president, varying the duration of each scene in direct
work such as Moellering’s illustrated the “power of proportion to the magnitude of the victory in each
map animation,” as “the result was a clear representa- election.
tion of the daily cycle of accidents, with peaks during Rate of change is defined as m/d, where m is the
rush hour periods and troughs between those times, magnitude of change (in position and attributes of enti-
as well as the spatial pattern of weekday versus week- ties) between frames or scenes, and d is the duration of
end accidents.” Although Moellering continued to each frame or scene. For example, Figure 1 illustrates
work with animation, there were few other efforts in different rates of change for the position of a point fea-
cartographic animation prior to the 1990s. Craig ture (A) and the attribute of circle size (B). The
Campbell and Stephen Egbert (1990, 24–25) argued smoothness (or lack thereof) in an animation is a func-
that this was due to the lack of funds in geography tion of the rate of change. Either decreasing the mag-
departments, the lack of a “push toward the leading nitude of change between frames or increasing the
edges of technology” in these departments, a negative duration of each frame will decrease the rate of change
view of academics who develop and market software, and hence make the animation appear smoother.
and the difficulty of disseminating animations in stan- Order is the sequence in which frames or scenes are
dard print outlets. In the following sections, we will presented. Normally, of course, animations are pre-
see that many of these limitations have been over- sented in chronological order, but DiBiase et al. argued
come. that knowledge also can be gleaned by reordering
frames or scenes. For example, they modified not only
* DiBiase et al. borrowed the term scene from the work of Szego (1987).
11:52:56.
Map Animation
animation emphasizes (1) change in either position or an

A Frame 1 Frame 2 Frame 3 Frame 4 Frame 5 attribute, (2) the location of some phenomenon, or
Slow (3) the spatial distribution of an attribute.
Rate of
Change
Animations Emphasizing Change
For animations emphasizing change, DiBiase et al.
Fast
Rate of focused on changes in either position or attribute. As
Change
an example of change in position, we can envision a
moving point representing the mean center of the U.S.
B Frame 1 Frame 2 Frame 3 Frame 4 Frame 5
population for each decennial census. A change in an
attribute would be represented by the animated choro-
Slow
Rate of pleth map of New Mexico population mentioned
Change
previously. More recently, Mark Harrower (2001)
expanded the notion of animations emphasizing
Fast change, and suggested that we also consider change
Rate of
Change
in the shape, size, or extent of a phenomenon; for
instance, we might animate changes in the shape and
areal extent of a hurricane over time.
FIGURE 1 The visual variable rate of change is illustrated DiBiase et al. suggested that animations emphasizing
for (A) geographic position and (B) circle size. Rate of change change be further divided into three types: time series,
is defined as m/d, where m is the magnitude of change re-expressions, and fly-bys.A time series, or an animation
between frames or scenes, and d is the duration of each frame that emphasizes change through time, is most common.
or scene. For these cases, duration is presumed to be constant DiBiase et al. illustrated several time series, including a
in each frame. (After DiBiase et al. 1992, 205.) cartogram of world population growth and the spread of
AIDS in Pennsylvania. Borrowing from the work of
Tukey (1977), DiBiase et al. defined re-expression as an
the duration of frames associated with the presidential “alternative graphic representation . . . whose structure
election data, but also reordered the frames such that has been changed through some transformation of the
lower magnitudes of victory were presented first. original data” (p. 209). Re-expression might involve
Alan MacEachren (1995) extended the visual variables choosing subsets of a time series (brushing), reordering a
for animation to include display date, frequency, and syn- time series, or changing the duration of individual frames
chronization. He defined display date as the time some within a time series. To illustrate brushing, DiBiase et al.
display change is initiated. For example, in an animated animated the longest sequence of victories by candidates
map of decennial population for U.S. cities, we might note of a single party in U.S. presidential elections (the
that a circle first appears for San Francisco in 1850. Democratic Party won all elections from 1932 to 1948).
Frequency is the number of identifiable states per unit To illustrate reordering a time series and changing the
time, or what MacEachren called temporal texture. To duration of frames, they animated five global-climate-
illustrate frequency, he noted its effect on color cycling, in model predictions for Puebla, Mexico. In reordering the
which various colors are “cycled” through a symbol, as is temperature predictions from those that varied the least
often done on weather maps to show the flow of the jet to those that varied the most, and then pacing them such
stream. Synchronization deals with the temporal corre- that those with the greatest variation had the greatest
spondence of two or more time series. MacEachren duration, they found “a trend of maximum uncertainty
(pp. 285–286) indicated that if the “peaks and troughs of . . . during the spring planting season of April, May, and June,
[two] time series correspond, the series are said to be ‘in a pattern that had previously gone unnoticed” (p. 211).
phase,”’ or synchronized. For example, if we examine ani- In a fly-by (or fly-over), the user is given the feel-
mations of precipitation and greenness, we would expect ing of flying over a 3-D surface. A classic example was
them to be out of sync because greenness might not the video L.A. The Movie, which illustrated a high-
immediately respond to precipitation changes. speed flight over Los Angeles, California, and vicinity.
Traditionally, creating fly-bys was beyond the scope of
2.2 Categories of Animation most mapmakers because of the considerable com-
puting time required (L.A. The Movie required 130
In addition to introducing visual variables for animation, hours of CPU time on a VAX 8600 mainframe).
DiBiase and his colleagues developed a useful catego- Today, this is no longer the case because map-
rization of animated maps as a function of whether the makers now can create their own fly-bys on desktop
441
11:52:56.
Map Animation
computers. A wide range of software is available for directional tick marks (or a floating compass), and
this purpose. In addition to empowering mapmakers heads-up display text. Color Plate 1 illustrates some of
to create their own fly-bys, the most sophisticated these solutions. In more recent work, Harrower and
hardware and software also now enable users to con- Sheesley (2007) tested the effectiveness of variations
trol the flight path while the surface is being viewed, of some of these approaches with users, finding that a
as opposed to having the flight path governed by the horizon compass and monorail were effective orienta-
mapmaker, which was the case with L.A. The Movie. tion cues, and that a grid was effective for directional
Although fly-bys can produce eye-catching graphics, tasks (p. 27). The monorail was similar in concept to
Mark Harrower and Ben Sheesley (2005) noted a num- the spotlight path, but served as a “physical represen-
ber of problems with them, including their oblique per- tation of the actual flight path” (personal communica-
spective, their focus on realism and thus information tion, Mark Harrower, 2007).
overload, visual occlusion, and user disorientation. The
oblique perspective of fly-bys is problematic because it Animations Emphasizing Location
causes the scale to change across the image, making it
Although a static map can be used to indicate loca-
difficult for viewers to estimate relative sizes, distances,
tion, animation can assist in emphasizing location.
and directions. One solution to this problem suggested
For example, just the topmost map in Figure 2 could
by Harrower and Sheesley is to superimpose a grid on
be used to indicate the location of major earth-
the scene. An example of this can be found in
quakes. Alternatively, one could use flashing-point
Wilhelmson et al.’s depiction of thunderstorm develop-
symbols (as suggested by the set of three maps in
ment discussed in section 3.1. Although developers of
Figure 2) to emphasize the location of the earth-
fly-bys have focused on creating realistic flights (espe-
quakes. Such flashing symbols would not necessarily
cially because the power of computers has increased),
help users interpret the distribution, but they would
Harrower and Sheesley argued that abstraction and
attract attention, and for earthquake data, would
generalization are often necessary in cartography. They
serve as a warning signal (“It happened here once, it
stated that “cartographic abstraction in 3-D fly-overs
could happen here again”).
can be advantageous if it permits viewers to mentally
organize and structure the landscape, for example, when
roads are exaggerated, forest boundaries are high- Animations Emphasizing the Spatial Distribution of an
lighted, voice-overs are added, [and] visual hierarchy is Attribute
established. . . .” Animations can also emphasize the spatial distribu-
Visual occlusion in fly-bys is closely associated with tion of an attribute by highlighting selected portions
the oblique perspective, as a low-angle view causes of it. An example would be sequencing, in which a map
portions of the image to be obscured. One solution to is displayed piece by piece. For instance, the program
this problem is to have the flight path circle key fea- ExploreMap can be used to present choropleth map
tures in the landscape as opposed to simply flying by classes in order from low to high. Later in this chapter
them. A second solution is to provide a static plani- we will consider a similar concept utilized by Gershon
metric overview prior to showing the fly-by—the (1992) for isarithmic maps.
notion is that if the viewer can retain this planimetric
overview in short-term memory while viewing the fly-
by, his or her interpretation of the fly-by should be 3 EXAMPLES OF ANIMATIONS
enhanced. A final solution is to provide both the plani-
metric overview and fly-by views simultaneously. Sven In this section we consider a range of animations that
Fuhrmann and Alan MacEachren (2001) have found have been developed by cartographers and other
this approach effective for navigating virtual environ- graphic designers. Because these animations are impos-
ments. sible to depict here (although we approximate a few
User disorientation is often a problem for viewers with small multiples), we encourage you to examine
of fly-bys, in part because viewers generally do not the home page for this text, which provides informa-
have control over the fly-by. Harrower and Sheesley tion on where these animations can be viewed (or
suggested numerous solutions for disorientation, acquired), along with links to numerous other anima-
including the superimposed grid and overview maps tions. We have divided the animations into two
mentioned previously, landmarks and labels, a jet con- categories: those having little or no interaction and
trail (which shows the flight path already completed), those characterized by substantial interaction. The cur-
a spotlight path (which indicates the flight path that is rent trend is toward interactive animations and related
ahead and the flight path that has been completed), data exploration concepts.
11:52:56.
Map Animation
numerical model of a developing thunderstorm via

attractive ribbon-like streamers and buoyant balls
(online versions can be found at http://www.mediaport.
net/CP/CyberScience/BDD/fich_050.en.html). Although
Wilhelmson et al.’s animation was developed primarily
for research purposes, it is apparent that similar anima-
tions could be used to teach students basic concepts
about thunderstorm development and other meteoro-
logical concepts. This animation required the effort of
several researchers knowledgeable about storm devel-
opment and the resources of the Visualization Services
and Development Group at the National Center for
Supercomputing Applications. Development required
11 months: Four scientific animators were involved, as
were scriptwriters, artistic consultants, and postproduc-
tion personnel. Also required was a substantial hard-
ware component: “about 200 [hours] of computer time
on a Silicon Graphics Iris 4D/240 GTX using Wavefront
software and special software written by the visualiza-
tion group” (Wilhelmson et al. 1990, 33). Given today’s
improved software and hardware environment, such ani-
mations would take far less effort now, although sophis-
ticated animations are still likely to require a group
effort and a funding source.
Treinish’s Portrayal of the Ozone Hole

Another striking early animation was Lloyd Treinish’s
(1992) Climatology of Global Stratospheric Ozone,
which depicted the development of the ozone hole over
Antarctica. The online version of this animation can be
viewed at http://www.research.ibm.com/people/l/lloydt/
ozone_video.mpg, and a frame from the animation is
shown in Color Plate 2A.* In viewing the animation,
you will see that levels of ozone (in columns of the
Earth’s atmosphere) are symbolized using three redun-
dant attributes: height above the Earth’s surface (as a
21/2-D surface), a traditional rainbow spectral scheme
(red–orange–yellow–green–blue), and the degree of
opacity. Thus, low ozone values are blue, close to the
Earth, and transparent, whereas high ozone values are
red, far from the Earth, and opaque. The ozone symbol-
FIGURE 2 Using animation to emphasize location. Point ization is wrapped around a global representation of
symbols are flashed off and on to highlight the location of the Earth, which is colored to represent the Earth’s
major earthquakes from 1900 to 1990. (After DiBiase et al. topography (Color Plate 2A). The complete animation
1992. First published in Cartography and Geographic consists of 4,700 frames, one for each day over the
Information Systems 19(4), p. 207. Reprinted with permission period between 1979 and 1991.
from American Congress on Surveying and Mapping.) Although Treinish’s ozone animation is striking, it
also illustrates some of the difficulty of interpreting ani-
mations over which the user has little control. There
3.1 Animations Having Little or No Interaction
appear to be three reasons why this particular anima-
tion is difficult to follow. First, the entire 3-D structure
Wilhelmson et al.’s Depiction of Thunderstorm
(i.e., globe plus ozone symbolization) is frequently
Development
Robert Wilhelmson and his colleagues (1990) developed * For further discussion of the ozone data on which the animation is
one of the more dramatic early animations, depicting a based, see Treinish and Goettsche (1991) and Treinish (1992).
443
11:52:56.
Map Animation
rotated during the animation to show interesting areas maps.unomaha.edu/books/IACart/book.html). For uni-
at certain times of year; as a result, it is not possible to variate data, Peterson advocated animating both the
focus on a specific region over the entire period of classification method (say, mean–standard deviation,
observation. Second, because the satellite that collected equal intervals, quantiles, and natural breaks) and the
the data for the animation functioned on reflected sun- number of classes. For example, rather than displaying
light, the Antarctic winter appears as a gap in the ani- all four classification methods at once (as in Figure 3),
mation; unfortunately, it is difficult to differentiate this MacChoro II displayed them in succession as an anima-
gap from the true ozone hole. Third, in its original tion. For multivariate data, Peterson suggested using
videotape form (Treinish 1992), the user could not eas- animation to examine the geographic trend of related
ily focus on a particular time segment of the animation. attributes (i.e., those measured in the same units
This latter problem has diminished somewhat in the and part of a larger whole), and to compare unrelated
online version, as users can now stop and restart the attributes (i.e., those measured in dissimilar units and
animation and move to an arbitrary frame, but the ani- not part of a larger whole). With respect to related
mation is still difficult to grasp. In spite of these prob- attributes, we might animate the percentage of popula-
lems, Treinish (1992) claimed “the region of Antarctic tion in various age groups (0–4 years old, 5–9 years old,
ozone depletion each spring can be easily seen growing etc.) within census tracts of a city. Peterson argued that
in size and severity over the last decade.” We did not such data “will usually show a clear regionalization in a
see this feature, but this might be due to our lack of city with older populations closer to the center and
familiarity with the ozone data, as DiBiase et al. (1992, younger populations nearer the periphery. An anima-
213) argued that those with expert knowledge about a tion . . . from low to high depicts high values ‘moving’
problem are often more likely to benefit from anima- from the periphery to the center” (pp. 41–42). Two unre-
tions. Those viewing the ozone animation might also lated attributes would be median income and percent of
have had access to small multiples of the data (Color the adult population with a high school education.
Plate 2B), which would make it possible to compare one Peterson argued that an animation of such attributes
year directly with another and thus note that the ozone would detect similarities or differences in the two distri-
hole does appear larger in the last few years of the ani- butions (he recommended switching rapidly between
mation. the attributes).
More recently, Peterson (1999) promoted the idea
Animation of Urban Growth of “active legends,” in which the user controls the
order of the animation by selecting individual legend
Several researchers have developed animations of urban
boxes representing the elements to be animated
growth, including William Acevedo and Penny Masuoka
(Figure 4), thus moving away from a “movie” to
(1997) and Michael Batty and David Howes (1996a;
a more interactive form of animation. Although
1996b). In contrast to the other animations we have dis-
Peterson’s ideas for animating nontemporal data are
cussed, these animations seem relatively simple to fol-
intriguing, we found his animations difficult to follow.
low because of the smooth transition between frames
It seems that a small-multiples approach is more
(Color Plate 3). The challenge does not seem to be in
effective than animation for analyzing such nontem-
understanding the animation, but rather in acquiring the
poral data, primarily because the transitions between
data and creating the frames necessary for the anima-
frames of the animation are not smooth. In fact,
tion. For instance, Acevedo and Masuoka discussed the
Peterson (p. 41) demonstrated his ideas in static form
problem of creating intermediate frames depicting
using small multiples (as in Figure 3).
urban growth—a simple algorithm based on linear dis-
Nahum Gershon (1992) also experimented with non-
tance produced “star-like outward growth along search
temporal animations. One of Gershon’s approaches
directions” (pp. 427–428), and so a more complex algo-
involved an “animation of segmented components” in
rithm based on a cellular automaton model of urban
which users were shown an isarithmic map in a
growth was necessary (Clarke et al. 1997).
sequenced presentation: A high, narrow range of tem-
perature appeared first, followed by a progressively
Peterson’s and Gershon’s Nontemporal Methods wider range, until the entire range to be focused on
The preceding animations that we described all involved appeared; then the process was reversed with progres-
temporal data. In contrast, several researchers have sively narrower ranges of temperature displayed until
experimented with animating nontemporal data. For just the highest narrow range appeared again. Gershon
instance, Michael Peterson’s (1993) MacChoro II soft- indicated that users “perceived the existence of struc-
ware permitted animating nontemporal univariate tures, their locations and shapes better than in the static
and multivariate data via choropleth symbolization display of the data” (p. 270). This result is a bit surpris-
(representative animations can be found at http:// ing, given the lack of success some have found with
11:52:56.
Map Animation
Percent of Births to Mothers Cancer Deaths by County

Nebraska, 1996
Under Twenty
Standard
Deviation Move the cursor over a
Percentage of square to see that map
Total Population
Cerebro-
Above 28.80 Cancer vascular
22.90 - 28.80
18.64 - 22.89 Heart Chronic
Disease Lung
14.82 - 18.63
Below 14.82 Total - Unclassed
Total - Classed
FIGURE 4 Peterson’s (1999) notion of active legends: Users

Equal can control the order of frames within an animation by click-
Interval ing on the desired legend boxes. (Courtesy of Michael P.
Peterson.)
sequencing on choropleth maps. Gershon’s superior

results might be a function of his particular sequencing
strategy; or, possibly, sequencing is more appropriate for
contour than for choropleth maps.
Gershon also had people examine the correlation
between two attributes as he quickly switched between
maps of the attributes. This approach is analogous to
Peterson’s suggestion for animating more than one
Quantile attribute. In contrast to the difficulty we had in under-
standing Peterson’s animations, Gershon argued that
his audience could “visually correlate structures appear-
ing in both [maps]” (p. 272). We suspect that Gershon’s
superior results were due to the similar structure of his
maps: They were both of January sea surface tempera-
tures, but for different years. As a result, users could
easily focus on the differences between the maps.
Lavin et al.’s Animation of Map Design Parameters

Stephen Lavin and his colleagues (1998) presented the
Natural notion of animating map design parameters, focusing on
Breaks
the effect that varying interpolation parameters has on an
isarithmic map.* For instance, one of the parameters
involved in inverse-distance interpolation is the power
FIGURE 3 A small multiple illustrating four classification
(or exponent) to which distances are raised. Those who
methods that Peterson animated. (After Peterson 1993.
First published in Cartography and Geographic Information * Lodha et al. (1996a) and Mitas et al. (1997) also discussed using ani-
Systems 20(1), p. 41. Reprinted with permission from the mation to illustrate the effect that varying interpolation parameters
American Congress on Surveying and Mapping.) has on isarithmic mapping.
445
11:52:56.
Map Animation
have never worked with interpolation are often unsure of

what effect a particular exponent will have. Lavin and his
colleagues argued that we can examine the effect of the
exponent by creating a series of animated maps in which
each map of the series has a slightly different exponent.
Figure 5 illustrates this notion for six frames of a 120- WT=1
frame animation; in the figure, the exponent varies from 1
to 6 in increments of 1, whereas in the actual animation,
the exponent varied from 0.05 to 6, with increments of
0.05. When this was done, Lavin and his colleagues found
that the vertical relief increased as the exponent
increased, but that exponents above 4 produced little
WT=2
change, and that these findings held regardless of control
point search and selection methods. Although Lavin and
his colleagues focused on the inverse-distance parameter,
they stressed that animation could be applied to other
interpolation parameters, and more generally to other
mapping techniques.
WT=3
Goldsberry’s Examination of Rate of Change
Most animations use a fixed duration between map
frames so that map display time is directly propor-
tional to the real-world time being represented. Kirk
Goldsberry (2004) argued that this approach compli-
cates the interpretation of animations because spatial WT=4
data do not necessarily change uniformly over time.
As a solution to this problem, Goldsberry developed
a choropleth map animation of home ownership in
the United States that stabilized the rate of change
over time. This was accomplished using the rate of
change formula presented previously in section 2.1:
m/d, where m is the magnitude of change between WT=5
frames, and d is the duration of each frame. Since the
home ownership data were available at 10-year inter-
vals, Goldsberry computed the magnitude of change
for each 10-year interval by determining how many
states had changed classes during the 10-year period.
For example, in Figure 6 we see home ownership maps
for 1940 and 1950, and a matrix in between them that WT=6
indicates which states remained in the same class (the
diagonal elements) and which states changed classes
(the off-diagonal elements). We see that 25 of the 48
contiguous states changed classes. When the magni-
tude of change (the number of states in the off-diago-
nal elements) was high, Goldsberry increased the FIGURE 5 Lavin et al.’s (1998) notion of using an animated
duration by increasing the number of frames between set of maps to depict the effect of the exponent in inverse-dis-
the decennial census years—the net result was that tance interpolation. The numbers indicate the weight of the
the rate of change was constant for each 10-year inter- exponent. (Courtesy of North American Cartographic
val. Information Society.)
Goldsberry argued that traditional animations
(which have a variable rate of change) answer the fol- In contrast, he argued that stabilized rate of change
lowing sorts of questions: maps answer the following questions:
• What is the overall trend in the attribute over time? • When did the attribute change occur?
• When did the attribute change? • How much did the attribute change during time
• Where did the attribute change? period X?
11:52:56.
Map Animation
(http://maps.unomaha.edu/AnimatedFlightAtlas/).
Home Ownership, 1940-1950 This atlas illustrates a typical 24-hour cycle of air
traffic for weekdays using small plane icons (when
the distance across the display is less than 1,000
1940 miles) and dots (when the distance is greater than
1,000 miles). These plane icons and dots move in a
fashion similar to how a plane travels between cities.
In contrast to some of the animations we have dis-
cussed, this animation of movement seems to be very
effective, as there is a congruence between the real-
world movement of planes and the animation of them
(Tversky et al. 2002). Peterson and Wendel provide a
broad range of options for animation, including all
flights in either the United States, Canada, or the
Caribbean; types of aircraft; major airlines; traffic
1950
associated with particular airports; major flight corri-
Percentage of Ownership 1 34 13 dors (e.g., West Coast–Hawaii); regions (e.g., the
Northeast); and individual states.
28.30 - 45.53 13 1 12 0 Although it is unclear who exactly might benefit
1940
45.54 - 62.76
35 0 22 13 from using this atlas, a number of interesting spatiotem-
62.77 - 80.00 poral patterns arise. For instance, in examining the West
Equal Interval Classification 0 0 0 0 Coast–Hawaii corridor, we saw the following: little traf-
fic between approximately 12 and 2 AM, a distinct
Magnitude of Change = 25
increase in traffic from Hawaii to the mainland at about
2 AM, little traffic between 7 and 10 AM, a distinct
1950 increase in traffic from the mainland to Hawaii around
10 AM, and traffic going both directions between 6 and
12 PM. Some of these patterns we saw ourselves, while
others were prompted by text that is included with each
animation. As another example, the animation of FedEx
traffic shows very distinctive patterns associated with
Memphis, Tennessee, which is the hub for FedEx.
Around 4 PM, there is a distinct trend in outgoing
flights from Memphis—this corresponds to planes leav-
ing “to collect express mail and packages from airports
across the country.” Around 10 PM planes are begin-
ning to return to Memphis, and this trend continues
FIGURE 6 Information that Goldsberry used to compute a
until about 1 AM. The text associated with the anima-
stabilized rate of change; when the magnitude of change (the
tion indicates that from 11 PM to 3 AM, packages are
number of states in the off-diagonal elements) was high,
sorted and distributed to the appropriate planes. You
Goldsberry increased the duration by increasing the number
then see a major exodus of planes from Memphis
of frames between decennial census years. (Courtesy of Kirk
between 3 and 5 AM. Although some of the animations
Goldsberry.)
in the atlas are fuzzy due to poor contrast, the anima-
tions certainly catch your attention and make you ask
We encourage you to examine both the variable and the why particular spatiotemporal patterns exist.
stabilized rate of change animations, which are available
at http://www.geog.ucsb.edu/~kirk/alternative_animation. Animations for Courseware and Lectureware
htm and http://www.geog.ucsb.edu/~kirk/thesis_animation. The bulk of the animations that we have discussed
htm, respectively, and think about the types of questions were developed in association with academic publica-
that each animation might answer. tions, which typically cover the approaches used to
create the animation and often comment on its pre-
Peterson and Wendel’s Animated Atlas of Air Traffic sumed effectiveness. You should recognize, however,
Michael Peterson and Jochen Wendel have devel- that numerous animations have been created outside
oped an intriguing animated atlas of air traffic over this framework, particularly for educational pur-
the United States, Canada, and the Caribbean poses. In general, two forms of educational software
447
11:52:56.
Map Animation
are possible: courseware and lectureware. Courseware enumeration units making up British census districts
assists students in learning concepts outside the could be examined:
classroom, whereas lectureware is used in a more tra-
ditional lecture setting (Krygier et al. 1997). In prac- What emerged was a picture of the way in which areas of
tice, it appears that the bulk of educational software is affluence simultaneously surrounded, looped around,
developed as courseware, but then is also used in a lec- and appeared inside the places where the less prosperous
ture setting. Introductory physical geography text- lived. . . . One of the most interesting patterns occurs in inner
London, where a thin, snakelike belt of “professionals” winds
books, such as Christopherson (2006) and Strahler and
from north to south, between areas sharply differentiated as
Strahler (2006) commonly include courseware CDs or
being dominated by the housing of those working in the low-
links to online resources, which include animations. est status occupations. (p. 217)
Courseware materials, many of which contain anima-
tions, also have been released on the Web; one exten- Although Dorling used the term “animation” to
sive source is the Digital Library for Earth System describe his interaction with the map, we consider it a
Education (http://www.dlese.org). Because a wide form of data exploration: The instantaneous change in
range of courseware has been developed, it is not pos- the data is a form of animation, but the control exerted
sible to easily classify this software in either of our two by the user makes exploration a more appropriate
major categories based on the level of interaction. term.
However, we have placed it in the little-or-no-interaction Dorling used the term “three-dimensional anima-
category because many of the associated animations tion” to describe the animation of data in 3-D space; for
are characterized by limited interaction. instance, the software IDV can be used to create 3-D
animations of weather-related phenomena. As with ani-
3.2 Animations Characterized by Substantial mations of color at fixed locations, Dorling found 3-D
Interaction animations confusing, arguing that although a computer
can easily create 3-D graphics, it is more appropriate to
Dorling’s Work represent data in two-dimensional form.
Dorling appeared to concur with our suggestion of
Daniel Dorling (1992) provided some interesting
using small multiples as a substitute for animation, but
thoughts on animation by experimenting with it in
he indicated that this might require an animation of
three realms, which he designated “animating time,” space (i.e., data exploration). He stated:
“animating space,” and “three-dimensional anima-
tion.” Although some of the animations he described Comparison requires at least one simultaneous view. The
appeared to fit our little-or-no-interaction category, distribution of successive years’ unemployment rates is
we include them in the substantial-interaction cate- often best shown on a single map, where symbols can
gory because he supported the move toward greater become very small when a large number of areas are
interaction. Dorling’s idea of animating time roughly involved. Animating space can allow detailed spatial inves-
equated to DiBiase et al.’s (1992) time-series anima- tigation of such a complex picture, which incorporates a
tion. Dorling argued that animating time is not suc- temporal attribute statically. (p. 224)
cessful when animations involve changes in color at
fixed locations due to “the brain’s poor visual mem- In summary, although Dorling criticized certain aspects
ory” (p. 223). His conclusion concurs with our diffi- of animation, he felt it could be especially useful for
culty of interpreting some of the animations we have enlivening presentations.
considered (e.g., Peterson’s animation of nontempo-
ral data associated with choropleth maps). However, Monmonier’s Graphic Scripts
Dorling argued that animating movement over time Mark Monmonier (1992a) defined a graphic script as a
can be effective, noting that his most effective anima- series of dynamic maps, statistical graphics, and text
tion involved moving dots within a triangle that blocks used to tell a story about a particular set of data.
represented the proportion of votes for three major Monmonier intended graphic scripts to be a form of data
political parties (pp. 218–219). Peterson’s animation exploration, and thus we have placed them in the “inter-
of air traffic discussed in the preceding section is action” category. His initial implementation of these
another good example of the ease of animating move- scripts, however, was as fixed animations (“movies”),
ment over time. given that users could not control their speed or direc-
Animating space involves “panning and zooming . . . tion of play. To illustrate a graphic script, consider the
a large two-dimensional static image” (p. 216). To portion of a script shown in Figure 7, which Monmonier
illustrate, Dorling described how a map of three used to examine the distribution and relationship of the
occupational attributes in the more than 100,000 attributes “Female Percentage of Elected Local
11:52:56.
Map Animation
control over the scripts (remember, they were “movies”).

Act I: Introduce the Attributes
I-1. Introduce dependent attribute “Female Percentage of
As a result, Monmonier and Gluck suggested that graphic
Elected Local Officials, 1987,” with brief title “Female Officials” scripts be stopped automatically at key points, continu-
and signature hue of red. ing only when the viewer pushes a “resume” key.
Text block: full title, brief title, and description; pause for Alternatively, they suggested that a “rerun” key would
reading . . . provide greater flexibility by allowing a user to repeat the
Large map above rank-ordered bar graph, with each poly-
gon linked to a bar.
immediately preceding sequence. To justify our place-
Upward sweep by rank [Figure 8]; downward sweep by rank. ment of scripts in the “substantial-interaction” category,
Upward sweep by value; downward sweep by value. capability would also have to be added to permit users to
Highlight highest fifth; highlight lowest fifth. explore individual sections of the script. For example,
Classify by quintiles (equal fifths); then blink-highlight each when Figure 8 is presented, it should be possible for a user
category (in sequence from highest category to lowest).
Classify by equal thirds; then blink-highlight each category.
to easily select either an upward sweep by rank or a
Quintiles again; blink-highlight each of the nine census downward sweep by rank.
divisions . . . One problem that we encountered while viewing the
I-2. Introduce independent attribute “Female Labor Force scripts was the lack of a verbal commentary that would
Participation Rate, 1987,” with brief title “Females Working” and explain what was being shown. Monmonier (1992a)
signature hue of blue.
Same layout and scenario as in I-1.
indicated that the lack of verbal commentary was a
I-3. Compare geographic patterns by rapid alternation of the two function of the large storage space required by high-
attributes on a single map. quality speech. Monmonier and Gluck avoided this
Large map above hue-coded titles for both attributes. problem by having a trained moderator provide a run-
Classify by quintiles; 13 cycles: display dependent attribute on ning commentary during the viewers’ evaluation of the
map, then display independent attribute on map.
scripts. Apparently, this was essential, as one of the par-
ticipants remarked, “It was all logical, but without the
FIGURE 7 A portion of a graphic script. (After Monmonier, narrative it would have been difficult. . . . If I had to
M. 1992a. First published in Cartography and Geographic read the description and watch what was going on, I
Information Systems 19(4), p. 254. Reprinted with permission
from the American Congress on Surveying and Mapping.)
Female Percentage of Elected Local Officials, 1987
Officials” and “Female Labor Force Participation Rate”

for the 50 states of the United States. The complete
script consisted of three “acts”:
Act I: Introduce the Attributes
Act II: Variation and Covariation in Geographic Space
and Attribute Space
Act III: Explore the Relationship with a Regression
Model
Figure 7 lists just the script for Act I, which, as you can
see, consists of three major steps: introduce the depen- high
dent attribute (I-1 in Figure 7), introduce the indepen-

dent attribute (I-2), and compare geographic patterns (I- low
3). Note that each step involves a series of animated Upward Sweep, by rank
maps. To illustrate, Figure 8 portrays a frame from

“Upward sweep by rank; downward sweep by rank.” The FIGURE 8 A frame from a graphic script illustrating
heights of the individual bars beneath the map corre- “Upward sweep by rank; downward sweep by rank.”
spond to each of the values symbolized on the map. Individual bars at the bottom correspond to values for each of
During an upward sweep, a bar would be lit in the graph the data shown on the map. During an upward sweep, a bar is
and then on the map, beginning with the lowest ranked lit in the graph and then on the map, beginning with the low-
value and progressing to the highest; the opposite would est ranked value and progressing to the highest; the opposite
be true for a downward sweep. This approach is similar is true for a downward sweep. (After Monmonier, M. 1992a.
to the sequencing option used in the software First published in Cartography and Geographic Information
ExploreMap. Systems 19(4), p. 248. Reprinted with permission from the
Monmonier and Gluck (1994) found that users had dif- American Congress on Surveying and Mapping and Mark
ficulty understanding graphic scripts because they had no Monmonier.)
449
11:52:56.
Map Animation
would have had a hard time” (p. 43). Fortunately, those of Figure 9 shows examples of linear and cyclic legends,
wishing to develop graphic scripts today will find that respectively. Important conclusions from Harrower et al.’s
storage space is no longer an issue, given the large disk study were that students who understood the brushing
drives that are available. and focusing tools performed the best, and that those
without access to the tools performed better than those
who did not understand the tools, suggesting that instruc-
Harrower’s Work
tors must take great care in introducing such tools to stu-
Mark Harrower has undertaken considerable work with dents.
animation; here we consider three studies that he has In a second study, Harrower (2001) promoted ani-
been involved in. In the first, Harrower and his colleagues mation as a means for understanding satellite-based
(2000) analyzed EarthSystemsVisualizer (http://www. time-series data. He stated:
geovista.psu.edu/grants/VisEarth/animations1.html),
which was developed to facilitate undergraduate Current change-detection techniques are insufficient for
students’ learning about global weather (Figure 9). the task of representing the complex behaviors and
Rather than simply showing “canned” animations, motions of geographic processes because they emphasize
Harrower et al. wanted students to explore the underly- the outcomes of change rather than depict the process of
ing data using brushing and focusing tools. Brushing change itself. Cartographic animation of satellite data is
allows the user to highlight an arbitrary set of individual proposed as a means of visually summarizing the complex
entities; for instance, the user might wish to see only ani- behaviors of geographic entities. (p. 30)
mation frames associated with 6 AM. In contrast, focusing
allows the user to focus on a particular subrange of data; To illustrate the potential of animation for satellite-
for instance, you might wish to see an animation of only a based imagery, Harrower developed a prototype system
48-hour period, as opposed to an entire week. An impor- known as VoxelViewer (see “portfolio” at http://www.
tant characteristic of EarthSystemsVisualizer is the geography.wisc.edu/~harrower/Flash_Splash6.swf). Two
attempt to represent different kinds of time (i.e., linear key elements of VoxelViewer are tools that allow a tem-
and cyclic) with corresponding linear and cyclic legends poral resolution ranging from 10 to 80 days and a spatial
(based on the work of Edsall and Peuquet (1997)). Linear resolution (pixel size) ranging from 10 to 250 km. Such
time refers to the notion of a general change over time, tools are useful for reducing spatial heterogeneity and
such as population growth for a city, whereas cyclic time short-term temporal fluctuations, ultimately producing
refers to temporal changes that are repetitive in nature, smoother animations that capture general trends in the
such as diurnal changes in temperature.The bottom portion data. In an intriguing application, Harrower described
FIGURE 9 The interface for

EarthSystemsVisualizer (original
appears in color). (Courtesy of Mark
Harrower.)
11:52:56.
Map Animation
how such tools can be used to examine the behavior of whether the user steps frame by frame or the system
the Sahel, a semi-arid region in northern Africa. Using a automatically displays the frames. Below the map is a
spatial resolution of 10 km and a temporal resolution of graph depicting the change in each individual enu-
40 days, Harrower stated: meration unit over time. In this case, we see that
Finland is highlighted, and in the graph we can see the
The northern and southern boundaries do not move in uni- trend for forest exports from Russia to Finland from
son. Rather there is a temporal lag in which the southern 1992 to 1997. Color Plate 5B further highlights the
boundary moves northward first (since the monsoon rains considerable flexibility of the software. Note that
arrive from the south) compressing the region before the “Enable visual comparison” has been checked in the
northern boundary expands into the desert. The southern
upper left corner. Because “Fixed time moment” has
boundary is also the first to retreat southward with the
also been selected, and because the bar below that is
onset of the dry season. (p. 38)
positioned above “1992,” each frame of the animation
Harrower noted that such a phenomenon could not can be compared with the year 1992. Also notice that
be seen at other resolutions and would be missed if the graph below the map allows us to compare each
animation were not used. year against 1992. Finally, note that a diverging color
In a third study, Harrower (2002; http://www.geography. scheme has been selected to enhance the comparison.
wisc.edu/~harrower/dissertation/index.html) developed The complete flexibility of the software can be
the notion of visual benchmarks, or reference points with fully appreciated by experimenting with the online
which other frames of an animation can be compared. version at http://www.ais.fraunhofer.de/descartes/time/
Fixed visual benchmarks remain constant from frame to AreaAnalysis/app/efis/.
frame; for example, Color Plate 4 illustrates fixed visual Another interesting piece of software developed by
benchmarks for a proportional circle map—yellow and Andrienko and her colleagues (2000) is the depiction of
red circles represent, respectively, the minimum and the movement of storks across the continent of Africa
maximum numbers of new AIDS cases for each country using directed line segments of differing colors. This
over the time period of the animation (1993–2000). topic is rather difficult to depict in static form but pro-
Dynamic visual benchmarks are more complicated vides an interesting illustration of how linear move-
because they change from frame to frame; for example, a ments can be depicted in an animation. Andrienko and
“ghost image” of a previous frame might appear superim- her colleagues introduce the notion of varying both the
posed on the current frame. Harrower implemented visual interval of time depicted and the time step between
benchmarks for both proportional symbol and isarithmic individual frames; the result is that the directed line seg-
maps, finding that benchmarks were more effective on ments move across the map and change in length,
proportional symbol maps. Although Harrower found depending on how long a stork stops at a particular
that benchmarks were not as effective as he had hoped, he location. An online version can be viewed at http://
noted that their effectiveness might be enhanced through www.ais.fraunhofer.de/and/java/birds/.
further use and training. Deep Thunder
Andrienko et al.’s Work Deep Thunder is a project undertaken by Lloyd
Treinish and his colleagues at IBM to develop sophisti-
Natalia Andrienko and her colleagues (2000; 2001a) cated prediction models and associated visualizations
argued that traditional “fixed” animations are insufficient: for weather-sensitive business operations (http://www.
research.ibm.com/weather/DT.html; Treinish 2005). Two
A simple “movie”, e.g., showing the growth of city popula-
sample maps from Deep Thunder are shown in Color
tion by expanding circles that represent cities, is often
Plate 6. As you can see, the resulting images are typi-
enough for demonstration of a known temporal trend. . . .
This is insufficient, however, for supporting exploration, cally 3-D visualizations of multiple atmospheric attrib-
i.e. revealing unknowns. An analyst needs special tools that utes. For instance, in the case of Color Plate 6A, we see
can help to compare states at different moments, detect the following: a base map of the Atlanta and Savannah,
significant changes, assess magnitudes and speeds of the Georgia, vicinity; arrows that have orientation and color
changes etc. (Andrienko et al. 2000, 217) indicating wind direction and speed at the surface;
isarithmic information depicting precipitation; and a
A software package that Andrienko et al. developed cloud structure depicted as a white, translucent isosur-
for this purpose is illustrated in Color Plate 5. face, with cyan isosurfaces in the interior indicating
Consider Color Plate 5A—on the right is a set of ani- forecast rain shafts. Although Deep Thunder includes a
mation controls that allow considerable flexibility in substantial interactive component, an important ele-
animating choropleth maps, such as the step size ment is animations, which are often viewed without
between frames, the delay between frames, and extensive interaction. For example, Color Plate 6A was
451
11:52:56.
Map Animation
part of an animation that was used to predict the available in Brownrigg’s software. In this case, each
weather for the 1996 Centennial Olympic Games in vertical prism corresponds to a city, and the intensity of
Atlanta. Although Treinish and his colleagues obviously the gray corresponds to the magnitude of the popula-
feel that such animations are effective, their complexity tion, with the constraint that population values above 1
again raises the issue of the usefulness of animation. million are symbolized as a solid black, allowing us to
see more differentiation at the bottom end of the scale.
Note that the z axis (the third dimension) is used to
4 USING 3-D SPACE TO DISPLAY show changes over time. Obviously, one difficulty with
TEMPORAL DATA using 3-D space to show temporal changes is that cities
often block one another; thus, interpreting the display
The bulk of the animations we have described show requires the user to interact with it.
geospatial changes over time by varying images in two- Brownrigg argued that he was “able to detect all of the
dimensional space. A potential disadvantage of this trends in the data reported in Slocum et al. (2000) [using
approach is that at any instant, we can see only one MapTime], as well as some additional facts not mentioned
point in time. An alternative approach is to use the third and likely difficult to detect with an animation” (p. 92).
dimension to show the temporal changes. Thus, in the- One example of these additional facts is illustrated in
ory, we should be able to see not only spatial patterns Color Plate 7A, where a spectral color scheme is used
for individual moments in time, but also changes over (i.e., red, orange, yellow, green, blue, indigo, and violet,
time. Although this approach is not technically a form with red representing the highest values). MapTime
of animation (i.e., it requires a great deal of interaction reveals an apparent drop in population for some of the
with the display), we include it in this chapter because it largest northeastern cities from about 1950 to the present.
provides a distinct alternative to animation. As an illus- In Color Plate 7A, we see that Cleveland, Ohio, actually
tration of the 3-D approach, we consider the work of experienced two declines: a brief one in the 1930s and
Richard Brownrigg (2005), who developed what he then a longer one beginning in 1950.
termed spacetime maps. Brownrigg also discovered that several cities had pop-
We will focus on just one of the data sets that ulations of zero sometime after they were initially
Brownrigg mapped, the population of major U.S. cities founded. For instance, in Color Plate 7B we see that the
from 1790 to 1990. Figure 10 provides an overview of prism for Salem, Oregon, has been selected and that the
the population data using a gray-tone symbolization population in 1880 was 2,538. Just slightly higher on the
FIGURE 10 An example of a space-

time map. The population for major
U.S. cities from 1790 to 1990 is
depicted using vertical prisms shaded
with an intensity of gray propor-
tional to the population. (Courtesy of
Richard Brownrigg.)
11:52:56.
Map Animation
prism, we see a light gray area indicating zero population, if they could have interacted with the animated maps.
and above that point, the prism is colored, indicating an A study by Patton and Cammack (1996) also revealed
increasing population. Brownrigg indicated, “It is not support for animation on sequenced choropleth maps.
known whether these are gaps in the census data at those In contrast, studies by Slocum and Egbert (1993),
times, or whether the data truly reflect these cities’ brief Johnson and Nelson (1998), and Cutler (1998) have
histories as ghost towns” (p. 95). Although some found little difference between animated and static
researchers have expressed caution in using 3-D space maps. For instance, in comparing an animated map
(Dorling above, and Ware and Plumlee 2005), it is appar- with a small multiple, Cutler found that the animated
ent that information can be gleaned from 3-D views that version of a shaded isarithmic map resulted in a
is difficult or impossible to obtain from an animation. significantly lower percentage of correct answers and
a slightly slower processing time than the small
5 DOES ANIMATION WORK? multiple. Cutler noted that “the strongest indicators
of comprehension were not the type of map viewed,
but the subjects’ reading levels and their prior
At several points in this chapter, we have raised the
knowledge” (p. 63).
question of whether animation works. Although those
More recently, Amy Griffin and her colleagues
developing animations are often enamored of anima-
(2006) have responded to some of Tversky et al.’s con-
tions’ capability, we (and others) often find them diffi-
cerns by conducting a carefully controlled study of ani-
cult to understand. Is it possible that the capability of
mation versus small multiples for the detection of
animation has been overhyped? Such thoughts were
spacetime clusters. Figure 11 is a small multiple
shared by Barbara Tversky and her colleagues (2002),
illustrating an example of a spacetime cluster; these
who undertook a meta-analysis of animation efforts
abstract hexagonal displays were used to control for
that did not involve geospatial information. Although
the effect that the structure of a real geographic region
researchers suggested that the animations Tversky and
might have on the results. Griffin et al. found that ani-
her colleagues evaluated were more effective than their
mated maps did enable “users to more often correctly
static counterparts, Tversky et al. found that the anima-
identify whether a particular type of pattern was
tions often contained microsteps that were not avail-
present than did the static small-multiple representa-
able in the static displays, and thus they argued that a
tion” (p. 749). An interesting finding of their study was
fair comparison was not possible. Tversky et al. also
that for animated spacetime clusters, males were signif-
noted that although interactivity clearly has the poten-
icantly more likely to find patterns than were females.
tial to enhance an animation, studies have not yet been
Griffin and her colleagues suggested that this differ-
done that fully analyze the effects of such interactivity
ence might be a function of the enhanced “visual atten-
relative to static maps. Finally, Tversky et al. made the
tion skills” of boys, resulting from their more common
following observation:
use of video games.
Animations must be slow and clear enough for observers
to perceive movements, changes, and their timing, and to
understand the changes in relations between the parts and
the sequence of events. This means that animations should
lean toward the schematic and away from the realistic, an
inclination that does not come naturally to many program-
mers, who delight in graphic richness and realism.
These seem to be useful thoughts for those of you

considering developing animations for others.
Experimental studies of the effectiveness of map
animations have produced mixed results. We have
already mentioned the success that Gershon (1992)
appeared to have with his subjects’ understanding ani-
mations. A study often used to illustrate the benefits FIGURE 11 A small multiple illustrating an example of a
of map animation is the work of Koussoulakou and spacetime cluster used in Griffin et al.’s experiment. (From
Kraak (1992), who found that although the percent- Griffin, A. L., MacEachren, A. M., Hardisty, F., Steiner, E. and
age of correct answers did not differ, animated maps Li, B. (2006) “A comparison of animated maps with static
were processed significantly faster than static maps small-multiple maps for visually identifying space-time clus-
when depicted as a small multiple. Koussoulakou and ters.” Annals of the Association of American Geographers 96,
Kraak stressed that users might have done even better no. 4:740–753. Courtesy of Blackwell Publishing.)
453
11:52:56.
Map Animation
Sara Fabrikant and her colleagues (Fabrikant 2005; the map). In the psychological literature, this failure to
Fabrikant and Goldsberry 2005; Fabrikant et al. 2007; detect changes in the visual field is known as change
http://www.geog.ucsb.edu/~animeye/) have begun a blindness. In one famous experiment (Simons and
research agenda to examine the effectiveness of Chabris 1999), participants were shown a video of two
animated maps and small-multiple displays using an basketball teams, one wearing white shirts and the other
eye-movement data-collection method. The eye-move- wearing black shirts, and were asked to count how many
ment method enables Fabrikant and her colleagues to times the ball was passed between members of one team.
determine where the map reader is looking (and for While the ball was being passed, a person in a gorilla suit
how long) while viewing an animated map or a small walked among the players, stopped to face the camera
multiple. For example, Figure 12 shows an individual and thump its chest, and then walked away. Only about
map reader’s eye-movement patterns for a small multi- 50 percent of the participants even noticed that the
ple for two different tasks. The graduated circles show gorilla had appeared! Although we don’t have gorillas
where the reader’s eye was fixated (the larger the circle, walking across our maps, this humorous example illus-
the longer the fixation), and the connecting dark gray trates the likelihood that we may miss changes in one
lines represent eye movements between the fixations. portion of a map display while attending to others.
The task for Figure 12A was “to gain an overall impres- Clearly, more research is necessary if we are to
sion of the small multiple display, and to verbally determine whether animation is truly effective. We
describe the patterns that are discovered during its suspect that the effectiveness of animation will be a
visual exploration” (Harrower and Fabrikant in press). function of numerous factors, including the type of
In contrast, the task for Figure 12B was to compare the symbology used (e.g., depicting temporal change on a
two maps shown in the lower right corner of the small proportional symbol map seems easier than depicting
multiple. Obviously, these tasks led to dramatically dif- temporal change on a choropleth map); how the ani-
ferent eye movements. Using results such as these, it mation is viewed (a “movie” in which the user has no
can be argued that “small multiple displays will never control is obviously quite different from an anima-
be informationally equivalent to non-interactive animation in which the user can move forward, backward,
tions as Tversky and colleagues (2002) are implying” and jump to different frames); whether the user has
(Harrower and Fabrikant). expertise in the domain being mapped (animated
Several cartographers who have studied map anima- maps created using Deep Thunder seem complex to
tions have pointed out the difficulty of interpreting them, us, but they might not be to the trained meteorolo-
noting that information in an animation can be missed, gist); and experience with animation (does one who
especially when the animation is complex (e.g., when uses animation on a regular basis become more adept
changes are happening at five or six different places on at using it?).
A B
FIGURE 12 An individual map reader’s eye-movement patterns for a small multiple for two different tasks. Graduated circles
show where the reader’s eye was fixated (the larger the circle, the longer the fixation), and connecting dark gray lines represent
eye movements between the fixations. (Courtesy of Sara Fabrikant.)
11:52:56.
Map Animation
SUMMARY nontemporal methods, and Peterson and Wendel’s

animation of air traffic. Those who have developed ani-
In this chapter, we examined several matters pertaining mations involving substantial interaction include
to map animation. First we noted that although anima- Harrower and Andrienko and her colleagues. In addition
tion commonly is associated with recent technological to creating interactive animations, Harrower is notewor-
advances, the earliest map animations were produced in thy for developing the notion of visual benchmarks,
the 1930s, and cartographers began discussing the which enable reference points to be compared with other
potential of animation in the 1950s. The first computer- frames of an animation.
based map animations were produced in the 1970s, but The bulk of animations that we discussed are dis-
it was not until the 1990s that computer-based anima- played in 2-D space and thus do not make use of the
tions became common. third dimension. An alternative to animation is the use
The visual variables utilized for static maps can also of the z axis of 3-D space to depict time. Brownrigg
be used on animated maps; for example, we might uti- terms this a spacetime map and argues that it enables
lize proportional circles (the visual variable size) to the detection of information that cannot be detected or
depict changes in the population of census tracts over is difficult to detect in an animation. A downside of
time. Animated maps, however, also utilize other visual spacetime maps is that symbols for different locations
variables, including duration (the length of time that a block one another and require a great deal of user inter-
frame of an animation is displayed), rate of change action with the display.
(computed as m/d, where m is the magnitude of change In viewing animations for this chapter, we found
between frames or scenes, and d is the duration of each many of them difficult to understand, raising the ques-
frame or scene), order (the sequence in which frames tion of whether animation is truly effective. We are not
or scenes are presented), display date (the time at alone in our uncertainty, as Tversky and her colleagues
which some display change is initiated), frequency (the also raised this issue, noting that research supporting
number of identifiable states per unit time), and the use of animation over static renditions is often
synchronization (the temporal correspondence of two flawed (e.g., animations might contain microsteps that
or more time series). Animations can be categorized in are not available in static displays). It appears that
terms of whether they emphasize change in either posi- more research is needed to determine whether anima-
tion or an attribute, the location of a phenomenon, or tion (and interactivity) actually enhances visualiza-
the spatial distribution of an attribute. Animations tion. Sara Fabrikant and her colleagues have begun a
emphasizing change can be further divided into three research program that promises to illuminate the dif-
types: time series, re-expressions, and fly-bys. A time ferences between animated and small-multiple dis-
series is the most common form of animation because plays through an eye-movement data-collection
animating time can serve as a scale model of real-world method. We suspect that the effectiveness of anima-
time. Some of the most interesting animations, however, tion will be a function of numerous factors, including
do not involve time. For example, a fly-by, in which the the type of symbology used, the level of interaction
viewer is given the impression of flying over a land- the user is permitted, whether the user has expertise in
scape, can be particularly dramatic. the domain being mapped, and the user’s experience
We considered numerous examples of animations, with animation.
extending from those that involved little or no interac- Our focus in this chapter has been on stand-alone
tion to those having substantial interaction. Examples of animations and on interactive software in which ani-
animations involving little or no interaction include mation plays a major role. Animations can also play a
Wilhelmson et al.’s depiction of thunderstorm develop- part in data exploration software and in the visual-
ment, Treinish’s portrayal of the ozone hole, Peterson’s ization of uncertainty.
FURTHER READING
Acevedo, W., and Masuoka, P. (1997) “Time-series animation dynamic maps and graphics.” Computational Statistics
techniques for visualizing urban growth.” Computers & 16:417–433.
Geosciences 23, no. 4:423–435. Discusses highly interactive software for exploring spatiotem-
Discusses the development of animations to depict the growth of poral data.
urban areas. Also see Batty and Howes (1996a; 1996b). Blok, C. A. (2005) Dynamic Visualization Variables in
Andrienko, N., Andrienko, G., and Gatalsky, P. (2001a) Animation to Support Monitoring of Spatial Phenomena.
“Exploring changes in census time series with interactive Netherlands Geographical Studies 328. Utrecht: Enschede.
455
11:52:56.
Map Animation
Describes the development and user testing of software for moni- Reviews issues and research related to the development of suc-
toring vegetation change. cessful map animations.
Blok, C., Köbben, B., Cheng, T., and Kuterema, A. A. (1999) Karl, D. (1992) “Cartographic animation: Potential and
“Visualization of relationships between spatial patterns in research issues.” Cartographic Perspectives no. 13:3–9.
time by cartographic animation.” Cartography and Geo- Introduces a number of research issues associated with animation.
graphic Information Science 26, no. 2:139–151.
Krygier, J. B., Reeves, C., DiBiase, D., and Cupp, J. (1997)
Uses animation to examine the relationship between two attributes “Design, implementation and evaluation of multimedia
that change over time. resources for geography and earth science education.” Journal
Campbell, C. S., and Egbert, S. L. (1990) “Animated cartogra- of Geography in Higher Education 21:17–38.
phy: Thirty years of scratching the surface.” Cartographica 27, Discusses the development and evaluation of multimedia
no. 2:24–46. resources for geography and earth science education.
A review of early animation efforts in cartography; explores the Lodha, S. K., and Verma, A. K. (2000) “Spatio-temporal visual-
reasons for the lack of animation in cartography prior to 1990 and ization of urban crimes on a GIS grid.” Proceedings of the 8th
suggests future prospects for animation.
ACM International Symposium on Advances in Geographic
Cox, D. J. (1990) “The art of scientific visualization.” Academic Information Systems, Washington, DC, pp. 174–179.
Computing 4, no. 6:20–22, 32–34, 36–38, 40.
Illustrates several approaches for using 3-D space to depict spa-
Discusses a number of animations developed at the National tiotemporal data associated with urban crime.
Center for Supercomputing Applications, where so-called
Mitas, L., Brown, W. M., and Mitasova, H. (1997) “Role of
Renaissance Teams were formed.
dynamic cartography in simulations of landscape processes
Eddy, W. F., and Mockus, A. (1994) “An example of the esti- based on multivariate fields.” Computers & Geosciences 23,
mation and display of a smoothly varying function of time and no. 4:437–446.
space—The incidence of the disease mumps.” Journal of the
Discusses the use of animation to illustrate (1) simulations of
American Society for Information Science 45, no. 9:686–693.
landscape processes and (2) the effect of changing parameters
Discusses sophisticated mathematical approaches for creating a associated with interpolation methods.
smooth animation of spatiotemporal data for enumeration units.
Monmonier, M. (1990) “Strategies for the visualization of
Fabrikant, S. I., and Goldsberry, K. (2005) “Thematic relevance geographic time-series data.” Cartographica 27, no. 1:30–45.
and perceptual salience of dynamic geovisualization displays.”
Discusses methods for visualizing spatiotemporal data; animation
Proceedings, 22nd International Cartographic Conference, is considered one useful method.
A Coruña, Spain, CD-ROM.
Monmonier, M. (1992b) “Summary graphics for integrated
Introduces the work that Fabrikant and her colleagues are visualization in dynamic cartography.” Cartography and
beginning to do involving eye-movement studies of animated
maps and small multiples.
Presents static maps and graphics that can be used in association
Gahegan, M. (1998) “Scatterplots and scenes: Visualization
with animated maps.
techniques for exploratory spatial analysis.” Computers,
Environment and Urban Systems 21, no. 1:43–56. Monmonier, M. (1996) “Temporal generalization for
dynamic maps.” Cartography and Geographic Information
Describes sophisticated approaches for using animation to analyze
Systems 23, no. 2:96–98.
multivariate relationships at one point in time.
Describes methods for smoothing map animations.
Gersmehl, P. J. (1990) “Choosing tools: Nine metaphors of
four-dimensional cartography.” Cartographic Perspectives Ogao, P. J., and Kraak, M.-J. (2001) “Geospatial data explo-
no. 5:3–17. ration using interactive and intelligent cartographic anima-
tions.” Proceedings, 20th International Cartographic
Discusses various ways in which software can be used to create
animations; Gersmehl termed these “animation metaphors.”
Conference, International Cartographic Association, Beijing,
China, CD-ROM.
Harrower, M. (2003) “Tips for designing effective animated
maps.” Cartographic Perspectives, no. 44:63–65. Suggests the possibility of intelligent animation, in which features
within animations are created on the fly, as in a developing storm
Provides some tips for designing effective animated maps. system.
Harrower, M. (2004) “A look at the history and future of Peterson, M. P. (1995) Interactive and Animated Cartography.
animated maps.” Cartographica 39, no. 3:33–42. Englewood Cliffs, NJ: Prentice Hall.
Provides a historical overview of the development of animated This chapter has focused on the appearance of animations, as
maps and considers some of the remaining technological hurdles. opposed to how they are created. In contrast, Peterson’s text pro-
Harrower, M., and Fabrikant, S. (in press) “The role of map vides information on creating animations; see particularly
animation for geographic visualization.” In Geographic Chapters 7 through 9.
Visualization: Concepts, Tools and Applications. ed. Rana, S., and Dykes, J. (2003) “A framework for augmenting
M. Dodge, M. Turner, and M. McDerby. Chichester, UK: the visualization of dynamic raster surfaces.” Information
Wiley. Visualization 2:126–139.
11:52:56.
Map Animation
Illustrates numerous approaches for transforming continuous Sidney, L. R., and Edsall, R. M. (2005) “Challenging the
phenomena to create meaningful animations. conventions of interface design for education-oriented ani-
Rensink, R. A. (2002) “Change detection.” Annual Review of mated maps.” Research in Geographic Education 5:5–24.
Psychology 53:245–277. Describes an experiment that examined the effect of the direction-
Reviews the psychological literature on change detection, or “the ality of a temporal legend on the interpretation of an animation.
apprehension of change in the world around us” (246).
GLOSSARY
animated maps: maps characterized by continuous change, lectureware: software used to assist students in learning
such as the daily weather maps shown on television newscasts material in a traditional lecture setting.
that depict changes in cloud cover. order: one of the visual variables for animated maps; refers
brushing: when a user is able to highlight an arbitrary set of to the sequence in which frames or scenes of an animation are
spatial entities while exploring data; for instance, a user might presented.
display the percentage of Hispanics only for tracts that voted rate of change: one of the visual variables for animated
for a particular political party. maps; defined as m/d, where m is the magnitude of change
courseware: software used to assist students in learning con- between frames or scenes, and d is the duration of each frame
cepts outside the classroom. or scene.
display date: one of the visual variables for animated maps; re-expression: an animation created by modifying the origi-
refers to the time that some display change in an animation is nal data in some manner, such as by choosing subsets of a time
initiated (e.g., in an animation of population for U.S. cities, a series or by reordering a time series.
circle for San Francisco appears in 1850). sequencing: displaying a map piece by piece (e.g., displaying
duration: one of the visual variables for animated maps; refers the title, geographic base, legend title, and then each class in
to the length of time that a frame of an animation is displayed. the legend).
dye-sublimation printer: a printing device (normally, full- synchronization: one of the visual variables for animated
color) that incorporates heat sources to convert solid dyes maps; refers to the temporal correspondence of two or more
into gases, and applies them to the print medium. time series.
fly-by (or fly-over): a form of animation in which the viewer time series: an animation that emphasizes change through
is given the feeling of flying over a landscape. time.
focusing: when exploring data, a user is able to highlight a videocassette: an analog tape format for storing and playing
subrange of numeric values; for example, a user might focus moving images and sound.
on census tracts that are less than 25 percent Hispanic. visual benchmarks: reference points with which other
frequency: one of the visual variables for animated maps; frames of an animation can be compared (e.g., for a propor-
refers to the number of identifiable states per unit time, such tional-circle map animation of AIDS by country, yellow and
as color cycling used to portray the jet stream. red circles could represent the minimum and maximum num-
graphic script: a series of dynamic maps, statistical graphics, ber of new AIDS cases for each country over the time period
and text blocks that are used to tell a story about a set of data. of the animation).
REFERENCES
Acevedo, W., and Masuoka, P. (1997) “Time-series animation Andrienko, N., Andrienko, G., and Gatalsky, P. (2000)
techniques for visualizing urban growth.” Computers & “Supporting visual exploration of object movement.”
Geosciences 23, no. 4:423–435. Proceedings of the Working Conference on Advanced
Ambroziak, B. M., Ambroziak, J. R., and Bradbury, R. (1999) Visual Interfaces AVI, ACM Press, Palermo, Italy, pp.
Infinite Perspectives: Two Thousand Years of Three- 217–220, 315.
Dimensional Mapmaking. New York: Princeton Andrienko, N., Andrienko, G., and Gatalsky, P. (2001a)
Architectural Press. “Exploring changes in census time series with interactive
American Cartographic Association. (1986) Which Map Is dynamic maps and graphics.” Computational Statistics
Best: Projections for World Maps. Bethesda, MD: 16:417–433.
American Congress on Surveying and Mapping. Andrienko, N., Andrienko, G., Savinov, A., Voss, H., and
American Cartographic Association. (1988) Choosing a World Wettschereck, D. (2001b) “Exploratory analysis of spatial
Map: Attributes, Distortions, Classes, Aspects. Bethesda, MD: data using interactive maps and data mining.” Cartography
American Congress on Surveying and Mapping. and Geographic Information Science 28, no. 3:151–165.
457
11:52:56.
Map Animation
Anselin, L. (1998) “Exploratory spatial data analysis in a Fuhrmann, S., and MacEachren, A. M. (2001) “Navigation in
geocomputational environment.” In Geocomputation: A desktop geovirtual environments: Usability assessment.”
Primer, ed. by P. A. Longley, S. M. Brooks, R. McDonnell, Proceedings, 20th International Cartographic Conference,
and B. MacMillan, pp. 77–94. Chichester, England: Wiley. International Cartographic Association, Beijing, China,
Batty, M., and Howes, D. (1996a) “Exploring urban develop- CD-ROM.
ment dynamics through visualization and animation.” In Griffin, A. L., MacEachren, A. M., Hardisty, F., Steiner, E.,
Innovations in GIS 3, ed. by D. Parker, pp. 149–161. and Li, B. (2006) “A comparison of animated maps with
Bristol, PA: Taylor & Francis. static small-multiple maps for visually identifying space-
Batty, M., and Howes, D. (1996b) “Visualizing urban time clusters.” Annals of the Association of American
development.” Geo Info Systems 6, no. 9:28–29, 32. Geographers 96, no. 4:740–753.
Brownrigg, R. L. (2005) “Data visualization with spacetime Gershon, N. (1992) “Visualization of fuzzy data using gener-
maps.” Unpublished PhD dissertation, University of alized animation.” Proceedings, Visualization ’92, Boston,
Kansas, Lawrence, KS. MA, pp. 268–273.
Campbell, C. S., and Egbert, S. L. (1990) “Animated cartogra- Goldsberry, K. (2004) “Stabilizing rate of change in thematic
phy: Thirty years of scratching the surface.” Cartographica map animations.” Unpublished MA thesis, University of
27, no. 2:24–46. California, Santa Barbara, CA.
Christopherson, R. (2006) Elemental Geosystems (5th ed.). Gong, J., Lin, H., and Yin, X. (2000) “Three-dimensional
Upper Saddle River, NJ: Prentice Hall. reconstruction of the Yaolin Cave.” Cartography and
Geographic Information Science 27, no. 1:31–39.
Clarke, K. C., Hoppen, S., and Gaydos, L. (1997) “A self-
modifying cellular automaton model of historical urban- Harrower, M., and Fabrikant, S. (in press) “The role of map
ization in the San Francisco Bay area.” Environment and animation for geographic visualization.” In Geographic
Planning B: Planning and Design 24:247–261. Visualization: Concepts, Tools and Applications, ed. by M.
Dodge, M. Turner, and M. McDerby. Chichester, UK:
Cutler, M. E. (1998) “The effects of prior knowledge on
Wiley.
children’s abilities to read static and animated maps.”
Unpublished MS thesis, University of South Carolina, Harrower, M., MacEachren, A., and Griffin, A. L. (2000)
Columbia, SC. “Developing a geographic visualization tool to support
earth science learning.” Cartography and Geographic
DiBiase, D., Krygier, J., Reeves, C., MacEachren, A., and
Information Science 27, no. 4:279–293.
Brenner, A. (1991) “Elementary approaches to carto-
graphic animation.” Videotape. University Park, PA: Deasy Harrower, M., and Sheesley, B. (2005) “Moving beyond
GeoGraphics Laboratory, Department of Geography, novelty: Creating effective 3-D fly-over maps.”
Pennsylvania State University. Proceedings, 22nd International Cartographic Conference,
A Coruña, Spain, CD-ROM.
Dorling, D. (1992) “Stretching space and splicing time: From
cartographic animation to interactive visualization.” Harrower, M., and Sheesley, B. (2007) “Utterly lost: Methods
Cartography and Geographic Information Systems 19, for reducing disorientation in 3-D fly-over maps.”
no. 4:215–227, 267–270. Cartography and Geographic Information Science 34,
no. 1:19–29.
DiBiase, D., MacEachren, A. M., Krygier, J. B., and Reeves,
C. (1992) “Animation and the role of map design in Harrower, M. (2001) “Visualizing change: Using cartographic
scientific visualization.” Cartography and Geographic animation to explore remotely-sensed data.” Cartographic
Information Systems 19, no. 4:201–214, 265–266. Perspectives, no. 39:30–42.
Edsall, R., and Peuquet, D. (1997) “A graphical user interface Harrower, M. A. (2002) “Visual benchmarks: Representing
for the integration of time into GIS.” http://www.geovista. geographic change with map animation.” Unpublished
psu.edu/publications/ACSM97/index.html. PhD dissertation, The Pennsylvania State University,
University Park, PA.
Fabrikant, S. I. (2005) “Towards an understanding of geovisu-
alization with dynamic displays: Issues and prospects.” Johnson, H., and Nelson, E. S. (1998) “Using flow maps to
Proceedings, American Association for Artificial visualize time-series data: Comparing the effectiveness of
Intelligence, 2005 Spring Symposium Series, Reasoning with a paper map series, a computer map series, and anima-
Mental and External Diagrams: Computational Modeling tion.” Cartographic Perspectives, no. 30:47–64.
and Spatial Assistance, Stanford, CA, pp. 6–11. Koussoulakou, A., and Kraak, M. J. (1992) “Spatio-temporal
Fabrikant, S. I., and Goldsberry, K. (2005) “Thematic rele- maps and cartographic communication.” The Cartographic
vance and perceptual salience of dynamic geovisualization Journal 29, no. 2:101–108.
displays.” Proceedings, 22nd International Cartographic Lavin, S., Rossum, S., and Slade, S. R. (1998) “Animation-
Conference, A Coruña, Spain, CD-ROM. based map design: The visual effects of interpolation on
Fabrikant, S. I., Montello, D. R., and Rebich, S. (2007) the appearance of three-dimensional surfaces.”
“Seeing through small multiple map displays.” Abstract, Cartographic Perspectives, no. 29:26–34.
Association of American Geographers Annual Meeting, Krygier, J. B., Reeves, C., DiBiase, D., and Cupp, J. (1997)
San Francisco, CA, p. 192. “Design, implementation and evaluation of multimedia
11:52:56.
Map Animation
resources for geography and earth science education.” Slocum, T. A., Yoder, S. C., Kessler, F. C., and Sluter, R. S.
Journal of Geography in Higher Education 21:17–38. (2000) “MapTime: Software for exploring spatiotemporal
Lodha, S. K., Sheehan, B., Pang, A. T., and Wittenbrink, C. M. data associated with point locations.” Cartographica 37, no.
(1996a) “Visualizing geometric uncertainty of surface 1:15–31.
interpolants.” Graphics Interface, pp. 238–245. Strahler, A., and Strahler, A. (2006) Introducing Physical
http://www.cse.ucsc.edu/research/slvg/surf.html. Geography (4th ed.). Hoboken, NJ: Wiley.
MacEachren, A. M. (1995) How Maps Work: Representation, Szego, J. (1987) Human Cartography: Mapping the World of
Visualization, and Design. New York: Guilford. Man. Stockholm, Sweden: Swedish Council for Building
Mitas, L., Brown, W. M., and Mitasova, H. (1997) “Role of Research.
dynamic cartography in simulations of landscape processes Thrower, N. J. W. (1959) “Animated cartography.” The
based on multivariate fields.” Computers & Geosciences Professional Geographer 11, no. 6:9–12.
23, no. 4:437–446. Tobler, W. R. (1970) “A computer movie simulating urban
Moellering, H. (1976) “The potential uses of a computer growth in the Detroit region.” Economic Geography 46,
animated film in the analysis of geographical patterns of no. 2(Supplement):234–240.
traffic crashes.” Accident Analysis & Prevention 8:215–227. Treinish, L. A. (1992) “Climatology of global stratospheric
Monmonier, M. (1992a) “Authoring graphic scripts: ozone.” Videotape, IBM Corporation.
Experiences and principles.” Cartography and Geographic Treinish, L. A. (2005) “Web-based dissemination and visual-
Information Systems 19, no. 4:247–260, 272. ization of operational 3D mesoscale weather models.” In
Monmonier, M., and Gluck, M. (1994) “Focus groups for Exploring Geovisualization, ed. by J. Dykes, A. M.
design improvement in dynamic cartography.” MacEachren, and M.-J. Kraak, pp. 403–422. Amsterdam:
Cartography and Geographic Information Systems 21, Elsevier.
no. 1:37–47. Treinish, L. A., and Goettsche, C. (1991) “Correlative visual-
Patton, D. K., and Cammack, R. G. (1996) “An examination ization techniques for multidimensional data.” IBM
of the effects of task type and map complexity on Journal of Research and Development 35, no. 1/2:184–204.
sequenced and static choropleth maps.” In Cartographic Tukey, J. W. (1977) Exploratory Data Analysis. Reading, MA:
Design: Theoretical and Practical Perspectives, ed. by C. H. Addison-Wesley.
Wood and C. P. Keller, pp. 237–252. Chichester, England:
Tversky, B., Morrison, J. B., and Betrancourt, M. (2002)
Wiley.
“Animation: Does it facilitate?” International Journal of
Peterson, M. P. (1993) “Interactive cartographic animation.” Human–Computer Studies 57, 247–262.
Cartography and Geographic Information Systems 20,
Ware, C., and Plumlee, M. (2005) “3D geovisualization and
no.?1:40–44.
the structure of visual space.” In Exploring
Peterson, M. P. (1999) “Active legends for interactive carto- Geovisualization, ed. by J. Dykes, A. M. MacEachren, and
graphic animation.” International Journal of Geographical M.-J. Kraak, pp. 567–576. Amsterdam: Elsevier.
Wilhelmson, R. B., Jewett, B. F., Shaw, C., Wicker, L. J.,
Simons, D. J., and Chabris, C. F. (1999) “Gorillas in our midst: Arrott, M., Bushell, C. B., Bajuk, M., Thingvold, J., and
Sustained inattentional blindness for dynamic events.” Yost, J. B. (1990) “A study of a numerically modeled
Perception 28:1059–1074. storm.” The International Journal of Supercomputer
Slocum, T. A., and Egbert, S. L. (1993) “Knowledge acquisi- Applications 4, no. 2:20–36.
tion from choropleth maps.” Cartography and Geographic
Information Systems 20, no. 2:83–95.
459
11:52:56.
11:52:56.
Data Exploration
461
12:31:15.
Data Exploration
OVERVIEW
Data exploration, like the larger notion of geovisualiza- software, ExploreMap, and Aspens. This software either is
tion, is most frequently a private activity in which no longer available or has limitations when compared to
unknowns are revealed in a highly interactive environ- today’s software, but it provides a useful historical perspec-
ment. The first two sections of this chapter consider the tive. Next, we cover more recent software that was developed
goals and methods of data exploration. Broad goals primarily with geographers in mind: MapTime, Project
include: (1) identifying the spatial pattern associated with Argus, CommonGIS, and GeoDa. Finally, we consider soft-
a single attribute at one point in time; (2) comparing spa- ware that, although not explicitly developed for geographers,
tial patterns for two or more attributes at one point in time; clearly is useful for geographers: HealthVisPCP, ESTAT,
(3) identifying how spatial patterns for a single attribute and IDV. With the exception of the historical software, we
change over time; and (4) comparing spatial patterns for have attempted to select software that is available (one prob-
two or more attributes to see how they covary over time. In lem is that software developers in academia often create
a sense, these same goals apply to static paper maps, but prototypes but do not distribute them), and is inexpensive
the interactive environment of data exploration software (all are available for free to academics).
permits us to discover spatial patterns that might not be
seen in a single static map.
1 GOALS OF DATA EXPLORATION
The methods of data exploration include a host of
possibilities: manipulating the input data (e.g., standard-
izing data); varying the symbolization (e.g., changing the Like the larger notion of geovisualization, data explo-
color scheme on a choropleth map); manipulating the ration is most frequently a private activity in which
user’s viewpoint (as in a 3-D display); highlighting por- unknowns are revealed in a highly interactive environ-
tions of a data set (focusing and brushing are two ment. What are these unknowns, or, more specifically,
approaches); presenting multiple views (e.g., a small what are the goals of data exploration? The following
multiple, in which a large number of maps are shown); are four broad goals for data exploration; in each case,
using animation; linking with other forms of display we provide an example using attributes associated with
(e.g., tabular and graphical displays); providing access to census tracts in Los Angeles, California:
miscellaneous resources (e.g., via the Web); deciding how
symbols are assigned to attributes on a multivariate map; 1. Identify the spatial pattern for a single attribute at
and providing automatic map interpretation (e.g., data one point in time. (Example: What was the spatial
mining). No single data exploration software package pattern of housing costs in 2000?)
will include all of these features, but we would expect an 2. Compare spatial patterns for two or more attrib-
effective package to include many of them. utes at one point in time. (Example: How did the
The third section of the chapter covers examples of data spatial pattern for housing costs compare with the
exploration software. We begin by considering some early spatial pattern for the percentage of Hispanics in
data exploration software: Hal Moellering’s 3-D mapping the population in 2000?)
12:31:15.
Data Exploration
3. Identify how the spatial pattern for a single methods listed here were culled from the works of
attribute changes over time. (Example: How has the MacEachren et al. (1999c), Crampton (2002), Kraak
spatial pattern of housing costs changed with each (1998), and Buja et al. (1996).
decennial census?)
4. Compare spatial patterns for two or more attrib- 2.1 Manipulating Data
utes to see how they covary over time. (Example:
How has the relationship between housing costs Methods for manipulating data include standardizing
and percentage of Hispanics changed over time?) data (e.g., adjusting raw totals for enumeration units to
account for the area over which the totals are distrib-
You might argue that these goals are no different from uted), transforming attributes (e.g., taking logarithms of
those for static paper maps, and this is true. One differ- data), and various methods of data classification. The
ence, however, is that the character of spatial patterns first two of these methods can be handled in spread-
might not become clear until we manipulate the map in sheet programs, but it is easier for the user if they are
a highly interactive graphics environment. For instance, included within data exploration software.
if for the first goal we make a single choropleth map of
housing costs, our visual interpretation might be a func-
tion of the color scheme—we might visualize a quite 2.2 Varying the Symbolization
different pattern using a red–blue diverging scheme
than with a blue sequential scheme. In a data explo- One approach to varying the symbolization is to
ration environment, we can change the color scheme change the general form of symbolization. For instance,
instantaneously, and thus learn something about the different views of wheat harvested in Kansas could be
spatial pattern of housing costs that was “hidden” in the obtained using four different methods of symboliza-
single-map solution. tion: choropleth, isopleth, proportional symbol, and
Another difference between a paper map and a data dot. In a data exploration environment, such views can
exploration environment is that the latter provides link- often be generated quickly, although certain symbol-
ages to other views of the data in tables, graphs, and ization methods (e.g., the dot map) are not trivial to
numerical summaries. For instance, when viewing a create. It is also possible to vary the precise manner in
bivariate map of housing costs and the proportion of which the symbolization is implemented. For instance,
Hispanics in the population, you might also utilize a if a choropleth map is deemed appropriate, then
scatterplot depicting the relationship between the numerous color schemes can be employed. Alternative
attributes. Examining this scatterplot might help you color schemes can lead to dramatically different-looking
better understand the spatial pattern appearing on the maps. The power of data exploration software is that,
map. Alternatively, you might notice a group of points not only can the overall color scheme be changed, but
that cluster together on the scatterplot and wonder the detailed relation between color and data can also
where they occur on the map. To determine the geo- be modified dynamically. For instance, with the
graphic location of the clustered points, you could use a CommonGIS software (see section 3.6), we can drag a
mouse to select the clustered points in the scatterplot, break point separating two colors in a diverging
which would cause these same points to be highlighted scheme and have the result displayed instantaneously.
on the map. Clearly, this function could not be accom-
plished using a static map. 2.3 Manipulating the User’s Viewpoint
Manipulating the user’s viewpoint refers to modify-

2 METHODS OF DATA EXPLORATION ing the “view” that users have of the map, with the
data and method of symbolization held constant.
In this section we summarize a broad range of methods Zooming and panning are perhaps the most common
characteristic of data exploration software. Obviously, viewpoint manipulations. Such functions are critical,
individual software packages will not include all of the given the relatively small size of standard display
methods, as each piece of software normally has specific screens and the large databases that we sometimes
(and thus limited) purposes. As you read about (and, we wish to display. Sophisticated software will automati-
hope, use) some of the data exploration software that cally adjust the detail shown as a function of the
follows, you might find it useful to consider which of degree of zoom (e.g., the names of census tracts might
these methods have been implemented and how they not be visible at a small scale, but would appear at a
are utilized. It should be noted that there is no widely larger scale). The ability of users to change the view-
agreed-on set of methods for exploring spatial data. The point is of course critical with 3-D maps, as high
463
12:31:15.
Data Exploration
points on the surface can obscure lower points. Also 2.7 Linking Maps with Other Forms of Display
important with 3-D maps is the means by which a user
navigates through the landscape (e.g., using a mouse The ability to link maps with tables, graphs, and numerical
vs. using a joystick, and the associated on-screen con- summaries is a critical aspect of data exploration soft-
trols that are available), and how the user is provided ware. Users should be able to highlight data in tables,
a sense of orientation. graphs, or maps and have the highlighted specifications
carried through to the other forms of display. Numerical
summaries provide a useful complement to various forms
2.4 Highlighting Portions of a Data Set of graphical display. For a single attribute, numerical sum-
maries often include measures of central tendency and
Fundamental to data exploration is the ability to highlight dispersion, although a measure of spatial autocorrelation
portions of a data set via either focusing or brushing. also could be useful in determining whether a spatial pat-
Focusing involves highlighting a subrange of numeric val- tern might have resulted by chance. For multiple attrib-
ues. For example, you might want to focus on census tracts utes, more complex numerical summaries would be
that are less than 25 percent Hispanic. In a sense, this is a appropriate (e.g., for two numeric attributes, the user
form of data classification, because you divide the data might want to know the correlation coefficient).
into two classes: those tracts with less than 25 percent
Hispanic, and those with 25 percent or more. Brushing
involves highlighting an arbitrary set of spatial entities. 2.8 Access to Miscellaneous Resources
For example, you might wish to display the percentage of
Hispanics only for tracts that voted for a particular politi- Throughout this discussion, we have assumed that you
cal party. have already collected data that you wish to visualize.
In practice, as you explore spatial data, particularly with
either multiple attributes or spatiotemporal data, you
2.5 Multiple Views will find that you need additional information. In some
cases, this information will be in a format similar to
Multiple views refers to displaying more than one map what you have already acquired. For instance, in relat-
image on the computer screen at a time. One form of ing housing costs to percent of Hispanic population, you
multiple view is the small multiple, in which maps are might want to examine public expenditures for educa-
displayed for individual time steps or attributes. Thus, if tion. In other cases, you might wish to collect a quite
we have data on housing costs for decennial censuses, different form of information. For instance, in studying
we would include a map for each decennial census. housing costs, you might want to be able to click on a
Alternatively, for a particular decennial census, we might census tract and see what typical houses look like in
create separate maps for percent Hispanic, median that tract. Ideally, it would be desirable if you could
income, housing costs, and percentage of students readily import this sort of information directly into your
attending public schools. We know that the small multi- data exploration software, possibly via the Web.
ple is often an alternative to animation. Another form of
multiple view is to show the same data set using differ-
ent symbolization methods, as was discussed in section 2.9 How Symbols Are Assigned to Attributes
2.2. In general, multiple views take advantage of our
ability to compare things visually. With multiple attributes (i.e., a multivariate map), an
important decision often involves which attributes are
assigned to which symbols. For instance, for the com-
2.6 Animation bined star and snowflake glyph symbol, the appearance
will depend on the assignment of attributes to the rays
Animation is often an integral part of data explo- of the glyph. In Figure 1, the eight attributes on the left
ration software. The key with data exploration soft- are arbitrarily assigned, whereas on the right, the attrib-
ware is that the user has considerably more control utes have been ordered from smallest to largest begin-
over an animation than is possible with canned ani- ning at the top and progressing clockwise. As another
mation “movies.” With data exploration software, example, the order in which attributes are displayed on
users might be able to stop an animation, redefine the a parallel coordinate plot will influence the look of the
data–symbol relationship, and rerun the animation. plot. Remember that for highly positively correlated
We’ll see an example of this with the MapTime soft- attributes, line segments on the parallel coordinate
ware in section 3.4. plot tend to be parallel, whereas for highly negatively
12:31:15.
Data Exploration
1978) to demonstrate software capabilities. In addition

1 8
to illustrating surface exploration, the video portrayed
8
two animations of spatiotemporal data: the growth of
2 6 2
the U.S. population from 1850 to 1970 and the diffusion
7 3 5 4 of farm tractors in the United States from 1920 to 1970.
6 These animations were created using attractive, colored
4 1 3-D prism maps.
3
5 7 Although Moellering’s software suggested great
potential for both data exploration and animation, for
approximately 10 years following his effort, there was
FIGURE 1 Different ways of assigning attributes to the com- little development in data exploration (at least by geo-
bined star and snowflake glyph symbol. On the left, the attrib- graphers), in part because many geographers lacked the
utes are arbitrarily assigned; on the right, they are assigned in necessary hardware and software. The lack of progress
order from the lowest to the highest attribute. in animation, in particular, was lamented by Craig
Campbell and Stephen Egbert (1990). Fortunately, con-
correlated attributes, line segments tend to intersect one siderable advancements have taken place since 1990, as
another in the middle of the graph. this text illustrates. Today it is possible to generate maps
similar to Moellering’s on a personal computer right on
2.10 Automatic Map Interpretation one’s desktop.
In discussing data exploration thus far, we have focused 3.2 ExploreMap

on the ability of the user to interpret spatial patterns. In
complex multiple-attribute situations or with spa- Stephen Egbert and Terry Slocum (1992) developed
tiotemporal data, it might be useful to utilize methods ExploreMap to assist users in exploring data commonly
that automatically interpret the data, or at least assist mapped with choropleth symbolization. In developing
the user in finding useful patterns in the data. ExploreMap, their intention was not to create the ulti-
mate data exploration tool, but rather to provide a pro-
3 EXAMPLES OF DATA EXPLORATION SOFTWARE totype that others might improve on. In contrast to
Moellering’s video distribution, ExploreMap and other
software described in this chapter are distributed via
In this section, we discuss numerous examples of data diskette, CD-ROM, or over the Web.
exploration software. First, we provide a historical per- ExploreMap operates in two basic modes: Design and
spective by considering some early attempts at data Explore. Design is necessary for creating maps, whereas
exploration: Moellering’s 3-D mapping software, Explore includes methods for exploring data, and thus is
ExploreMap, and Aspens. Second, we consider more the focus of this chapter. Two of ExploreMap’s functions,
recent software that was developed primarily with geo- Areas and Overview, correspond to the two types of
graphers in mind as potential users: MapTime, Project information that can be acquired from a map: specific
Argus, CommonGIS, and GeoDa. Third, we consider and general. To illustrate how Areas provides specific
recent software that was not necessarily developed for information, consider two of its options: Single Area and
geographers, but that certainly could be useful to them: Ratio of Areas. The Single Area option is used to deter-
HealthVisPCP, ESTAT, and IDV. mine the exact value of an enumeration unit, either by
pointing at it with the mouse or by typing its name. In
3.1 Moellering’s 3-D Mapping Software contrast, on a traditional paper map, the value for an enu-
meration unit is determined by visually comparing a
Hal Moellering’s (1980) 3-D mapping software was an shade on the map with shades in the legend—for exam-
early attempt to demonstrate the potential of data ple, on a classed map, the best estimate is a range of val-
exploration. With Moellering’s software, a user could ues shown in the legend.† Most current data exploration
explore a 3-D digital elevation model (DEM) in real software includes an option similar to the Single Area
time, a process that Moellering termed “surface explo- option. The Ratio of Areas option is used to determine
ration.”* Because Moellering’s software depended on the ratio of values for two enumeration units: When a
expensive hardware, he developed a video (Moellering
†Data values on a paper map can be displayed within enumeration

*Technically, the data were 2 1/2-D but exploration was done in 3-D units, but this detracts from the visualization of the map pattern;
space. moreover, values cannot be placed within small units.
465
12:31:15.
Data Exploration
FIGURE 2 Using the Ratio of Areas option within FIGURE 3 The Classes option for ExploreMap. Any subset
ExploreMap to compare two enumeration units. The compared of classes can be selected by clicking on the legend boxes; in
units (counties in this case) are highlighted on the map, and the this case, classes 2 and 4 are shown.
ratio of their values is shown at the bottom of the display.
out reservation”; subjects used Classes not only to
identify the location of individual classes, but also to
user selects two units, they are highlighted on the map, examine patterns and trends. Although useful, the
and the ratio of their values is shown at the bottom of the Classes option is constrained by how the data are
display (see Figure 2). Such precise information is not classed (in Figure 3, a five-class optimal map is used).
available from paper maps, although it remains to be One solution to this constraint is the Reclass function,
seen how worthwhile such an option is to users. which permits changing the method and number of
The Overview function within ExploreMap includes classes. An option within Reclass, Compare Maps, per-
three options: Sequenced, Classes, and Subset. Rather mits presenting up to four maps simultaneously (each
than presenting the entire map at once (as on printed map can have a different method of classification or
maps), the Sequenced option builds the map piece by number of classes). For example, four methods of clas-
piece: The title, geographic base, and legend title and sification for five-class maps are equal intervals, quan-
boxes appear first (in sequence), and then each class is tiles, optimal based on the mean, and optimal based on
displayed (from low to high). Two arguments can be the median.
made for sequencing: (1) it should enhance map under- An alternative to reclassing the data is the Subset
standing by providing users “chunks” of information option, which permits focusing on an arbitrary range of
(Taylor 1987), and (2) it should emphasize the quantita- data. To assist users in selecting data to focus on,
tive nature of the data (i.e., that the data are ordered ExploreMap provides a dispersion graph (Color Plate
from low to high values). However, experimental tests 1A) illustrating the distribution of an attribute along
of sequencing have produced mixed results. A study by the number line; the current class breaks, mean, median,
Slocum et al. (1990) found that although more than 90 and standard deviation can be shown on the dispersion
percent of readers favored sequencing over the tradi- graph (only the median is shown in Color Plate 1A).
tional static (printed) approach, objective measures of Users can simply click on a desired range (or type a
map use tasks revealed no significant difference range of values), and the specified range is drawn in a
between sequenced and static approaches. As a result, highlighted color (blue is used in Color Plate 1B to dis-
Slocum et al. argued that sequenced maps might be pre- play all values less than the median).
ferred simply because of their novelty. A later study by Although Subset enables users to highlight selected
David Patton and Rex Cammack (1996) revealed that portions of a data set, it is limited because all highlighted
sequencing is more effective when the time to study the data are shown in a single color, and because real-time
map is limited. interaction with the dispersion graph is not possible. A
The Classes option within Overview permits any more flexible approach would permit a range of shades
combination of classes to be displayed; for example, for the highlighted color (say, a range of blues instead of
Figure 3 portrays two classes of a five-class map. In an a constant blue). This would enable users to focus on a
evaluation of ExploreMap, Egbert (1994, 87) found subset of the data while also seeing the variation within
that the Classes option “was liked by all subjects with- that subset. A more flexible approach would also allow
12:31:15.
Data Exploration
the map to change dynamically as the mouse cursor is 3.3 Aspens

dragged back and forth along the number line.
Interestingly, the latter capability was provided in the Barbara Buttenfield and Christopher Weber (1994)
early Xmap system developed at the Massachusetts developed Aspens to assist Dennis Jelinski (1987) in
Institute of Technology by Joseph Ferreira and Lyna exploring the growth in trembling aspen trees at
Wiggins (1990). Xmap used a “density dial” to highlight Waterton Lakes National Park in Alberta, Canada.
portions of a data set; Ferreira and Wiggins described More specifically, their purpose was to examine “the
the use of the density dial as follows: apparent contradiction that radial growth rates [in
aspens] were higher where local environmental condi-
What makes the density dial a truly interactive visualizations (elevation, precipitation, and soils) were more
tion tool is its ability . . . to change the group of cells that harsh” (p. 11). Although Aspens was designed solely to
are highlighted in red . . . as fast as we can move the handle Jelinski’s data, it was intended as a prototype for
mouse. We get an effect that is like watching a video. We illustrating the broader concept of proactive graphics, a
see a moving sequence of red-shaded polygons going from
term Buttenfield and Weber coined to describe software
the least dense tract to the most dense tract, or vice versa.
that enables users to initiate queries using icons (or sym-
This speed—and the sense of motion that comes with it—
enhances our ability to remember what we have just seen bolic representations) as opposed to words. Ideally, they
of the spatial pattern. (p. 71) argued, proactive graphics should avoid steep learning
curves and be “responsive to commands and queries . . .
Unfortunately, Xmap was not made generally available, not . . . anticipated by system designers” (p. 10).
and so it is unlikely that you will have access to it. We The philosophy of proactive graphics in Aspens is
will see, however, that the CommonGIS system pro- apparent in the opening screen of the study area (Color
vides similar functionality. Plate 2), where no pull-down menus (which, for exam-
Other major functions in ExploreMap include ple, appear in MapTime) are found. Rather, it is pre-
Regions, Tables, Graphs & Stats, and Utilities. Whereas sumed that users will become aware that links to infor-
Subset allows users to focus on a subset of data along mation are most often available via the color green
the number line, Regions permits users to focus on a found on various icons. For example, clicking on the
subregion of a mapped display of the data. For example, green “i” provides information about using the display,
a politician interested in counties comprising an eco- and clicking on the green symbol portion of the legend
nomic development region could use Regions to high- leads to further information about the legend.
light enumeration units falling within that region. The Interestingly, an animation begins as soon as the study
Tables and Graphs & Stats functions focus on tabular, area appears: Six leaves (corresponding to the six major
graphical, and statistical views of the data (as opposed to study sites) change in size and color as each year of
spatial views). For instance, Graphs & Stats can display growth is highlighted in a bar graph; one frame from
a histogram and statistical parameters (Figure 4). the animation is shown in Color Plate 2. Larger leaf
sizes represent larger total cumulative growth, and a
darker green leaf color represents a greater incremen-
tal growth for each year. Buttenfield and Weber argued
that having the animation start automatically is appro-
priate because “The sooner graphical activity is recog-
nized, the more quickly data will be explored” (p. 11).
A click on the name of a study site (e.g., Prairie 2 in
Color Plate 2) leads to a screen of descriptive statistics
and a histogram of the yearly incremental growth for that
site. As we have suggested, linking statistical and graphi-
cal displays with maps is critical for exploratory analysis.
Clicking on a leaf associated with a study site leads to a
more detailed view of that site: the aspen clones making
up the site (Color Plate 3). To distinguish the clones from
the generalized study site, treelike symbols rather than
leaves are used. The clones are animated in a manner
analogous to the general study site, and descriptive sta-
tistics and a histogram of yearly growth rates can also be
FIGURE 4 The Graphs & Stats option within ExploreMap. obtained by clicking on a clone identifier. Within any
Graphical and statistical information provide an alternate view clone, it is possible to jump to another clone by clicking
of the map of the data. on one of the icons in the upper right of the screen.
467
12:31:15.
Data Exploration
Buttenfield and Weber recognized that Aspens is not Animations in MapTime are constructed using a series
truly proactive because it is not possible to generate of key frames (i.e., frames associated with collected data)
queries and commands unanticipated by system design- and intermediate frames (i.e., frames associated with
ers. They also indicated that it would be desirable to interpolated data). A distinct advantage of MapTime is
incorporate photographs of study sites, add capabilities that the number of intermediate frames can be varied.
to reverse the animation and jump to an arbitrary year, Thus, if one has data for every year from 1900 to 1940, but
and implement functions for saving data, maps, and text only for every other year between 1940 and 1980, inter-
information to files. There is also a need for incorporat- mediate frames might be used just for the 1940–1980 data.
ing other data that might be related to tree growth, such In the context of temporal data, a small multiple con-
as elevation, precipitation, and soils. In spite of these sists of a set of maps, one for each time element for which
limitations, Aspens is illustrative of an early attempt at data have been collected. As an example, Figure 5
data exploration. portrays a small multiple of stream discharge for seven
collection stations within a hypothetical drainage basin at
3.4 MapTime 12-hour intervals over a four-day period (the data are
distributed with MapTime). It is presumed that the
MapTime permits users to explore temporal data associ- streams flow from west to east, that circle size is propor-
ated with fixed point locations using three major tional to stream discharge (in cubic feet per second), and
approaches: animation, small multiples, and change that heavy rainfall occurs between 6 PM Monday and
maps. Stephen Yoder (1996), the developer of MapTime, 3 AM Tuesday throughout the region. The small multiple
argued that animation is a logical solution for showing reveals an initial increase in discharge for upstream
changes over time because “. . . the cartographic presen- stations, followed by a later increase for downstream
tation [is not only] a scale model of space, but of time as stations; essentially, a pulse of water moves through the
well” (30). Animations of point data are easiest to inter- system. The pulse of water seen in the small multiple is
pret when data change relatively gradually over time. also detectable in an animation of the stream discharge
For example, an animation of the change in population data, but it is not easy to discern. Thus, small multiples are
of U.S. cities from 1790 to 1990 (distributed with often a necessary complement to animations. Small mul-
MapTime) is easy to follow because city populations tiples can also assist in contrasting two arbitrary points in
gradually increase and decrease. In contrast, an anima- time. For example, one might examine the beginning and
tion of the change in water quality at point locations ending multiples for stream discharge and note approxi-
along streams and rivers would exhibit much sharper mately how much change has taken place over the period
increases and decreases, and thus be harder to interpret. at each location.
FIGURE 5 A small multiple cre-

ated within MapTime. Each map
represents one point in time for
hypothetical stream discharge data
at selected points. (Courtesy of
Stephen C. Yoder.)
12:31:15.
Data Exploration
Change maps explicitly show the change that takes Like ExploreMap, the enumerated software is
place between two points in time; in MapTime, change intended for data collected in the form of enumeration
can be computed in terms of the raw data (magnitude units, but it extends well beyond ExploreMap’s
change), in percent form, or as a rate of change. The U.S. capabilities. For example, the software can symbolize
city population data from 1790 to 1990 provide a good data in more than one form (e.g., choropleth maps, pro-
illustration of the need for change maps. When these portional circle maps, and Dorling’s cartograms are
raw data are animated, one sees a major growth in city possible) and display up to three attributes at a time.
populations in the Northeast beginning in 1790, with an Interestingly, proportional circle maps are, by default,
apparent drop in population for some of the largest displayed using a redundant symbology of size and
northeastern cities from about 1950 to the present. In gray tones; the gray tones can, however, also depict a
contrast, a map showing the percent of population second attribute, thus producing a bivariate map.
change between 1950 and 1990 reveals a distinctive pat- To illustrate some of the enumerated software’s capa-
tern of population decrease throughout most of the bility, consider Color Plate 4, which portrays a view cre-
Northeast. ated with a sample data set for Illinois included with the
An interesting possibility is to create a small multi- software. The three attributes mapped are percent black,
ple of a series of change maps (an example of this for percent of the population 25 years and older with any
the U.S. population data from 1790 to 1990 is shown in college education, and percent below the poverty level.
Figure 6). In an evaluation of MapTime by users, Across the top of the screen, the attributes are shown as
Slocum and his colleagues (2004) found that a small individual choropleth maps. Choropleth maps are shown
multiple of change maps was favored over both anima- (as opposed to proportional circle maps) because the
tion and a simple raw small multiple for showing change maps are based on standardized data. By default,
over time. Animation, however, was deemed useful for unclassed choropleth maps are shown, although several
examining general trends and providing a sense of classification methods are available within the software.
change over time, whereas the raw small multiple was We know that the visual correlation of choropleth maps
useful for comparing arbitrary time periods. can be adversely affected by using unclassed maps, but
As with other exploration software, MapTime can unclassed maps are a useful starting point for visual
determine precise values associated with a particular analysis.
location or highlight a subset of the data. Additionally, In the middle portion of the screen, we see point
a zoom feature enables users to enlarge a portion of the graphs (one-dimensional scatterplots) of two of the
map. These features can be used to explore a single attributes: percent black and percent of the population
moment in time, or be implemented throughout an ani- 25 years and older with any college education. We can
mation. Those who wish to experiment with different compare these plots with the previous maps and see
circle-scaling methods also will find that circle-scaling that the overall light appearance of the percent black
exponents can easily be changed in MapTime. map is a function of a positive skew (i.e., outliers occur
toward the positive end of the number line). To the right
3.5 Project Argus of the point graphs is a scatterplot illustrating the rela-
tionship between the percent black and education
Project Argus, a collaborative venture of research labo- attributes (displayed along the x and y axes, respec-
ratories at the University of Leicester and Birkbeck tively). We can see that when percent black is low, a
College in England, was created to illustrate a broad large range of education percentages is possible, but
range of possible data exploration functions. Jason that the larger percent black values are associated with
Dykes was responsible for developing most of the soft- lower percent education values.
ware and wrote several papers (1995; 1996; 1997) sum- In the top right of the screen, we see a bivariate
marizing its capability. Two major pieces of software are choropleth map of the percent black and education
available at the Web site for Project Argus: one for data attributes. Higher values on these attributes are repre-
collected in the form of enumeration units (the “enu- sented by increasing amounts of yellow and blue,
merated” software) and one for tourist data collected respectively, whereas the lowest values are represented
over time (the “time-space” software). Here we focus by the absence of these colors, or black in the extreme
on the enumerated software.* case. Note that the two highest percent black values
shown in the scatterplot (they have low scores on the
education attribute) are depicted as a bright yellow in
the extreme southern part of Illinois. Overall, the map
*Technically, the enumerated software is termed the “cdv for enu- is relatively dark because of the concentration of both
merated data sets.” We use “enumerated software” here for simplicity. attributes in the lower left portion of the scatterplot.
469
12:31:15.
Data Exploration
FIGURE 6 A small multiple of change maps created using MapTime. Population increases are shown by darker circles, whereas pop-
ulation decreases are shown by lighter circles. The population decreases in the northeastern United States from 1950 to the present
were not easily recognized in either an animation or a raw small multiple of the data. (Courtesy of Stephen C. Yoder.)
Also note that the colors displayed on the map are these same observations are highlighted in green on
shown within dots on the scatterplot. other selected views. Also shown in Color Plate 4 is a
We have focused on how two of the three attributes parallel coordinate plot and Dorling’s cartogram
shown in Color Plate 4 might be analyzed. When actu- method (in this case illustrating the prominence of
ally using the software, a much more complete analy- Cook County). Like much of the software in this chap-
sis is, of course, possible. One interesting option is the ter, these various display methods can best be under-
use of a trivariate choropleth map, in which three stood by actually using the software.
choropleth maps are overlaid. It also must be recog-
nized that the static nature of a book prevents a full 3.6 CommonGIS
understanding of the dynamic capability of the enu-
merated software. For example, the software permits CommonGIS (http://www.commongis.com) was devel-
designating an outlier on a point graph and highlight- oped to give Web users easy access to visualizations of
ing that point on all other displays currently in view. attribute data associated with enumeration units or
Alternatively, somewhat similar to ExploreMap, it is point locations. Software development was funded by
possible to highlight a subrange of values on an the CommonGIS Consortium, which was composed of
attribute. To illustrate these concepts, in Color Plate 4 seven public and private companies from five countries
a small cluster of dots has a box drawn around it, and of the European Union. The roots of CommonGIS were
12:31:15.
Data Exploration
largely the Descartes software (formerly IRIS), which map nonnumerical data, CommonGIS assumes that the
was developed by Gennady and Natalia Andrienko data are nominal and that differing color hues or shapes
(1999; http://geoanalytics.net/and/). are appropriate. For instance, if a user wishes to map
An important characteristic of CommonGIS is the membership in the North Atlantic Treaty Organization
ability to automatically select an appropriate method of (NATO) using “TRUE” for Yes and “FALSE” for No,
symbolization, given a set of attribute data that a user CommonGIS permits only “Qualitative colouring” and
wishes to map. The choice of symbology is based on the an “Icons” option (Figure 7A). If we select the
nature of data input by the user and, in the case of Qualitative colouring option, we get the result shown in
numerical data, the user’s responses to a series of ques- Color Plate 5. Selecting the Icons option produces a
tions posed by CommonGIS. If the user attempts to map with either a circle or a square on each country.
A B
FIGURE 7 Menus from CommonGIS illustrating various mapping options available when the following kinds of data are input:
(A) nonnumerical data indicating membership in NATO, (B) population and population density, (C) a single numeric attribute—
population. (Courtesy of Knowledge Discovery Department Team, Fraunhofer Institute IAIS, http://www.iais.fraunhofer.de.)
471
12:31:15.
Data Exploration
To illustrate the types of questions that might be accomplished by dragging the small dark blue triangle
asked for numerical data, consider the following at the top of the point graph to just above the next
example. Imagine that you wish to map both “popula- highest observation. The result is shown in Color Plate
tion” and “population density” attributes. When you 6B, in which the outlier is depicted by a small brown
specify these attributes for European countries, triangle (in Albania) and a second point graph is drawn
CommonGIS provides the set of map options shown (to the left of the initial one) representing the reduced
in Figure 7B. Also imagine that you are unsure of data set. Now we see that Moldova, Iceland, and
which option to select and so experiment by selecting Macedonia clearly have the highest birth rates after
“Pies.” CommonGIS will indicate that you should Albania.
“Select the attribute which includes all others.” Since In the notion of diverging color schemes, two colors
neither attribute includes the other, you specify the diverge from a central light point. In CommonGIS,
option “No such attribute.” CommonGIS then asks diverging color schemes can be applied by selecting
whether either of the attributes “represent non-over- the “Compare by click in” option (see below the dot
lapping parts of some whole?” Since they do not, you plots in Color Plate 6), and then by clicking on the leg-
enter “No,” and CommonGIS terminates the ques- end or map to indicate the point from which colors
tions by responding, “No inclusion or additivity rela- should diverge. If the “Dynamic map update” option is
tionship between the attributes!” CommonGIS has also selected, the diverging point can be dragged,
thus kept you from mistakenly producing an inappro- allowing you to change the diverging point dynami-
priate map. Hopefully, you would now begin to think cally. Color Plate 7B suggests this notion for the per-
more carefully about your data and realize that you centage of the population greater than or equal to 65
should map each attribute separately: population as a in divisions of Portugal. As you move the horizontal
“Graduated circle” and population density as a marker along the legend from low to high data values,
“Classified choropleth map” (to use CommonGIS ter- you can see “how blue shades (corresponding to low
minology). values) spread from the areas around the biggest cities
Although CommonGIS can assist you in choosing of Portugal such as Lisbon and Porto, first along the
the correct mapping technique, there are limitations to coast and then into the inner parts of the country”
the question-and-answer technique. For instance, if you (CommonGIS documentation 2002). Note that this
decide to map just the population attribute, you will pattern is very difficult to detect in the nondiverging
receive the options shown in Figure 7C. If you select scheme shown in Color Plate 7A.
“Classified choropleth map,” a choropleth map will be An important component of version 2.2.0 of
produced without any questions being posed. Ideally, CommonGIS is its capability to display spatiotemporal
CommonGIS should ask you whether your data are data. In addition to providing visualizations similar to
standardized. If you have a chance to experiment with those we described for MapTime (e.g., animation, small
CommonGIS, you should think about what might con- multiples, and change maps), CommonGIS utilizes a
stitute an ideal set of questions to pose to users of such value flow diagram to display change in an attribute
data exploration software. over a time interval on a single map. Figure 8 displays
CommonGIS includes numerous interesting capa- value flow diagrams of the burglary rate from 1960 to
bilities for exploring spatial data. Consider the 2000 for the 48 contiguous U.S. states. Andrienko and
unclassed choropleth map of birth rate for European Andrienko (2004, 417) argue that value flow diagrams
countries shown in Color Plate 6A. To the right of the permit answering the following questions related to a
map, we see a vertical point graph (CommonGIS terms time interval:
this a “dot plot”) of the data displayed alongside the 1. How do attribute values at a given place vary over
range of possible colors representing birth rates. This time (local behavior pattern)?
point graph displays the distribution of the data along
the number line—it just happens to be a vertical plot For instance, in the case of Nevada, the burglary
rather than a horizontal plot. Note that the point graph rate is relatively low at the beginning of the period,
reveals a positively skewed distribution, largely due to increases to a maximum near the middle of the
a single outlier, Albania, which has a birth rate of 21.71. period, and then decreases to a level similar to
Because of this outlier, most of the countries appear to where it began.
be similar in color, thus making it difficult to visualize 2. Are the behaviors at locations l1 and l2 similar or
which countries have the next highest birth rates after different?
Albania. We can improve the ability to discrimi- If we contrast Nevada with North Dakota, we see
nate countries if we remove the outlier and extend the that North Dakota’s rate is much lower throughout
color scheme over the remaining countries. This is the time period.
12:31:15.
Data Exploration
North
Dakota
FIGURE 8 Value flow diagrams uti-
lized in CommonGIS; the diagrams
Nevada
depict the burglary rates in the
United States from 1960 to 2000. The
solid line encloses diagrams having a
similar temporal structure. Note that
those enclosed by the solid line are
all red in Color Plate 8 (i.e., they
have burglary rates above the
median). (Courtesy of Knowledge
Discovery Department Team,
Fraunhofer Institute IAIS, http:/
/www.iais.fraunhofer.de.)
3. How are different behavior patterns distributed edu/ ). Although ESDA tools are normally intended for
over the territory? Are there spatial clusters of those who wish to conduct a statistical analysis of spatial
similar behaviors? data, such tools frequently contain maps and graphical
To illustrate the notion of spatial clusters, we have drawn displays of interest to cartographers, and this is certainly
a line around several value flow diagrams in the west and the case for GeoDa.
southwest that appear similar in structure (i.e., relatively In addition to the standard quantiles and mean–
high crime rates, with a peak near the middle of the standard deviation methods of data classification,
period). This, of course, is a subjective process, given that GeoDa includes two classification techniques, the per-
our cluster differs slightly from the one that the centile map and the boxmap, that focus on extremes in
Andrienkos created (see Andrienko and Andrienko the data, which are often important in a statistical analy-
2006, 101). sis. These techniques are illustrated in Color Plate 9. The
percentile map is a special case of a quantile map in
To highlight the capability of CommonGIS to explore
which class limits are adjusted to accentuate extreme
the data, Color Plate 8 displays value flow diagrams
values. Specifically, the following six class limits are cre-
using the median of each year as a comparison value:
ated based on percentiles: < 1%, 1% to <10%, 10% to
values above the median are shown in red, and values
<50%, 50% to <90%, 90% to <99%, and >99%. In the
below the median are shown in blue. Here we can con-
case of the foreign-born data (see the window in Color
trast Nevada and North Dakota again, and note that
Plate 9 labeled “Percentile . . .”), we can see that the low-
Nevada is always above the median value, whereas
est value in the data set is in north central Florida
North Dakota is always below the median. We also note
(Madison County) and that the highest value is in the
that the states we identified as similar in structure in
extreme south (Dade County). Dade County is high-
Figure 8 all appear in red throughout the time period. It
lighted with a cross-hatched symbol on the percentile
is important to stress that we have only touched on some
map and is also highlighted in blue in the data table (the
of the capabilities of CommonGIS. Other options, such
lower left portion of Color Plate 9).The boxmap is a spe-
as linking the map of value flow diagrams with graphical
cial case of a four-class quantile map (i.e., a quartile
displays of the data for each enumeration unit, can best
map) in which outliers (values distinctly different from
be appreciated by actually using the software.
the rest of the data) are placed in classes by themselves.
The boxmap thus indicates whether the most extreme
3.7 GeoDa
values are truly different from the rest of the data. For
instance, we can define outliers as follows:
GeoDa, the latest incarnation of tools for exploratory
spatial data analysis (ESDA), was developed by
Luc Anselin and his colleagues at the University of Values > UQ + 3 * (UQ - LQ)
Illinois, Urbana-Champaign (https://www.geoda.uiuc. Values < LQ - 3 * (UQ - LQ)
473
12:31:15.
Data Exploration
where UQ and LQ are the upper and lower quartiles, sequenced (or animated, in GeoDa terms). Both Single
respectively. GeoDa actually uses somewhat different and Cumulative options are available. In the Single
formulas than these, but the principle is the same. option, individual enumeration units are highlighted
Looking at the boxmap in Color Plate 9 (the window in order from low to high data values, but only one
labeled “BoxMap . . .”), we can see that the value for highlighted unit appears on the map at any one time. In
Dade County is identified as an outlier, whereas Madison the Cumulative option, highlighted values remain high-
County is not similarly identified.This should make sense lighted until the end of the animation. In our opinion,
if you consider the histogram shown below the maps, the Cumulative option seems most useful because it is
where you can see that Dade County (highlighted in apparent which enumeration units have already been
yellow) is distinctly different from the rest of the data. animated. One limitation, however, is that all animated
GeoDa includes a range of approaches for visualiz- units are highlighted in the same color, making it diffi-
ing and exploring bivariate and multivariate data. For cult to remember the pattern of animation. In any case,
example, imagine that you wish to analyze the relation- it is interesting that the developers of GeoDa seemed
ship between percent foreign born and percent Catholic to find the sequencing approach useful. For a more
for counties in Florida in 1990. Color Plate 10 illustrates thorough discussion of GeoDa’s capabilities, see the
some of the maps and graphs that you might use in your paper by Anselin et al. (2006), which appeared in a spe-
analysis. Note first that we have displayed boxmaps of cial issue of Geographical Analysis focusing on a variety
each attribute (labeled as “PER_FOR_BOR” and of software for exploratory spatial data analysis.
“PER_CATHOLI”). This is reasonable, since the
boxmap is a variant of the quantile map, which is useful 3.8 HealthVisPCP
for map comparison. The overall patterns on these maps
suggest a similarity in the spatial distribution of these HealthVisPCP, developed by Robert Edsall (2003a;
attributes. Surprisingly, however, note that a scatterplot 2003b), is an extension of HealthVis, which was created
of the two attributes (shown to the lower left of the by Alan MacEachren and his colleagues (1998a) to
maps) does not fit a regression line very well. The rea- explore spatiotemporal health statistics data. We focus
son for this lack of fit is that the distribution of both on HealthVisPCP because Edsall was one of the first
attributes is positively skewed. The skew for percent geographers to experiment with parallel coordinate
foreign born was illustrated in the histogram in Color plots (PCPs). The basic interface for HealthVisPCP is
Plate 9, whereas the skew for percent Catholic is shown shown in Color Plate 11. In the upper left, we see a
in the histogram in Color Plate 10. When we transform bivariate choropleth map of heart disease mortality
both attributes (performing a square root transforma- rates (for white men) for two time periods (1980 and
tion for percent Catholic and a log transformation for 1992) in health service areas of the United States (an
percent foreign born), the result is a much more logical associated legend is shown to the right of the map in a
regression line (see the right-hand scatterplot in Color separate window). Below the legend is a scatterplot of
Plate 10). the heart disease attributes shown on the map; note that
In addition to being able to easily manipulate maps class breaks used on the bivariate map are also depicted
and graphs, a key aspect of GeoDa is the availability of on the scatterplot. Additionally, the scatterplot includes
dynamic linkages between maps, graphs, and tables. a faint histogram depicting the distribution of heart dis-
Thus, any object selected (or brushed) in one image is ease mortality for 1980, which is plotted along the x axis.
highlighted on all other images. For instance, we might In the lower left is the key feature of HealthVisPCP, a
highlight dots that do not fit a regression line (such as parallel coordinate plot, which provides a graphical por-
the three dots below the regression line in the raw scat- trayal not only of the attributes shown in the other win-
terplot in Color Plate 10) and see where these observa- dows, but also of additional attributes (those for heart
tions lie on other maps and graphics. Alternatively, we disease for white men in 1983, 1986, and 1989 are shown
might define a rectangle on a map; as we brush the rec- here, but others can be seen by moving the slider bar in
tangle over the map, enumeration units falling within the PCP Panner window). Note that the colors on the
this rectangle will be highlighted on the map and on all PCP correspond to those on the bivariate choropleth
other maps, graphs, and tables. Remember that we map, and thus enable the reader to see how certain
argued in section 2 that such linkages are a key compo- groups of health service areas change over time.
nent of data exploration software. When two attributes are highly positively corre-
Another intriguing aspect of GeoDa is the Map lated, the lines crossing the attribute axes on the PCP
Movie option, which provides a variant of the sequenc- will tend to be parallel. If we look at the left side of the
ing approach previously described for ExploreMap PCP in Color Plate 11, we see that this generally is the
(section 3.2). In the case of GeoDa, individual enumer- case here, indicating that the rank of health service
ation units (as opposed to classes of data) are areas in terms of heart disease mortality has remained
12:31:15.
Data Exploration
relatively constant over time. The general downward The upper left portion of Color Plate 12 portrays a
trend of the lines also indicates that heart disease mor- scatterplot of the percent of voters for George Bush
tality was less of a problem in 1992 than in 1980 (note (pct_Bush) against the percent of population claiming
that the range of values on the vertical axes is identical, adherence to Catholicism (pct_Cath). A negative
enabling the different temporal attributes to be readily correlation of -0.209 is seen for this pairing.* One can
compared). Although the overall trend in heart disease also see that there are many green and purple areas on
mortality is downward, we also note in the scatterplot the bivariate choropleth map (the right-hand portion of
that there are some health service areas that do not fit the figure), but few dark gray areas—indicating that few
this trend as well as others, and thus we might want to areas are high in both percent Catholic and percent who
focus on those. In the scatterplot, we have selected two voted for George Bush. The parallel coordinate plot in
of them (in red), and they have been highlighted in the the lower left of Color Plate 12 shows six attributes;
other windows. Note that in the PCP, these are lines from left to right, they are percent Baptist (pct_Bapt),
that indicate an increase in heart disease mortality percent votes for Bush (pct_Bush), percent Catholic
from 1980 to 1992. If additional data on health service (pct_Cath), percent votes for Kerry (pct_Kerry), per
areas were available, we would want to examine these capita income (pcincome), and percent female head of
two areas to see what might have created these poten- household (pctfemhh). Correlation values are plotted
tial “hot-spots.” in between the parallel coordinate axes; as before, we
can see that Bush voters are negatively correlated with
3.9 ESTAT Catholic adherents, but we can also see that Bush voters
are positively correlated with Baptist adherents.
ESTAT, the Exploratory Spatio-Temporal Analysis An alternative way of looking at the 2004 election
Toolkit (http://www.geovista.psu.edu/ESTAT/), is an data is to plot the percentage of Kerry voters against the
interactive geovisualization environment that was built percentage of adherence to Baptist sects. We would find
by the Penn State GeoVISTA Center (http://www. that the visualizations are more or less inverted from the
geovista.psu.edu/index.jsp). ESTAT was developed and pct_Bush and pct_Catholic combination, and a
designed to support exploratory cancer epidemiology pct_Kerry and pct_Baptist bivariate map would show a
for health analysts at the National Cancer Institute. It is few areas that are high in both attributes. Selecting the
intended to help users explore complex multivariate few areas that are highest in both attributes (i.e., the top
data, to create new hypotheses, and to refine hypothe- quantile in each case), we see that the bivariate map in
ses that may already exist. ESTAT features a scatter- Color Plate 13 reveals that the counties are all in the
plot, parallel coordinate plot, bivariate choropleth map, South, particularly along the southern Mississippi River.
and time-series graph. Each of these tools is dynami- Using the parallel coordinate plot, it is possible to
cally linked to one another to allow for rich interaction filter this pattern a bit further. In Color Plate 14, the per-
and real-time cross-filtering. cent of voters for the Kerry axis has been brushed from
The ESTAT application is based on the open-source its highest value to just slightly above 50 percent, the
GeoVISTA Studio (http://www.geovistastudio.psu.edu/) threshold necessary to win any county. This filters the
platform (Takatsuka and Gahegan 2002), a codeless, map, scatterplot, and parallel coordinate plot in the same
Java-based visual programming environment designed way. In the parallel coordinate plot, one can read across
to support the construction of geovisualization toolkits. (from left to right) to view how this selection plays out
In GeoVISTA Studio, applications are created by wiring across the attributes.The winning Kerry counties had low
together individual self-contained tools called Java percentages of Baptist adherents. These counties are also
Beans. A wide range of geographic visualization tools generally lower in their share of Bush voters (something
are available, including interactive maps, scatterplots, we should expect to see). Next, winning Kerry counties
data-mining tools, and spatial analysis methods. have a wide range of percentage of Catholic adherents,
In addition to health applications, ESTAT can be and per capita income is also quite diverse. Finally, a pos-
applied to virtually any type of multivariate spatial itive correlation appears (i.e., the value 0.46 appears in
data. Here we use ESTAT to explore voting patterns green between the axes in the parallel coordinate plot)
from the 2004 presidential election as they relate to pat- between percent female head of household and the
terns of religious adherence, as measured by the
Glenmary Research Center’s 2000 survey of religion
(Jones et al. 2002) and various other socioeconomic
* Itis important to note that the correlation value of -0.209 applies to
attributes from the 2000 U.S. Census. Color Plates 12 to
the more than 3,000 counties within the United States. If the technique
14 show several ESTAT images associated with these of geographically weighted regression were used (Fotheringham,
data. Because there are no time-series data in this Brunsdon, and Charlton 2002; Mennis 2006), we might find that the
example, the time-series plot is left out of these figures. correlation value varies spatially across the United States.
475
12:31:15.
Data Exploration
Kerry winning counties, potentially explaining some of forecast data and interesting historical data sets can be
the wins Kerry had in the South, despite that region’s accessed.
tendency to be Baptist and, therefore, not positively As another illustration of IDV’s capability, consider
correlated with high proportions of Kerry voters. The Color Plate 16, which portrays the Globe Display option,
county highlighted in Color Plate 14 appears as a red dot in which the Earth is displayed as seen from space. Here
in the scatterplot, as a red line in the parallel coordinate two attributes are displayed: wind speeds in excess of 55
plot, and as a tooltip in the bivariate map. By brushing meters per second and air temperature near the Earth’s
over different counties in any of these three views, you surface. In this case, wind speed is represented by an
can reveal the detailed data values for any particular isosurface, which is the 3-D analog of a contour line in a 2-
place. D plot. By clicking on “windspeed – Isosurface” in the leg-
end, we could specify higher wind speeds, which would
3.10 IDV occur within the core of the isosurfaces shown in the color
plate.Temperature in this case is represented by the Color-
Integrated Data Viewer (IDV), distributed by the Filled Contour symbolization used in Color Plate 15. In
Unidata Program Center (http://www.unidata.ucar.edu/ ), both of these illustrations, we have used the default spec-
is primarily intended for exploring three-dimensional tral color scheme (red, orange, yellow, green, and blue)
meteorological data, although it can be used to visualize provided by IDV. Users, do, however, have the option of
any three-dimensional geoscientific data, such as oceano- selecting alternative color schemes. As with the previous
graphic data (Murray et al. 2006). To understand the example, users can rotate the image to achieve different
character of IDV, it is useful to consider its precursor, views of the Earth, which is critical with the Globe
Vis5D. The “5-D” portion of Vis5D comes from the Display since the entire Earth cannot be seen at once.
notion that meteorological data have three spatial Alternatively, one can select an Auto-rotate option, and
dimensions (i.e., latitude, longitude, and altitude), a time see the Earth rotate automatically. And of course one can
dimension, and a dimension for the attributes to be dis- animate the attributes displayed, as we described for
played (e.g., pressure, temperature, and wind speed). the previous example; in the present case, since we have
Vis5D could display any data that could be expressed in two attributes, both would change simultaneously. In fact,
the 5-D format; for example, oceanographic data have one can select both the Auto-rotate and animation
three spatial dimensions (latitude, longitude, and depth), options: the result produces a rather complex visualiza-
a time dimension, and multiple attributes (e.g., tempera- tion, but we can imagine that it could be extremely useful
ture, salinity, and current speed; Hibbard and Santek to the experienced meteorologist.
1990). IDV shares the 5-D character of Vis5D while
enabling us to work in the Windows-based systems that 3.11 Other Exploration Software
we typically have on our desktops (whereas Vis5D was
intended for sophisticated workstation environments). From a historical perspective, some of the more inter-
To illustrate some of IDV’s capabilities, consider esting software includes Miller’s (1988) Great
Color Plate 15, which shows a time slice of wind speed at American History Machine (GAHM), MacDougall’s
an elevation above the surface where we typically find (1992) Polygon Explorer, and Roth et al.’s work with
the jet stream. In Color Plate 15A, wind speed is Visage, SAGE, and SDM (1997). Each of these is sum-
depicted by a “Color-Filled Contour,” whereas in Color marized briefly in the Further Reading section. The
Plate 15B, a “Contour Plan View” is used (note the leg- notion of information visualization is closely tied to the
end in the lower right). The Color-Filled Contour sym- ideas of geovisualization and data exploration. A con-
bolization is more useful for visualizing wind speed, but cise overview of information visualization can be found
the Contour Plan View allows us to easily see the under- in Keim et al. (2005), and more complete information
lying topography. The 3-D character of the image can be found in Card et al. (1999), Bederson and
becomes more obvious when you use the mouse to Shneiderman (2003), and Ware (2004). Examples of
rotate the image. Since this particular time slice comes work relevant to cartography include publications by
from a three-day forecast at six-hour intervals, it is pos- Shneiderman et al. (1999), Fredrikson et al. (2003), and
sible to animate the wind speed layer by using the but- Dang et al. (2003). Dang et al.’s paper discusses
tons to the upper right of the images. IDV allows users Dynamaps, a dynamic mapping system that is available
to connect to remote servers so that the most recent via the Web (http://www.cs.umd.edu/hcil/census/).
12:31:15.
Data Exploration
SUMMARY CommonGIS is its ability to assist the user in selecting an

appropriate method of symbolization. Other key features
Data exploration, like the larger notion of geographic include eliminating the effect of outliers by extending a
visualization, is a private activity in which unknown color scheme over only a portion of the data set, varying
spatial patterns are revealed in a highly interactive envi- the break point of a diverging color scheme (and seeing
ronment. Data exploration is achieved by using a vari- the result depicted dynamically), and using a value flow
ety of methods, including varying the symbolization diagram to display changes in an attribute over a time
(i.e., moving the break point separating two colors in a interval on a single map. Other exploration software,
diverging color scheme); highlighting subsets of data via however, provides a variety of useful features. For
either focusing (for a subrange of the data) or brushing instance, MapTime enables users to explore spatiotem-
(for an arbitrary set of spatial entities); creating multi- poral data at fixed point locations via animation, small
ple views (e.g., the small multiple); using animation; and multiples, and change maps; GeoDa enables the creation
linking with other forms of display (e.g., tabular and of percentile maps and boxmaps, and provides fully
graphical displays). The methods for data exploration dynamic linkages among maps, graphs, and tables;
continue to evolve—for instance, consider the notion of ESTAT features dynamic linkages among a scatterplot,
data mining, which enables patterns to be detected parallel coordinate plot, and bivariate choropleth map;
automatically. and IDV enables users to explore true 3-D data that have
In this chapter, we have looked at a variety of software both temporal and multivariate components. Since a full
that has been developed for exploring spatial data. One appreciation of data exploration requires that you inter-
program particularly illustrative of data exploration act with the data, we encourage you to experiment with
capabilities is CommonGIS. An interesting feature of several of the software packages that we have described.
FURTHER READING
Andrienko, G. L., and Andrienko, N. V. (1999) “Interactive Geographical Analysis 38, no. 1, 2006 (entire issue).
maps for visual data exploration.” International Journal of This special issue focuses on recent advances in software for
Geographical Information Science 13, no. 4:355–374. exploratory spatial data analysis.
Discusses the development of the Descartes software; many ideas Heywood, I., Oliver, J., and Tomlinson, S. (1995) “Building an
developed for Descartes have been implemented in CommonGIS. exploratory multi-criteria modelling environment for spatial
Andrienko, N., and Andrienko, G. (2006) Exploratory Analysis decision support.” In Innovations in GIS 2, ed. by P. Fisher,
of Spatial and Temporal Data: A Systematic Approach. Berlin: pp. 127–136. Bristol, PA: Taylor & Francis.
Springer-Verlag.
Describes the development of specialized software that can assist
An extensive treatise on techniques that the Andrienkos have in exploring data commonly analyzed within a GIS.
developed for exploring spatial and spatiotemporal data. The bulk
Jankowski, P., Andrienko, N., and Andrienko, G. (2001) “Map-
of illustrations in the book were produced using the CommonGIS
system.
centered exploratory approach to multiple criteria spatial
decision making.” International Journal of Geographical
Card, S. K., Mackinlay, J. D., and Shneiderman, B., eds. (1999) Information Science 15, no. 2:101–127.
Readings in Information Visualization: Using Vision to Think.
San Francisco, CA: Morgan Kaufmann. Presents a sophisticated approach for exploring attributes
involved in decision making.
A collection of readings on information visualization; also see
Bederson and Shneiderman (2003) and Ware (2004). Keim, D. A., Panse, C., and Sips, M. (2005) “Information visu-
alization: Scope, techniques and opportunities for geovisual-
Dykes, J. A. (1997) “Exploring spatial data representation with
ization.” In Exploring Geovisualization, ed. by J. Dykes, A. M.
dynamic graphics.” Computers & Geosciences 23, no.
MacEachren, and M.-J. Kraak, pp. 23–52. Amsterdam: Elsevier.
4:345–370.
Provides a concise overview of information visualization relevant
Discusses using the programming language Tcl/Tk to create
to geovisualization.
Project Argus.
Kraak, M.-J. (1998) “Exploratory cartography: Maps as tools
Edsall, R. M. (2003b) “The parallel coordinate plot in action:
for discovery.” ITC Journal 1, no. 1:46–54.
Design and use for geographic visualization.” Computational
Statistics and Data Analysis 43, no. 4: 605–619. Considers numerous issues related to data exploration.
Describes using the parallel coordinate plot to explore data; MacDougall, E. B. (1992) “Exploratory analysis, dynamic statisti-
for more advanced work, see Andrienko and Andrienko cal visualization, and geographic information systems.” Carto-
(2005). graphy and Geographic Information Systems 19, no. 4:237–246.
477
12:31:15.
Data Exploration
Describes the data exploration software Polygon Explorer, which Questions the ethics of using one map to display a set of spatial
was developed for handling both univariate and multivariate data. data and proposes data exploration as one of six solutions to
Readers with knowledge of cluster analysis will find the capability showing only one map.
to handle multivariate data intriguing. Roth, S. F., Chuah, M. C., Kerpedjiev, S., and Kolojejchick, J. A.
MacEachren, A. M., and Ganter, J. H. (1990) “A pattern iden- (1997) “Toward an information visualization workspace:
tification approach to cartographic visualization.” Carto- Combining multiple means of expression.” Human-Computer
graphica 27, no. 2:64–81. Interaction 12, no. 1/2:131–185.
A classic paper on the need for visualization and data exploration Covers several sophisticated pieces of software (Visage, SAGE,
in discovering patterns in spatial data. and SDM) for exploring spatial data.
Miller, D. W. (1988) “The great American history machine.” Slocum, T. A., Yoder, S. C., Kessler, F. C., and Sluter, R. S.
Academic Computing 3, no. 3:28–29, 43, 46–47, 50. (2000) “MapTime: Software for exploring spatiotemporal
Summarizes the data exploration software GAHM, which was data associated with point locations.” Cartographica 37,
developed to encourage undergraduates to think like professional no. 1:15–31.
historians. Provides a detailed discussion of the development of MapTime
Monmonier, M. (1991b) “Ethics and map design: Six strate- and suggests potential improvements. For a user evaluation of
gies for confronting the traditional one-map solution.” MapTime, see Slocum et al. (2004).
Cartographic Perspectives no. 10:3–8.
GLOSSARY
boxmap: a variant of quartile data classification in which key frames: in an animation, the frames (or maps) associ-
outliers are placed in classes by themselves; thus, six rather ated with collected data; contrast with intermediate frames.
than four classes are used. percentile map: a method of data classification in which
brushing: when a user is able to highlight an arbitrary set of class limits are adjusted to accentuate extreme values; six
spatial entities while exploring data; for instance, a user might classes are used and the limits are <1%, 1% to <10%, 10% to
display the percentage of Hispanics only for tracts that voted <50%, 50% to <90%, 90% to <99%, and >99%.
for a particular political party. proactive graphics: software that enables users to initiate
data exploration: examining data in a variety of ways to queries using icons (or symbolic representations) as opposed
develop different perspectives of the data; for example, we to words.
might view a choropleth map in both its unclassed and its sequencing: displaying a map piece by piece (e.g., displaying
classed forms. Data exploration is sometimes equated with the title, geographic base, legend title, and then each class in
geovisualization. the legend).
focusing: when exploring data, a user is able to highlight a small multiple: many small maps are displayed to show the
subrange of numeric values; for example, a user might focus change in an attribute over time or to compare many attrib-
on census tracts that are less than 25 percent Hispanic. utes for the same time period.
intermediate frames: in an animation, the frames (or maps) value flow diagram: a graph that displays change in an
associated with interpolated data; contrast with key frames. attribute over a time interval; multiple value flow diagrams
isosurface: a surface of equal value within a true 3-D phe- permit the analysis of spatiotemporal data on a single static
nomenon (e.g., a surface representing a wind speed of 100 map.
kilometers per hour in the jet stream).
REFERENCES
Andrienko, G. L., and Andrienko, N. V. (1999) “Interactive Andrienko, N., and Andrienko, G. (2006) Exploratory
maps for visual data exploration.” International Analysis of Spatial and Temporal Data: A Systematic
Journal of Geographical Information Science 13, Approach. Berlin: Springer-Verlag.
no. 4:355–374. Anselin, L., Syabri, I., and Kho, Y. (2006) “GeoDa: An intro-
Andrienko, G., and Andrienko, N. (2005) “Blending aggega- duction to spatial data analysis.” Geographical Analysis 38,
tion and selection: Adapting parallel coordinates for the no. 1:5–22.
visualization of large datasets.” The Cartographic Journal Bederson, B. B., and Shneiderman, B. (eds.). (2003) The Craft
42, no. 1:49–60. of Information Visualization: Readings and Reflections. San
Andrienko, N., and Andrienko, G. (2004) “Interactive visual Francisco, CA: Morgan Kaufmann.
tools to explore spatio-temporal variation.” Proceedings of Buja, A., Cook, D., and Swayne, D. F. (1996) “Interactive
the Working Conference on Advanced Visual Interfaces, high-dimensional data visualization.” Journal of
Gallipoli, Italy, pp. 417–420. Computational and Graphical Statistics 5, no. 1:78–99.
12:31:15.
Data Exploration
Buttenfield, B. P., and Weber, C. R. (1994) “Proactive Hibbard, B., and Santek, D. (1990) “The VIS-5D system for
graphics for exploratory visualization of biogeographical easy interactive visualization.” Proceedings, Visualization
data.” Cartographic Perspectives, no. 19:8–18. ’90, San Francisco, CA, pp. 28–35.
Campbell, C. S., and Egbert, S. L. (1990) “Animated cartogra- Jelinski, D. E. (1987) “Intraspecific diversity in trembling
phy: Thirty years of scratching the surface.” Cartographica aspen in Waterton Lakes National Park, Alberta: A bio-
27, no. 2:24–46. geographical perspective.” Unpublished PhD dissertation,
Card, S. K., Mackinlay, J. D., and Shneiderman, B. (eds.). Simon Fraser University, Burnaby, British Columbia,
(1999) Readings in Information Visualization: Using Vision Canada.
to Think. San Francisco, CA: Morgan Kaufmann. Jones, D. E., Doty, S., Grammich, C., Horsch, J. E., Houseal,
Crampton, J. W. (2002) “Interactivity types in geographic R., et al. (2002) Religious Congregations and Membership
visualization.” Cartography and Geographic Information in the United States: 2000. Nashville, TN: Glenmary
Science 29, no. 2:85–98. Research Center.
Dang, G., North, C., and Shneiderman, B. (2003) “Dynamic Keim, D. A., Panse, C., and Sips, M. (2005) “Information
queries and brushing on choropleth maps.” In The Craft of visualization: Scope, techniques and opportunities for
Information Visualization: Readings and Reflections, ed. by geovisualization.” In Exploring Geovisualization, ed. by J.
B. B. Bederson and B. Shneiderman, pp. 38–45. San Dykes, A. M. MacEachren, and M.-J. Kraak, pp. 23–52.
Francisco, CA: Morgan Kaufmann. Amsterdam: Elsevier.
Dykes, J. (1995) “Cartographic visualization for spatial analy- Kraak, M.-J. (1998) “Exploratory cartography: Maps as tools
sis.” Proceedings of the 17th International Cartographic for discovery.” ITC Journal 1, no. 1:46–54.
Conference, Volume 1, Barcelona, Spain, pp. 1365–1370. MacEachren, A. M., Boscoe, F. P., Haug, D., and Pickle, L.
Dykes, J. (1996) “Dynamic maps for spatial science: A unified W. (1998a) “Geographic visualization: Designing manipu-
approach to cartographic visualization.” In Innovations in lable maps for exploring temporally varying georefer-
GIS 3, ed. by D. Parker, pp. 177–187, Color Plates 4–5. enced statistics.” Proceedings, Information Visualization
Bristol, PA: Taylor & Francis. ’98, IEEE Computer Society, Research Triangle Park,
NC, pp. 87–94.
Dykes, J. A. (1997) “Exploring spatial data representation
with dynamic graphics.” Computers & Geosciences 23, no. MacEachren, A. M., Wachowicz, M., Edsall, R., and Haug, D.
4:345–370. (1999c) “Constructing knowledge from multivariate spa-
tiotemporal data: Integrating geographical visualization
Egbert, S. L., and Slocum, T. A. (1992) “EXPLOREMAP: An
with knowledge discovery in database methods.”
exploration system for choropleth maps.” Annals of the
International Journal of Geographical Information Science
Association of American Geographers 82, no. 2:275–288.
13, no. 4:311–334.
Edsall, R. M. (2003a) “Design and usability of an enhanced
Mennis, J. (2006) “Mapping the results of geographically
geographic information system for exploration of multi-
weighted regression.” The Cartographic Journal 43,
variate health statistics.” The Professional Geographer 55,
no. 2:171–179.
no. 2:146–160.
Miller, D. W. (1988) “The great American history machine.”
Edsall, R. M. (2003b) “The parallel coordinate plot in action:
Academic Computing 3, no. 3: 28–29, 43, 46–47, 50.
Design and use for geographic visualization.”
Computational Statistics and Data Analysis, 43, no. Moellering, H. (1978) “A demonstration of the real time
4:605–619. display of three dimensional cartographic objects.”
Videotape. Columbus: Department of Geography, Ohio
Egbert, S. L. (1994) “The design and evaluation of an interac-
State University.
tive choropleth map exploration system.” Unpublished
PhD dissertation, University of Kansas, Lawrence, KS. Moellering, H. (1980) “The real-time animation of three-
dimensional maps.” The American Cartographer 7,
Fotheringham, A. S., Brunsdon, C., and Charlton, M. (2002)
no. 1:67–75.
Geograhically Weighted Regression: The Analysis of
Spatially Varying Relationships. West Sussex, England: Murray, D., McWhirter, J., Wier, S., and Emmerson, S. (2006)
John Wiley. “The Integrated Data Viewer (IDV) – A Web-enabled
application for scientific analysis and visualization.” 19th
Ferreira, J. J., and Wiggins, L. L. (1990) “The density dial: A
International Conference on IIPS for Meteorology,
visualization tool for thematic mapping.” Geo Info Systems
Oceanography, and Hydrology.
1, no. 0:69–71.
http://ams.confex.com/ams/pdfpapers/57870.pdf.
Fredrikson, A., North, C., Plaisant, C., and Shneiderman, B.
Patton, D. K., and Cammack, R. G. (1996) “An examination
(2003) “Temporal, geographical and categorical aggrega-
of the effects of task type and map complexity on
tions viewed through coordinated displays: A case study
sequenced and static choropleth maps.” In Cartographic
with highway incident data.” In The Craft of Information
Design: Theoretical and Practical Perspectives, ed. by C. H.
Visualization: Readings and Reflections, ed. by B. B.
Wood and C. P. Keller, pp. 237–252. Chichester, England:
Bederson and B. Shneiderman, pp. 22–30. San Francisco,
Wiley.
CA: Morgan Kaufmann.
479
12:31:15.
Data Exploration
Shneiderman, B. (1999) “Dynamic queries for visual informa- Taylor, D. R. F. (1987) “Cartographic communication on
tion seeking.” In Readings in Information Visualization: computer screens: The effect of sequential presentation of
Using Vision to Think, ed. by S. K. Card, J. D. Mackinlay, map information.” Proceedings of the 13th International
and B. Shneiderman, pp. 236–243. San Francisco, CA: Cartographic Conference, Volume 1, Morelia, Mexico, pp.
Morgan Kaufmann. 591–611.
Slocum, T. A., Robeson, S. H., and Egbert, S. L. (1990) Ware, C. (2004) Information Visualization: Perception for
“Traditional versus sequenced choropleth maps: An exper- Design (2nd ed.). San Francisco, CA: Morgan Kaufman.
imental investigation.” Cartographica 27, no. 1:67–88. Yoder, S. C. (1996) “The development of software for
Takatsuka, M., and Gahegan, M. (2002) “GeoVISTA Studio: exploring temporal data associated with point locations.”
A codeless visual programming environment for geoscien- Unpublished MA thesis, University of Kansas,
tific data analysis and visualization.” Computers & Lawrence, KS.
Geosciences 28, no. 10:1131–1144.
12:31:15.
Visualizing Uncertainty
OVERVIEW
We often think of maps as truthful representations of most commonly used, although several software
reality. For example, imagine viewing a state-level choro- packages have been developed for exploring uncer-
pleth map of the United States entitled “Percent of the tainty. Section 3 considers various visual variables that
Population That Exercises Regularly.” If a shade in the can be utilized to depict uncertainty. Broadly speaking,
legend has class limits of 21 to 25 percent, then we are apt we can group these into intrinsic and extrinsic visual
to presume that states with that shade do, in fact, fall in the variables, depending on whether the variable is visually
21 to 25 percent range. In reality, this presumption proba- separable from the variable depicting the actual attribute
bly would not be correct because the map likely would be of interest (extrinsic variables are separable, whereas
based on a sample—thus, there would be a margin of intrinsic variables are not). We will see that although
error around the sampled value. Such error in maps is a some traditional visual variables (e.g., size and shape)
form of uncertainty; in this chapter, we consider can be utilized, specialized visual variables have been
approaches for visualizing uncertainty. developed for depicting uncertainty, including crispness,
In section 1, we consider basic elements of uncertainty. resolution, and transparency.
We will see that uncertainty can arise from a variety of In section 4, we cover applications for visualizing
sources, including the raw data on which a map is based, uncertainty. By looking at numerous applications, we
the manner in which these data are processed, and the hope to give you a feel for the broad range of approaches
manner in which the visualization is created. Although that have been developed for visualizing uncertainty.
the term uncertainty is frequently used in the cartographic Although cartographers and other graphic designers have
literature, the terms “reliability” and “quality” are also created a multitude of approaches for visualizing uncer-
used (i.e., if data are uncertain, they are also unreliable tainty, you might find yourself asking whether these meth-
and of poor quality). The National Institute of Standards ods are effective: Can users understand the notion of
has proposed five categories for assessing data quality: uncertainty and its various visual depictions, and, maybe
lineage, positional accuracy, attribute accuracy, more important, can they make reasonable decisions
completeness, and logical consistency. The categories based on this uncertainty? In this context, section 5 deals
of positional accuracy and attribute accuracy have with some studies explicitly developed to examine the
received the most attention in the literature pertaining to effectiveness of methods for depicting uncertainty.
visualization.
In section 2, we briefly describe the general ways in
which uncertainty can be depicted: Separate maps are 1 BASIC ELEMENTS OF UNCERTAINTY
created for an attribute and its associated uncertainty
(i.e., maps are compared); the attribute and its uncer- Uncertainty can arise from a variety of sources, including
tainty are displayed on the same map (i.e., maps are the raw data on which a map is based, the manner in
combined); and data exploration tools are utilized. which these data are processed, and the manner in which
Generally, the “maps combined” approach has been the visualization is created (Pang et al. 1997). As an
481
12:32:38.
example of uncertainty in the raw data, consider the includes “information about selection criteria, definitions
attribute “percent foreign born,” which is commonly pro- used and other relevant mapping rules. For example, geo-
vided by the U.S. Bureau of the Census. If we see a value metric thresholds such as minimum area or minimum
of 25.6 percent foreign born in a census publication for a width must be reported” (National Institute of Standards
particular census tract, we are apt to think that this is the and Technology 1994, 24). These five categories and two
“correct” value for the tract, but an attribute such as this additional ones (semantic accuracy and temporal infor-
is based on sampling approximately one of every six mation) were described in detail in Elements of Data
housing units (U.S. Bureau of the Census 1994, A-2), and Quality, which was published by the International
thus values are estimates of population values. Moreover, Cartographic Association (ICA) Commission on Spatial
such values are recorded at a particular point in time, Data Quality (Guptill and Morrison 1995). The cate-
and, of course, are subject to errors in estimating the size gories of positional accuracy and attribute accuracy have
of the population itself. As an example of uncertainty in received the most attention in the literature pertaining to
processing data, consider the problem of interpolating visualization, and thus we focus on them here.†
values between known control points. There are a variety Our focus in this chapter is on methods for visualiz-
of possible algorithms, each producing a potentially dif- ing the uncertainty in spatial data. You should be aware
ferent set of interpolated values. Thus, for any particular that sometimes you might need to purposely create
location on a map, we can think of the set of interpolated uncertainty in the data. For example, you may wish to
values as constituting uncertainty in the data. Finally, an map the locations of deaths due to certain diseases or
example of uncertainty in the visualization is the various crimes, but find that there are laws protecting the confi-
approaches that are used to illuminate 3-D scenes (Pang dentiality rights of the victims. Along these lines,
et al. 1997). Michael Leitner and Andrew Curtis (2004, 2006) have
Although the term uncertainty is commonly used in developed guidelines for masking the locations of con-
the literature to describe the potential variation in values fidential point data and thus purposely introduced
of an attribute at a spatial location, the terms quality and uncertainty into the point data.
reliability are also frequently used: Uncertainty is equated
with unreliability and poor quality. The term quality 2 GENERAL METHODS FOR DEPICTING
should be used with care, as it has been used to refer to a UNCERTAINTY
variety of data characteristics. In particular, the U.S.
Federal Information Processing Standard 173 (FIPS 173,
Alan MacEachren (1992) suggested three general
National Institute of Standards and Technology 1994)
methods for depicting uncertainty. First, individual
lists five categories for assessing data quality: lineage,
maps can be shown for both an attribute and its associ-
positional accuracy, attribute accuracy, completeness, and
ated uncertainty. For example, we might create one map
logical consistency.* Lineage refers to the historical
showing the “Percent of the Population That Exercises
development of the digital data. For example, if the data
Regularly” and a second map showing a confidence
were acquired from a paper map, you might want to
interval for each enumeration unit depicted on the first
know the scale and projection of the map, whether digi-
map. Second, the attribute and its uncertainty can be
tizing or scanning was used to generate the digital data,
displayed on the same map—given the appropriate
and when the data were converted from analog to digital
visual variables (e.g., gray tones for one and parallel
form. Positional accuracy refers to the locational accu-
lines for the other), we can conceive of overlaying one
racy of geographic features, both horizontally and verti-
map on the other. Note that these first two methods cor-
cally. On a USGS topographic sheet, you might wish to
respond to the “maps compared” and “maps combined”
examine the accuracy of a stream’s position or the spot
approaches in bivariate and multivariate mapping. The
height of a mountain. Attribute accuracy refers to the
third method is to use interactive data exploration tools
accuracy of features found at particular locations. For
that allow us to easily manipulate the display of both
example, in remote sensing, you might be interested in
the attribute and its uncertainty. For example, the user
the accuracy of a land use/land cover classification.
might be permitted to toggle back and forth between
Logical consistency describes “the fidelity of relation-
the display of the attribute and its uncertainty, or be
ships encoded in the data structure of the digital spatial
permitted to view an animation reflecting uncertainty
data” (National Institute of Standards and Technology
in the data. These methods will be further illustrated in
1994, 23). For instance, we might ask whether all poly-
section 4, where we consider examples of efforts to visu-
gons close correctly (in GIS terminology, this would be
alize uncertainty.
termed topological correctness). Finally, completeness
*For the most recent version of the standards, see http://mcmcweb.er. † For other ways of categorizing data quality, see MacEachren et al.
usgs.gov/sdts/standard.html. (2006, 142–146).
12:32:38.
3 VISUAL VARIABLES FOR DEPICTING

UNCERTAINTY
If we assume the “maps combined” approach, visual

variables for depicting uncertainty can be broken down
into two broad groups: intrinsic and extrinsic (Gershon
1998). As the name implies, intrinsic visual variables are
intrinsic to the display; for example, to depict uncer-
tainty, we might vary the saturation of colored tones on
a choropleth map. Extrinsic visual variables involve
adding objects to the display, “such as dials, thermome-
ters, arrows, bars, [and] objects of different shapes”
(Gershon, 44). Because many of the approaches devel-
oped by cartographers appear to have involved intrinsic
visual variables, we consider those first.
One approach for depicting uncertainty is to utilize
the basic visual variables. In some cases these visual
variables are suitable, whereas in other cases they can
be confusing. A suitable usage would be Tissot’s indica-
trix (distortion ellipse), in which the visual variables size
FIGURE 2 The visual variable size (e.g., the width of a line)

depicts data uncertainty associated with stream position. Here
the level of data uncertainty might be misinterpreted as
stream discharge. (After McGranaghan 1993, 17.)
and shape represent the ability of various map projec-

tions to maintain correct size and angular relationships
at point locations (see Figure 1). Potentially confusing
would be utilizing the visual variable size to depict the
uncertainty of stream position (e.g., using a wider line
to indicate greater uncertainty), because a wide line
normally would be associated with greater discharge
(see Figure 2).
Of all the visual variables present, MacEachren
(1992) argued that saturation is particularly logical for
depicting uncertainty, with “pure hues used for very
certain information [and] unsaturated hues for uncer-
tain information” (14). Referencing the work of Brown
and van Elzakker (1993), MacEachren (1995, 440–441)
noted that three levels of saturation can be used for up
to 12 individual hues. Although portraying only three
levels of data uncertainty might appear to be limiting,
an analyst could utilize data exploration tools to apply
the three levels to only a portion of the data.
In addition to the standard visual variables,
MacEachren (1995, 275–276) proposed the visual
FIGURE 1 The visual variables size and shape represent the variable clarity, which, he argued, could be subdivided
ability of various map projections to maintain correct size and into three additional visual variables: crispness, reso-
angular relationships at point locations. Here the visual variables lution, and transparency. Crispness refers to the
are easily understood. (Abler, Ronald F., et al., eds., Geography’s sharpness of boundaries (or area fills): A crisp bound-
Inner Worlds: Pervasive Themes in Contemporary American ary would represent reliable data, whereas a fuzzy
Geography. Copyright © 1992 by Rutgers, the State University. boundary would represent uncertain data. For exam-
Reprinted by permission of Rutgers University Press.) ple, Figure 3 illustrates the uncertain nature of the
483
12:32:38.
Where Is The Grits Line?
High Resolution
FIGURE 3 The visual variable crispness depicts the uncer-

tain nature of the boundary of grits in the United States.
boundary of where grits is consumed in the United

States (grits is a southern food made of corn meal).
Resolution is the level of detail in the spatial data
underlying an attribute, where a lower (coarser) reso-
lution equals greater uncertainty. For example, Figure
4 illustrates different resolutions for a raster database.
Transparency is the ease with which a theme can be
seen through a “fog” placed over that theme; reliable
data can be easily seen through the fog, whereas
uncertain data cannot (see Figure 5).* Low Resolution
Alex Pang and his colleagues (1997) at the Santa
Cruz Laboratory for Visualization & Graphics have
been major proponents of using extrinsic visual vari-
ables to depict uncertainty. Wittenbrink et al. (1996)
described their approach as “verity visualization,” as
it “suggests the quality of the data that is exactly what
it purports to be or is in complete accord with the FIGURE 4 Example of differing resolutions for a raster data-
facts.” Figure 6B is an example of a glyph that base; the low-resolution image would have greater uncer-
Wittenbrink et al. created to show the uncertainty in tainty. (© Copyright 1995 by the Guilford Press. Reprinted
wind speed direction and magnitude. Uncertainty in with permission.)
direction is indicated by width of the arrowhead, and
extra arrowheads indicate uncertainty in magnitude.
Note how this glyph compares with the traditional interpolation methods. They stated:
wind arrows shown in Figure 6A, which do not depict
uncertainty. Although their glyph might appear com- Smoothly interpolated surfaces, while visually pleasing and
plex, Wittenbrink and his colleagues showed that valuable for analysis of data, sometimes imply that the data
people could extract useful information from it. has greater spatial resolution than is the case; by actually
marking on the surface the location, value, and/or uncer-
Another example of glyphs is Helen Mitasova and
tainty of each measurement, the visualization contains
her colleagues’ (1993) efforts to map the uncertainty of much more valuable information. The use of glyphs helps
to highlight which sampling sites were most important in
* calculating the interpolated concentrations . . . . Areas
For additional discussion on visual variables for depicting uncer-
tainty, see McGranaghan (1993). An alternative visual variable that
where the glyphs are small and round indicate areas of low
he proposed is realism, which refers to the “photorealism of the uncertainty, while areas where the glyphs are elongated
image” (p. 10). indicate areas where the data is sparse.
12:32:38.
A
10
20
30
40
50
FIGURE 5 The visual variable transparency is used to depict

the uncertainty of the data. A light “fog” indicates reliable
data, whereas a dark fog indicates uncertain data. (After B
MacEachren 1992, 15.)
In the following section, we see additional ways in

which visual variables can be used to depict uncertainty.
4 APPLICATIONS OF VISUALIZING
UNCERTAINTY
4.1 Howard and MacEachren’s R-VIS Software
David Howard and Alan MacEachren (1996) devel-

oped the software R-VIS (Reliability VISualization) to
enable environmental scientists and policy analysts to
examine levels of dissolved inorganic nitrogen (DIN) in
Chesapeake Bay. Basic data input to R-VIS was a set of
49 DIN values collected at point locations by a scien-
tific vessel. Because DIN is presumed to be a smooth,
continuous phenomenon, the data for each point in
time could be contoured using the approaches triangu-
lation, inverse distance, and kriging. Howard and
FIGURE 6 (A) A traditional approach for depicting wind
MacEachren chose kriging for R-VIS because kriging
speed and direction; (B) Wittenbrink et al.’s glyph approach
provides an uncertainty map expressed as confidence
for depicting uncertainty: the width of the arrowhead indicates
intervals, in addition to a basic isarithmic map of the
uncertainty in direction, and extra arrowheads indicate uncer-
theme (DIN in the case of R-VIS).
tainty in speed. (After Wittenbrink, et al. (1996) “Glyphs for
The default display for R-VIS compared the
visualizing uncertainty in vector fields” IEEE Transactions on
interpolated DIN data for a particular point in time
Visualization and Computer Graphics 2, p. 272; © 1996 IEEE.)
with an associated uncertainty map. Initially, Howard
485
12:32:38.
and MacEachren chose a lightness scheme (light to

dark red) for the contoured data and a saturation
scheme (desaturated to saturated red) for the uncer-
tainty map, but the schemes “did not provide enough
contrast . . . with the saturation range being particu-
larly ineffective” (p. 68). Thus, they ultimately used
different hues for the two maps (red for the contour
map and blue for the reliability map) and varied both
the lightness and the saturation within each map
(Color Plate 1).
To combine the isarithmic and uncertainty maps,
two basic approaches were used in R-VIS. One was
termed the “overlay method,” in which area shading
was used for the isarithmic map and weighted isolines
were used for the uncertainty map (Figure 7). Howard
and MacEachren indicated that the overlay method
“emphasize[s] the data while allowing analysts to
check the reliability in map areas that seem to be par-
ticularly good or bad in terms of meeting the EPA
dissolved inorganic nitrogen targets” (p. 70). Another
approach for combining the isarithmic and uncertainty
maps in R-VIS, termed a “merged display,” involves
displaying each map with a unique color (or symbol).
For example, using the bivariate choropleth method,
the isarithmic map could be shown with a lightness
scheme and the uncertainty map with a saturation
scheme. Alternatively, a merged display could be cre-
ated by showing the isarithmic map as a lightness/sat-
uration scheme (or as shades of gray) and the uncer-
tainty map as varying amounts of fill within a raster
grid cell. Howard and MacEachren noted that merged
displays emphasize the relationship between the
isarithmic and uncertainty maps.
Several approaches also were available within
R-VIS for exploring uncertainty. One was to alternate
isarithmic and uncertainty maps at the same screen
location. This alternation could be under the control
of the user or it could be animated. With respect to
the animation approach, Howard and MacEachren
indicated, “it is with multiple viewing that relation- FIGURE 7 The “overlay method” within R-VIS. The isarith-
ships between data and reliability distributions (if mic map is shown in shades of gray, and the uncertainty map
there are any) begin to become apparent” (p. 71). A is shown as weighted isolines. (Courtesy of David Howard
second exploration approach was to highlight some and Alan MacEachren.)
portion of the data set by dynamically specifying the
range of either the isarithmic or the uncertainty val- 4.2 Fisher’s Animation of Dot Map Uncertainty
ues to be shown. For example, users might be inter-
ested in targeting areas that exceed a particular DIN Dot maps are sometimes problematic to interpret
concentration. A third exploration approach was to because the reader is inclined to think that the phe-
animate the isarithmic and uncertainty maps over nomenon being mapped is actually found where the
time. Although Howard and MacEachren did not dots are located. To illustrate, imagine a dot map of
release R-VIS for general use (Howard, personal bears in the United States, where each dot represents 25
communication 1997), you will find it useful to study bears. The naive viewer might assume that bears are
their description of the software, and thus consider found wherever dots are located on the map. Where
the capability that might be included in software dots are highly concentrated, there would be a high
packages that depict uncertainty. probability of bears being found, but where dots are
12:32:38.
dispersed, there would be a relatively low probability

(high uncertainty) that a bear would be found at the
location of a particular dot. Peter Fisher (1996) devel-
oped a method to depict this uncertainty via animation.
To understand Fisher’s approach, we must first
review how dot maps are constructed. The first step is
to delineate regions (polygons) within which the phe-
nomenon being mapped is located; ideally, this step con-
siders ancillary information that can assist in determin-
ing appropriate locations for dots. Second, decisions are
made concerning dot size and unit value (i.e., the count
represented by each dot). Third, dots are placed within
each region. Fisher followed these basic steps, with the
exception that he did not consider ancillary information
because of the slow computer hardware he used. First,
he created an initial dot map by placing dots randomly
within a minimum bounding rectangle that surrounded
each enumeration unit. A random dot falling within the
bounding rectangle and within the enumeration unit
was plotted on the map, whereas a dot falling within the
bounding rectangle but outside the enumeration unit
was not plotted. He then animated this initial map by
selecting one dot at random, deleting that dot, and
locating a replacement dot within the enumeration unit
containing the deleted dot. He intended the movement
of dots on the resulting animated map to illustrate the
uncertainty associated with dot placement.
Although Fisher’s notion of animating dot maps is
intriguing, the animations themselves were disconcert-
FIGURE 8 Two maps from an animation used to illustrate
ing. One problem was that relatively dense enumera-
uncertainty. The number of dots within each data collection
tion units appeared to be just as uncertain as less
unit (state, in this case) is held constant, but the dots are placed
dense units, although intuitively it seems that the posi-
at different random locations. (After Fisher, P. F. 1996. First pub-
tion of dots within a dense area should appear more
lished in Cartography and Geographic Information Systems
certain. A second problem was that a purely random
23(4), p. 200. Reprinted with permission from the American
placement of dots could lead to strange concentrations
Congress on Surveying and Mapping and Peter F. Fisher.)
of dots (note the circular arrangement of dots in the
western portion of Connecticut in the top map in
Figure 8). These problems might be ameliorated by a
Hunter and Michael Goodchild (1995) indicated that one
revised dot placement algorithm such as the modifica-
important application of the resulting contours is the
tion of Lavin’s (1986) dot-density method. It is also
specification of a particular threshold elevation. For
possible that a map comparison approach might be
example, in floodplain mapping we might specify the con-
useful in stressing areas of greater uncertainty; for
tour above which building is permitted.Although contour
example, compare the two animation frames shown in
maps are useful, users often are unaware of the uncer-
Figure 8, and note the quite different appearance of
tainty associated with the corresponding DEMs, which
dot distributions in the state of Vermont (an area of
can be measured using the root mean square error
low density), and the relatively greater stability for
(RMSE) as follows:
New York State (an area of high density).
n
* 2
4.3 Hunter and Goodchild’s Mapping a (Zi - Zi )
i=1
of DEM Uncertainty Q n
Topographic data are commonly available in the form of where Zi = elevation of the DEM at a sampled point
an equally spaced gridded network known as a digital ele- Zi* = true elevation at the sampled point
vation model (DEM). Such data can be readily contoured
using standard mapping software such as Surfer. Gary n = number of sampled points
487
12:32:38.
Hunter and Goodchild (1995) described how the gray tone used to mask cells outside the 2.5 to 97.5 per-
RMSE associated with the DEM could be used to spec- cent probability values is identical to one of the gray
ify a probability that the actual elevation value was tones falling within the 2.5 to 97.5 percent probability
above a particular threshold elevation. Their method range. As a solution to this problem, Hunter and
was based on the assumption that the distribution of Goodchild suggested several alternative color schemes
error around a particular contoured value is normal (see Plate 1 of their paper). If the 50 percent probabil-
and that the standard deviation of this normal distribu- ity value were treated as a logical break point in the
tion is equal to the RMSE. This concept is illustrated in data, then the diverging color schemes proposed by
Figure 9. Here, Zt is the threshold elevation value of Brewer also would be a logical solution to portraying
interest (e.g., the elevation above which building is per- such data.
mitted in a floodplain). Note that if the DEM value for
a raster cell (Zcell ) is equal to the threshold (as in 4.4 MacEachren et al.’s Depiction of Uncertainty
Figure 9A), then the probability of being above the in Health Statistics Data
threshold is 0.5. If, however, the DEM value for a cell is
below or above the threshold (as in Figures 9B and C), Epidemiologists often use maps to detect patterns in
then the area under the normal curve above the thresh- health-statistics-related data. For example, Deborah
old must be computed. This can be accomplished using Winn and her colleagues (1981) described a situation in
probability tables published in basic statistics books which a map of oral cancer death rates among women
(e.g., Burt and Barber 1996, 194–198), and can be in the southern United States prompted a study that
implemented in GIS software (Hunter and Goodchild identified snuff dipping as a major risk factor for this
indicated that the GIS software IDRISI provided this type of cancer. Although maps such as Winn et al.’s can
capability). be useful in detecting patterns, they must be treated
Once the basic probabilities have been computed for with caution when the populations for individual enu-
each cell, contour maps of these probabilities can be meration units are small, because the resulting death
constructed. Hunter and Goodchild illustrated several rates and ratios are unstable. To illustrate, consider two
different ways of visualizing these probabilities. A sim- enumeration units with populations of 100 and 10,000:
ple black-and-white approach is shown in Figure 10. An A single death due to cancer in each population pro-
obvious problem with this approach is that the neutral duces dramatically different rates (1 out of 100 vs. 1 out
of 10,000). There are several solutions to this problem.
One is to aggregate small populations, but this elimi-
nates small-scale variation that we might be interested
in. A second solution is not to symbolize small popula-
tion areas (i.e., to leave them blank), but Lewandowsky
P(Z > Zt) = 0.84
P(Z > Zt) = 0.50 et al. (1995) found that this interfered with pattern
recognition on the resulting map. A third solution is to
P(Z > Zt) = 0.16 size each enumeration unit as a function of the popula-
Z cell tion (i.e., use Dorling’s cartogram), but as we have
Zt Z cell
1 x RMSE noted, Dorling’s cartograms can be difficult to interpret.
(threshold)
1 x RMSE Given these problems, Alan MacEachren and his col-
Z cell leagues (1998b) promoted a fourth solution: embedding
the uncertainty directly within the map. MacEachren
and his colleagues utilized two approaches for embed-
ding uncertainty. In one approach, they made the data
and its uncertainty visually separable so that users could
view one component and ignore the other. MacEachren
A B C and his colleagues indicated that they experimented
with dozens of possibilities, finally arriving at a set of
“parallel lines split down their length, with one side
FIGURE 9 Determining the probability that the elevation black and the other white” (p. 1,551). This is illustrated
for a raster cell in a DEM (Zcell ) is greater than a particular in Color Plate 2, where we can see that dark lines
threshold (Zt ). The process involves computing the area appear if the underlying tone is light and that light lines
under the normal curve above the threshold value. The stan- appear if the underlying tone is dark. In the second
dard deviation of the normal curve is a function of the root approach, MacEachren and his colleagues attempted to
mean square error (RMSE) for the DEM. (After Hunter and make the data and its uncertainty visually integral so
Goodchild 1995, 532.) that users could view the components as a unit. They
12:32:38.
A B
FIGURE 10 (A) The contour line associated with a particular threshold and (B) the probabilities associated with exceeding that
particular threshold. (Reproduced with permission, the American Society for Photogrammetry and Remote Sensing,
Photogrammetric Engineering and Remote Sensing 61, no. 5, p. 529–537.)
accomplished this by either shifting the hue of diverging Uncertainty in remotely sensed imagery arises from
color schemes or changing the saturation of spectral and several sources including subpixel mixing, spatial mis-
sequential color schemes. In experiments with potential registration, and biases of particular sensors (Bastin et
users, MacEachren and his colleagues found the over- al. 2002). Subpixel mixing refers to the fact that indi-
laid parallel line approach to be more successful for vidual square pixels in a remotely sensed image are
depicting uncertainty, as the uncertainty information often composed of a mixture of various land use/land
could “be disregarded when necessary in order to notice cover categories; for example, a 30-meter pixel might
potentially important data clusters” and the method include portions of a house, a grassy area, and a pond.
prompted “map readers to recognize those maps, or Obviously, this produces uncertainty if we are trying
map sections, in which data should be used with cau- to classify the pixel in one of three categories, say,
tion” (p. 1,559). buildings, grassland, and water. From a data processing
standpoint, one solution to this problem is to develop
4.5 The FLIERS Project a probabilistic measure for each pixel belonging to
particular land use categories; thus, the following prob-
The FLIERS (Fuzzy Land Information from Environ- abilities might be computed for a pixel: building (40
mental Remote Sensing) Project was an attempt by percent), grassland (10 percent), and water (50 per-
researchers in several European universities to cent). One approach for determining such probabili-
develop software that can depict the uncertainty in ties is known as fuzzy classification (Foody 1996).
remotely sensed images (information on FLIERS can Spatial misregistration refers to the notion that a
be found at http://www.soton.ac.uk/~hglewis/research/ remotely sensed image might not be perfectly regis-
fliers/). Although the FLIERS group was not the first tered with other images and maps of interest. For
to develop such software (van der Wel et al. 1998 instance, we might wish to check our remotely sensed
provided an early flexible package), their software classification against a high-resolution photograph. If
provides a wide range of visualization options, is avail- these two images are not registered precisely, then
able for free at the site just mentioned, and has been uncertainty will arise in our interpretation. Biases of
evaluated in user studies. particular sensors can be illustrated by the fact that
489
12:32:38.
the pixel-level information stored in a database does 4.6 Pang et al.’s Work with Uncertainty
not actually match what was recorded by the sensor
on the Earth’s surface. In the case of Landsat TM, a Alex Pang and his colleagues (1997) have worked
traditionally popular sensor, a stronger signal is extensively with uncertainty, in both geographic and
recorded in the center of the pixel, with surrounding nongeographic contexts. We have already mentioned
pixels also providing input to the signal associated their work with glyphs for vector-based data
with the central pixel. (Wittenbrink et al. 1996). Another major area of
Although the FLIERS Project developed approaches research has been the depiction of uncertainty in
for visualizing these three sources of uncertainty, interpolation on isarithmic maps (Lodha et al.
for simplicity we discuss just the approaches used to 1996a). We touched on the notion of uncertainty in
examine subpixel mixing. In considering these interpolation earlier in this chapter (see section 3),
approaches, we need to recognize how conventionally when we considered Lavin et al.’s use of animation to
classified remotely sensed images are symbolized— depict the effect of the exponent in inverse-distance
normally, distinct colors are used for each land use/land interpolation. Rather than look at one parameter of a
cover category. With this in mind, three approaches were particular interpolation approach (as Lavin et al.
used to examine the uncertainty arising from subpixel did), Pang et al. focused on comparing one interpola-
mixing: a static grayscale image, serial animation of a tion approach with another (e.g., we might compare
grayscale image, and random animation. The static inverse distance with kriging). Unfortunately, you
grayscale image was shown separately from the usual will find Pang et al.’s work difficult to follow because
colored classified image and simply consisted of gray we have not covered some of the interpolation meth-
tones, with dark areas representing more uncertain ods they examined. Still, it is useful to consider some
information. Serial animation of a grayscale image of their approaches for visualizing uncertainty,
involved animating a series of grayscale images, where including displacement mapping, spot mapping, and
each image portrayed a progressively larger portion of glyphs. Displacement mapping involves randomly
the probability values for a particular land use/land perturbing one of the interpolated surfaces in pro-
cover. For instance, we might begin with a 0–0.1 proba- portion to the uncertainty (Figure 11), whereas in
bility range for a particular land use/land cover and spot mapping, regions of high relative difference
progress to a 0–1.0 range: In the beginning only a few
pixels would appear, and in the end we would see uncer-
tainty for all pixels. Random animation integrated the
visualization of uncertainty with the colored raw land
use/land cover map by randomly selecting pixels and
then assigning a land cover as a random function of
probability values (i.e., a pixel with high reliability would
be more likely to remain constant, whereas a pixel with
high uncertainty would continually change).
A study with users of FLIERS by Blenkinsop et al.
(2000) supported some of our concern about the use-
fulness of animation. Overall, Blenkinsop et al. found
that users tended to prefer the static grayscale image
approach for depicting uncertainty. They noted, how-
ever, that animations were helpful “in providing a
very effective first impression of uncertainty” and that
“animations may, therefore, have their greatest
usefulness at an early stage in the exploration of a
data set” (p. 11). They also stressed the exploratory
capability of FLIERS, noting that no single method
could answer all desired questions and that the
performance of users might improve with train-
ing and experience in issues of visualization and FIGURE 11 Displacement mapping: In comparing two inter-
uncertainty. polated surfaces, one surface is randomly perturbed in pro-
portion to the uncertainty. (Courtesy of Suresh K. Lodha.)
12:32:38.
for this region—this visualization is in contrast to

Tissot’s indicatrix, which reveals a perfect circle for the
Mercator projection because there is no angular distor-
tion at a point.
4.7 Cliburn et al.’s Visualization of Uncertainty

in Water Balance Models
Daniel Cliburn and his colleagues (2002) have utilized

various visualization techniques to examine the
uncertainty of water balance models for terrestrial
regions of the world. The basic output of a water bal-
ance model is an indication of whether each geo-
graphic location has a water surplus or deficit, along
with an associated magnitude. Uncertainty arises
because modelers must determine appropriate input
parameters for the model (e.g., which historical tem-
perature and precipitation data should be used) and
which global circulation models (GCMs) should be
used to predict the future water balance. Although
GCMs do generally predict global warming (and thus
FIGURE 12 Volume-filling glyph: The volume between two less water available), there is spatial variation within
interpolated surfaces is filled by spheres with radii that are models, with cooling occurring at some locations (and
proportional to the difference between the two surfaces. thus more water available).
(Courtesy of Suresh K. Lodha.) Following Gershon’s notion of intrinsic and extrin-
sic visual variables, Cliburn and his colleagues pro-
between the surfaces appear spotted. Figure 12 illus- posed a range of intrinsic and extrinsic methods for
trates one of several glyphs they developed—in this depicting uncertainty, with an emphasis on 3-D (tech-
case, a volume-filling glyph in which the volume nically, 2 1/2-D) surfaces. Color Plate 3 illustrates some
between two surfaces is filled by spheres with radii intrinsic approaches that Cliburn et al. utilized. In
that are proportional to the difference between the Color Plate 3A, the height of the 3-D surface provides
two surfaces. Lodha et al. (1996b) have undertaken an indication of water availability based on historical
efforts to utilize sound to depict the uncertainty of temperature and precipitation data. Areas of the sur-
interpolation. face above 0 mm (note the scale in the lower left) rep-
Another area that Pang and his colleagues have resent a water surplus, and areas below 0 mm represent
worked on is the visualization of map projection dis- a deficit. Obviously, it is not easy to see both surpluses
tortion. We can think of different map projections as a and deficits in this static image; Cliburn et al. handled
form of uncertainty, as the nature of geographic repre- this problem by permitting users to manipulate the
sentation varies as a function of the map projection. In image (e.g., rotate and zoom in) via a wall-sized dis-
this context, Brainerd and Pang (2001) developed the play. Note that Color Plate 3A is colored using an
Interactive Map Projections (IMP) software (http:// orange–purple diverging scheme, indicating how the
www.cse.ucsc.edu/research/slvg/map.html). A key ele- average of GCMs compares with the historical data; in
ment of IMP is the floating ring tool, which is displayed this case, orange and purple indicate less and more
on both the globe and a desired projection (a Mercator water, respectively, in comparison to historical data.
in the case of Figure 13). On the globe, the floating ring Recall that Brewer found the orange–purple scheme
is displayed as a fixed-size circle, indicating no distor- particularly effective from the standpoint of color nam-
tion, whereas on the Mercator projection, the ring ing, and also found that it avoided problems of color
reflects the nature of the distortion. For instance, in vision impairment and simultaneous contrast. Color
Figure 13B we see that the floating ring is relatively Plate 3B depicts the uncertainty in the GCM predic-
large near the pole, reflecting the greater areal distortions using transparency (here, the more transparent
tion in the Mercator projection as one moves away the image, the greater the variation among the GCMs,
from the Equator. Also note that the floating ring is and hence the greater the uncertainty). Note how this
not perfectly circular on the projection and that its affects our view of orange and purple peaks in the
center is offset, indicating that there is shape distortion northeast part of Color Plate 3A—apparently, there is
491
12:32:38.
FIGURE 13 An image from the

Interactive Map Projection (IMP)
software; note how the Mercator
projection distorts the size and shape
of the floating ring tool in the north-
ern latitudes. (Courtesy of Jeffrey
Brainerd and Alex Pang.)
disagreement among the models regarding what will 5 STUDIES OF THE EFFECTIVENESS OF METHODS
happen at these locations. FOR VISUALIZING UNCERTAINTY
Color Plate 4 is an example of an extrinsic method
utilized by Cliburn et al. In this case, orange and pur- In this section, we consider several studies that tested
ple bars represent the range of GCM predictions at a the effectiveness of methods for visualizing uncertainty.
particular location. For example, if only an orange bar Although several of the preceding applications also
appears, then at least one GCM predicted less water involved tests of effectiveness, those tests were gener-
available (and none predicted more water available), ally not the major emphasis of the application.
whereas if both orange and purple bars appear, then at
least one GCM predicted less water and at least
another GCM predicted more water. Small pyramid- 5.1 Evans’s Study
like symbols at the end of the bars denote which
GCMs were associated with the extreme low or high Beverley Evans (1997) was one of the first to explore
point on a bar (note the legend in the lower left). the effectiveness of displays of uncertainty from the
Obviously, the complexity of this image requires that standpoint of the map user. In an experiment designed
users be able to manipulate it, an option that Cliburn to examine various methods for depicting uncertainty,
et al. did provide. Evans utilized four maps: a land use/land cover map, a
12:32:38.
map depicting areas of the land use/land cover map Leitner and Buttenfield analyzed the results
that had a reliability of 95 percent or more, a static in terms of both the number of correct sitings for the
bivariate map that combined the basic land use/land park and airport and the speed of response for the sit-
cover map with the 95 percent reliable map, and a ing. In terms of the number of correct sitings, the
dynamic map in which the land use/land cover map was addition of uncertainty information significantly
“flickered” with the bivariate map. Evans asked three increased the correct number, but only if a lighter
basic questions: value or a finer texture was chosen to display greater
uncertainty. Leitner and Buttenfield deemed this
1. Will map users access and comprehend reliability result surprising, given that a “darker value is usually
information in graphic form when it is provided? perceived as being more prominent” (p. 13). They
2. Which of four versions of graphically depicted argued, however, that their finding was likely a func-
reliability information will the map user prefer: tion of displaying the maps on a CRT rather than on
flickering . . ., [the] static composite . . ., a map paper.
depicting only “95% reliable” information, or an From the standpoint of speed of response, Leitner
interactive toggling between the mapped data and and Buttenfield found that including uncertainty
highly reliable information? information resulted in either the same or a faster
response time than when only basic wetland informa-
3. Is graphically depicted reliability information tion was shown. This might seem surprising, given the
equally useful . . . for novices and experts? greater information content of the display, but Leitner
(p. 410) and Buttenfield stated, “It would seem that map
Regarding the first question, users did access and certainty is understood as clarification rather than
comprehend the reliability information; moreover, adding complexity to a map display” (p. 14). Regarding
they claimed to be willing to access it in similar situa- speed of response, they also found that a more satu-
tions in the future. Concerning the various methods rated color (for greater uncertainty) produced the
for depicting reliability, the flickering and static meth- fastest response.
ods were deemed about equally useful, although some
found the flickering annoying. The 95 percent reliable
map was deemed difficult to use (because it was miss- 5.3 Edwards and Nelson’s Study
ing information), although some users found it useful
to toggle between it and the land use/land cover map, Laura Edwards and Elisabeth Nelson (2001) exam-
thus creating their own sort of flickering. Finally, ined methods for depicting uncertainty on proportional
Evans found little difference between novices and symbol maps. In their test maps, the basic phenome-
experts, indicating that both groups were able to make non being mapped (“Total Reports of Disease D in
use of reliability information.* Costa Rica” in Figure 14) was depicted using circle size.
Uncertainty was illustrated using one of four
approaches: a legend statement (Figure 14A), a relia-
5.2 Leitner and Buttenfield’s Study bility diagram (Figure 14B), focus (Figure 14C), and
value (Figure 14D). Edwards and Nelson anticipated
Michael Leitner and Barbara Buttenfield (2000) exam- that the legend and reliability approaches would be the
ined how the decision-making process is affected when most difficult to use because of their separation from
uncertainty information is added to a map. Their deci- the mapped phenomenon of interest. Moreover, they
sion-making process involved locating first a park and felt that the legend statement would be particularly dif-
then an airport, with the key information being how ficult to process because of its nongraphic nature.
wetlands were represented. On one map, all wetlands Regarding focus and value, Edwards and Nelson
were grouped into one class and no uncertainty infor- felt that value would be more effective because of its
mation was included. On a second map, the wetlands familiarity and graphic “punch.” They also noted the
were split into three classes of information, but, again, subtlety of the focus approach from a figure-ground
no uncertainty information was included. On six addi- perspective (p. 23).
tional maps, wetlands were again shown, with greater Edwards and Nelson performed two basic experi-
uncertainty depicted by the following visual variables: ments with maps similar to those shown in Figure 14.
lighter value, darker value, coarser texture, finer tex- In the first experiment, users performed a rapid pat-
ture, less saturated color, and more saturated color. tern detection task by identifying areas in which data
values were perceived to be highest and areas in which
*Our summary of Evans’s study is based on a summary of her work data were perceived to be most certain. In the second
by Cliburn (2001). experiment, users answered multiple-choice questions
493
12:32:38.
FIGURE 14 Approaches that Edwards and Nelson (2001) used to depict uncertainty on proportional symbol
maps. (Courtesy of North American Cartographic Information Society.)
about the variation in the mapped phenomenon and SUMMARY

its uncertainty, and indicated their confidence in their
answers. The result from the experiments that stood In this chapter, we have considered the notion of
out most clearly was the poor performance of the uncertainty and various approaches for visualizing this
legend statement. A surprising result, however, was uncertainty. Uncertainty can arise from a variety of
that the focus method outperformed the value method sources, including the raw data on which a map is based,
for depicting uncertainty. Edwards and Nelson the manner in which these data are processed, and the
explained this result as follows: “More certain data manner in which a visualization is created. The creation
gets the graphic punch, at the expense of less certain of an isarithmic map depicting precipitation recorded at
data, so much so that perhaps it becomes a more weather stations provides a good illustration of various
effective means of displaying the two data sets in sources of uncertainty. Uncertainty in the raw data is
tandem” (p. 34).
12:32:38.
illustrated by the possibility that incorrect values can be (1997) used for depicting the uncertainty of wind speed
recorded at individual weather stations; uncertainty due and direction.
to processing is reflected in the variety of interpolation We covered a broad range of applications involving
methods that can be utilized to estimate values between the visualization of uncertainty, including environmental
weather stations (the control points); and uncertainty of data at point locations (DIN), dot maps, contours associ-
visualization is a function of whether a classed or an ated with DEMs, health statistics data, remotely sensed
unclassed isarithmic map is created. data, Pang and his colleagues’ work (e.g., with map
There are three general methods for depicting uncer- projections), and water balance models. Although the
tainty: separate maps can be created for an attribute of applications that we have discussed illustrate that many
interest and its associated uncertainty (i.e., maps can be cartographers have attempted to visualize uncertainty,
compared); the attribute and its uncertainty can be dis- our impression is that the notion of uncertainty is often
played on the same map (i.e., maps can be combined); disregarded or forgotten. Software for the analysis and
and data exploration tools can be utilized. Generally, the display of spatial data generally does not provide rou-
“maps combined” approach has been most commonly tines expressly intended for visualizing uncertainty;
used, although a number of data exploration tools (includ- rather, it is up to the user of the software to recognize
ing map animation capability) have been developed. uncertainty and to explicitly map this uncertainty.
Visual variables for depicting uncertainty can be split It also must be recognized that relatively little is
into intrinsic and extrinsic categories. Intrinsic visual known about how users respond to the visualization of
variables are intrinsic to the display; for example, uncer- uncertainty. In this chapter we have touched on some
tainty might be depicted by varying the transparency of studies of visualizing uncertainty, but these studies are
the colors composing an isarithmic map. Extrinsic visual all relatively recent and do not span the full range of
variables involve adding objects to the display, such as uncertainty portrayal. Certainly, there is more work that
the glyphs that Wittenbrink et al. (1996) and Pang et al. needs to be done in this area.
FURTHER READING
Aerts, J. C. J. H., Clarke, K. C., and Keuper, A. D. (2003) Provides a thorough treatment of the errors that lead to uncer-
“Testing popular visualization techniques for representing tainty when visualizing demographic data.
model uncertainty.” Cartography and Geographic Information Ehlschlaeger, C. R., Shortridge, A. M., and Goodchild, M. F.
Science 30, no. 3:249–261. (1997) “Visualizing spatial data uncertainty using animation.”
Describes tests of the effectiveness of visualizing the uncertainty Computers & Geosciences 23, no. 4:387–395.
associated with SLEUTH, an urban growth model. Uses animation to visualize the uncertainty in digital elevation data.
Beard, M. K., Buttenfield, B. P., and Clapham, S. B. (1991) Fisher, P. F. (1996) “Animation of reliability in computer-
NCGIA Research Initiative 7: Visualization of Spatial Data generated dot maps and elevation models.” Cartography and
Quality. Technical Paper 91–26. Santa Barbara, CA: National Geographic Information Systems 23, no. 4:196–205.
Center for Geographic Information and Analysis.
An example of some of the work that Peter Fisher has done to
An early work that discusses many of the issues involved in visu- visualize uncertainty. Also see Fisher (1993; 1994a; 1994b).
alizing uncertainty.
Goodchild, M., Chih-chang, L., and Leung, Y. (1994) “Visual-
Cartographica 30, nos. 2/3, 1993. izing fuzzy maps.” In Visualization in Geographical Information
Includes several articles dealing with the visualization of Systems, ed. by H. M. Hearnshaw and D. J. Unwin, pp. 158–167.
uncertainty. Chichester, England: Wiley.
Clarke, K. C., Teague, P. D., and Smith, H. G. (1999) “Virtual Describes efforts to visualize the uncertainty in remotely sensed
depth-based representation of cartographic uncertainty.” images.
Proceedings of the International Symposium on Spatial Data Howard, D., and MacEachren, A. M. (1996) “Interface design
Quality, Hong Kong, pp. 253–259. for geographic visualization: Tools for representing reliabil-
Discusses the possibility of using apparent visual depth to repre- ity.” Cartography and Geographic Information Systems 23, no.
sent uncertainty via a head-mounted display (HMD). 2:59–77.
Comenetz, J. (1999) Cartographic Visualization of the Quality In addition to covering R-VIS (see section 4.1), this paper dis-
of Demographic Data. Unpublished PhD dissertation, cusses general considerations in designing interactive software for
University of Minnesota, Minneapolis, MN. visualizing spatial data.
495
12:32:38.
Jervis, J. (2002) “Visualizing uncertainty in Earth Observing Cartography and Geographic Information Science 32, no.
System satellite data.” Gridpoints 3, no. 2:6–9. 3:139–160.
Summarizes research at the NASA Advanced Supercomputing Provides an overview of issues related to visualizing uncertainty in
Division on visualizing the uncertainty in remotely sensed data. geographic data and proposes seven research challenges.
Lucieer, A., and Kraak, M.-J. (2004) “Interactive and visual Pang, A., Wittenbrink, C., and Lodha, S. (1997) “Approaches
fuzzy classification of remotely sensed imagery for explo- to uncertainty visualization.” The Visual Computer 13, no.
ration of uncertainty.” International Journal of Geographical 8:370–380.
Covers a broad range of methods for visualizing uncertainty for
Describes interactive software for exploring the uncertainty in both geographic and nongeographic data. For a more comprehen-
remotely sensed imagery based on fuzzy classification. sive overview and more recent work, see http://www.cse.ucsc.edu/
MacEachren, A. M. (1992) “Visualizing uncertain informa- research/slvg/unvis.html.
tion.” Cartographic Perspectives no. 13:10–19. Wittenbrink, C. M., Pang, A. T., and Lodha, S. K. (1996)
Covers basic principles for visualizing uncertainty; for more recent “Glyphs for visualizing uncertainty in vector fields.” IEEE
work, see MacEachren (1995). Transactions on Visualization and Computer Graphics 2, no.
MacEachren, A. M., Robinson, A., Hopper, S., Gardner, S., 3:266–279.
Murray, R., et al. (2005) “Visualizing geospatial information Introduces a range of glyphs for visualizing the uncertainty in vec-
uncertainty: What we know and what we need to know.” tor-based data.
GLOSSARY
attribute accuracy: one means of assessing data quality; intrinsic visual variables: when mapping uncertainty, these
refers to the accuracy of features found at particular locations are visual variables that are intrinsic to the display; for exam-
(e.g., the accuracy of a land use/land cover classification). ple, we might vary the saturation of colored tones on a choro-
bounding rectangle: a rectangle that just touches but pleth map.
completely encloses an arbitrary shape. lineage: one means of assessing data quality; refers to the
clarity: a term used to summarize the following visual vari- historical development of the data.
ables for depicting data uncertainty: crispness, resolution, and logical consistency: one means of assessing data quality;
transparency. describes the fidelity of relationships encoded in the structure
completeness: one means of assessing data quality; includes of spatial data (e.g., are polygons topologically correct?).
information about selection criteria and definitions used (e.g., positional accuracy: one means of assessing data quality;
the minimum area needed to define a water body as a lake). refers to the correctness of location of geographic features.
crispness: a visual variable for depicting data quality; a sharp resolution: (1) for CRT displays, the number of addressable
(or crisp) boundary would represent reliable data, whereas a pixels; for printers, the number of dots per inch; (2) a visual
fuzzy boundary would represent uncertain data. variable for depicting data quality; for example, coarser raster
data quality: the notion that data for thematic maps are cells indicate greater uncertainty in the data.
often subject to some form of error; see uncertainty. transparency: a visual variable for depicting data quality;
extrinsic visual variables: when mapping uncertainty, these refers to the ease with which a theme can be seen through a
are visual variables that involve adding objects to the display, “fog” placed over that theme.
such as dials and thermometers. uncertainty: the notion that data for thematic maps are
often subject to some form of error; uncertain data are also
considered to be unreliable and of poor quality.
REFERENCES
Blenkinsop, S., Fisher, P., Bastin, L., and Wood, J. (2000) Brassel, K. (1974) “A model for automatic hill shading.” The
“Evaluating the perception of uncertainty in alternative American Cartographer 1, no. 1:15–27.
visualization strategies.” Cartographica 37, no. 1:1–13. Brown, A., and van Elzakker, C. P. J. M. (1993) “The use of
Bastin, L., Fisher, P. F., and Wood, J. (2002) “Visualizing colour in the cartographic representation of information
uncertainty in multi-spectral remotely sensed imagery.” quality generated by a GIS.” Proceedings, 16th Conference
Computers & Geosciences 28, no. 3:337–350. of the International Cartographic Association, Cologne,
Brainerd, J., and Pang, A. (2001) “Interactive map projec- Germany, pp. 707–720.
tions and distortion.” Computers & Geosciences Burt, J. E., and Barber, G. M. (1996) Elementary Statistics for
27:299–314. Geographers (2nd ed.). New York: Guilford.
12:32:38.
Cliburn, D. C. (2001) “Representing multiple uncertainties Leitner, M., and Curtis, A. (2006) “A first step towards a
and evaluating usability in a spatial decision support framework for presenting the location of confidential
system.” Unpublished PhD dissertation, University of point data on maps—Results of an empirical perceptual
Kansas, Lawrence, KS. study.” International Journal of Geographical Information
Cliburn, D. C., Feddema, J. J., Miller, J. R., and Slocum, T. A. Science 20, no. 7: 813–822.
(2002) “Design and evaluation of a decision support Lodha, S. K., Wilson, C. M., and Sheehan, R. E. (1996b)
system in a water balance application.” Computers & “LISTEN: Sounding uncertainty visualization.” IEEE
Graphics 26:931–949. Visualization ’96, San Francisco, CA, pp. 189–195.
Evans, B. J. (1997) “Dynamic display of spatial data-reliabil- Lewandowsky, S., Behrens, J. T., Pickle, L. W., Herrmann, D.
ity: Does it benefit the map user?” Computers & J., and White, A. (1995) “Perception of clusters in mortality
Geosciences 23, no. 4:409–422. maps: Representing magnitude and statistical reliability.”
Edwards, L. D., and Nelson, E. S. (2001) “Visualizing data In Cognitive Aspects of Statistical Mapping, ed. by L. W.
certainty: A case study using graduated circle maps.” Pickle and D. J. Herrmann, pp. 107–132. Washington, DC:
Cartographic Perspectives, no. 38:19–36. Centers for Disease Control, National Center for Health
Statistics.
Foody, G. M. (1996) “Approaches to the production and
evaluation of fuzzy land cover classification from MacEachren, A. M. (1992) “Visualizing uncertain informa-
remotely-sensed data.” International Journal of Remote tion.” Cartographic Perspectives, no. 13:10–19.
Sensing 17, no. 7:1317–1340. MacEachren, A. M. (1995) How Maps Work: Representation,
Fisher, P. F. (1993) “Visualizing uncertainty in soil maps by Visualization, and Design. New York: Guilford.
animation.” Cartographica 30, no. 2&3:20–27. MacEachren, A. M., Brewer, C. A., and Pickle, L. W. (1998b)
Fisher, P. F. (1994a) “Hearing the reliability in classified “Visualizing georeferenced data: Representing reliability
remotely sensed images.” Cartography and Geographic of health statistics.” Environment and Planning A
Information Systems 21, no. 1:31–36. 30:1547–1561.
Fisher, P. F. (1994b) “Visualization of the reliability in McGranaghan, M. (1993) “A cartographic view of spatial
classified remotely sensed images.” Photogrammetric data quality.” Cartographica 30, no. 2&3:8–19.
Engineering and Remote Sensing 60, no. 7:905–910. Mitasova, H., Brown, W., Gerdes, D. P., Kosinovsky, I., and
Gershon, N. (1998) “Visualization of an imperfect world.” Baker, T. (1993) “Multidimensional interpolation, analysis
IEEE Computer Graphics and Applications 18, no. and visualization for environmental modeling.” GIS/LIS
4:43–45. Proceedings, Volume 2, Minneapolis, MN, pp. 550–556.
Guptill, S. C., and Morrison, J. L. (1995) Elements of spatial National Institute of Standards and Technology. (1994)
data quality. Kidlington, Oxford, England: Elsevier Federal Information Processing Standard 173 (Spatial Data
Science. Transfer Standard Part 1, Version 1.1). Washington, DC:
U.S. Department of Commerce.
Hunter, G. J., and Goodchild, M. F. (1995) “Dealing with
error in spatial databases: A simple case study.” Pang, A., Wittenbrink, C., and Lodha, S. (1997) “Approaches
Photogrammetric Engineering and Remote Sensing 61, no. to uncertainty visualization.” The Visual Computer 13, no.
5:529–537. 8:370–380.
Howard, D., and MacEachren, A. M. (1996) “Interface design U.S. Bureau of the Census. (1994) County and City Data
for geographic visualization: Tools for representing relia- Book: 1994. Washington, DC: U.S. Government Printing
bility.” Cartography and Geographic Information Systems Office.
23, no. 2:59–77. van der Wel, F. J. M., van der Gaag, L. C., and Gorte, B. G. H.
Leitner, M., and Buttenfield, B. P. (2000) “Guidelines for the (1998) “Visual exploration of uncertainty in remote-sens-
display of attribute certainty.” Cartography and ing classification.” Computers & Geosciences 24, no.
Geographic Information Science 27, no. 1:3–14. 4:335–343.
Lodha, S. K., Sheehan, B., Pang, A. T., and Wittenbrink, C. M. Winn, D. M., Blot, W. J., Shy, C. M., Pickle, L. W., Roledo, A.,
(1996a) “Visualizing geometric uncertainty of surface and Fraumeni, J. F. J. (1981) “Snuff dipping and oral cancer
interpolants.” Graphics Interface, pp. 238–245. among women in the southern United States.” New
http://www.cse.ucsc.edu/research/slvg/surf.html. England Journal of Medicine 304, no. 13:745–749.
Leitner, M., and Curtis, A. (2004) “Cartographic guidelines Wittenbrink, C. M., Pang, A. T., and Lodha, S. K. (1996)
for geographically masking the locations of confidential “Glyphs for visualizing uncertainty in vector fields.” IEEE
point data.” Cartographic Perspectives, no. 49: 22–39. Transactions on Visualization and Computer Graphics 2,
no. 3:266–279.
497
12:32:38.
12:32:38.
Web Mapping
OVERVIEW
This chapter focuses on the World Wide Web (or “the that allow individuals to modify the map (e.g., change the
Web”) and how this technology has fundamentally data classification system or choose a new color scheme).
changed the way in which mapping activities take place. In The second continuum distinguishes between Web
section 1, we provide a short history of the Web, its com- sites that offer low levels of interaction and those that
ponents, and how it works. In that section we also explain offer high levels of interaction. TypeBrewer, an online
the differences between the commonly confused terms “the tool to help mapmakers select type for maps, is one Web
Web” and “the Internet.” Common ways in which the Web site that offers low interaction levels. Google Earth and
is used and related technologies such as FTP, IP, and World Wind are two Web-deliverable products that offer
HTTP are also covered. In section 2, we introduce a clas- users high levels of interaction with maps and data.
sification of cartographically themed Web sites. In our The third continuum considers those Web sites that dis-
classification, we attempt to survey those Web sites that are tribute data versus those that rely upon user-contributed
particularly interesting from a cartographic standpoint data. At one end of the continuum are those sites that
and demonstrate how the distribution of geospatial data distribute geospatial data such as DEMs or DRGs.
has changed as a result of the Web. Our classification orga- Federal, state, and commercial agencies such as the U.S.
nizes various cartographically themed Web sites into five Geological Survey’s National Map offer users the ability
continua that begin to describe the rich variety of to view geospatial data online as well as to download the
approaches that have been used to distribute geospatial data (sometimes at no cost) and incorporate it into a GIS
data. In developing this chapter, we purposefully chose environment. At the other end of the continuum are sites
those Web sites that are stand-alone applications (i.e., are that are entirely built around data that are received from
able to run by themselves across the Web) and do not individuals (i.e., user-contributed data). In most cases,
require the purchase of any additional software. these data are then posted for others to view or interact
The first continuum categorizes Web sites as contain- with. The New York City Subway smell map is one such
ing either preexisting maps or custom-tailored maps. The site where the information content of the site is expressly
distinction is between those Web sites that contain maps controlled and contributed by the population at large.
that are already created (historically, preexisting maps The fourth continuum divides Web sites on the basis
were the first to appear on Web sites) versus those Web of how frequently the mapped data are updated.
sites that allow users to interact with the data to create a AirportMonitor is one Web site that presents updated data
map that is custom-tailored to the users’ specifications. showing near-real-time displays of incoming and outgo-
For example, the National Park Service offers a Web site ing flights from selected airports around the world. On the
that allows users to download preexisting maps of vari- other hand, the Perry-Castañeda Map Library from the
ous national parks in the United States, but does not offer University of Texas at Austin contains an online catalog of
users any opportunities to change a map’s appearance. In thousands of non-updated maps. In other words, once the
contrast, American Factfinder offers Web-based tools map is posted for downloading, it is never updated.
499
12:35:00.
Web Mapping
The fifth continuum involves a distinction between tally different. The Internet is a series of interconnected
animated maps and static maps. Animation is another networks whose purpose can be thought of as transfer-
facet of how spatial information is displayed on the Web ring files back and forth. The first network that formed
and plays a significant role in the continuing evolution of the basis of what we know today as the Internet can be
Web mapping. One matter-of-fact Web site uses animation traced back to the space race between the United States
to show the daily occurrence of fatalities of U.S. and coali- and the Soviet Union during the 1950s. It was during
tion troops, going back to the start of the Iraq war. Opposite that time that U.S. researchers working for the
the Web sites with animated maps are those sites that dis- Advanced Research Projects Agency (ARPA) devel-
play static maps. David Rumsey’s superb collection of sta- oped a countrywide network to assist U.S. scientists in
tic maps is one Web site that fits this end of the continuum. communicating with one another. In the original config-
In section 3 we bring together the various continua uration developed by ARPA, sending a file from one
and explore how one particular Web site from the “networked” computer to another required the infor-
National Weather Service can be classified along these mation to be stored in an Internet Protocol (IP) packet
five continua. Obviously, the Web sites that we discuss in with an appropriate address. This protocol was com-
this chapter are only a small fraction of the number of monly used until 1983,* when the TCP/IP (Transmission
sites that we could have selected. Moreover, there are Control Protocol/Internet Protocol) was introduced,
other possible continua that could be developed to clas- which offered the ability of a globally based network.
sify the rich amount of mapping activity that takes place How this globally based network has evolved and
on the Web. For additional examples of cartographic- how it will continue to evolve matter a great deal
themed Web sites, we encourage you to examine this text’s because the Internet has fundamentally changed soci-
Web site at http://www.prenhall.com/slocum. etal, economic, cultural, political, and spatial relations
and interactions. Martin Dodge and Rob Kitchen
1 A BRIEF HISTORY OF WEB MAPPING (2001b) discuss eight continua in which the Internet has
impacted society. We focus on the four of their eight
continua that we feel are the most relevant to Web map-
For centuries, maps were routinely printed on paper. ping. The first continuum, space-spacelessness, focuses
Although useful and aesthetically pleasing, paper maps on how the Internet has changed the concept of geo-
were not always easily accessible and were often too graphic space. This idea suggests that the instantaneous
expensive for the map consumer. In many cases, univer- nature of the Internet has collapsed the concept of tra-
sity libraries were the best places to locate paper maps. ditional spatial boundaries. In this respect, geographic
But the Web has fundamentally changed how maps are distance is no longer as important as it once was. For
accessed, distributed, and produced. According to instance, you no longer have to drive to a store to make
Michael Peterson (2003), the Web has evolved over purchases (e.g., you can buy a globe online); you can
three distinct periods that have fundamentally changed experience a distant location via a live Webcam without
how maps are distributed: leaving your own environment; and you can peruse the
1. During the Web’s early years, existing paper maps holdings of a distant map library.
were digitally converted into .jpg or .gif images that The second continuum focuses on how the sense of
were stored and retrieved via the Web. place-placelessness has changed. The Internet has effec-
tively removed the concept of geographic space from
2. As the Web matured, developments in technology
the process of individuals mingling together. Consider a
and the evolution in computer languages allowed
specific neighborhood and the people that comprise that
interactivity and animation to take place via the Web.
neighborhood. You will likely find commonalities such
3. In its current state, the Web has the potential to as income, educational attainment, and career choices
deliver a considerable range of mapping products. among the people who live there. In other words, people
As technology leaps forward with increasing sophis- tend to locate in neighborhoods or communities that are
tication, the kinds of mapping products that will be tied to geographic space. However, through using the
available in the future are left to the inventiveness Internet, people all over the world can meet and discuss
of the larger map user community. matters that are important to them without having to
physically meet in geographic space. For example,
1.1 The Internet and the World Wide Web Cartotalk (http://www.cartotalk.com/) is an online forum
in which individuals interested in cartography can meet
Before we proceed with our discussion of Web map-
ping, it is important to consider how the Web came to * ARPA switched to the TCP/IP protocol on January 1, 1983, and
be. Technically speaking, the World Wide Web and the many consider this to be the date on which the Internet was “born”
Internet are not synonyms. In fact, they are fundamen- (Peterson 2003).
12:35:00.
Web Mapping
in virtual space to discuss and exchange ideas about a it. It is commonplace today to find many telecommuters
wide range of cartographic topics. that perform a day’s worth of work from their own
The third continuum, broadcaster-listeners, deals with home by remotely connecting to their office computer
the distribution and accessibility of information. Here from their home. The fourth component is collabora-
the focus is on the exchange of information between the tion between individuals. Applications such as instant
broadcaster and the listeners. With traditional media messaging and videoconferencing allow multiple indi-
(e.g., television and radio), the listener is passive and has viduals to simultaneously engage in discussion or
little input or control over what information is exchanged exchange ideas over their computers at distant loca-
via the broadcast. Now, however, the Internet makes it tions. The last component, and the one of special focus
possible for “listeners” to request or post information in this chapter, is the World Wide Web, or simply the
specific to their own needs. For instance, users can Web. The Web was developed by scientists at the
request driving directions and a map for a specific travel Conseil Européen pour la Recherche Nucléaire (CERN)
itinerary from a map generator (a one-to-one broadcast); near Geneva, Switzerland, in 1989. The Web is a series
an individual may have created a new map and posted it of interconnected hypertext documents that are shared
on a Web site for anyone to view or download (a one-to- over the Internet between a computer user and a
many broadcast); or many individuals may post their own remote computer and are viewable through a Web
map creations to a Web site where others can access them browser (e.g., Microsoft’s Internet Explorer or Mozilla’s
(a many-to-many broadcast). Firefox). In short, the Web makes it possible for a com-
The fourth continuum, public-private, deals with the puter user (the “client”) to specify a URL (uniform
level at which information is accessible by the global resource locator) through a Web browser interface. The
community over the Internet. In some cases, information URL is changed into an Internet Protocol (IP) address,
is considered public and thus accessible by anyone. For which is then used to access the remote computer (the
the most part, U.S. government documents (e.g., hard- “server”) on the Internet network. The server delivers
copy maps and digital data) are considered to be in the the requested information through HTTP (hypertext
“public domain” and are obtainable via the Internet, transfer protocol) back to the Web browser on the
many without cost. On the other hand, some information client’s computer, which takes the hypertext markup
is considered private and thus is limited in its accessibil- language (HTML) and configures it so that the client’s
ity. For instance, the Census Bureau collects information computer can display it on-screen.
at the individual household level and uses it for a variety
of internal purposes. While the Census Bureau makes 1.3 The Web’s Importance to Mapping
available over the Web a considerable amount of data
that it collects, the Census Bureau considers the house- Cartography has also evolved along with the develop-
hold data it collects to be “private” information and does ments of the Internet and computer technology. Dodge
not release it to the public. and Kitchin (2001b) describe three evolutions that car-
tography experienced toward the end of the twentieth
1.2 Common Usage of the Web century that facilitated the view that the Web offers
unique opportunities for cartography. First, technologi-
Of course, today the Internet constitutes an important cal advances in computing fundamentally changed the
part of daily life for many people. Peterson (2003) way in which mapmaking was conducted. For instance,
describes five main components of the Internet that geographic information systems (GIS), remote sensing,
facilitate its usability. The first component is communi- and global positioning systems (GPS) have become the
cation directly between computer users via e-mail, primary technological means through which many car-
which is certainly one of the more popular ways people tographic products are now produced. Increasingly,
use the Internet. The second component is sharing files these technological tools have become more common-
between computer users, which is commonly accom- place and available to individuals with little formal
plished using FTP (File Transfer Protocol). For training in their use and application, thus facilitating the
instance, the Earth Resources Observation Science mapmaking process. Second, new methods of geo-
(EROS) data center in Sioux Falls, SD, allows users to graphic visualization have evolved. For example, inter-
obtain geospatial data (e.g., digital elevation models, activity (and data exploration), animation, and virtual
digital raster graphics, etc.) via the Internet using FTP environments are now possible and are the direct result
(http://edc2.usgs.gov/geodata/index.php). of advances in technological developments. Software is
The third component is remote access, in which users available to create a virtual landscape complete with
remotely log in to a networked computer and have buildings and forest, which can be placed upon a digital
access to that computer as if they were sitting in front of elevation model that the user can fly through, pan
501
12:35:00.
Web Mapping
around, and zoom to different views. Third, attitudes scanned and converted into a digital format or pro-
toward maps have also changed. Traditionally, maps duced through specific mapping software. In either case,
were produced by trained mapmakers, who carried out these maps are saved in a format such as .gif or .jpg and
their craft under the established rules and guidelines of then posted on a Web site for viewing but offer no inter-
the agency in which they were employed. The maps that activity. Spatial database catalogs and libraries are sites
were produced were seen as rather sterile, scientific end that offer various types of raw geospatial data (e.g., dig-
products. Dodge and Kitchin (2001b, 69) indicate that ital orthophoto quadrangles (DOQs), digital elevation
while many view the mapmaking process as an “objec- models (DEMs), and digital line graphs (DLGs)), along
tive, scientific pursuit, like all sciences, it is a process of with associated metadata. Ultimately, users can down-
construction shaped by many forces.” Thus, maps should load the geospatial data for further manipulation in, for
not be viewed as neutral objects, but rather as reflec- example, a GIS environment. With map generators,
tions of societal attitudes, tools of political persuasion, users enter a location (usually in the form of a street
and embodiments of power. address), and a detailed map is produced that is cen-
In the context of mapmaking, it has been possible to tered on the point specified by the user. Real-time map
transfer geospatial data across computers since the late browsers offer users the ability to display individual the-
1970s through FTP.* However, the ability to create matic data layers. In addition, users can query the map
maps from this geospatial data required access to and for attribute information, which is similar to the func-
knowledge of how to use specialized software, which tionality that is present in a GIS environment. Of
greatly limited the appeal and usability of the Web for course, at the time of Calkins’s work, the Web was still
mapping purposes. Today, the Web breaks down many in its infancy, and many of the technologies that we are
barriers to the production, distribution, and accessibil- accustomed to today didn’t exist; however, many of his
ity of maps. In what has been called a “convergence of early Web site types still exist.
spatial technologies,” remote sensing, GIS, GPS, and More recently, William Cartwright (2003) discussed
digital cartography have seamlessly merged their capa- six cartographic services available via the Web: (1) map
bilities and products into new kinds of map-based prob- and image collections, (2) downloadable data, (3) infor-
lem solving (Crampton 1998). The Web also makes it mation services with maps, (4) online map-generation
possible for individuals to create maps “on demand.” services, (5) Web atlases, and (6) mixed distribution.
Using specific mapping-based Web sites, users can indi- Map and image collections are similar to Calkins’s
vidually tailor data sets, symbolization, data classes, graphic snapshots. One important distinction is that this
color schemes, etc., and produce maps to suit their indi- service also points users to other Web sites that provide
vidual mapping requirements in real time. Along simi- geospatial information. Downloadable data are similar
lar lines, the Web has also ushered in the democratiza- to spatial database catalogs and libraries in the sense
tion of cartography, which facilitates the mapping that users can locate and select raw geospatial data to
process by permitting anyone with a computer and download. An information service with maps is a new
access to the Internet the opportunity to create a vari- kind of Web site in which maps are incorporated into a
ety of maps. Increasingly, the Web makes it possible for variety of services rather than exist as stand-alone prod-
individuals with little or no formal cartographic educa- ucts. For example, many online travel assistants provide
tion to practice the art and science of mapmaking. users with a variety of cartographic services such as
a map of where a particular hotel is located with
2 CARTOGRAPHIC WEB SITES: respect to various attractions or a map of which
A CLASSIFICATION hotels are located along a particular highway. Online
map-generation services provide users with the ability
to select various thematic layers of geospatial data,
In order to organize the various mapmaking Web sites
essentially allowing them to construct a custom-tailored
and discuss their cartographic importance, we first sur-
map in real time. Web atlases are another new category
vey three classification schemes and then use them to
that essentially places the paper atlas into the digital
create our own scheme. In early work, John Calkins
domain. In some cases, this digital conversion takes the
(1996) distinguished among graphic snapshots, spatial
very simple approach of collecting a series of graphic
database catalogs and libraries, map generators, and
snapshots. In other cases, Web atlases offer new deliv-
real-time map browsers. In his conceptualization,
ery options by providing a spectrum of content through
graphic snapshots are hard-copy maps that are either
various media styles (e.g., video and sounds). Mixed
distribution services are provided by those Web sites
*For an overview on the history of FTP, see the following URL: that combine different types of distribution media. For
http://www.tcpipguide.com/free/t_FTPOverviewHistoryandStandards.htm. example, atlases such as the Atlas of Oregon (Loy et al.
12:35:00.
Web Mapping
2001) are distributed not only via the Web, but also example, offer software that you must purchase as a CD
through CD-ROM/DVD and as hard copy. (or download directly) and install on a local computer
In another approach for classifying Web sites, via the Web (although these are certainly additional,
Langer (2002) crafted a classification that mimics the important components of Web mapping).
cartography-cubed idea developed by MacEachren
(1994b), in which map use is presented on three axes
2.1 Preexisting Maps versus Custom-Tailored Maps
and includes high versus low interactivity, revealing
knowns versus presenting unknowns, and public ver-
This continuum differentiates maps based on when the
sus private. Langer distinguished cartographic Web
cartography is executed. In the case of preexisting
sites according to the levels of user interactivity (inter-
maps, the cartography is completed by a mapmaker
active vs. view only), whether the maps are dynamic or
prior to the user’s viewing the Web site. These maps are
static (animated vs. static), and the degree of GIS func-
described as static in nature because once they are pro-
tionality (such as whether the user can query attrib-
duced, they are not updated. In addition, users have no
utes, add a data set, or annotate existing data sets).
(or very limited) ability to zoom in/out or pan around
Langer emphasized the latter point because of techno-
these maps. These maps can be created through spe-
logical advances and the merging of GIS-like function-
cialized mapping software, and the map is then
ality with cartographic Web sites. Increasingly, carto-
exported as an image file in .gif, .jpg, or .pdf format.
graphic Web sites are offering sophisticated levels of
Alternatively, these maps can be created by scanning
GIS-type functionality that go beyond a simple point-
an existing hard-copy paper map and saving it in a dig-
and-click map.
ital image format. The image files are then posted to a
We have borrowed several ideas from Calkins’s,
Web site, where users can download or view them.
Cartwright’s, and Langer’s classifications and expanded
Custom-tailored maps are found on Web sites that
on them to develop our own set of five continua that
allow users some control over the cartographic process.
describe Web sites and how they incorporate cartogra-
These processes include, but are not limited to, specify-
phy. Our first continuum (preexisting maps vs. custom-tai-
ing the kinds of data shown, deciding on the map
lored maps) is based upon the notion that while some
design, choosing the scale, and modifying the coordi-
Web sites provide maps that are already made, other Web
nate system. Although fewer in number than preexist-
sites offer users the ability to create their own custom-
ing map sites, custom-tailored map sites offer users the
tailored map (e.g., through a real-time map browser or a
ability to explore a variety of data sets and to experi-
map generator). Our second continuum (low interaction
ment with map design options not afforded with other
vs. high interaction) stems from Langer’s distinction
cartographic Web sites.
between the types of interactivity levels that many carto-
graphic Web sites offer. The third continuum (distributed
data vs. user-contributed data) reflects the spatial data- Preexisting Map: National Park Service
base catalogs and libraries and downloadable data idea There are probably tens of thousands of Web sites that
previously discussed, but to it adds individuals’ having post preexisting maps for users to view and download.
the opportunity to contribute their own data to a Web One particularly interesting site is that of the U.S.
site for others to download. This continuum also includes National Park Service’s Interpretative Center at
the idea that the Internet blurs the distinctions between Harper’s Ferry, WV (http://home.nps.gov/applications/
private and public access to and distribution of data. Our hafe/hfc/carto.cfm). When visiting this site, users can
fourth continuum (updated data vs. non-updated data) is view, print, and download preexisting maps that show
a new distinction that reflects the timeliness of the data the various parks throughout the U.S. National Park
available to the user. The fifth continuum (animated system. The maps are viewable as a .pdf document, and
maps vs. static maps) is another newly created continuum users can also download the map in Adobe Illustrator
that incorporates the fact that the Internet makes it pos- (.ai) format. Figure 1 shows a portion of an
sible for the temporal component of a data set to be Appalachian Trail map highlighting the Maryland,
viewed via animation. It should be noted that we do not West Virginia, and Virginia area. All of the maps avail-
consider these continua an exhaustive way of describing able from this site have the same design and layout
every cartographic Web site; certainly, other continua specifications that are followed to create the printed
could be developed. For instance, we make no distinction maps people receive as a brochure when they visit
as to whether Web sites present two-dimensional or each national park. Although this site offers only the
three-dimensional visualizations. maps for downloading and not the entire brochure,
In the discussion that follows, it is our intent to sur- users can gain additional information about each park
vey those Web sites that provide maps and mapmaking and available activities by clicking on the links to the
capability directly. We do not include Web sites that, for different national parks.
503
12:35:00.
Web Mapping
FIGURE 1 An example of a preexisting map: a U.S. National Park Service map showing a portion of the Appalachian Trail (the
thick dashed line) that runs from Maine to Georgia. (Source: http://home.nps.gov/applications/hafe/hfc/carto.cfm; the original
appears in color.)
Custom-Tailored Map: American Factfinder Thematic also have the ability to specify the number of classes and
Mapping Tool choose from a limited set of color schemes; a map title
American Factfinder (http://factfinder.census.gov/) is an can also be specified. One limitation of this site is that the
example of a custom-tailored mapping site. At this site, county and state outlines are always shown in black,
users are presented with broad U.S. Census data cate- which often makes visualizing the spatial pattern difficult
gories: people, housing, and business & government. in very dense areas. The finished map can then be down-
Within each category, users can select from a variety of loaded in .pdf format. Color Plate 1 is an example of a
thematic data sets (e.g., population density, total number map of Health Care and Social Assistance Receipts by
of households, and retail sales). For illustrative purposes, County for 2002, using quantiles as the data classification
we chose to create a custom-tailored map of health care method, five classes, and a green color scheme.
and social assistance receipts for 2002. Once a data set is
selected, users are provided a considerable array of 2.2 Low Interaction versus High Interaction
options for creating a customized map: Users have the
ability to select from various geographic levels (e.g., con- This continuum focuses on the level of user interactivity.
gressional districts, counties, metropolitan statistical areas, Some Web sites allow the user no interactivity. Other
and census tracts), and a variety of base information Web sites permit limited user interactivity, such as zoom-
(roads, hydrology, parks, point features, etc.) may also be ing in and out, panning, and re-centering the map. Web
displayed. In terms of map design, users are permitted to sites with high levels of interactivity include tools
experiment with four data classification methods: equal that are usually associated with GIS software, such as
interval, quantiles, natural breaks, and user defined. Users being able to specify a coordinate system, add or remove
12:35:00.
Web Mapping
various data layers, query the data attributes, measure TypeBrewer, developed in Adobe’s Flash environment,
distances, and so forth. follows in the footsteps of ColorBrewer, which offers
mapmakers assistance in the appropriate selection of
Low Interaction: TypeBrewer color schemes. Thus, TypeBrewer allows mapmakers to
explore type in a cartographic framework. Figure 2
An intriguing Web site that offers a low level of inter- provides a sample of TypeBrewer’s interface design
activity is TypeBrewer (http://www.typebrewer.org/). and overall appearance.
According to the Web site: TypeBrewer begins with a template screen that allows
users to choose a broad type category (e.g., formal, infor-
TypeBrewer is a free help tool that gives non-specialist mal, classic, and contemporary) and then within each
mapmakers a chance to explore typography in a semi-
category to choose specific type families (e.g., the
structured environment. It is not mapmaking software.
Instead of providing the functionality of a graphic design
Formal|Zinfadel template uses the type families of
program or GIS, TypeBrewer offers a quick and easy way Georgia and Trebuchet). Once chosen, a new screen
to explore typographic alternatives and see the impact that appears showing a map of the Italian Peninsula and its
various elements of type have on the overall look and feel surrounding region. Outside the map, four options are
of a map. TypeBrewer is designed for mapmakers who available: Explore, Get Specs, Compare, and Export.
want to learn more about map typography and get practi-
cal design specifications for starting a map project.
FIGURE 2 A screen capture of TypeBrewer (http://www.typebrewer.org/), showing a combination of Meridien and Frutiger
type families using small size, high density, and medium tracking. (From http://www.typebrewer.org/; the original is in color;
courtesy of Ben Sheesley.)
505
12:35:00.
Web Mapping
Users can Explore type on the map by using slider bars Maps, Multimedia Maps, Printed Maps, Printable Maps,
to change the type size, density, and tracking. Each slider and Map Layers Data Warehouse. Interactive Maps
has three settings: high, medium, and low. By moving allows the user to map numerous types of broad
one of the sliders, all type on the map changes according thematic categories, including agriculture, biology, cli-
to that slider definition. For instance, by moving the size mate, environment, geology, history, people, transporta-
slider, all type on the map either enlarges or shrinks in tion, and water. Each of these broad categories includes
size. While an interesting approach, it would be unusual subcategories; for instance, the climate category
for a mapmaker to choose to increase all type uniformly includes average annual precipitation, major hurricanes
across a map. Rather, the type specs of individual fea- that move over land, and tropical cyclones. Color Plate
tures such as all capital cities would likely differ from the 2 illustrates a map of major hurricanes that passed over
type specs of other features. Type density, or the number land (“landfalling” hurricanes) in the 2000s. Note that
of type categories shown on the map, can be changed each hurricane’s path is represented by different colors
from low density (e.g., showing countries and water bod- that suggest the varying wind speeds along a given
ies), to medium density (e.g., showing capital cities, rivers, storm’s track. Given our discussion of color in this text,
and islands), to high density (e.g., showing major cities, you might want to consider whether the colors used to
physical features, and marginalia). Tracking involves depict different speeds of the hurricane are appropri-
changing the letter spacing (also known as kerning) and ate. In addition to being able to select from hundreds of
is useful to help reinforce the geographic extent of fea- data sets to map, this site allows basic interactivity lev-
tures. Low tracking shows the individual letters com- els of zooming in and out, to a specific state or to all 48
pressed together, while higher tracking shows the letters or 50 states. Interactivity is further enhanced through
farther apart. By changing the size, density, and tracking, the ability to identify various data attributes. For
users get a sense of how each typographic variable alone instance, by choosing the Identify button and clicking
or in combination can impact the overall map design. In on a hurricane track, a small window appears giving a
Figure 2, note that we elected to view the description of the hurricane.
Formal|Cabernet type template, which includes the type One limitation of the The National Atlas of the
families of Meridien and Frutiger. We also chose a small United States is that data for enumeration units are
text size, high density, and medium tracking. There are invariably mapped using the choropleth method. As we
also options to change the appearance of the base map. have seen with American FactFinder discussed above,
For example, the terrain can be toggled on or off. Type this problem is not peculiar to this atlas. Like American
highlight or the use of a thin white highlight around indi- Factfinder, The National Atlas of the United States con-
vidual letters can also be toggled on and off. Another tains a variety of base data, such as cities and towns,
interesting option is having the type be the only visible roads, and streams. A key characteristic of the atlas (and
element on the map, which demonstrates the importance one shared by several other atlases we have examined)
of how type reinforces the layout of geographic space. is its ability to automatically control the amount of
information displayed as a function of scale. For exam-
Medium Interaction: National Atlas ple, when “cities and towns” is selected and the entire
of the United States country is shown, only a few major cities appear. As the
In 1970, the U.S. Geological Survey produced The user zooms in, more and more cities are displayed. This
National Atlas of the United States of America, a paper feature also illustrates the problem of labeling cities
atlas depicting national conditions in the United States that are located closed together: all labels in the atlas
in the mid-1960s. Although this large and weighty atlas are placed just above a city location; more elegant label-
(all 12 pounds of it!) proved to be useful and contained ing routines are possible.
many intriguing maps, it can be argued that a national The Dynamic Maps section of the The National Atlas
paper atlas would be limiting in today’s rapidly chang- of the United States (http://www.nationalatlas.gov/
ing digital age. Thus, in 1997, the United States began dynamic.html) utilizes animation and interactivity to
developing an electronic national atlas, known as The examine a variety of interesting topics. Particularly
National Atlas of the United States. The Web site for the intriguing is the Tapestry of Time and Terrain, which
resulting atlas (http://www.nationalatlas.gov/) indicates combines Pike and Thelin’s digital shaded relief map
that one of the key driving goals was to produce an atlas with a geologic map of the United States. One way to
“intended as an essential reference for all computer combine these maps is to gradually blend them in an
users—not just the scientists and decision-makers who animation: the view thus begins with just the geologic
are our traditional customers.” map shown, and then relief information is gradually
The National Atlas of the United States contains five added; finally, the geologic information is removed until
major categories of mapping capability: Interactive
12:35:00.
Web Mapping
only the relief information appears. Although we found Vector Graphics); however, Google Earth can also run
ourselves questioning the utility of this animation, we on the Mac OS and Linux environments.† Google Earth
also found that the interactive portion of Tapestry of promotes three cost-based versions: Plus, Pro, and
Time and Terrain provides many useful features, includ- Enterprise, each offering additional features such as
ing the following options: The Two Maps, Description of increased performance and the ability to import GPS
Features, Color Legend, Boundaries, Play Puzzle of points. World Wind, on the other hand, is completely
Regions, and Panorama Movie. Color Legend illustrates free. Both programs are downloadable via the Web and
the power of interactive mapping, as one can click on a require a live Internet connection to operate. While both
portion of the geologic time scale (Color Plate 3A) and programs offer the same basic features of being able to
see the location of associated rocks in the United States zoom and fly around the virtual globe as well as view a
(Color Plate 3B). Other topics in the Dynamic Maps variety of geospatial data, they have differences in inter-
section illustrate the potential of linking text and pho- face design, available data, level of interactivity, access
tos with maps, and, of course, the potential of linking to source code, and data manipulation. Rather than dis-
information via the Web (for example, see the cuss each software package separately, we will review
Animated Invasive Species Map, which depicts the both products simultaneously to provide examples of
invasion of zebra mussels into rivers and lakes of the common and unique features.
United States).
Printed Maps (http://www.nationalatlas.gov/printable. Common Features. The most basic aspect of both
html) allows users to order a variety of large-format products is their ability to zoom to a location when the
printed maps, such as a shaded relief map of North user types in the name of a geographic entity (e.g., a city
America and a map depicting electoral votes by politi- or county), supplies coordinate locations for a point, or
cal party and state for all presidential elections from enters a street address. The navigational controls can
1789 to 2000. Printable Maps allows users to print also be used to zoom into an arbitrarily chosen loca-
maps of interest that are suitable for home printing tion. With Google Earth, street-level address searching
(examples of topics include West Nile Virus and bound- is available for a limited number of countries such as
aries for congressional districts). Finally, Mapping the United States, Canada, China, and portions of Eu-
Professionals (http://www.nationalatlas.gov/pros.html) rope. World Wind includes street-level address search-
allows users to download actual data, which can then be ing for the United States and Australia and is powered
imported into other programs such as ArcView and by Yahoo! directory services. Both programs show sim-
ArcGIS; this is desirable for those who wish to do more ilar geospatial data such as state and county boundaries,
sophisticated analyses and visualization. With this broad place-names, transportation routes, streams and lakes,
range of features, The National Atlas of the United States satellite imagery, and elevation models. Google Earth
illustrates the considerable flexibility that electronic focuses more on displaying commercial ventures and
atlases can provide, especially when they are imple- providing users with the ability to view the location of
mented in a Web format. specific commercial enterprises such as restaurants,
lodging, and gas stations. On the other hand, World
High Interaction: Google Earth and World Wind Wind displays only features that are in the public do-
Google Earth and World Wind are two exciting new Web- main or are noncommercial, such as schools, churches,
based mapping products. Keyhole, Inc., a satellite imagery and post offices.
company, developed Earth Viewer* in 2001. Google pur- Both products display satellite imagery. Google Earth
chased Keyhole, Inc. in 2004 and renamed Earth Viewer uses imagery from a variety of commercial sources (e.g.,
as Google Earth (http://earth.google.com), which was Digital Globe, which supports the Quickbird sensor)
launched in 2005. World Wind (http://worldwind.arc. and government sources (e.g., the U.S. Landsat pro-
nasa.gov/), developed by NASA to educate the public gram, which includes Thematic Mapper 7 ETM+), and
about environmental issues, was offered to the public therefore offers a variety of imagery resolutions.
in 2004. Both Google Earth and World Wind can be However, the imagery does not exhibit a uniform tem-
considered virtual globes in the sense that users can fly poral, spectral, or spatial resolution. Since each satellite
around, zoom into places or addresses, and view a variety sensor records in different portions of the electromag-
of geospatial data. Both products are free of charge and netic spectrum, the color varies from image to image,
run under the Windows 2000, XP, and Vista environments and the crispness of the imagery is also inconsistent. The
and take advantage of open standards such as XML coarsest spatial resolution that Google Earth makes
(eXtensible Markup Language) and SVG (Scalable available, which covers most of the Earth, is 15-meter, but
* For more information on Keyhole, see http://www.keyhole.com/ † A beta version of World Wind is circulating that will deploy in the
index.html and http://en.wikipedia.org/wiki/Keyhole,_Inc. Mac and Linux operating environments.
507
12:35:00.
Web Mapping
higher resolutions (e.g., 2-meter, 70-centimeter, 12-inch, (e.g., the location of Mount Saint Helens’ volcanic
6-inch, and, amazingly, 1-inch) are available for selected eruption) during their own World Wind session.
cities around the world. One-inch resolution is available
in only a few locations (e.g., London), but is so crisp that Distinctive Features. While both products offer
you can see individual chairs positioned around a table. similar features, it is the uniqueness of each that helps
One problem, however, is that some of the imagery is up to set them apart. Google Earth includes 3-D buildings
to three years old. in selected U.S. cities as well as in Japan and London.
World Wind uses satellite imagery primarily from Once active, the 3-D building layer produces gray,
Blue Marble and various government satellite pro- blocky geometric shapes, which are a bit unpleasant to
grams. The imagery also is available at different tempo- see. No details such as rooftops, bricks, or windows are
ral, spectral, and spatial resolutions. For instance, Blue added to the shapes, although a plug-in allows users to
Marble, Lansdat 7 Global Mosaic, and MODIS imagery interact with the shapes and add details to buildings,
is available at 1-kilometer, 15-meter, and 250-meter res- making their appearance more realistic. An interesting
olutions, respectively. Higher resolution imagery such aspect of the 3-D buildings feature in Google Earth
as 1-meter USGS orthophotos and SRTM data also is is the user’s ability to change the viewing angle and
available. With both products, the differing resolutions perspective to achieve dramatic views. Another feature
can be exciting to view, but if users are not conscious of unique to Google Earth is users’ ability to generate
the differing resolutions when panning or flying around driving directions, similar to MapQuest.com or
the virtual globe, they may become confused when YahooMaps.com, but with a distinct difference: In
scenes are blurred (due to coarse spatial resolutions) or Google Earth, users can enter a “from” and a “to” ad-
when they must zoom out to a specified altitude before dress or city, and the route will appear on a map. The
the imagery appears. specific route directions, as well as mileages for each leg
Both the Google Earth community (http://bbs. of the trip, are also provided in written form. Once the
keyhole.com/) and World Wind Central (http://forum. route is shown, users can choose to display various land-
worldwindcentral.com/) offer extensive forums for dis- marks along the route (such as gas stations, restaurants,
cussion with and assistance from other users. It is and lodging) overlaid on top of the satellite imagery. By
through these forums that users learn about the variety clicking on these landmarks, users can learn specific de-
of customization options. With Google Earth, users can tails such as the street address and phone number of a
create “tours” of places around the world and then post hotel or restaurant. Because World Wind is not intend-
these tours on the Google Earth Community forum for ed to promote commercial ventures, the addition of such
others to view. For example, the entire 2005 Tour de a feature is not likely to occur. However, World Wind of-
France bicycle race route is available for viewing (see fers Place Finder, an interesting feature not found in
http://www.gearthblog.com/blog/archives/2005/09/ Google Earth. With Place Finder, users can search for
watching_sports_1.html). When viewing this tour, users geographical features by using keywords such as
begin the “race” at the departure city and then are “stream” or “populated place.” Information on the
whisked along each stage of the race. Place markers name of each feature, the country in which it is located,
are located along the route, indicating significant points and its latitude/longitude coordinates is then listed, and
in the race (e.g., a small mountain icon represents a users can click on any feature and zoom to it on the
steep climb up a mountainside). Users can click on map. When we searched for “dam” in World Wind, more
these place markers to learn additional information than 58,000 names from all over the Earth were pre-
about the specific locations and what the cyclists sented. Color Plate 4 shows one of these dams, the
encountered along the way. Eventually, viewers cross Grand Coulee, which is located along the Columbia
the finish line at the end of the day’s race. River in Washington State.
Similar in concept to Google Earth, World Wind As described earlier, World Wind was designed as a
also allows its user community to create and post tool to educate the public about the environment. To
“hot-spots” on World Wind Central (http://www. help achieve this goal, World Wind incorporates the
worldwindcentral.com/hotspots/). Hot-spots are char- research efforts of Goddard Space Flight Center’s
acterized as interesting places around the world and Scientific Visualization Studio (http://svs.gsfc.nasa.
are categorized into topics such as national parks, gov/), whose aim is to “promote a greater understand-
waterfalls, natural disasters, etc. Users create a screen ing of Earth and Space Science research activities.”
capture in World Wind of the location of interest and The Center accomplishes this goal through producing
then post the screen capture to World Wind Central, hundreds of visualizations covering the entire spec-
along with a short textual description of the screen trum of the Earth’s environment, such as climate
capture’s importance. Other users can then click on change, magnetic fields, and CO2 emissions. These
the “hot-spot” and instantly view the same location visualizations are seamlessly integrated into World
12:35:00.
Web Mapping
Wind and are easily accessible through a menu. As an the World Wind community through World Wind
illustration, Figure 3 shows a screen capture of an ani- Central, which in turn helps shape the kinds of func-
mation displaying the break up of Iceberg B-15A in tionality that are continually being incorporated into
Antarctica’s McMurdo Sound. This iceberg was 3,000 the software. Even though Google Earth’s source code
square kilometers, or about the size of Long Island, remains proprietary, users do have the ability to inter-
NY, and was the largest fragment of a much larger act with the Google Earth software through KML
iceberg that broke away from the Ross Ice Shelf (keyhole markup language). This is a tagged language
in March 2000. Users are provided with an animation (similar in concept to HTML) that allows geographical
control device that permits some choice over the ways data to be displayed. KML, like HTML, is rather simple
in which the visualization is presented, as well as a to use, and once a KML file is created, it can be shared
brief description so that they can learn more about the among users, who can then integrate the KML file into
animation. their own Google Earth browser. Possibly the most
When users download World Wind, they have the popular aspect of KML is its ability to “annotate”
opportunity to obtain the program’s source code, which the Earth’s surface with place markers that identify
makes it possible for them to change World Wind’s locations on the Earth’s surface, create different camera
interface, specify new data servers, or add functionality (viewing) positions for features, overlay images onto
to enhance the program. The results can be shared with the Earth’s surface, and display textured 3-D buildings.
FIGURE 3 A screen capture of a visualization from the Goddard Space Flight Center’s Scientific Visualization Studio showing the
break up of Iceberg B-15A in Antarctica’s McMurdo Sound (located slightly below the white crosshair), as viewed through World
Wind. Note that the top window shows a partial listing of the visualizations from which World Wind users can choose. (The original
is in color; see http://worldwind.arc.nasa.gov/).
509
12:35:00.
Web Mapping
2.3 Distributed Data versus User-Contributed Data Federal-Distributed and User-Contributed Data:
The National Map
This continuum concentrates on geospatial data and Imagine, if you will, the ability to openly access, select,
how they are distributed. At one end of this contin- and view a wide variety of high-quality, seamless geospa-
uum, geospatial data are collected, prepared, and dis- tial data at any level of geographic detail from federal,
tributed under the auspices of a federal, state, or local state, and local government offices across the United
government agency. Many commercial ventures also States. Now imagine that once you have selected the
have Web sites devoted to providing geospatial data to desired data at a specific scale, you are able to freely
the broader Web community. Collectively, these Web download that data to your computer and continue with
sites are commonly referred to as geospatial data clear- your mapping project. The U.S. Geological Survey
inghouses because they distribute a variety of geospa- (USGS) has taken this dramatic concept and created
tial data types. In most instances the data are freely The National Map (http://nmviewogc.cr.usgs.gov/).
available, but in some instances, the data may be costly, The National Map is an example of a combination of
such as when the data are recent, have high accuracy, user-contributed and federal-distributed data. Currently,
or have a large amount of attribute information the USGS is the largest contributor of geospatial data
attached to them. In addition, users of data clearing- through its mapping efforts, but the organization hopes
house sites generally have no ability to contribute that more entities will be able to contribute to the build-
their own geospatial data to the data that are redis- ing of The National Map in the near future.
tributed. Here we focus on three levels of distributed The National Map utilizes ESRI’s ArcIMS (Internet
data: state level, private business level, and federal Mapping Server) technology to offer a high level of
level. User-contributed (and ultimately redistributed) interactivity that enables users to browse through the
data are supplied through Web sites that allow individ- myriad of geospatial data. When The National Map
uals or agencies to contribute their own data for oth- loads, users see a map of the United States flanked by a
ers to view and download. list of data sets on the right and a list of tools on the left.
The available data sets, grouped by theme, include but
State-Level Distributed Data: West Virginia’s are not limited to administrative boundaries, elevation,
Department of the Environment hydrology, imagery, land use/land cover, topography, and
Many states offer a Web site devoted to distributing the transportation. Inside each theme are numerous options
geospatial data of that particular state. For instance, from which to choose. For example, inside the trans-
West Virginia’s Department of the Environment portation theme, users can view miscellaneous trans-
(WVDEP) (http://gis.wvdep.org/) offers users a range of portation options (e.g., airports, heliports, seaplane
geospatial data focusing on West Virginia. On bases), railroads, and road network data contributed by
WVDEP’s main page is a link to “Mappings, GIS or the Bureau of Transportation Statistics, USGS, and the
GPS,” which takes the user to the WVDEP GIS server. U.S. Forest Service. Once the geospatial data are dis-
The available GIS data include DRG (digital raster played, the tool listing provides users with a consider-
graphics), aerial photography, mining permit data, able range of GIS functionality. In addition to the com-
SRTM (shuttle radar topographic mission), USGS mon zoom, pan, and re-center tools, users can get
NED (a national elevation data set), FEMA floodplain information about features shown on the map, query the
maps, and historical maps (i.e., scanned versions of map to find the elevation at a specific point, measure
USGS 1:62,500 topographic quadrangles). Not surpris- distances between points, bookmark the current map
ingly, this data clearinghouse tailors its data to the inter- display, change user preferences (e.g., select measure-
ests of the WVDEP; for instance, one data set of partic- ment units), and mouse over coordinates. Color Plate 5
ular interest to WVDEP deals with coal mining and shows an example of the interface and data layer possi-
related activities such as mine reclamation. bilities included in The National Map. The data shown in
The data distributed through the WVDEP site have Color Plate 5 include elevation data and the class-1 rail-
been derived from outside sources such as the USGS roads along the Front Range of the Rocky Mountains of
but manipulated so that they focus entirely on West the United States. Additional layers of data are easily
Virginia. For instance, all of the data sets have been selectable along the right-hand side of the map.
assigned a common coordinate system of the UTM, Changing scales is accomplished by clicking on the ver-
zone 17, NAD83 datum. Also, any data that fall out- tical bars at the top right. Various tools are listed along
side of West Virginia have been removed. The aerial the left-hand side of the screen and offer a considerable
photography is from USGS DOQ, and any DOQ from range of functionality.
surrounding states that does not fall partially within In partnership with a host of government and
West Virginia’s borders is not available. commercial agencies, the USGS has promoted The
12:35:00.
Web Mapping
National Map as a “one-stop shop” for geospatial data, sites update data on a regular time interval (e.g., every
where each member entity (e.g., a local government) 15 minutes). Non-updated Web sites take the opposite
works on its day-to-day mapping/GIS operations while approach in that once a map is posted on the site, it is
at the same time sharing this data with the global never changed.
geospatial community via The National Map. The even-
tual goal is to make The National Map a nationwide Updated Data: AirportMonitor
warehouse of geospatial data that is accessible to the AirportMonitor 2.0 (http://www.passur.com/am_airport.
widest community of geospatial data users. htm) is an example of a Web site that posts information
in real time. This is a free Web site that shows the current
User-Contributed Data: New York City Subway status of aircraft arrivals and departures at selected
Smell Map airports across the United States. AirportMonitor is a
Many maps are built and updated using information software solution designed and produced by Passur,
supplied by the broader Web community. One example, which provides airports with custom software solutions
the New York City Subway Smell Map (http://www. that assist in air traffic management, safety, and security.
gawker.com/maps/smell/), presents the subway routes, Current aircraft departures, arrivals, and those in transit
station stops, and associated smells throughout the sub- across the airport’s control space are shown on a map by
way system in lower Manhattan. The base map is pow- small airplane icons in green, blue, and black hues,
ered by Google Maps. What makes this Web site unique respectively; additionally, there is a separate icon for
is that it relies entirely upon users to supply the data. In helicopters. As the aircrafts’ positions change, the icons
other words, subway station users are asked to send in reposition themselves automatically across the map.
their descriptions of smells they encountered when they According to the AirportMonitor Web site, aircraft posi-
last visited a specific subway station. tions and associated information are updated every 4.6
When a user positions her or his mouse over a station seconds. Users can click on an aircraft to learn of its flight
icon on the map, a small balloon pops up identifying the identification, type, altitude, origin, and destination.
station name and the different subway trains that stop at There is also a zoom option that allows users to set the
that station. Along with the station information, various range of viewable airspace around the target airport at 4,
icons may appear showing the kinds of smells that were 10, 20, 40, and 90 miles. The base map is powered by
reported by users. There are 10 different icons ranging Mircrosoft and includes streets, hydrography, and popu-
from pleasant smells (e.g., food) to unpleasant smells (e.g., lated places that are commensurate with the chosen
sewage). Color Plate 6 illustrates the smells associated zoom level. Users also have the ability to view all airline
with the Spring Street/Lafayette Street station. In this case, traffic and associated information for a given airport
the smell icons indicate that an individual reported body over the past three months by selecting a specific day
odor and food odors wafting from this particular station. and time from the available pull-down menus.
In addition to reporting the kinds of smells emanating
from the subway station, individuals can also submit Updated Data: i Vidlog
commentary and a short statement on their personal expe- An interesting approach that combines static maps with
rience while exposed to the smell. To see the personal video is iVidlog (http://www.dot7.state.pa.us/ividlog/
statements, Web site users simply double-click on the sta- video_locate.asp), provided by the Pennsylvania
tion dot. Obviously, the utility of such a Web site is limited Department of Transportation (PennDOT). The goal of
to those individuals who take the New York City Subway this site is to assist local planners and other agencies in
and frequent a specific station.Although users do not have monitoring the pavement conditions of key roadways
the ability to download data per se, this site demonstrates across the state by utilizing static maps and virtual dri-
how the contributions of map information by nonspecial- ving videos of the Pennsylvania road network. Users
ized individuals can benefit the greater community. begin by selecting a location to start their virtual trip
(e.g., by entering a street name, city, or latitude/longi-
2.4 Updated Data versus Non-Updated Data tude). In response, a static map of the location entered is
displayed depicting a variety of base information. The
This continuum distinguishes between Web sites that road network is divided into small color-coded segments
update data on a regular schedule and Web sites that do (usually, less than 1 mile) that users can select. Streets
not. The concept of a regular schedule suggests that the that appear in red can be seen as a video (streets with-
data or other map-based information are posted to a out video appear in black). Once the user selects a seg-
specific Web site shortly after an event took place. ment, three side-by-side windows appear that display a
While several Web sites post data as an event happens, video of what the route looks like when traveling along
in more practical terms, many so-called “real-time” Web it in a vehicle. The screen in front shows the road directly
511
12:35:00.
Web Mapping
ahead of the virtual driver, and the left and right screens Color Plate 8 shows a map of Iraq ethnic groups as they
show the scenes off to the driver’s left and right sides, existed in 1978 that was downloaded from the site.
respectively. Color Plate 7 shows the view as one crosses
the steel-grate bridge into Glencoe, PA. In combination 2.5 Animated Map versus Static Map
with the video displays, a map appears that also shows
the segment over which the virtual driver is traveling, This continuum concentrates on Web sites that have the
highlighted in green. ability to offer maps that demonstrate a temporal change
The iVidlog interface is designed to allow users to or a change in geographic position versus Web sites whose
change the frame rate of travel, essentially slowing maps show only a static representation of a data set.
down or speeding up the driving experience. Users can Animation as a way to visualize data has been a focal
also travel the route one frame at a time, change the point of cartographic activity for the past few decades.
frames displayed (e.g., show every third frame), or One benefit of this activity is the considerable progress
“back up” along the route. Since the video runs in only that has been made in the ability to incorporate map ani-
one direction, users cannot “turn” their virtual vehicle mation into our daily lives. Animated maps can show
around and travel the other direction. When the vir- changes in a data set over time, such as watching Doppler
tual driver has reached the end of the segment, users radar depict precipitation movement across a local area
can continue onto the next segment, turn at the inter- during the past six hours. Animated maps can also illus-
section, or back up along the previously driven seg- trate a change in geographic position such as “flying over”
ment. Given the scale of the project, not all of the a landscape or “zooming into” a geographic location.
video was captured in the same time period. Some Static maps are, as their name implies, single maps on
parts of the state are shown at the peak of vegetation which no change in the data or map focal point is
(such as in Color Plate 7), whereas other sections are depicted. Static maps in printed form have been around
shown without vegetation. This inconsistency can be for thousands of years and, despite the prevalence of tech-
significant—for instance, traveling along a route and nological advances, continue to be the most prevalent
then going suddenly from summer to fall provides a type of map found on the Web.
jarring contrast.
Aside from the administrative utility of this product Animated Map: Iraq Coalition Fatalities
for city planners, this site offers users an interesting While there are numerous sites that include animated
view of Pennsylvania’s road network. For instance, maps, one site that shows the number of U.S. and other
imagine that you are relocating to a town in coalition fatalities in Iraq (http://www.obleek.com/iraq/)
Pennsylvania and need to look for housing. You could illustrates many aspects of an interesting and timely ani-
use this site to take a virtual tour of some areas of the mated map. The goal of the map is to show the location
city or countryside, look at the condition of the schools, and temporal nature of all coalition fatalities in the Iraqi
and survey the downtown area. Tourists may also use Theater of Operations since the start of the Iraq war in
this site to explore scenic roadways, noting where, for 2003. Figure 4 shows a snapshot of the animation on
example, antique shops are located. Teachers and his- August 2, 2003. The animation runs day by day beginning
tory buffs may also find this site interesting because March 20, 2003, and continues through the last day that
they can, for example, take a virtual tour of the roads the map was updated. As the animation runs, viewers see
around the Gettysburg National Military Park, where a single large red “flash” with a small white dot in the
the battle of Gettysburg was fought in 1863. center that represents a single fatality, which is intended
to grab the viewer’s attention. The white dot quickly
Non-Updated Data: Perry-Castañeda Map Library changes to black and after approximately 30 days, fades
The Perry-Castañeda Map Library at the University of to gray, while at the same time the brilliant red “flash”
Texas at Austin (http://www.lib.utexas.edu/maps/) is an fades away. There is also a date indicator in the top right-
example of a collection of maps that are not updated. In hand portion of the screen, indicating the current date of
fact, most of the maps available through this map the animation. As each fatality appears, a distinctive “tic”
library are in the public domain (i.e., the bulk of these sound is heard. The volume of the “tic” depends on the
maps come from U.S. and other governmental sources). number of simultaneous fatalities that occurred on a par-
Users can browse approximately 11,000 maps that have ticular day. In other words, the more fatalities that
been digitally created from among the 250,000 map occurred on one day, the louder the “tic,” while fewer
sheets that exist in the printed collection and are avail- daily fatalities produce a muffled “tic” sound.
able either in .jpg or .pdf format. Users can select a map The animation was created using a multimedia
categorized according to a geographic region (e.g., authoring program called Flash, which is one of sev-
world, county, Africa, Asia, Europe), a topic (e.g., avian eral multimedia authoring tools that have found con-
flu), and cartographic references such as a gazetteer. siderable popularity with Web developers who want to
12:35:00.
Web Mapping
FIGURE 4 A snapshot of an animation of U.S. and other coalition fatalities during the Iraq war, which began in 2003. (From http:/
/www.obleek.com/iraq/; the original is in color; courtesy of Tim Klimowicz.)
include multimedia content in their Web applications. browser, or integrated into a presentation as self-
Much of the Web’s content, especially advertisements executing animation.
that contain animation or integrated video and sound, While this site certainly evokes a range of emotions,
is made possible by Flash. Technically, “Flash” is a we take issue with the lack of control users are permitted
generic term for two separate products: the software to have over the animation. Aside from turning on or off
that develops Web-based multimedia applications and the place-names and base information and selecting a
the plug-in for Web browsers that play Flash movies. particular country’s fatalities, the user has no control
The software development tool Flash is an example of over the progression of the animation. In some cases,
an IDE (integrated development environment), which allowing users to slow down or speed up the animation
is a set of tools that allows programmers to compile would be a useful addition. In a similar light, rather than
source code, build graphical user interfaces, and debug having to view the entire animation from start to finish,
programs. Flash uses a scripting language called users might prefer to specify a date range that corre-
ActionScript to enhance animations or graphical user sponds with some significant event that happened in
interface options. For example, ActionScript can be Iraq. Users may also be interested in learning about the
used to direct action that the animation takes when a details of the fatalities in a given area. Thus, not being
user presses a specific button on the graphical user able to zoom in on the map is another problematic fea-
interface. Once an application is built using Flash, the ture of this site. For obvious reasons, specific details
animation (usually called a Flash file) can be exported about each fatality (e.g., name, country of origin, etc.) are
as a .swf file and played on its own, played through a omitted from the site, but interactivity allowing users to
513
12:35:00.
Web Mapping
FIGURE 5 A map of Yosemite from

1879 (part of the Wheeler Survey of
the western United States) overlaid
on a modern DEM. The David
Rumsey Map Collection permits
users to fly through this environ-
ment. (Courtesy of the David
Rumsey Map Collections, www.
davidrumsey.com.)
click on individual dots to learn about the fatality could that “Many historical maps were created before the
be provided. It would also be useful if the different coun- availability of modern mapping and surveying
tries were color-coded in order to give a spatial perspec- technologies, and some of the items were tourist maps
tive to the distribution of fatalities by country of origin. that were visually surveyed or based on an artist’s ren-
dering.” One nicety of the GIS browser is that when one
Static Map: David Rumsey Collection zooms in on any of the four maps, all four maps are
The David Rumsey Map Collection (http://www. shown at the same enlarged scale.
davidrumsey.com/) is an online set of more than 8,800 Another intriguing capability provided in Rumsey’s
(primarily) static maps (and map-related items such as Map Collection is that of being able to create three-
globes) from David Rumsey’s personal collection of dimensional images by overlaying some of the historical
more than 150,000 eighteenth- and nineteenth-century images on digital elevation models (DEMs). For
maps of North and South America (Rumsey 2003). For instance, Figure 5 shows a map of Yosemite Valley in
those interested in historical maps, this is a tremendous 1879 overlaid on a DEM. Users who have the necessary
resource, since most historical maps traditionally have graphics hardware can fly through these 3-D images and
been available only in paper form in libraries. Color produce a virtual environment.
Plate 9 illustrates an example from this impressive col-
lection—in this case, William Henry Holmes’s 1882 ren-
dition of the Grand Canyon. 3 TYING TOGETHER THE FIVE CONTINUA
In addition to making the historical maps available
online, Rumsey has linked some of the maps to modern In the previous section, we discussed five continua rele-
geographic data, such as satellite images, and to a variety vant to describing the nature of maps and geospatial
of vector and raster data available from the U.S. data found on the Web. It is important to note that these
Geological Survey. For instance, Color Plate 10 shows a five continua are not necessarily orthogonal to one
QuadViewer option available within Rumsey’s GIS another and that any particular Web site will have some
browser. In this case, four historical maps of Boston are “score” along each continuum. In this section, we will
shown with the modern-day street network (in red) over- tie together aspects of the five continua to demonstrate
laid on them. Note that this image reveals some of the how one Web site can exhibit characteristics of several
difficulty of linking historical and modern data, as the continua. For illustrative purposes, we will work with
modern streets do not line up precisely with the histori- the National Weather Service’s (NWS) Web site (http://
cal data. Rumsey (2003) recognized this problem, noting www.nws.noaa.gov/). The NWS displays two types of
12:35:00.
Web Mapping
maps on this Web site: current watches and warnings, and saving the image. Most of the images are available in
and Doppler radar. Both maps encompass the United .gif format. This site further permits the radar images to
States (including Alaska and Hawaii). In addition, be downloaded and incorporated into GIS software. An
approximately 50 NWS regional offices cover different enhanced version of the radar images, called RIDGE
portions of the United States. For instance, the Sterling, (Radar Integrated Display with Geospatial Elements), is
VA, radar shows Washington DC, Maryland, and defined in a coordinate system that can easily be incor-
Delaware along with portions of West Virginia, Virginia, porated into a GIS. This Web site also allows users to link
Pennsylvania, and New Jersey (Color Plate 11). Each to other governmental sites and download historical and
regional office provides current watches and warnings current climate data in text-based form. Thus, we con-
as well as Doppler radar images for these smaller sider this site to offer a considerable range of federally
geographic areas. distributed data.
The watches and warnings map shades affected U.S. All information on this site is updated frequently.
counties using various colors to indicate the kind of However, different features have different update fre-
weather threat that is looming (e.g., a red hue indicates quencies. Special weather-related announcements are
a severe thunderstorm warning). The Doppler radar made as the weather situation warrants (e.g., new
image also uses a color sequence to indicate the inten- watches and warnings are posted to the maps as soon as
sity of precipitation (e.g., green means light rain while they are made). The Doppler radar images are updated
red means very intense rain). On these small-scale every 30 minutes, and the text-based forecasts are
maps, there is no opportunity for the map user to updated on the hour. From this standpoint, the Web site
change the symbolism in any way (e.g., the colors are scores high on the updated side of the updated/non-
predefined); thus the maps are completely preexisting. updated continuum.
However, users can customize other maps on this site. This site contains both animated and static maps.
For instance, users can customize NWS regional maps Although the Web site contains a wide range of static
of the Doppler radar images. By default, the Doppler maps, most of the static maps can be animated. The
radar image of the regional map appears in conjunction nationwide watches and warning maps can be ani-
with different data layers (e.g., roads, cities, county mated to show when the watches and warnings were
boundaries, rivers, and elevation). The legend and storm issued. In a similar manner, all of the Doppler radar
warnings are also shown on their own layers. Map users images can be animated. For instance, the map showing
have the ability to toggle each map layer “on” or “off” the Doppler radar image of the United States can be
to show data they feel are important when creating “looped” to show the past 30 minutes of precipitation
their own weather maps (Color Plate 12). Despite the movement. Users also have the option to view an ani-
ability to turn layers on and off, we feel that this level of mation of satellite images of the North American
customization is rather minimal compared to the cus- continent in 24-, 12-, 6-, or 3-hour time loops. Other
tomization possible on other Web sites we have dis- maps of weather-related data (e.g., precipitation, tem-
cussed. perature, and wind speed/direction) let users view
The user has some level of interaction with the NWS weather-related data displays in three-hour time inter-
maps. For instance, on the U.S. map, the user can click vals (Color Plate 12). Thus, this Web site spans the sta-
on a given NWS regional office and zoom into a larger tic/animated map continuum.
scale map of that region. If the user clicks on the NWS When considering mapping products available
Washington, DC–Baltimore regional office, a new map through other Web sites, we suggest that you use our
appears showing the geographic area that this office five continua to determine how the products “score”
covers. On this larger scale map, users can see the along each continuum as we have done here. You also
watches and warnings that have been applied to indi- may wish to develop other continua (such as two-
vidual counties. Users can also pan to other NWS dimensional vs. three-dimensional maps) to see how the
regions of coverage to view current watches and warn- products score on those continua.
ings. Users also have the opportunity to point and click
on a given location within a county to read a detailed
five-day forecast for that location. While not providing SUMMARY
a sophisticated level of interactivity, this site does offer
a minimum level of interactivity appropriate for the In this chapter, we have examined the World Wide Web
diverse users that typically visit the site. and considered how this technology has impacted the
In addition to allowing users to view radar images and manner in which individuals interact with, distribute, and
read weather forecasts, the NWS site allows users to view mapped data. We began with a distinction between
download data in several forms. Users can also download the World Wide Web and the Internet. The Internet can
all of the map-based images on this site by right-clicking be thought of as a series of interconnected networks that
515
12:35:00.
Web Mapping
enable information to be passed back and forth. These The third continuum focuses on Web sites that
networks were developed in the 1950s as a result of distribute geospatial data versus those sites whose data
scientists needing to interact with one another in a timely are fully acquired from user contributions. The U.S.
and efficient manner. Early technological developments Geological Survey’s National Map Web site is an exam-
such as File Transfer Protocol (FTP) allowed networks to ple in which geospatial data are freely distributed for
communicate with one another by passing files back and viewing and downloading. Whereas the data housed in
forth. The World Wide Web is a series of interconnected the National Map site are contributed by federal, state,
HTML documents that are shared across the Internet and and local agencies, other Web sites exist that allow the
viewable through a Web browser. We noted that the Web public at large to contribute their own “mapped data.”
has not only changed the way in which users interact with The New York City Subway Smell Map is one such Web
data, but, more importantly, has also changed the ease site—in this case, users specify the “odor” they experi-
with which data are made available and distributed to enced while traveling through one of the many subway
Web users worldwide. stations, thus linking that odor to a map location. Other
We presented a classification of types of cartograph- users can then view a map showing the kinds of smells
ically related Web sites. Our objective in classifying Web that were encountered in the various subway stations
sites was to provide an overview of the more interesting throughout the New York City Subway System.
approaches in which cartographic products have been The fourth continuum distinguishes between Web
made publicly available across the Web. Our classifica- sites that continually update data versus those that offer
tion contains five continua. The first continuum divides only non-updated maps. AirportMonitor is a Web site
those maps that are preexisting from those that are that offers near-real-time traffic situations at several
custom-tailored. We highlighted the National Park airports across the United States. Users can see inbound
Service’s Web site as a site that offers a considerable and outbound flights as well as those flights in transit
range of preexisting maps that are freely downloadable within 50 miles of selected airports. The Perry-
in .pdf format. Contrasting with this site is the American Castañeda Map Library from the University of Texas at
Factfinder Web site, where users have control over Austin is an example of a Web site that offers thousands
options such as the data to be mapped, the color of static, non-updated maps for free downloading.
scheme, and the method of data classification. The fifth continuum focuses on those Web sites that
The second continuum distinguishes between Web offer animated maps versus those that display static maps.
sites that provide users a low level of interaction with a We presented a compelling animation showing the daily
given map and associated data and those sites that pro- number of fatalities for U.S. and other coalition forces
vide users a high level of interaction with a given map during the Iraq war in order to illustrate the power that
and associated data. TypeBrewer is one Web site for animation has when presenting a data set. Contrasting
which users have a rather low ability to interact with a with animation is the amazing collection of static maps
map. For example, users have a limited set of options provided through the David Rumsey Collection.
that assist in the selection of type (e.g., choosing a type We concluded the chapter with an examination of
family and modifying type sizes). On the other hand, how the five continua apply to the National Weather
Web sites such as Google Earth and World Wind Service’s Web site. In a similar fashion, we encourage
offered a considerably greater range of interactivity you to use the five continua to examine your favorite
options. Both Google Earth and World Wind allow cartographically related Web sites to see how they fit
users to specify a geographic location to zoom in and along each of the continua. Bear in mind that these five
pan around, to display a variety of base information, continua are not exhaustive, so you may find other con-
and to customize the information content of maps. tinua useful in distinguishing among Web sites.
FURTHER READING
Brown, M. (2006) Hacking Google Earth and Google Maps. One of the earlier attempts at classifying cartographically themed
New York: John Wiley. Web sites.
Provides a technical overview on designing custom interfaces and Cartwright, W. (2003) “Maps on the Web.” In Maps and the
integrating new data into Google Earth and Google Maps. Internet, ed. by M. Peterson, pp. 35–56. Amsterdam: Elsevier.
Calkins, John. (1996) “The Internet and GIS Data.” From Discusses various kinds of cartographically related products that
Pima County Arizona Department of Transportation are delivered via the Web
Geographic Information Services Division (http://www.dot.co. Cartwright, W., Peterson, M., and Gartner, G., eds. (1999)
pima.az.us/gis/webdev/; accessed August 29, 2007). Multimedia Cartography. Berlin: Springer-Verlag.
12:35:00.
Web Mapping
Focuses on the technological developments that allow various Kraak, M.-J., and Brown, A., eds. (2001) Web Cartography:
kinds of cartographically related Web sites to function. Developments and Prospects. London: Taylor & Francis.
Cartwright, W., Peterson, M. P., and Gartner, G., eds. (2007) Covers the impact that cartographic products appearing on the
Multimedia Cartography, 2nd ed. Berlin: Springer-Verlag. Web have had on map use, map users, and map design.
Updated discussion on various facets of the Web as a delivery Langer, M. (2002) Gebruikerswensen bij online kaartbebruik
medium of multimedia cartography. Chapters 9 through 11 are of [User requirements in online map use]. Unpublished Master’s
particular relevance to our discussion of the Web because they Thesis, Utrecht, The Netherlands: Utrecht University.
explore various electronic atlases such as the Atlas of Canada,
which is delivered over the Web (http://atlas.nrcan.gc.ca/site/ Presents a classification of the types of maps that are found on
english/index.html). the Web.
Crampton, J. (1998) “The convergence of spatial technologies.” Peterson, M., ed. (2003) Maps and the Internet. Amsterdam:
Cartographic Perspectives no. 30:3–5. Elsevier.
Explores several issues relating to the convergence of digital Covers various issues regarding how cartography has been
cartography, GIS, and the Web. incorporated into the Web
Crampton, J. (2000) “A history of distributed mapping.” Plewe, B. (1997) GIS Online: Information Retrieval, Mapping,
Cartographic Perspectives no. 35:48–65. and the Internet. Sante Fe, NM: OnWord Press.
Discusses the history of how GIS, the Web, and other technologies An early text on how the Internet plays a role in distributing
have changed the way in which cartographic products are distributed. geographic information to a wide audience.
Crampton, J. (2003) The Political Mapping of Cyberspace. Plewe, B. (2007) “Web cartography in the United States.”
Chicago: University of Chicago Press. Cartography and Geographic Information Science 34, no.
2:133–136.
A detailed discussion of the political, social, and ethical
issues that surround the Web, with particular attention given to Describes various Web services such as Google Earth and EROS
cartography. (Earth Resource Observation and Science) Data Center that are
available in the United States.
Dodge, M., and Kitchin, R. (2001a) Atlas of Cyberspace.
Harlow, England: Addison-Wesley. Stefanakis, E., Peterson, M. P., Armenakis, C., and Delis, V., eds.
(2006) Geographic Hypermedia: Concepts and Systems. Berlin:
Presents a variety of intriguing visualizations of cyberspace and
Springer-Verlag.
its interconnectedness. Much of the book’s contents are available
online at http://personalpages.manchester.ac.uk/staff/m.dodge/ Describes the concepts and technologies that facilitate how the
cybergeography/atlas/atlas.html. Web, hypermedia (e.g., graphics, animation, and video), and geo-
graphic information sciences are successfully integrated under a
Dodge, M., and Kitchin, R. (2001b) Mapping Cyberspace.
new paradigm called “geographic hypermedia;” focuses mostly on
London: Routledge. theoretical and research issues.
A discussion of how the Web has changed the distribution and Taylor, D. R. F., ed. (2005) Cybercartography: Theory and
conceptualization of spatial information.
Practice. Amsterdam: Elseiver.
Harder, C. (1998) Serving Maps on the Internet: Geographic
Although not specific to the Web, includes discussions on how the
Information on the World Wide Web. Redlands, CA: ESRI, Inc. Web and other technologies have changed the ways in which maps
A nontechnical overview of how GIS information is distributed organize, communicate, and present geospatial information;
via the Web. largely focuses on theory and research issues.
Harrower, M., Keller, P., and Hocking, D. (1997) “Cartography van Elzakker, C. (2000) “Use and users of maps on the Web.”
on the Internet: Thoughts and a preliminary user survey.” Cartographic Perspectives no. 37:34–50.
Cartographic Perspectives no. 26:27–38. Discusses the characteristics that comprise those using the Web
Early research on Web users’ perceptions of maps delivered via for map-related products. In general, issue no. 37 of Cartographic
the Web. In general, issue no. 26 of Cartographic Perspectives Perspectives contains interesting and relevant articles dealing with
offers some interesting, but early, articles on the Web. the Web and cartography.
GLOSSARY
animated maps: maps characterized by continuous change, graphical user interface (GUI): a type of computer interface
such as the daily weather maps shown on television newscasts in which the user specifies tasks to be performed by pointing
that depict changes in cloud cover. to the desired task, typically by using a mouse.
File Transfer Protocol (FTP): a set of methods defining the hypertext transfer protocol (HTTP): a set of methods defin-
manner in which electronic files and documents are sent and ing the manner in which electronic files and documents are
retrieved via the Internet. sent and retrieved via the World Wide Web.
517
12:35:00.
Web Mapping
Internet: a global network of computers that permits shar- static, as their information content does not change while the
ing of data and resources. user looks at the map.
Internet Protocol (IP): a method of transferring data across World Wide Web: a system of Internet file servers that sup-
a network. port the hypertext transfer protocol.
static maps: maps whose information content is unchanging,
such as a traditional printed map; many online maps are also
REFERENCES
Calkins, J. (1996) “The Internet and GIS Data.” From Pima MacEachren, A. M. (1994b) “Visualization in modern cartog-
County Arizona Department of Transportation raphy: Setting the agenda.” In Visualization in Modern
Geographic Information Services Division Cartography, ed. by A. M. MacEachren and D. R. F. Taylor,
(http://www.dot.co.pima.az.us/gis/webdev/; accessed on pp. 1–12. Oxford, England: Pergamon.
8/29/07). Peterson, M. P. (ed.). (2003) Maps and the Internet.
Cartwright, W. (2003). “Maps on the Web” In Maps and the Amsterdam: Elsevier.
Internet, ed. by M. Peterson, pp. 35–56. Amsterdam: Pingel, T. J., and Clarke, K. C. (2005) “Assessing the usability
Elsevier. of a wearable computer system for outdoor pedestrian
Crampton, J. (1998). “The convergence of spatial navigation.” Proceedings of AutoCarto 2005, Las Vegas,
technologies.” Cartographic Perspectives no. 30:3–5. NV. http://www.geog.ucsb.edu/~pingel/documents/pingel-
Dodge, M., and Kitchin, R. (2001b) Mapping Cyberspace. autocarto2005-final.pdf.
London: Routledge. Rumsey, D. (2003) “Tales from the vault: Historical maps
Langer, M. (2002) Gebruikerswensen bij online kaartbebruik online.” Common-place 3, no. 4. http://www.common-
(User requirements in online map use). Unpublished place.org/vol-03/no-04/tales/.
Masters Thesis, Utrecht University, Utrecht, The
Netherlands.
Loy, W. G. (ed.), Allan, S., Buckley, A. R., and Meacham, J. E.
(authors). (2001) Atlas of Oregon. Eugene: University of
Oregon Press.
12:35:00.
Virtual Environments
519
12:36:59.
OVERVIEW
From reading daily newspapers, watching television, or navigating through the VE and understanding one’s cur-
playing computer games, you are probably familiar with rent orientation and location. Information intensity
the term virtual reality (VR)—it likely brings forth the deals with the level of detail (LOD) present in the VE,
image of donning a head-mounted display (HMD) and and intelligence of objects refers to the idea that objects
experiencing a simulation of reality. Although this notion within a VE often exhibit some degree of behavior or
is correct, in this chapter we will see that other terminol- intelligence.
ogy is often used and that a broader set of technologies Various examples of GeoVE applications are covered
are now possible. In section 1, we define terms that are in section 4, including virtual cities, the Virtual Field
closely associated with VR. A virtual environment (VE) Course, the notion of a Digital Earth (and Digital
is a 3-D computer-based simulation of a real or imag- Globes), water-resource applications, and virtual forests.
ined environment that users are able to navigate through Given the novelty of GeoVEs, relatively little is known
and interact with; for example, you might “fly through” about how they should be designed and used. Thus,
the Grand Canyon while sitting at your computer. VEs of section 5 considers a number of research issues related to
interest to geographers and other geoscientists can be GeoVEs.
termed “geospatial virtual environments” (GeoVEs). Section 6 covers ongoing developments in MEs of
A mixed environment (ME) (or a mixed reality (MR)) interest to geographers. We specify “ongoing” because
combines a real-world experience with a virtual these are recent developments that have the potential to
representation, such as overlaying the boundaries of fundamentally transform the way geographers do field-
“old-growth” forests on your view of a mountainous work and interpret maps. Although geographers are
environment. Most commonly, MEs are referred to as excited about the potential offered by virtual and mixed
augmented reality because the real world is dominant environments, we should recognize that there are health,
(i.e., virtual representations are used to augment reality). safety, and social issues associated with their use—these
Section 2 describes some of the technologies for creat- issues are considered in section 7.
ing VEs including desktop, wall-sized, head-mounted,
drafting table–format, and room-format displays.
Because these technologies are likely to change over time, 1 DEFINING VIRTUAL AND MIXED
you should check the home page of this text for links to ENVIRONMENTS
the most recent developments. Section 3 discusses four
“I” factors that are relevant to designing GeoVEs: Early notions of virtual reality (VR) promoted the idea
immersion, interactivity, information intensity, and intel- of simulating the tangible real world—the everyday
ligence of objects. Immersion is closely tied to presence, world that we see, hear, touch, smell, taste, and move
which deals with one’s sense of being part of a VE. through. One of the first researchers to recognize the
Important issues associated with interactivity include potential of VR for geography was Ian Bishop (1994).
12:36:59.
Although Bishop did not use the term virtual reality and the Web. Here the structure is based, not on geographic
focused largely on vision, it was clear that this was what location, but rather on the relationship between topics
he had in mind: presented via the Web, with more similar topics shown
together. Each sphere represents a Web page, with the
The non-scientific audience . . . wants abstraction mini- links between spheres representing the links between
mized, information content maximized . . . with the whole Web pages and the size of a sphere representing a func-
package digestible and non-threatening. This suggests the tion of the number of linking Web pages. The idea is
use of a visual realism approach [that] shows information that we could utilize the resulting 3-D space to avoid
consumers what will or might happen under a variety of
becoming “lost in hyperspace” (Dodge and Kitchin
conditions and permits them to explore the alternative envi-
ronments using their natural sensory perceptions. (p. 61)
2001b, 126). In this case, the ultimate reality is a set of
zeros and ones in the computers used to depict this
Color Plate 1 is an example of some of Bishop’s early information, and thus the term virtual environment is
work—a simulation of a proposed lake created as a result again appropriate.
of filling an existing open-cut coal mine with water. Although we choose to use the term virtual environ-
Although it makes sense to simulate the tangible real ment (VE) rather than VR, you should recognize that
world, there are instances in which we might need to dis- researchers, even within geography, do not agree on the
play something intangible. For instance, imagine depict- definitions and uses of these terms. For instance, the
ing the concentration of ozone in the atmosphere both recent book Virtual Reality in Geography, edited by
vertically and horizontally at some particular point in Peter Fisher and David Unwin, largely utilizes the term
time. Because we cannot see ozone, we must create an VR as opposed to VE. Others, such as Alan MacEachren
abstract representation of it. To achieve this, we might and his colleagues (1999b), have utilized VE, defining
immerse individuals in a computer-simulated environ- the subset of interest to geographers as geospatial vir-
ment in which they travel through a virtual atmosphere, tual environments (GeoVEs).
seeing the concentration of ozone as they move. For our purposes, we define a virtual environment as
The concentration of ozone could be depicted by the a 3-D computer-based simulation of a real or imagined
density of small balls having a uniform size and color, environment that users are able to navigate through
with a high density of balls indicating a high ozone and interact with. Normally, VEs are experienced visu-
concentration. Here the term virtual reality seems inap- ally, although ideally it would be possible to utilize the
propriate, given that humans cannot see ozone; as a full range of senses, including sound, touch, smell, taste,
result, researchers have chosen to use the term virtual and body movement. Using this definition, 3-D anima-
environment to encompass both tangible and intangible tions in which the user does not control the point of
simulations. view (e.g., the classic L.A. The Movie would not be con-
As a more abstract example, consider Figure 1, which sidered VEs.
was developed by Andrew Wood and his colleagues Although early efforts focused on the potential of
(1995) to portray the spatial structure of a portion of immersing oneself in a VE, more recent work has sug-
gested that a combination of reality and VEs might also
be useful. This notion is depicted in Figure 2, where we
see a real-world-to-VE continuum. The “real world”
FIGURE 2 The real-world-to-virtual-environment contin-

uum. (After Drascic and Milgram 1996. In their illustration,
Drascic and Milgram used the term “virtual reality” rather
FIGURE 1 A depiction of the spatial structure of a portion than “virtual environment” and the term “mixed reality”
of the Web. (Courtesy of Bob Hendley.) rather than “mixed environment.”)
521
12:36:59.
end of the continuum represents what we perceive in common approach for depicting VEs. Popular software
the real world, whereas the VE end represents a simu- packages for desktop displays, such as ArcGIS’s 3D
lation of an environment. In between, we see mixed Analyst and ERDAS Imagine’s VirtualGIS, include
environments (MEs) (or mixed realities (MRs)), which interactive 3-D mapping options, and hundreds of pack-
represent a combination of the real world and a virtual ages have been developed for creating 3-D displays,
representation. The notion of MEs can be further many of which provide an interactive capability (see
subdivided into augmented reality and augmented http://www.tec.army.mil/research/products/TD/tvd/
virtuality, depending on whether the real world or the survey/). Particularly important has been the develop-
virtual representation is emphasized. ment of the Virtual Reality Modeling Language
In this chapter, we’ll discuss two examples of MEs that (VRML), which enables VEs to be disseminated over
are of interest to geographers. One is Project Battuta the Web. A specialized version of VRML intended for
(http://dg.statlab.iastate.edu/dg/), which involved develop- the representation of geographical data is also available
ing approaches for collecting and using geospatial data in (see http://www.ai.sri.com/geovrml/).
the field. Utilizing wearable computers, users were able The VE capability of standard desktop displays can
to see computer-simulated information overlaid on top be enhanced by presenting separate images to the left
of their real-world view. For instance, in theory an urban and the right eyes, thus enabling a stereoscopic view,
geographer could go to the upper floor of a tall building, just as we have in the real world. This can be accom-
look out at the city, and see a map of crime incidence plished in a variety of ways (Slocum et al., 2007). One is
overlaid on the city. We’ll see that a system marketed by to add a polarizing panel to the desktop monitor and to
Microvision now provides the capability to make such have the user wear polarized glasses, producing what is
geographic augmented reality a distinct possibility rather known as passive stereo. A second approach is to cycle
than wishful thinking. The second example of an ME is left- and right-eye views on the display, and have the
the work of Nick Hedley (2001; 2003), who developed an user wear shutter glasses that are in sync with the dis-
approach for displaying a VE on a large card held in the play—this is known as active stereo. A third approach is
user’s hands. Users could rotate the card, just as they to use an autostereoscopic display in which no special-
would rotate an object in the real world, thus providing a ized glasses are required (Dodgson 2005). Although
potentially more intuitive approach for examining 3-D autostereoscopic displays sound ideal, they are more
landscapes than traditional mouse-based interfaces. expensive and are characterized by a limited number of
Although Hedley and those involved with Project viewing angles known as “sweet spots.”
Battuta have described their work as a form of aug- A wall-sized display covers a large portion of a wall,
mented reality, it can be argued that Hedley’s work typically by tiling images created by multiple projectors.
places greater emphasis on interpreting the VE, and thus For instance, one display at the University of Kansas
might be considered a form of augmented virtuality. measures 25 * 6 feet, covers 120° of the visual field,
and provides a 5,760 * 1,200 pixel resolution (Color
Plate 2). As with desktop displays, wall-sized displays
2 TECHNOLOGIES FOR CREATING VIRTUAL can often be viewed without any special devices (e.g.,
ENVIRONMENTS the head tracker and eyeglasses mentioned later for the
CAVE).* Although a traditional keyboard and mouse
Technologies for creating VEs are important because can be used to interact with a wall-sized display, other
they impact how we experience those environments. control devices either are being used or are in the
Nick Hedley (2001; 2003) argued that technology for process of being developed. For instance, Guimbretière
creating VEs can be broken down into three parts: the et al. (2001) and Davis and Chen (2002) described pen-
display, the hardware controls, and the graphical user based and laser pointer approaches, respectively. Such
interface (GUI). For instance, with today’s desktop com- novel control devices are essential, given the varied dis-
puters, we might use an LCD display, a mouse and key- tances of users from the display and the need to handle
board as controls, and Microsoft Windows® as the GUI. simultaneous users. An overview of technical issues
Normally, in describing the different technologies for associated with wall-sized displays (e.g., combining
visualization purposes, emphasis is placed on the display, images from multiple projectors) can be found in the
and so we do that here. You should bear in mind, how- July/August 2000 issue of IEEE Computer Graphics
ever, that the affordances possible with a particular dis- and Applications.
play might be a function of the controls and GUI. The GeoWall (http://geowall.geo.lsa.umich.edu/home.
VEs can be created using five basic forms of display html) is a specialized form of a wall-sized display. In its
technologies: desktop, wall-sized, head-mounted, room-
format, and drafting table–format. The desktop display
(e.g., an LCD display or a laptop) is by far the most * An exception is the Infinity Wall (Czernuszenko et al. 1997).
12:36:59.
FIGURE 3 A typical GeoWall

setup; key elements include a com-
puter with a dual-head graphics card,
two DLP projectors on an adjustable
stand, polarizing filters for the pro-
jectors, a display screen capable of
preserving the polarization, and pas-
sive glasses for users. (Courtesy of
Electronic Visualization Laboratory,
University of Illinois at Chicago.)
basic form, a GeoWall is comprised of three components one-third of undergraduate nonmajor Earth science
(see Figure 3; Slocum et al. 2007): students in the US already use a GeoWall in their
coursework” (p. 12). Although the GeoWall is not yet
• A computer with a dual-head graphics card common in geography, researchers at the University
• Two DLP projectors on an adjustable stand of Minnesota and the University of Kansas have pro-
• Polarization equipment, including filters for the promoted its potential (Anthamatten and Ziegler 2006;
jectors, a display screen capable of preserving the Slocum et al. 2007). One disadvantage of the GeoWall
polarization, and glasses for users is that in its basic form, resolution is limited; for exam-
ple, a typical resolution is 1,024 × 768 pixels. To
Costing under $10,000, the GeoWall was developed as address this issue, Johnson and his colleagues devel-
a low-cost version of room-format displays (Johnson oped the GeoWall-2, which uses an array of LCD
et al. 2006). The GeoWall is becoming popular in panels powered by several high-end PCs for display
geology; Johnson et al. indicate that “more than (Figure 4). The result is an inexpensive solution for
FIGURE 4 The GeoWall-2, which

uses an array of LCD panels for
display; the result is an inexpensive
solution for creating a high-resolution
wall-sized display, albeit without stereo
capability. (Courtesy of Electronic
Visualization Laboratory, University
of Illinois at Chicago.)
523
12:36:59.
creating a high-resolution wall-sized display, albeit

without stereo capability. A
The term head-mounted display (HMD) derives
from the means used to create such a display—placed
on the user’s head is a helmet-like device that shields
the real-world view and provides images of the VE to
each eye (Figure 5A). Sophisticated HMDs provide
separate images to the left and the right eyes, thus
enabling the user to have a stereoscopic view. The
HMD includes a head-tracking device so that the view
of the virtual representation is a function of the posi-
tion of the user’s head. Because the real world is not B
visible when using HMDs, nontraditional control
devices such as a dataglove are utilized. HMDs have
sometimes been criticized because they are a burden
for users to wear (due to their weight), their resolu-
tion is low (a typical high resolution has been
640 * 480 addressable pixels), the field of view is
narrow (the view is rectangular like on a traditional
computer screen), and one cannot see other users (as
in a collaborative environment). You should be aware
that these criticisms are apt to diminish as technology
improves.
Room-format and drafting table–format displays
were developed in response to the limitations
of HMDs. Two well-known examples are the CAVE
and the ImmersaDesk, which were developed by
researchers at the Electronic Visualization Laboratory C
at the University of Illinois at Chicago. The CAVE
provides a room-sized view of a VE by projecting
images onto three walls and the floor (Figure 5B).* As
users stand or move in the CAVE, a head tracker on
one user governs the view that all users see through
stereo glasses; the tracker and glasses are much less
obtrusive than traditional HMDs. Interaction with the
VE is achieved through a 3-D mouse, which consists
of a joystick for navigation and programmable but-
tons for interaction. One advantage of the CAVE is
that users can see each other—in that sense, the
CAVE can be considered an ME, except that the
emphasis is clearly on viewing the VE; thus it makes
sense to specify it as VE display technology. The
ImmersaDesk was developed as a less expensive sub-
stitute for the CAVE. With the ImmersaDesk, the VE
is projected onto a single screen that is tilted at a 45°
angle (the logic is that looking down in the CAVE is
important—the tilt helps simulate this; see Figure 5C).
Again, a head tracker on one user governs the view
that other users see, and a specialized 3-D mouse can
be used to interact with the simulated environment. FIGURE 5 Examples of various displays for creating VEs: (A)
head-mounted, (B) room-format (CAVE), (C) drafting
table–format (ImmersaDesk). (Figure A courtesy of William
Winn; Figure B Image of the CAVE® virtual reality theater
courtesy of Milana Huang, Electronic Visualization Laboratory,
* The software will permit projection onto six surfaces, but normally University of Illinois at Chicago; Figure C courtesy of Electronic
only four are used. Visualization Laboratory, University of Illinois at Chicago.)
12:36:59.
3 THE FOUR “I” FACTORS OF VIRTUAL An interesting perspective on immersion and the
ENVIRONMENTS various forms of display was offered by Edward Verbree
and his colleagues (1999), who considered immersion in
Based on the work of Michael Heim (1998), Alan the context of the landscape planning process in the
MacEachren and his colleagues (1999a) proposed four Netherlands. Verbree and his colleagues identified three
“I” factors important to creating geospatial VEs: stages in the landscape planning process and an associ-
immersion, interactivity, information intensity, and ated set of relevant maps: (1) orientation—2-D maps
intelligence of objects. with a bird’s-eye view; (2) design and modeling—3-D
maps with abstract symbols; and (3) presentation and
final decision making—3-D maps with photorealistic
3.1 Immersion quality. Corresponding to these three stages and maps,
they proposed three levels of immersion: a relatively
Immersion is defined as “. . . perceiving oneself to be nonimmersive, traditional desktop display; an immersive
enveloped by, included in, and interacting with an envi- display using a drafting-table format; and the more fully
ronment that provides a continuous stream of stimuli and immersive CAVE. They noted, however, that for conven-
experiences” (Witmer and Singer 1998, 225). The notion ience, all mapping techniques should be capable of being
of immersion is closely tied to that of presence, which is viewed on any of the hardware. Clearly, research is
defined “as the subjective experience of being in one needed to determine in what instances each type of
place or environment when one is physically in another” display will be most useful for a broad range of possible
(Witmer and Singer 1998, 225). Because a high level of geospatial tasks.
immersion leads to a sense of presence, those discussing How important is a sense of immersion? One
VEs often speak of the level of immersion, and we will argument says that systems providing greater immer-
do the same. Keep in mind, though, that the notion of sion will be more effective because users will be more
presence is important—an indication of its importance is likely to utilize real-world cognitive processing strate-
that one journal dealing with VEs is named Presence gies (Buziek and Döllner 1999). A counterargument
(http://www.mitpressjournals.org/loi/pres). says that cartography traditionally has been success-
Each of the displays that we have discussed provides ful because it is abstract—that is, we need a separa-
a different sense of immersion. Those desktop displays tion between the real world and its representation to
in which stereoscopy is not used provide the lowest make sense of the real world (Slocum et al. 2001).
sense of immersion, because both the real world out- William Cartwright (1997) argued that the “reality”
side the computer screen and the VE depicted on the suggested by highly immersive VEs could be mislead-
screen are readily visible, and because the images do ing, as it might be confused with “real reality.” He
not look truly 3-D (because they are not presented stated, “using information derived in such a manner
stereoscopically). For this reason, such desktop dis- may lead the user to believe that Surreal Reality is
plays are sometimes termed nonimmersive, but we pre- in fact truly real. [As a result] knowledge-building
fer to think of them as semi-immersive, since viewers and decision-making could be made in an artificial
often feel that they are part of the VE. For instance, in world completely devoid of nature’s checks and
using the Virtual Field Course software that we will balances” (p. 451).
describe subsequently, Jason Dykes (2002, 71–72) The display systems that we have described rely pri-
argued that the desktop display provides a “strong marily on vision, and, to a limited extent, sound. It is
sense of immersion, [in which] users frequently move important to recognize that a sense of immersion is a
with the surface, duck whilst navigating around obsta- function of more than vision and sound—it involves all
cles, and express their desire to ‘go’ somewhere, or the of the senses, including touch, smell, taste, and body
fact that they are ‘lost.’” movement. Outside the field of geography, sophisti-
A wall-sized display provides a greater sense of cated systems have been developed that incorporate
immersion than a desktop display because in the former, multiple senses. For instance, the $50 million National
a greater portion of the visual field is covered, especially Advanced Driving Simulator (NADS) housed at the
when one stands or sits close to the screen.This still must University of Iowa can be used to simulate the “feel”
be termed a semi-immersive system, though, as images of driving (http://www.nads-sc.uiowa.edu/; Samuels
do not look truly 3-D (unless multiple projectors are 1999), and thus can provide a safe means of evaluating
used to create stereoscopic images and specialized view- the effects of alcohol, drugs, and fatigue on driving.
ing glasses are used). In contrast, the HMD, the Although geographers probably do not need such
ImmersaDesk, and CAVE all provide a fuller sense of sophisticated systems, work needs to be done to
immersion, because the real world can be hidden from explore the roles of the other senses in GeoVEs. Some
view, images are presented stereoscopically, and senses of the senses currently being explored include sound,
other than vision (e.g., sound) are often used. touch, hand gestures, and body movements.
525
12:36:59.
3.2 Interactivity Another important issue concerning interactivity

in VEs is determining appropriate tools for interact-
A basic issue in interacting with VEs is navigating ing with and modifying the VE. As a starting point, it
through them and understanding one’s current orienta- seems reasonable to provide tools similar to those
tion and location; for instance, users often report feeling used outside VEs. For instance, we presume that tools
lost in VEs, as Dykes (2002, 71–72) noted when describ- for brushing and focusing should be available.
ing users of the Virtual Field Course. This is an issue However, the 3-D character of the VE implies that
that has yet to be solved by those developing VEs, other tools should also be available, such as those for
although some research has been done. For example, picking up and rotating objects. Although a suitable
consider the work of Sven Fuhrmann and Alan set of tools for geography has yet to be developed,
MacEachren (2001), who observed geoscientists navi- Joseph Gabbard and Deborah Hix (1997) summarized
gating a desktop GeoVE with a standard VRML numerous interaction techniques for VEs that might
browser. Because Fuhrmann and MacEachren found serve as the basis for developing such tools.
that users were often disoriented and thus not satisfied A third issue concerned with interactivity in VEs is
with the navigation and orientation capabilities pro- the nature of the control devices used to interact with
vided, they suggested an alternative GUI based on a the VE. As we indicated in the discussion of display
flying saucer metaphor. The notion of using metaphors technologies, various unconventional control devices
is widely accepted in computer software development are typically used to interact with VEs. Ideally, we need
(for instance, a “folder” is used to store documents in to evaluate the effectiveness of these devices and to
the typical Windows environment). Fuhrmann and consider the possibility of using speech to interact with
MacEachren provided several reasons for the flying the VE. This takes us into the realm of multimodal inter-
saucer metaphor, including: faces, because it also applies outside VEs.
• Its movement characteristics are commonly under-
3.3 Information Intensity
stood, and movement in all directions is expected to
be possible.
Information intensity deals with the level of detail (LOD)
• In science fiction, flying saucers operate in geo-
presented in the VE. One issue relevant to LOD is how
graphic environments.
much detail should be shown at a particular scale. If the
• Expertise in controlling flying saucers is not an issue
purpose is to simulate reality, then the inclination is to
because no user has training using the GUI.
show considerable detail, but how much detail is appro-
• Because a flying saucer has no apparent moving
priate? For instance, in section 4.1, we will see that devel-
parts, users think of it as sophisticated technology
opers strive for realistic images in virtual cities; but is such
with simple-to-learn controls.
realism always necessary? Martin Reddy (2001) devel-
• Flying saucers are simple in shape, providing a clean
oped an approach for specifying the amount of detail that
user interface.
should be shown, presuming that greater detail should
Fuhrmann and MacEachren tested the usability of their appear where humans’ vision is most sensitive to detail
GUI and found that navigation was easier than with the (in the fovea). His approach, however, is largely percep-
standard VRML browser and that users liked the inter- tual as opposed to cognitive, meaning that it does not con-
face. A shortcoming, however, was that many users still sider the information content of the image.
claimed to be “lost.” The problem was that while users A second issue relevant to LOD is how it should
had a natural egocentric view of the VE, they were not change as scale changes. There are a number of rules that
able to orient themselves in the flying saucer GUI are appropriate for generalizing abstract symbolization
because of missing spatial information. To solve this for two-dimensional maps, but these rules do not neces-
problem, Fuhrmann and MacEachren added an sarily extend to realistic-looking VEs. Those working with
overview map to the user interface showing the overall the notion of a Digital Earth (see section 4.3) have devel-
extent of the displayed landscape and a directional cone oped approaches for changing detail with scale, but these
indicating the user’s current position and orientation approaches deal largely with technical issues (e.g., Reddy
(Color Plate 3). In conducting a range of usability stud- et al. 1999), not with what users will necessarily need or
ies with this prototype, they found that users were less want. Similarly, those working with forested landscapes
likely to be disoriented in a desktop GeoVE when pro- have developed approaches for changing detail with scale
vided with both the natural egocentric view and an (e.g., Kumsap et al. 2005), but, again, these do not con-
enhanced overview map. sider the detail that users may need or want.
12:36:59.
Those in the field of landscape and urban planning cannot interact with the representation or select the
have begun to conduct experimental studies on the particular route of navigation), but such work paved
LOD appropriate for various applications (e.g., Bishop the way for later systems that do meet our definition
and Rohrmann 2003; Williams et al. 2007). A key point of a VE.
is that the LOD will depend on the purpose of the visu- A widely cited example of a virtual city is Virtual Los
alization. For instance, the conservationist who is inter- Angeles, which is being developed by William Jepson
ested in biodiversity may require a different LOD than and his colleagues at UCLA (1998; 2001). The goal of
someone who is evaluating a landscape for scenic Virtual Los Angeles is to develop a virtual view of
beauty (Williams et al. 2007, 215). the entire 4,000-square-mile Los Angeles Basin (Color
Plate 5). Because this is such a large undertaking,
3.4 Intelligence of Objects Virtual Los Angeles is being developed in stages, with
each stage encompassing 1 to 15 square miles. Jepson
Intelligence of objects refers to the notion that objects and his colleagues are able to extend well beyond mere
within a VE should exhibit some degree of behavior or representational cities by utilizing a combination of a
intelligence, just as we would expect of real-world CAD package (AutoCAD), a 3-D modeling environ-
objects. For instance, if we attempt to move a small ment (MultiGen/Creator), a simulation tool (based on
branch on a tree in a VE, we would expect the branch IRIS Performer and OpenGL), and a GIS (ArcView).
to move. Similarly, if we encounter a person (known as For instance, the system can depict changes over time
an avatar or an intelligent agent) in a VE, we would like (e.g., the growth of trees in the urban landscape), track
that person to exhibit intelligent behavior such as guid- vehicles in real time and generate views from those
ing us to a desired location. Outside the field of geogra- vehicles, and perform basic query and display functions
phy, avatars have been utilized for a variety of purposes (e.g., highlight all buildings having a specified number
such as teaching people how to work with machinery of employees). Users can move through Virtual
(Rickel and Johnson 1999) or for advertising and pres- Los Angeles in walk-through, drive-through, or fly-
entation (Noll et al. 1999). Within geography, using through modes. The Web site for Virtual Los Angeles
avatars is not yet common, although researchers have (http://www.ust.ucla.edu/ustweb/projects.html) contains
noted the potential that avatars provide (e.g., see numerous examples of projects that have been under-
Cartwright 1999 and Crampton 1999). taken and sample videos.
Another example illustrating the sophistication of
virtual cities is CommunityViz (http://www.simcenter.
4 APPLICATIONS OF GEOSPATIAL VIRTUAL
org/Projects/CommunityViz/communityviz.html), which
ENVIRONMENTS
was developed under the direction of the Orton
Family Foundation (http://www.orton.org) and the
In this section, we consider a range of applications for Environmental Simulation Center (http://www.simcenter.
GeoVEs. Numerous additional applications can be org; Kwartler and Bernard 2001). In contrast to large-city
found in Peter Fisher and David Unwin’s text Virtual cases such as Virtual Los Angeles, CommunityViz was
Reality in Geography. developed to assist rural communities in dealing with the
rapid changes often caused by suburbanization.
4.1 Virtual Cities CommunityViz consists of three modules built on the
ArcView GIS system: Scenario Constructor, Policy
Probably the most common form of geospatial VEs is Simulator, and TownBuilder 3D. Scenario Constructor
the virtual city, which typically takes the form of a “permits nontechnical users to create land use scenarios,
highly detailed visual representation of buildings (and and to evaluate those scenarios against community objec-
related structures) within the city. An early example tives and constraints” (p. 289). Policy Simulator allows
was the virtual city developed for Bath, England, users to forecast land use changes, along with associated
under the direction of Alan Day (1994; see Color demographic and economic changes, given a prospective
Plate 4 for some images). The purpose of the Bath vir- policy change (e.g., a tax incentive to industry). Of the
tual city was to assist in planning new developments three modules, TownBuilder 3D is the most relevant
in this historically sensitive city. For instance, Day (p. to our discussion because it permits photorealistic 3-D
375) described how a series of views and animations models that can be examined in real time. To illustrate
were set up to determine the most appropriate loca- some of the capability of CommunityViz, consider Color
tion for a new sports hall (see http://www.bath.ac.uk/ Plate 6, which illustrates the results of a proposed land
casa/completed/views.html for an example of an ani- development policy. Note that the image provides a con-
mation). Strictly speaking, fixed views and animations ventional thematic map as a bird’s-eye view (on the left),
would not meet our definition of a VE (because users some simple graphs (upper right), and a realistic view of a
527
12:36:59.
portion of the landscape (lower right). CommunityViz is modeling the 3-D character of cities, UrbanModeller also
promoted as a tool that can assist members of the enables the animation of pedestrian flows. UrbanModeller
community in becoming involved in planning develop- is a component of the Virtual Field Course, which we cover
ment—presumably, part of the reason community in the next section.
members find this tool useful is the ease of understanding
the realistic views CommunityViz provides. More recent- 4.2 The Virtual Field Course
ly, the Environmental Simulation Center has expanded
their work in VEs by working in urbanized areas (for The Virtual Field Course project (http://www.geog.le.ac.
example, see their work with the Baltimore region on uk/vfc/), a collaborative effort among several researchers
their Web site).* and universities in the United Kingdom, aims to intro-
Another example of ongoing work with virtual cities duce VEs and information technology to traditional
is that of researchers at the Centre for Advanced field-based geography classes. Geographers from the
Spatial Analysis (CASA) at University College London University of Leicester, Birkbeck College, and City
(http://www.casa.ucl.ac.uk/). In one research effort University London argue that a Virtual Field Course can
known as Virtual London: Online Participation, CASA assist in all facets of fieldwork, including preparation, col-
is developing a large-scale 3-D virtual model of Greater lection of data, analysis of data, and debriefing (Dykes
London that is intended not only for professionals but et al. 1999). Here we summarize two major software com-
also for the general public. CASA researchers are ponents (traVelleR and panoraMap) that can assist in
experimenting with how free software such as Google examining these facets.
Earth and ArcExplorer can be used to distribute this traVelleR comes closest to our notion of a VE, as it
virtual model; they envision users being “able to roam permits users to explore the 3-D character of the nat-
around a Virtual Gallery as Avatars . . . and explore the ural landscape where fieldwork is going to be under-
issues relating to London in game-like space.” In taken or has already been done. It contains a number of
another effort known as GeoVUE: Geographic Virtual features, however, that provide considerably more
Urban Environment, CASA researchers have under- information and analytic capability than a simple visual
taken the ambitious plan of allowing users to develop analysis of a VE would offer. traVelleR has the ability
their own VEs using online tools. We encourage you to track users’ movements through the VE on a two-
to examine their Web site (http://www.casa.ucl.ac.uk/ dimensional map, vary the characteristics of the scene
projects/index.asp), where numerous other research and query its properties, link multimedia representa-
projects related to urban geography and emerging tions, and display an animated route.
computer technologies are discussed. panoraMap uses linked georeferenced panoramic
Some of those working with virtual cities have empha- images to represent geographic locations. Color Plate 7
sized abstract phenomena and thus diminished the impor- demonstrates how seamless 360° imagery can be used
tance of realism. For example, Kate Moore (2002) empha- in this way. Discrete locations from which panoramic
sized abstract phenomena in mapping the vertical images have been taken are represented by orange
dimension of the virtual city. Moore argued that a 3-D VE symbols on the map (in the upper left of the Plate).
provides an opportune means for examining the 3-D When particular panoramic images are selected from
character of a city. Traditionally, cartographers mapped the map, a pair of wide arrows on the map depicts the
cities in two dimensions even though cities exhibit spatial direction and angular extent of the views, and a portion
variation vertically (e.g., rents charged on the ground of each image is shown (in the right portion of the
floor differ from those charged on the top floor). Moore Plate); thus, in this case the three images shown relate
felt that the traditional horizontal outlook was partly a to the three pairs of orange arrows. Because the views
function of insufficient tools for visualizing the vertical are interactive, moving the arrows on the map or drag-
component. Lack of data could also be considered a rea- ging the images with the mouse will result in panning.
son for not mapping the vertical dimension (e.g., the U.S. Moving the mouse in any panorama produces a longer,
Bureau of the Census collects data for census blocks, not thinner arrow in the image and an associated symbol
for layers within census blocks), but Moore suggested that on the map showing the viewing angle of the identified
the type of data collected has likely been a function of our feature. This means that features in the panoramas can
traditional use of two-dimensional maps. In response, be associated with data recorded on the map. The
Moore developed UrbanModeller to assist in visualizing arrows at the top of each image in Color Plate 7 show
the vertical structure of the virtual city. In addition to the directions and distances (i.e, by the length of the
arrows) of other panoramas. These can be touched with
* For an evaluation of CommunityViz in the context of public partic- the mouse to highlight the locations on the map (in
ipatory geographic information systems (PPGIS), see Aggett and green), and clicked on to display the relevant view.
McColl (2006). The user is thus able to “navigate” between discrete
12:36:59.
locations across the landscape. The three panoramas Windows operating systems is insufficient for interacting
shown in Color Plate 7 constitute a user-specified vir- with the vast amount of geographic data now available
tual route across part of Dartmoor National Park in (e.g., 1-meter satellite imagery). Instead, Gore called for
the United Kingdom. a new metaphor—a multiresolution, 3-D Digital Earth:
As with traVelleR, the real power of panoraMap is
its ability to link other kinds of information with these Imagine . . . a young child going to a Digital Earth exhibit
panoramic images and to explore this information in a at a local museum. After donning a head-mounted display,
spatial context. Data can be georeferenced by clicking she sees Earth as it appears from space. Using a data
glove, she zooms in, using higher and higher levels of reso-
on the map and using a selection interface to choose
lution, to see continents, then regions, countries, cities, and
any file on the computer. The location of each file is finally individual houses, trees, and other natural and man-
then represented by a yellow symbol, which reveals the made objects. Having found an area of the planet she is
data when clicked on. For instance, Color Plate 8 shows interested in exploring, she takes the equivalent of a “magic
a session from an exercise on footpath erosion. Two carpet ride” through a 3-D visualization of the terrain.
panoramic views are centered on Haytor Rocks, an Of course, terrain is only one of the many kinds of data with
important and much visited local feature. A georefer- which she can interact. Using the systems’ voice recognition
enced digital movie (lower right) containing audio and capabilities, she is able to request information on land
video collected in the field is displayed, along with a cover, distribution of plant and animal species, real-time
hypertext document (upper right) offering a series of weather, roads, political boundaries, and population. She
images, links, and qualitative information on erosion can also visualize the environmental information that she
and other students all over the world have collected . . . .
around Haytor Rocks as recorded by the National
She is not limited to moving through space, but can also
Parks Authority. panoraMap explores the links between travel through time. After taking a virtual field-trip to Paris
reality and VR by enabling users to plan routes for use to visit the Louvre, she moves backward in time to learn
with GPS receivers and plot them after returning from about French history, perusing digitized maps overlaid on
the field. The software also contains capabilities for the surface of the Digital Earth, newsreel footage, oral
graphical data analysis, meaning that numeric informa- history, newspapers and other primary sources.
tion can be assessed within the context provided by the
qualitative data sources. So in the exercise shown in Although it will likely be some time before Gore’s full
Color Plate 8, quantitative data recorded in the field vision is achieved, several pieces of software now provide
can be visualized in combination with secondary quali- much of the capability that Gore proposed. Probably
tative data in a spatial context. The use of the GPS data best known is Google Earth (http://earth.google.com/).
before and after visiting the field emphasizes the links Although you may have utilized Google Earth, you may
between the map and the physical environment and not have experienced its full capability to create a VE. To
supports the “reality” metaphor of the interface. experience this, you will need to view Google Earth
Both traVelleR and panoraMap have been designed images in stereo and use a tool that allows you to move
to take advantage of the VFC Hub mechanism (Dykes intuitively through 3-D space. This can be accomplished
et al. 1999), which connects software to the Internet by linking the TriDef Visualizer software (http://www.
and ensures that data stored on remote servers can be tridef.com/promotions/google-earth.html) with Google
incorporated into a session. For example, traVelleR can Earth using the Space Navigator (http://www.3dconnexion.
load maps and multimedia data found online (in one com/) for interacting with the display, and viewing the
example application, audio files containing an expert’s images on a stereoscopic display (we have used a
explanations were georeferenced at the locations of GeoWall). Our experience is that this provides a very
important geomorphological features, such as Haytor intuitive approach for flying through 3-D landscapes
Rocks), and panoraMap can search online VFC data- (such as the Grand Canyon) that are available in Google
bases for panoramas, add symbols to show their loca- Earth.
tions, and download and display the views when the A variety of other software has been associated with
symbols are clicked.* the notion of a Digital Earth. Historically, TerraVision
(http://www.ai.sri.com/TerraVision/) was one of the first
4.3 Digital Earth and Digital Globes systems to permit access to large data sets of varying res-
olution via the Web. TerraVision contains some unusual
Former Vice President Al Gore (1998) argued that the navigation capabilities: a terrain-following option that
“desktop metaphor” employed by the Macintosh and can maintain a particular height above the Earth’s sur-
face and account for the Earth’s curvature, and an alti-
tude-based velocity that varies speed as a function of the
*For a discussion of panoraMap in the context of data exploration height above the surface (e.g., when traveling high above
and multivariate data, see Dykes and Mountain (2003, 595–597). the surface, it makes sense to go faster). Normally, scenes
529
12:36:59.
created with TerraVision are viewed on a desktop been created (Middleton et al. 2005). An interesting
display with a mouse as the control device; it is possible, research question is whether animations shown on
however, to use HMDs and a 6-degrees-of-freedom con- Digital Globes are more effective than those shown on
trol device (a device that allows great flexibility in navi- two-dimensional map projections.
gating 3-D space). Those of you with experience in GIS
may have used ArcGlobe (http://www.esri.com/news/ 4.4 Water-Resources-Related Applications
arcnews/summer03articles/introducing-arcglobe.html), a
software component of ArcGIS, the flagship product of Several VEs have been developed for water-resources-
ESRI, which is arguably the worldwide leader in GIS related applications. Two of these, the Chesapeake Bay
software. ArcGlobe not only allows users to zoom into Virtual Environment (CBVE) and Virtual Puget
regions via remotely sensed images of varying resolu- Sound (VPS) have dealt with estuaries. We discuss
tion, but also permits the creation of realistic landscapes CBVE here and consider VPS in a subsequent section
using the procedures discussed in this chapter. The of this chapter. Glen Wheless and his colleagues
Globe Program (http://www.globe.gov/) provides inter- (1996a; 1996b) developed CBVE for the ultimate pur-
active 3-D visualizations of the Earth via the Web, but its pose of visualizing potential changes in the ecosystem
focus is on looking at patterns on the Earth’s surface, of Chesapeake Bay over both space and time. Ches-
not on providing the detail at different resolutions that apeake Bay is the largest estuary in the United States,
Gore was envisioning. serving as nursery grounds and spawning areas for
Closely associated with the notion of a Digital Earth many commercially important fish and shellfish.
are Digital Globes, which show digital spatiotemporal Wheless and his colleagues argued that understanding
information about the Earth or other celestial bodies as this ecosystem is critical, given the increasing urban-
an undistorted three-dimensional globe (Riedl 2007). ization, overfishing, and diseases that have affected
Most Digital Globes are virtual (such as the default some species of fish.
view in ArcGlobe), but particularly intriguing are CBVE consisted of three modules: numerical simula-
Digital Globes that can be touched and also permit the tion, visualization, and virtual reality. The purpose of the
display of digital images. For example, the National numerical simulation model was to examine the effect
Oceanic and Atmospheric Administration (NOAA) has of tides, freshwater discharge, and winds on fish living in
developed Science On a Sphere (SOS), which provides this fragile ecosystem. The visualization module was
a room-sized global display (Figure 6) that permits the based on the Vis5D system, which was a precursor to the
animation of Earth processes such as storms, ocean IDV system. The virtual reality model utilized both
temperature, and climate changes. Although normally an ImmersaDesk and CAVE. Although they did not
viewed in nonstereo, a stereo version of SOS also has conduct any usability tests of the software, Wheless and
FIGURE 6 A classroom viewing

NOAA’s Science On a Sphere, an
example of a Digital Globe that can
be used to animate Earth processes
such as storms, ocean temperature,
and climate change. (NOAA Photo
by Will von Dauster.)
12:36:59.
his colleagues clearly were enthralled with the Dunbar et al. focused on, largely to illustrate the
ImmersaDesk and CAVE: landscape change following the large fires of 1988.
Dunbar et al. visualized this landscape change at two
Aspects of the data not clearly recognizable with static two- levels: the landscape/stand level, to “highlight landscape
dimensional or three-dimensional images became immedi- characteristics such as the spatial arrangement of stand
ately apparent simply because of our ability to intuitively types and stand structure” (106), and the plot level, to
navigate through the virtual environment so that features highlight “unique structural characteristics of a specific
could be examined from different viewpoints. The complex
forest plot” (109). Color Plate 9 compares snapshots of
relief of the bay bathymetry and associated topographical
their landscape/stand approach to a more traditional
effects on the circulation in the bay became obvious . . . .
(1996a, 207) abstract cartographic visualization. Dunbar et al. argued
that their photorealistic approach made “the informa-
Interestingly, Wheless and his colleagues also indicated tion more accessible to forest managers, ecologists, and
that sound “heightened” their understanding of ecosys- the public.” Color Plate 10A illustrates the realistic
tem processes: detail that is possible at the stand level; note how this
compares with the actual photograph of the same area
We were able to navigate through the virtual environment shown in Color Plate 10B. Dunbar et al. argued that
to a location where we could clearly see an increase in “Remote sensing allows us to precisely locate objects
biological distributions and hear the change in pitch or on the earth’s surface without physically visiting that
frequency corresponding to salinity value, thereby linking location, while visualization provides the means to real-
the biological behavior with a physical variable in a more istically illustrate what this information actually repre-
enlightening way. (1996a, 205) sents.” A full appreciation of Dunbar et al.’s approach
can be achieved by viewing their animations of change
Another example of a water-resource-related VE is the over time, which are available at http://www.kars.ku.
Digital River Basin, which is being developed by the edu/projects/visualization/.
Mississippi RiverWeb Museum Consortium (http://
www.riverwebmuseums.org/consortium/index.htm) to 4.6 Other Applications
illustrate the dynamics of the Mississippi River and its
associated watershed processes. For instance, the Other applications of VEs are planning and virtual her-
Science Museum of Minnesota has installed a River itage.An excellent source for early work on VEs relevant
Pilot Simulator that enables visitors to experiment with to planning is volume 54 (in 2001) of Landscape and
the task of barge navigation. By utilizing a 52-inch Urban Planning entitled “Our Visual Landscape:
plasma display and providing users with “tiller” and Analysis, Visualization and Protection.” More recently,
“throttle” controls and engine “sounds,” the simulator Ian Bishop and Eckart Lange have edited the book
gives users a sense of actually piloting a barge (Johnston Visualization in Landscape and Environmental Planning,
et al. 2000). Although this simulator does not provide which provides a wealth of material related to using VEs
the immersive experiences of other systems (e.g., in planning. Virtual heritage refers to VEs that enable
CBVE), it suggests the direction in which museums are users to experience the world’s cultural and natural her-
going to illustrate complex physical processes to the itage without having to be at the actual locations. The
general public. April/June 2000 issue of IEEE Multimedia provided
early work on this topic, and the June 2006 issue of the
4.5 Virtual Forests journal Presence focuses on more recent developments.
Although virtual heritage is only marginally related to
Virtual forests are another common application for VEs. geography, developments in this area are illustrative of
Although a wide range of virtual forest applications the technological developments taking place in VEs.
have been developed (e.g., Stoltman et al. 2004; Reljic
et al. 2005), we’ll focus on the one created by Matt
Dunbar and his colleagues (2004) because of its geo- 5 RESEARCH ISSUES IN GEOSPATIAL VIRTUAL
graphic emphasis and the fact that it used Visual Nature ENVIRONMENTS
Studio, one of the most flexible software packages for
visualizing forested landscapes. The considerable skill From the preceding sections, it should be clear that
required to construct virtual forests is reflected by the many people are excited about the capability that VEs
fact that Dunbar et al. used not only Visual Nature are now providing or might provide in the future.
Studio, but also four other types of software: remote Although we share this excitement, readers must recog-
sensing image analysis, GIS, image editing, and video nize that VEs are novel, and thus little is known about
editing. Yellowstone National Park was one area that how they should be designed and effectively used.
531
12:38:11.
In response to these concerns, those affiliated with the • They increase students’ engagement in learning
International Cartographic Association’s Commission tasks.
on Visualization and Virtual Environments have devel- • They enable the rescaling of objects and processes in
oped the following set of research issues related to time and space (e.g., a daily tidal cycle can be com-
GeoVEs (Slocum et al. 2001): pressed to 5 seconds).
• Determine the situations in which (and how) immer- • They allow students to make errors that might be dis-
sive technologies can assist users in understanding astrous in the real world (e.g., experimenting with
geospatial environments. fighting forest fires).
• Develop methods to assist users in navigating and Disadvantages of VEs include the following:
maintaining orientation in GeoVEs. • They require some training before learning can take
• Develop suitable methods for interacting with place.
objects in GeoVEs. • They are not effective in developing fine motor skills
• Determine ways in which intelligent agents can assist (because the skills used are often different from
users in understanding GeoVEs. those used in the real world).
• Determine ways in which we can mix realism and • Their side effects are unknown (e.g., students believ-
abstraction in representations to influence cognitive ing that something is real when in fact it is not could
processes involved in knowledge construction. lead to confusion between fantasy and reality).
• Develop support for interpreting and understanding • Temporal scaling may lead to confusion; for instance,
spatial trends and patterns in GeoVEs. when compressing a 24-hour tidal cycle down to 8
Researchers have begun tackling some of these issues. seconds, some students felt that they were watching
In section 3.2, we mentioned Sven Fuhrmann and Alan waves rather than tides.
MacEachren’s work on designing user interfaces to sup- Although work by Winn and his colleagues forms a use-
port navigation and wayfinding in desktop GeoVEs. ful framework for studying VEs, they have only begun
William Winn and his colleagues at the Human Interface to deal with the research issues identified earlier, and so
Technology Laboratory (HIT Lab; http:// exciting research remains to be done.
www.hitl.washington.edu/home/) at the University of Others who have begun evaluating the effectiveness
Washington have undertaken considerable research of VEs include Candice Luebbering and Laurence
(Windschitl et al. 2002;Winn 2002;Winn et al. 2002;Winn Carstensen (2007) and Rob Edsall and Kelli Larson
2005). Much of their research focuses on Virtual Puget (2006). Luebbering and Carstensen have compared the
Sound (VPS), a VE that allows students to examine effectiveness of a wall-sized display consisting of up to
water movement, the behavior of tides, and salinity nine monitors and a single monitor display. Overall,
levels in Puget Sound. When using an HMD environ- they have found that people perform better when more
ment, Winn and his colleagues have found that more monitors are available, but that a single monitor was
successful students use a more systematic approach, are sufficient for basic map reading tasks. Edsall and
more animated and active, use more virtual tools, com- Larson compared the effectiveness of the Decision
ment more frequently on their progress, have less diffi- Theatre at Arizona State University with a two-dimen-
culty with the interface, and are more physically active. sional PowerPoint presentation shown in a standard
In comparing HMD and desktop approaches, Winn and university classroom. Edsall and Larson described the
his colleagues found that HMD users reported a greater Decision Theatre as follows:
sense of presence, which in turn led them to learn more
about the movements of water (e.g., the concept that A ten-sided windowless room (the “drum”) approximately
water in the Sound circulates both horizontally and 20 feet in diameter with an 8-foot ceiling. On seven of the
vertically). In the HMD, students could simply look ten walls are rear-projection high-definition 10’ x 8’ display
around to see the water movement, whereas in the desk- screens. Onto these screens are projected computer dis-
top display, they had to manipulate a control device. plays rendered in stereo . . . . Soft swivel chairs with cast-
Based on their research and on the work of others, ers are positioned around the carpeted floor, and tables
Winn and his colleagues noted a number of advantages can be brought in if the application warrants . . . . [T]he
size, set up, and furnishings . . . are designed for comfort
and disadvantages of VEs. Advantages of VEs include
and to facilitate interaction among participants.
the following:
• They help students understand phenomena that are The Decision Theatre is thus a room-format display
not directly accessible to the senses (e.g., varying similar to that of the CAVE.
salinity levels with water depth). To compare the Decision Theatre with a tradi-
• Immersion helps students understand dynamic 3-D tional PowerPoint presentation, Edsall and Larson had
processes, but not static 2-D ones. two different groups of people view presentations on
12:38:11.
two environmental issues (groundwater overdraft and

the urban heat island) that would “require the concep-
tualization of abstract three-dimensional concepts.” In
theory, one might hypothesize that the Decision Theatre
would be more effective because its large-screen stereo-
scopic capability corresponds more naturally to how we
see and conceptualize three-dimensional phenomenon
in the real world. Although Edsall and Larson’s results
were preliminary (they conducted a “pilot” study), they
found the Decision Theatre to be “effective for sup-
porting understanding and decision-making about com-
plex three-dimensional environmental issues, in both
experts and non-experts,” but “less clear in showing that
. . . [it] led to more understanding, or different decisions
and conclusions, than the PowerPoint presentation.” In
our opinion, since the Decision Theatre was not clearly
more effective, one could question the considerable
amount of money that is spent on designing systems like
the Decision Theatre.
6 DEVELOPMENTS IN MIXED ENVIRONMENTS
In this section, we consider some recent developments

in MEs that are of interest to geographers. This section
is short primarily because the notion of MEs is novel.
As the technology matures, such environments might
become more important than GeoVEs. Also keep in
mind that MEs are apt to become more important as
location-based services (LBS) and the associated realm
of telecartography expand.
6.1 Project Battuta and the UCSB FIGURE 7 The wearable computer developed by Keith
Wearable Computer Clarke and his colleagues as part of Project Battuta. (Courtesy
of Thomas Pingel.)
Project Battuta was a joint research project of Iowa State
University and the University of California at Santa
Barbara (UCSB) to develop modern approaches for col- colleagues (Clarke 2001; Clarke et al. 2003; Pingel and
lecting and using geospatial data in the field. The project Clarke 2005) at UCSB as part of Project Battuta. The
was named after an early geographer, Ibn Battuta, who user of this wearable computer can readily see the
in the 1300s traveled to 44 countries, in the process real world, which is obvious by the fact that his eyes
covering more than 75,000 miles. Two key components are unimpeded by the apparatus related to the com-
of Project Battuta were a wearable computer and an puter. The exception is the small clip-on device
associated augmented reality display package that attached to his glasses, which is a miniature computer
enabled users to overlay geospatial data and images on display that provides the augmentation to the real
the landscape as they travel through the environment. world. In his hand, he holds a trackball-style mouse.
Ultimately, users will have real-time access to a vast array On his left arm is a small keyboard that enables access
of distributed resources via a wireless Internet connec- to all of the keys found on a normal keyboard
tion. When wearable computers and augmented reality (although access to some keys requires that a combi-
displays are fully developed, a geographer will not only nation key also be pressed). On his left shoulder, we
be able to interpret the landscape as Battuta did, but will see a miniature GPS receiver with an embedded
also be able to utilize a wealth of geospatial data that will antenna that continuously records his position to
assist in that interpretation. within 15 meters, depending on sampling conditions.
Figure 7 shows a version of the wearable computer On his back, unobtrusively enclosed in his fishing-
that was developed by Keith Clarke and his vest-like jacket, is a computer that drives the system.
533
12:38:11.
This computer weighs only 1.75 pounds and measures One experiment that Hedley has conducted com-
6" × 5" × 2". Power is provided by two lithium-ion pares a standard desktop display with an augmented
rechargeable batteries that support up to 3 hours of reality (AR) system. With the desktop display, users
system run time each. manipulate 3-D surfaces using a VRML browser and a
Although the UCSB wearable computer is not yet mouse. The technology for the AR system is illustrated
used for practical geographic applications outside of in Figure 8. In the lower left and upper right of Figure
UCSB, it is important to recognize that the technology 8, we see users wearing what appears to be an HMD,
for wearable computers is evolving rapidly. In 2001, but this HMD is fundamentally different from those
Clarke described a promising new technology from described earlier in this chapter. In this case, a minia-
Microvision (http://www.mvis.com/) that would paint a ture video camera is mounted on the HMD. The video
computer image directly onto a person’s retina. Clarke stream captured by the camera is combined with a
indicated that this approach would be advantageous computer-generated 3-D image and displayed to each
because it would provide a high-resolution, high-con- eye via the HMD. The image that appears when look-
trast image that would not block the user’s view. Today ing through the HMD (see the lower right of Figure 8)
this technology is being marketed by Microvision; results from real-time tracking of the patterns shown
although not explicitly intended for the geographic on the card in the upper left of Figure 8 by AR soft-
applications that Clarke had in mind, it certainly has ware (ARToolkit). The AR software recognizes unique
that potential and is being used for a variety of com- marker patterns in the view from the camera, and
mercial and military purposes (http://www.microvision. places virtual 3-D objects on them in the output view.
com/wearable_displays/index.html). The result is that users can attach virtual 3-D objects to
real, manipulable objects (e.g., cards, paper), and hold
6.2 Research by Nick Hedley the card in their hands and manipulate it just as they
would manipulate objects in the real world. Contrast
In addition to its work on Virtual Puget Sound, the this notion with how we interact with 3-D objects in a
HIT Lab at the University of Washington is doing desktop display—to rotate an object, we click a button
intriguing research on advanced interface environ- and move the mouse. Novice users often find this task
ments. In this chapter, we focus on research that Nick awkward, presumably because it does not correspond
Hedley (2001; 2003) has done with individual users of to how they would examine a 3-D object in the real
“mixed reality” environments. world.
FIGURE 8 The augmented reality

system utilized by Nick Hedley in the
HIT Lab at the University of
Washington. (Courtesy of Nick
Hedley.)
12:38:11.
In his experiment, Hedley spent about 250 hours occlude a feature in the landscape that must be seen for
studying a total of 100 people who examined 3-D sur- safe navigation. We could also ask how your interaction
faces using these two different approaches. His findings with other people would change if others know that you
suggest that AR interfaces allow people to complete are seeing information they cannot see.
basic spatial problem-solving activities more quickly
and more accurately than when using desktop inter-
faces for the same 3-D geographic visualizations. His SUMMARY
findings also reveal that different combinations of inter-
face, user, and 3-D visualization can affect spatial prob- This chapter has examined how interactive computer
lem solving. From an interface perspective, AR is a 3-D graphics technologies can assist us in better under-
display (i.e., it is in the room with you) of a 3-D model, standing 3-D geospatial environments. The approaches
whereas a desktop interface is a 2-D display of a 3-D range from those that are a simulation of the real world
model. This means that in AR, we can inspect the vir- (a virtual environment (VE)) to those in which our view
tual 3-D model like we would inspect a real object, gain- of the real world is combined with computer-based
ing a better understanding of it by rotating it in our information (a mixed environment (ME)). An example
hands, or leaning closer to inspect it. This raises some of a geospatial VE would be your taking a tour through
interesting possibilities for visualizing spatial phenom- a simulation of how the 3-D landscape changed during
ena and gaining deeper insight into them through more major glaciations of North America. (Technically, we
natural interfaces. Hedley has also worked with other restrict the term VE to the situation in which you have
colleagues to develop and study AR interfaces for some control over the tour; if you had no control, then
geographic education (Shelton and Hedley 2002) and the tour would be termed an animation.) An example
public visualization tools (Woolard et al. 2003). of an ME would be your looking at the present-day
landscape and seeing past glacial features overlaid on
7 HEALTH, SAFETY, AND SOCIAL ISSUES the landscape. Thus far, VEs are more common than
MEs, but in the future, MEs might prove to be more
useful, particularly for geographic fieldwork.
In section 5 on the research issues for GeoVEs, we
We considered several technologies for creating VEs,
noted that one disadvantage of VEs is that their side
including desktop, wall-sized, head-mounted, drafting
effects are unknown. This idea is related to the broader
table–format, and room-format displays. A distinct
notion of health, safety, and social issues associated with
advantage of the latter three is that they allow the user
VEs and MEs. Health and safety issues can range from
to become more immersed in the VE—he or she might
something as simple as tripping over a power cord while
actually feel a sense of presence, of being in the VE,
immersed in a VE to discomfort associated with wear-
even though it is artificial. Today, most VEs are experi-
ing an HMD to cybersickness, a form of motion sick-
enced visually, although it is possible to utilize the full
ness that can result from exposure to VEs. Although
range of senses, including sound, touch, smell, taste, and
cybersickness and related problems are commonly asso-
body movement.
ciated with particular hardware (notably HMDs), such
We discussed four “I” factors that are relevant to
problems are not necessarily restricted to HMDs
designing geospatial VEs: immersion, interactivity,
(Wann and Mon-Williams 1997, 54).
information intensity, and intelligence of objects.
The use of VEs raises numerous social issues, such as
Although there is a tendency to assume that a higher
(Stanney et al. 1998):
degree of immersion is desirable (as in donning an HMD
• How will interaction in the virtual world modify that shields you from the real world), some researchers
behavior? have found that users seem to work effectively with stan-
• Will people turn their backs on the real world and dard desktop systems, which have a low degree of immer-
become discontented zombies wandering around sion. Key issues in interacting with the geospatial VE are
synthetic worlds that fulfill their whims? navigating through it and understanding one’s current
• Will people avoid reality and real social encounters orientation and location. Presumably this is difficult
with peers and become addicted to escapism? because when a user enters a VE, she or he loses the nor-
Because MEs are relatively new, little has been said mal benchmarks found in the real world. Information
about health, safety, and social issues related to them. intensity refers to the level of detail (LOD) present in the
Presumably, they would engender some of the same GeoVE. Our lack of experience with VEs makes it diffi-
concerns of VEs. For instance, imagine the potential cult to judge what an appropriate LOD should be or how
danger of trying to walk through a landscape in which a it should change with scale. Intelligence of objects refers
computer image is draped on the actual landscape. It is to the notion that objects in a VE can exhibit intelligence,
certainly conceivable that the computer image might just as they do in the real world. Potentially useful are
535
12:38:11.
avatars (images of people) that could help guide us Battuta is a form of augmented reality (AR) in which
through a VE and answer our questions about the VE. the real world is dominant and computer-based infor-
We looked at several applications of GeoVEs. One mation is overlaid on it (e.g., the preceding example
common application is the virtual city, which often in which glacial features were overlaid on the real-
stresses providing realistic visual representations of world landscape). A commercial system marketed by
buildings (and related structures). This approach has Microvision now provides the capability that appears
been criticized because it fails to consider the processes to make geographic augmented reality a distinct pos-
operating in the city, although some applications, such sibility rather than wishful thinking. Hedley’s work is
as Virtual Los Angeles and CommunityViz, clearly pro- also a form of AR, but in that case, greater emphasis
vide more than just realistic representations. is placed on interpreting the VE rather than the real
Given the novelty of VEs, there are numerous asso- world (i.e., the VE is displayed on a large card held in
ciated research issues. The International Cartographic the users’ hands, and users can rotate the card, and
Association’s Commission on Visualization and Virtual thus the VE, just as they would rotate an object in the
Environments has identified many research issues, and real world).
William Winn and his colleagues at the Human Finally, we touched on health, safety, and social issues
Interface Technology Laboratory at the University of associated with VEs and MEs. Health and safety issues
Washington have conducted a substantial amount of can range from something as simple as tripping over a
research with GeoVEs. Although Winn and his col- power cord to cybersickness, a form of motion sickness.
leagues have been able to identify a number of advan- Social issues deal with how the use of VEs and MEs will
tages and disadvantages of geospatial VEs, you should affect our views of the real world and our interaction
keep in mind that their research has focused largely on with others. For instance, will there be a tendency for
HMDs; thus, more research is needed. some to escape into VEs, and will it be difficult to inter-
In the realm of MEs, we looked at two applications: act with someone who utilizes an ME when you do not
Project Battuta and the work of Nick Hedley. Project have access to that same environment?
FURTHER READING
Bishop, I. D., and Lange, E., eds. (2005) Visualization in Goodchild, M. F. (2000) “Cartographic futures on a Digital
Landscape and Environmental Planning: Technology and Earth.” Cartographic Perspectives, no. 36:3–11.
Applications. London: Taylor & Francis. Considers what the future of digital cartography might hold and
Provides an overview of using VEs in planning, with an emphasis some issues associated with a Digital Earth. Also see the papers
on various applications; for earlier work, see Landscape and by Pickles and Rhind that appear in issue no. 37 of this journal.
Urban Planning Vol. 54, 2001.
Hansen, C. D., and Johnson, C. R., eds. (2005) The Visual-
Cartwright, W., Peterson, M. P., and Gartner, G., eds. (2007) ization Handbook. Amsterdam: Elsevier.
Multimedia Cartography, 2nd ed. Berlin: Springer-Verlag.
Chapters 21–24 and 44 deal with VEs.
Chapters 16–21 deal with VEs.
IEEE Computer Graphics and Applications 20, no. 4, 2000
Computers & Graphics 20, no. 4, 2000 (entire issue). (bulk of issue).
Although a bit dated, this special issue focuses on technical con-
Focuses on wall-sized displays and deals largely with the technical
cerns related to “virtual reality and 3D GIS.”
issues involved in generating them.
Costello, P. (1997) “Health and safety issues associated with
IEEE Computer Graphics and Applications 26, no. 4, 2006
virtual reality—A review of current literature.” http://www.
(bulk of issue).
agocg.ac.uk/reports/virtual/37/report37.htm.
Focuses on geovisualization; several papers discuss issues relevant
Reviews the early literature on health and safety issues associated
to VEs.
with VEs.
Feiner, S. K. (2002) “Augmented reality: A new way of seeing.” Johnson, A., Leigh, J., Morin, P., and Van Keken, P. (2006)
Scientific American 286, no. 4:48–55. “GeoWall: Stereoscopic visualization for geoscience research
and education.” IEEE Computer Graphics and Applications
Provides an overview of technical issues associated with creating
26, no. 6:10–14.
augmented reality, along with a range of applications.
Fisher, P., and Unwin, D., eds. (2002) Virtual Reality in Describes a range of geoscience applications for the GeoWall.
Geography. London: Taylor & Francis. MacEachren, A. M., Edsall, R., Haug, D., Baxter, R., Otto, G.,
Covers a range of applications for VEs, including “virtual land- et al. (2000) “Virtual environments for geographic visualiza-
scapes,” “virtual cities,” and “‘other’ worlds.” The emphasis is on tion: Potential and challenges.” In Workshop on New
desktop displays. Paradigms in Information Visualization and Manipulation
12:38:11.
(NPIVM ’99), ed. by D. S. Ebert and C. D. Shaw, pp. 35–40. Slocum, T. A., Blok, C., Jiang, B., Koussoulakou, A., Montello,
New York: The Association for Computing Machinery. D. R., Fuhrmann, S., and Hedley, N. (2001) “Cognitive and
Introduces the four “I” factors of geospatial VEs and describes a usability issues in geovisualization.” Cartography and
demonstration project for the ImmersaDesk involving the rela- Geographic Information Science 28, no. 1:61–75.
tionship of temperature to topography and precipitation. Introduces some research issues related to GeoVEs.
Raper, J. (2000) Multidimensional Geographic Information Stanney, K. M., Mourant, R. R., and Kennedy, R. S. (1998)
Science. London: Taylor & Francis. “Human factors issues in virtual environments: A review of
Pages 219–224 consider the use of VEs to depict geomorphologi- the literature.” Presence 7, no. 4:327–351.
cal change on Scolt Head Island, England; much of this book deals Reviews human factors issues related to VEs.
with theoretical issues related to 3-D mapping, including the
temporal aspect.
GLOSSARY
augmented reality: a combination of a real-world experi- enveloped by, included in, and able to interact with an envi-
ence with a virtual representation in which the real world is ronment that provides a continuous stream of stimuli and
dominant, such as overlaying the boundaries of “old-growth” experiences.
forests on your view of a mountainous environment; a subset information intensity: one of the “I” factors of geospatial
of a mixed environment. virtual environments (GeoVEs); deals with the level of detail
augmented virtuality: a combination of a real-world experi- (LOD) presented in the VE.
ence with a virtual representation in which the virtual repre- intelligence of objects: one of the “I” factors of geospatial
sentation is dominant; a subset of a mixed environment. virtual environments (GeoVEs); refers to the idea that objects
autostereoscopic display: a method of creating a stereo- within a VE often exhibit some degree of behavior or intelli-
scopic display in which specialized glasses are not necessary. gence.
avatar: a representation of a person in a virtual environ- interactivity: one of the “I” factors of geospatial virtual envi-
ment. ronments (GeoVEs); an important element of interactivity is
cybersickness: a form of motion sickness that can result developing suitable approaches for navigating through the
from exposure to virtual environments. VE.
desktop display: a technology in which a display device is mixed environment (ME) (or mixed reality (MR)): com-
placed on a desktop or table (e.g., an LCD display or a lap- bines a real-world experience with a virtual representation;
top). most commonly, the real world is dominant—this is termed
augmented reality.
Digital Earth: digital spatiotemporal information about the
Earth should be readily available via a multiresolution 3-D presence: the subjective experience of being in one place or
VE. environment when one is physically in another.
Digital Globe: digital spatiotemporal information about the room-format display: provides a room-sized view of a vir-
Earth or another celestial body is displayed in the undistorted tual environment by projecting images onto walls and the
three-dimensional globe, as either a virtual or a tactile display. floor (e.g., the CAVE); a head tracker on one user governs the
view that all users see through stereo glasses.
drafting table–format display: a virtual environment is
achieved by projecting images onto a large table-like screen virtual environment (VE): a three-dimensional computer-
that is tilted at a 45° angle; a head tracker on one user governs based simulation of a real or imagined environment that users
the view that the other users see, and a specialized 3-D mouse are able to navigate through and interact with.
can be used to interact with the simulated environment. virtual reality (VR): the use of computer-based systems to
head-mounted display (HMD): helmet-like device that create lifelike representations of the real world.
shields a person from the real-world view and provides images wall-sized display: a technology in which the graphic display
of a virtual environment to each eye. covers a large portion of a wall; this is typically accomplished
immersion: one of the “I” factors of geospatial virtual envi- by combining multiple projected images.
ronments (GeoVEs); a psychological state in which a user is
REFERENCES
Anselin, L., Syabri, I., and Kho, Y. (2006) “GeoDa: An intro- Anthamatten, P., and Ziegler, S. S. (2006) “Teaching geogra-
duction to spatial data analysis.” Geographical Analysis 38, phy with 3-D visualization technology.” Journal of
no. 1:5–22. Geography 105:231–237.
Aggett, G., and McColl, C. (2006) “Evaluating decision sup- Bishop, I. (1994) “The role of visual realism in communicat-
port systems for PPGIS applications.” Cartography and ing and understanding spatial change and process.” In
Geographic Information Science 33, no. 1:77–92. Visualization in Geographical Information Systems, ed. by
537
12:38:11.
H. M. Hearnshaw and D. J. Unwin, pp. 60–64. Chichester, Dykes, J., Moore, K., and Wood, J. (1999) “Virtual environ-
England: Wiley. ments for student fieldwork using networked compo-
Bishop, I. D., and Lange, E. (eds.). (2005) Visualization in nents.” International Journal of Geographical Information
Landscape and Environmental Planning: Technology and Science 13, no. 4:397–416.
Applications. London: Taylor & Francis. Dykes, J. A., and Mountain, D. M. (2003) “Seeking structure
Buziek, G., and Döllner, J. (1999) “Concept and implementa- in records of spatio-temporal behaviour: Visualization
tion of an interactive, cartographic virtual reality system.” issues, efforts and applications.” Computational Statistics &
Proceedings of the 19th International Cartographic Data Analysis 43, no. 4:581–603.
Conference, Ottawa, Canada, pp. Section 5:88–99 (CD- Edsall, R., and Larson, K. L. (2006) “Decision making in a
ROM). virtual environment: Effectiveness of a semi-immersive
Cartwright, W. (1997) “New media and their application to ‘Decision Theater’ in understanding and assessing human-
the production of map products.” Computers & environment interactions.” AutoCarto 2006, Vancover, WA,
Geosciences 23, no. 4:447–456. CD-ROM.
Cartwright, W. (1999) “Extending the map metaphor using Fuhrmann, S., and MacEachren, A. M. (2001) “Navigation in
web delivered multimedia.” International Journal of desktop geovirtual environments: Usability assessment.”
Geographical Information Science 13, no. 4:335–353. Proceedings, 20th International Cartographic Conference,
International Cartographic Association, Beijing, China,
Clarke, K. C. (2001) “Cartography in a mobile Internet age.”
CD-ROM.
International Cartographic Conference, Beijing, China, CD-
ROM. Gabbard, J. L., and Hix, D. (1997) A Taxonomy of Usability
Characteristics in Virtual Environments. Blacksburg:
Clarke, K. C., Du, Q., Nuernberger, A., and Pingel, T. (2003)
Virginia Polytechnic Institute and State University. http://
“A prototype cartographic user interface for wearable
people.cs.vt.edu/~jgabbard/publications/framework97.pdf.
computing.” Proceedings of the 21st International
Cartographic Conference, Durban, South Africa, pp. Gore, A. (1998) “The Digital Earth: Understanding our
1430–1438. planet in the 21st century.”
http://www.earthscape.org/p1/goa01/.
Crampton, J. W. (1999) “Online mapping: Theoretical context
and practical applications.” In Multimedia Cartography, ed. Guimbretière, F., Stone, M., and Winograd, T. (2001) “Fluid
by W. Cartwright, M. P. Peterson, and G. Gartner, pp. interaction with high-resolution wall-size displays.”
291–304. Berlin: Springer-Verlag. UIST’2001, Volume 2002, Orlando, FL. http://graphics.
stanford.edu/~francois/Papers/UIST2001/PostBrainstorm.p
Day, A. (1994) “From map to model: The development of an
df.
urban information system.” Design Studies 15,
no. 3:366–384. Hedley, N. R. (2001) “Virtual and augmented reality inter-
faces: Empirical findings and implications for spatial
Czernuszenko, M., Pape, D., Sandin, D., DeFanti, T., Dawe,
visualization.” Proceedings, 20th International Cartographic
G. L., et al. (1997) “The ImmersaDesk and Infinity Wall
Conference, International Cartographic Association,
Projection-based virtual reality displays.” Computer
Beijing, China, CD-ROM.
Graphics 31, no. 2:46–49.
Hedley, N. R. (2003) “3D geographic visualization and spatial
Davis, J., and Chen, X. (2002) “LumiPoint: Multi-user laser-
mental models.” Unpublished PhD dissertation, University
based interaction on large tiled displays.” Displays 23, no.
of Washington, Seattle, WA.
5: 205–211. http://graphics.stanford.edu/papers/multiuser/.
Heim, M. (1998) Virtual Realism. New York: Oxford
Drascic, D., and Milgram, P. (1996) “Perceptual issues in aug-
University Press.
mented reality.” Proceedings, SPIE, Volume 2653, San Jose,
CA, pp. 123–134. Johnson, A., Leigh, J., Morin, P., and Van Keken, P. (2006)
http://gypsy.rose.utoronto.ca/people/david_dir/SPIE96/SPI “GeoWall: Stereoscopic visualization for geoscience
E96.html. research and education.” IEEE Computer Graphics and
Dunbar, M. D., Moskal, L. M., and Jakubauskas, M. E. (2004)
“3D visualization for the analysis of forest cover change.” Johnston, D. M., Curtis, D., and Sonin, J. (2000) “The Digital
Geocarto International 19, no. 2:103–112. River Basin: A science and technology program for infor-
mal education.” Proceedings, Hydroinformatics 2000, Iowa
Dodge, M., and Kitchin, R. (2001b) Mapping Cyberspace.
City, IA.
London: Routledge.
http://www.gis.uiuc.edu/riverweb/DRB/HydroInfo.pdf.
Dodgson, N. A. (2005) “Autostereoscopic 3D displays.”
Jepson, W., and Friedman, S. (1998) “SimCity of Angels.”
Computer 38, no. 8:31–36.
Civil Engineering 68, no. 6:44–47.
Dykes, J. (2002) “Creating information-rich virtual environ-
Jepson, W. H., Liggett, R. S., and Friedman, S. (2001) “An
ments with geo-referenced digital panoramic imagery.” In
integrated environment for urban simulation.” In Planning
Virtual Reality in Geography, ed. by P. Fisher and D.
Support Systems, ed. by R. K. Brail and R. E. Klosterman,
Unwin, pp. 68–92. London: Taylor & Francis.
pp. 387–404. Redlands, CA: ESRI Press.
12:38:11.
Kumsap, C., Borne, F., and Moss, D. (2005) “The technique of Samuels, S. (1999) “Driving forward.” Spectator 33, no. 1:1.
distance decayed visibility for forest landscape visualiza- Shelton, B. E., and Hedley, N. R. (2002) “Using augmented
tion.” International Journal of Geographical Information reality for teaching earth–sun relationships to undergradu-
Science 19, no. 6:723–744. ate geography students.” Proceedings of the First IEEE
Kwartler, M., and Bernard, R. N. (2001) “CommunityViz: An International Augmented Reality Toolkit Workshop,
integrated planning support system.” In Planning Support Darmstadt, Germany. http://depts.washington.edu/pettt/
Systems, ed. by R. K. Brail and R. E. Klosterman, pp. papers/shelton-hedley-art02.pdf.
285–308. Redlands, CA: ESRI Press. Slocum, T. A., Blok, C., Jiang, B., Koussoulakou, A., Montello,
Luebbering, C. R., and Carstensen, L. W. (2007) “Geospatial D. R., Fuhrmann, S., and Hedley, N. R. (2001) “Cognitive
analysis by experts: High-resolution monitors and and usability issues in geovisualization.” Cartography and
improved performance.” Abstract, Association of Geographic Information Science 28, no. 1:61–75.
American Geographers Annual Meeting, San Francisco, Slocum, T. A., Dunbar, M. D., and Egbert, S. L. (2007)
CA, pp. 398–399. “Evaluating the potential of the GeoWall for geographic
MacEachren, A. M., Kraak, M.-J., and Verbree, E. (1999b) education.” Journal of Geography 106, no. 3: 91–102.
“Cartographic issues in the design and application of Stanney, K. M., Mourant, R. R., and Kennedy, R. S. (1998)
geospatial virtual environments.” Proceedings of the 19th “Human factors issues in virtual environments: A review
International Cartographic Conference, Ottawa, Canada, of the literature.” Presence 7, no. 4:327–351.
Section 5: pp. 108–116 (CD-ROM).
Stoltman, A. M., Radeloff, V. C., and Mladenoff, D. J. (2004)
http://www.geovista.psu.edu/publications/ica/ica99/.
“Forest visualization for management and planning in
MacEachren, A. M., Edsall, R., Haug, D., Baxter, R., Otto, G., Wisconsin.” Journal of Forestry 102, no. 4:7–13.
Masters, R., Fuhrmann, S., and Qian, L. (1999a) “Exploring
Verbree, E., Maren, G. V., Germs, R., Jansen, F., and Kraak,
the potential of virtual environments for geographic
M.-J. (1999) “Interaction in virtual world views—Linking
visualization.” http://www.geovista.psu.edu/publications/
3D GIS with VR.” International Journal of Geographical
aag99vr/fullpaper.htm.
Middleton, D., Scheitlin, T., and Wilhelmson, B. (2005)
Wann, J. P., and Mon-Williams, M. (1997) “Health issues with
“Visualization in weather and climate research.” In The
virtual reality displays: What we do know and what we
Visualization Handbook, ed. by C. D. Hansen and C. R.
don’t.” Computer Graphics 31, no. 2:53–57.
Johnson, pp. 845–871. Amsterdam: Elsevier.
Wheless, G. H., Lascara, C. M., Valle-Levinson, A., Brutzman,
Moore, K. E. (2002) “Visualizing data components of the
D. P., Sherman, W., Hibbard, W. L., and Paul, B. E. (1996a)
urban scene.” In Virtual Reality in Geography, ed. by
“The Chesapeake Bay Virtual Environment (CBVE):
P. Fisher and D. Unwin, pp. 257–269. London: Taylor &
Initial results from the prototypical system.” The
Francis.
International Journal of Supercomputer Applications and
Noll, S., Paul, C., Peters, R., and Schiffner, N. (1999) High Performance Computing 10, no. 2/3:199–210.
“Autonomous agents in collaborative virtual environ-
Wheless, G. H., Lascara, C. M., Valle-Levinson, A., Brutzman,
ments.” IEEE 8th International Workshops on Enabling
D. P., Sherman, W., Hibbard, W. L., and Paul, B. E. (1996b)
Technologies: Infrastructure for Collaborative Enterprises
“Virtual Chesapeake Bay: Interacting with a coupled
(WET ICE ’99), IEEE Computer Society, Stanford, CA,
physical/biological model.” IEEE Computer Graphics and
pp. 208–215.
Reddy, M. (2001) “Perceptually optimized 3D graphics.”
Williams, K. J. H., Ford, R. M., Bishop, I. D., Loiterton, D.,
IEEE Computer Graphics and Applications 21, no. 5:2–9.
and Hickey, J. (2007) “Realism and selectivity in data-dri-
Reddy, M., Leclerc, Y., Iverson, L., and Bletter, N. (1999) ven visualizations: A process for developing viewer-ori-
“TerraVision II: Visualizing massive terrain databases in ented landscape surrogates.” Landscape and Urban
VRML.” IEEE Computer Graphics and Applications 19, Planning 81, no. 3:213–224.
no. 2:30–38.
Windschitl, M., Winn, W., Hedley, N., Fruland, R., and
Reljic, Z., Sawada, M., Poitevin, J., and Saunders, G. (2005) Postner, L. (2002) “Using immersive visualizations to
“Integrating GIS and 3D visualization for dynamic promote the understanding of complex natural systems:
landscape representation in Canada’s national parks.” Learning inside Virtual Puget Sound.” http://depts.washing-
Proceedings of the 98th Annual Canadian Institute of ton.edu/edtech/jrst.html.
Geomatics Conference, Ottawa, Canada, CD-ROM.
Winn, W. (2002) “What can students learn in artificial
Rickel, J., and Johnson, W. L. (1999) “Animated agents for environments that they cannot learn in class?” First
procedural training in virtual reality: Perception, cognition, International Symposium, Open Education Faculty,
and motor control.” Applied Artificial Intelligence 13, Anadolu University, Turkey.
no. 4/5:343–382.
Winn, W. (2005) “What we have learned about VR and
Riedl, A. (2007) “Digital Globes.” In Multimedia Cartography learning and what we still need to study.” Proceedings,
(2nd ed.)., ed. by W. Cartwright, M. P. Peterson, and G. Laval Virtual 05: 7th International Conference on Virtual
Gartner, pp. 255–266. Berlin: Springer-Verlag. Reality, Laval, France, CD-ROM.
539
12:38:11.
Winn, W., Windschitl, M., Fruland, R., and Lee, Y.-L. (2002) Proceedings of the Third International World Wide Web
“Features of virtual environments that contribute to Conference, Darmstadt, Germany, pp. 21–25.http://www.
learners’ understanding of earth science.” Annual Meeting igd.fraunhofer.de/archive/1995_www95/proceedings/posters/
of the National Association for Research in Science 35/index.html.
Teaching, New Orleans, LA. Woolard, A., Lalioti, V., Hedley, N., Carrigan, N., Hammond,
Witmer, B. G., and Singer, M. J. (1998) “Measuring presence M., and Julien, J. (2003) “Case studies in application of
in virtual environments: A presence questionnaire.” augmented reality in future media production.”
Presence 7, no. 3:225–240. Proceedings of the Second IEEE Symposium on Mixed
Wood, A., Drew, N., Beale, R., and Hendley, B. (1995) and Augmented Reality (ISMAR ’03), Tokyo, Japan, pp.
“HyperSpace: Web browsing with visualization.” 294–295. http://csdl.computer.org/comp/proceedings/
ismar/2003/2006/00/20060294.pdf.
12:38:11.
Trends in Research and
Development
541
12:38:11.
Trends in Research and
Development
OVERVIEW
The discipline of cartography has changed considerably individual is involved in the visualization process. For
since the 1960s, evolving from a discipline based on pen instance, you might be located at one university and wish to
and ink to one based on computer technology. As the discuss with a researcher at another university the visual-
field continues to evolve, it is important to keep pace with ization aspects of a climate-change model. Ideally, you
ongoing research and new developments. The purpose of should both be able to manipulate the model from your
this chapter is to examine some of these developments; respective locations and see the results of what the other
specifically, we will consider the following: Daniel Carr person is doing. We consider two applications of collabo-
and his colleagues’ work with linked micromap plots and rative geovisualization: Nick Hedley and his colleagues’
conditioned choropleth maps, using senses other than work with augmented reality (AR) and a specialized
vision to interpret spatial patterns, collaborative geovisu- projection table known as HI-SPACE; and Alan
alization, multimodal interfaces, information visualiza- MacEachren and his colleagues’ work with ImmersaDesks
tion, spatial data mining, visual analytics, and mobile and HI-SPACE.
mapping and location-based services (LBS). Although today, most of us are still working with
Linked micromap plots (LM plots) focus on a series Windows and associated mouse-based interfaces,
of small maps termed micromaps, which divide a single researchers are experimenting with novel interfaces that
spatial distribution into pieces. LM plots focus on local utilize speech, lip movements, pen-based gestures, free-
pattern perception as opposed to overall spatial pattern hand gestures, and head and body movements.
perception, and permit the display of statistical informa- Multimodal interfaces are novel interfaces that involve
tion such as confidence intervals for each observation. In two or more of these techniques; for instance, you might
a fashion similar to LM plots, conditioned choropleth point to a location on a screen without touching the
maps (CCmaps) split a spatial pattern for a single screen (a free-hand gesture) and say, “Show me all
attribute into a series of choropleth maps. In the case of homes within 100 meters of this location.” Some of the
CCmaps, rows and columns of maps are used, with the most exciting work in multimodal interfaces is being
rows and columns corresponding to two attributes that done at the GeoVISTA Center at Penn State University,
might explain the attribute displayed in the choropleth where researchers are developing collaborative geovisu-
map series. alization systems for emergency management situations.
In this chapter, we will consider how other senses (e.g., Information visualization involves the visualization
sound, touch (or haptics), and even smell) might be used and analysis of nonnumeric abstract information such as
to assist in visualization, or be an alternative to visual- the nature of topics that are discussed on the front page
ization (for those who are blind or visually impaired). of a newspaper over a monthlong period. Spatialization
Collaborative geovisualization (or geocollaboration) is the process of converting such abstract information to
refers to geovisualization activities in which more than one a spatial framework in which visualization is possible.
12:38:11.
Trends in Research and Development
We consider some work that two geographers, André statistical ideas. In this section, we consider two of their
Skupin and Sara Fabrikant, have done in this realm. contributions: linked micromap plots and conditioned
Spatial data mining is the term used for spatial appli- choropleth maps.
cations involving knowledge discovery in databases
(KDD), which refers to exploratory approaches for find-
1.1 Linked Micromap Plots
ing interesting patterns in databases. Spatial data mining
is particularly useful for large databases because the
Linked micromap plots (LM plots) (Carr and Pierson
computational approaches that are used can uncover
1996; Carr et al. 1998) consist of three basic elements, as
patterns that visual processing might miss. We’ll look at
illustrated in Color Plate 1: a series of small maps
two applications of spatial data mining, one utilizing a
termed micromaps (the column labeled “Maps”), a leg-
sophisticated cluster analysis method to process data that
end (the column labeled “States”), and one or more
would normally be displayed as a choropleth map, and
related graphics (the third and fourth columns). The pri-
the other helping us determine which thematic attributes
mary attribute of interest (in this case, “Unemployment
are associated with particular regions within a city.
Rate By State”) is displayed in the micromaps and plot-
Visual analytics is a term developed by the U.S.
ted as colored dots in the column labeled “Rates and
National Visualization and Analytics Center (NVAC).
95% CI.” For instance, if we look at the bottom
Recognizing that large multivariate spatiotemporal data
micromap, we can see that North Dakota has an unem-
sets cannot be interpreted by visualization methods
ployment rate between 3 and 4 percent (as indicated by
alone, NVAC developed the notion of visual analytics,
the position of the green dot in the “Rates and 95% CI”
which involves a synthesis of visualization and analytical
column). The term linked is derived from the notion
methods; as such, to some extent it overlaps the notion of
that information in one column can be related to infor-
spatial data mining. Although the notion of visual ana-
mation in another column because of the correspond-
lytics was developed primarily for homeland security
ing color; thus, the micromap gives us a location
purposes (in response to the September 11 terrorist
for North Dakota, the first graph gives North Dakota’s
attacks on the United States), advancements in visual
unemployment rate, and the last graph gives the
analytics can also be applied to a broad range of areas,
number unemployed.
such as disease epidemiology and regional planning.
Rather than providing a depiction of the overall spa-
Mobile mapping refers to the notion that we can get
tial pattern (as on a traditional choropleth map), the
access to maps and related geospatial information as we
purpose of LM plots is to focus on local pattern percep-
move through our environments. It is closely associated
tion, which is achieved by ordering the data from high
with location-based services (LBS), which deliver data
to low (from top to bottom in Color Plate 1) and by
and information services as a function of a user’s current
depicting subsets of these ordered data in each
or projected location. Mobile mapping and LBS have
micromap. Again, by looking at the bottom map, we can
become possible because of the proliferation of personal
see that four of the states with the lowest unemploy-
data assistants (PDAs), smartphones, and wearable com-
ment rates are adjacent to one another and are located
puters, all of which provide access to the Web and deter-
in the north central United States.
mine the geographic location of these devices (e.g., via
Note that the 50 states (and Washington, DC) have
the global positioning system (GPS)). The book
been split into groups of five, with the exception of
Location Based Services and TeleCartography provides
Illinois, which is placed by itself in the middle. Carr
an overview of recent research in this realm.
and Pierson (1996, 19) argued “that groups of five
One of the difficulties of writing a text about a rapidly
facilitate counting and still allow quick label and value
changing discipline is that it is virtually impossible for
matching by relative position.” Placing Illinois by itself
the authors to provide up-to-date information. To assist
is logical if you remember that the middle value in an
you in acquiring information about ongoing develop-
ordered set of data is the median. We can thus easily
ments, the last section of this chapter lists journals, con-
contrast states above the median from those below the
ferences, and useful Internet sites. You should also con-
median.
sult the home page for this text (http://www.prenhall.
In looking at the micromaps, you might have noticed
com/slocum) to find more recent information.
that individual states appear distorted, and that some
appear larger or smaller than we might expect on an
1 LINKED MICROMAP PLOTS AND equivalent (equal-area) map projection. Carr and
CONDITIONED CHOROPLETH MAPS his colleagues used a notion developed by Mark
Monmonier (1992a) and intentionally distorted the
Daniel Carr and his colleagues have developed a num- states so that smaller states could be more easily identi-
ber of intriguing visualization techniques that not only fied. Even with such modifications, however, small enu-
display spatial information, but also integrate important meration units will be difficult to detect; thus, Carr and
543
12:39:03.
his colleagues suggest that 50 enumeration units is Californians are affected. In contrast, Washington, DC,
about the largest number that should be handled in an has the highest unemployment rate, but fewer than
LM plot in static (printed) form. 100,000 of the people living there are affected.
Note that the “Rates and 95% CI” column includes a Carr and his colleagues (Wang et al. 2002; Chen et al.
set of gray lines on either side of the line connecting the 2006) have developed Web-based programs for exploring
colored dots. These are the 95 percent confidence inter- LM plots associated with cancer data for the United
vals, a measure of uncertainty in the data. Confidence States, and a variation of these programs is now available
intervals are necessary because the data are based on a via the National Cancer Institute’s (NCI) State Cancer
sample; thus, rather than saying that the unemployment Profiles Web site (http://statecancerprofiles.cancer.gov/
rate for Washington, DC, is 9 percent, we should say micromaps/). These interactive programs have a number
that we are 95 percent certain that the unemployment of advantages over static printed LM plots (as suggested
rate is between approximately 8 and 10 percent (see the by Figure 1), including the abilities to display a large num-
top row in the color plate). Including such information ber of enumeration units (via scrolling), to display differ-
illustrates the statistical bent of Carr and his colleagues. ent cancer attributes (e.g., prostate cancer vs. lung and
The logic of including the graph “No. Unemployed” is bronchus cancer), to view the data at either the state or
that it places the data in context, since one could argue the county level, to sort by various data (e.g., to sort by
that a high unemployment rate is more important if it rate as opposed to incidence), and to change the color
affects a large number of people. Thus, we can see not linking individual units (spectral, sequential, and diverg-
only that California (the green dot on the extreme right) ing schemes are possible, with the sequential scheme
has a high unemployment rate, but also that 1.2 million being most desirable when printing in black and white).
FIGURE 1 Example of a Web-based program for creating linked micromap plots (LM plots). (From the National Cancer
Institute’s State Cancer Profiles Web site: http://statecancerprofiles.cancer.gov/micromaps/.)
12:39:03.
These features are best appreciated by actually using the represent the lung cancer mortality rate data. We can
interactive programs, and so we encourage you to experi- see that a large number of health service areas with
ment with the NCI site. high lung cancer mortality rates are associated with
high precipitation and a high percent below the
1.2 Conditioned Choropleth Maps poverty level (the red region in the southeastern
United States, shown in the upper right map). This
Carr and his colleagues (2000b; 2005) developed association (or correlation) does not necessarily mean
conditioned choropleth maps (CCmaps), which permit that there is a cause-and-effect relationship, but it does
a visual examination of attributes that might explain suggest that these and other related attributes should
the spatial pattern of various forms of cancer. More be examined in order to explain the high mortality
generally, CCmaps can be used to develop hypotheses rates in the Southeast.*
about the relationship between one attribute (the As with LM plots, CCmaps are best appreciated by
dependent attribute) and two potential explanatory experimenting with associated software, which can be
attributes. To illustrate, consider Color Plate 2, which found at http://www.galaxy.gmu.edu/~dcarr/ccmaps/.
provides a set of CCmaps intended to assist in explain- There you will find that it is possible to dynamically
ing lung cancer mortality rates for white men ages 65 change the class limits on any of the three attributes,
to 74 for health service areas (i.e., counties or aggre- thus developing different visualizations of the relation-
gates of counties) in the United States. In the legend ships among the attributes. The ability to change class
along the top of the color plate, you can see that lung limits fits well with the notion of data exploration soft-
cancer mortality rates are depicted in three colors: ware.
blue for low values, gray for moderate values, and red
for high values. Values at the bottom of this legend are
2 USING SENSES OTHER THAN VISION
lung cancer mortality rates, whereas values above the
TO INTERPRET SPATIAL PATTERNS
legend are the percent of white men ages 65 to 74 in
each class. In this example, the attributes considered
(because of their relationship to lung cancer) are per- Map designers have begun considering how we might
cent below the poverty level and annual precipitation. use our other senses, either as an alternative or as a
Although precipitation might seem like a strange supplement to vision. Using senses other than vision
attribute to relate to lung cancer, Carr and his is obviously critical for blind (or visually impaired)
colleagues (2002) argued that high precipitation was individuals, but there are numerous reasons why
historically advantageous for growing tobacco and senses other than vision can be useful even to nor-
could lead to greater time spent indoors, and that the mally sighted individuals. First, other senses can be
associated high humidity could increase chemical useful when the work environment requires that
activity on the lung surface. More recently, Carr and vision be used for tasks besides map interpretation
his colleagues (2005) provided an alternative argu- (as in an aircraft cockpit). Second, other senses can
ment for using precipitation—that low precipitation assist in map interpretation when visual interpreta-
would be associated with high particulate levels in the tion is not clear. For instance, Suresh Lodha and his
atmosphere, which would be associated with higher colleagues (2000) found that the comparison of
levels of lung cancer. regions on choropleth maps was enhanced when both
The annual precipitation data are depicted in the vision and sound were used. Third, multiple senses can
three columns of maps (health service areas having be useful for interpreting multivariate data; for
low, medium, and high precipitation values are shown instance, Chris Harding and his colleagues (2002)
from left to right, with the values for each column used vision, sound, and touch to represent three
indicated in the legend at the bottom of Color Plate respective attributes (gravity, age of the oceanic crust,
2). In an analogous fashion, the percent below poverty and change of slope) associated with geological struc-
level data are shown in the three rows of maps (health tures along the Mid-Atlantic Oceanic Ridge. Below
service areas having low, medium, and high poverty we consider additional examples of how three of our
levels are displayed from bottom to top, as indicated senses (sound, touch, and smell) might be utilized to
in the legend to the right). Thus, similar to a scatter- interpret spatial patterns.
plot, low values of precipitation and poverty are rep-
resented in the lower left, and high levels of * The numbers in the upper right of each map are the weighted aver-
precipitation and poverty are represented in the upper ages of lung cancer mortality rates for health service areas appearing
right; the actual symbols shown on each map, however, on that map.
545
12:39:03.
2.1 Sound THE ABSTRACT SOUND VARIABLES

Nominal Ordinal
From a cartographic standpoint, John Krygier (1994) Data Data
LOCATION A
conducted some of the earliest work on using sound as The location of a sound in a A
A
A Possibly
A A Effective
two or three dimensional space A
an alternative sense to vision. Krygier noted two basic A

A
A
A
A Effective
A
forms of sound: realistic and abstract. He divided LOUDNESS
realistic sounds, those sounds whose meanings are The magnitude of a sound Not
based on our past experiences, into speech (or narra- A A A A A A A A A A A Effective
Effective
tion) and mimetic sound icons (or “earcons”). A simple PITCH

example of speech is when a computer speaks the value The highness or lowness Not
(frequency) of a sound C D E F G A B C Effective
Effective
of an enumeration unit rather than displays it as text.
Another example of speech is the audio component REGISTER
The relative location of a pitch Not
sometimes included in multimedia encyclopedias (e.g., in a given range of pitches CDEFGABCCDEFGABC
Effective
Effective
you might hear, “You are now looking at major Allied
troop movements during the Battle of the Bulge”). An TIMBRE
The general prevailing Not
example of a mimetic sound icon would be the sounds quality or characteristic A A A Effective
Effective
of a sound
of fire and wind in an animation illustrating changes in
DURATION
a forested landscape. The length of time a sound Not
is (or isn't) heard A A Effective
Abstract sounds have no obvious meaning, and thus Effective
require a legend to explain their use. For example, RATE OF CHANGE

imagine a map of census tracts with the title “Median The relation between the Not
durations of sound and A A A A A A A A Effective Effective
Income, 1997” in which different magnitudes of loud- silence over time
ness represent different incomes (e.g., a mouse click on ORDER
a high-income tract would produce a louder sound than The sequence of sounds Not
over time A B C D C A D B Effective
Effective
a mouse click on a low-income tract). To understand the
magnitudes of loudness, a legend indicating that a ATTACK/DECAY
The time it takes a sound to Not
higher magnitude represents a higher income would reach its maximum/minimum A A
Effective
Effective
have to be provided (if the reader were blind, someone
would have to tell the reader what each magnitude of
FIGURE 2 Abstract sound variables. (Reprinted from
loudness indicates or give the reader a tactile legend).
Krygier, J. B., Sound and Geographic Visualization, Copyright
The process of creating abstract sounds is sometimes
1994, p. 153, with permission from Elsevier Science.)
referred to as sonification.
Analogous to the visual variables, Krygier (1994,
154–146) identified the following abstract sound vari- Krygier (1993) created a video that illustrated sev-
ables: eral applications of how abstract sound variables might
be used. In one application, he combined a choropleth
• Location: The location of a sound in two- or three-
animation of AIDS in the United States from 1980 to
dimensional space
1995 with a loudness variable representing the total
• Loudness: The magnitude of a sound
number of AIDS cases in each year. When viewing the
• Pitch: The highness or lowness (frequency) of a
resulting animation, one gets the feeling of an impend-
sound
ing disaster, which is exactly what Krygier intended. In
• Register: The relative location of a pitch in a given
this same animation, Krygier also used pitch to repre-
range of pitches
sent the percentage increase of new AIDS cases each
• Timbre: The general prevailing quality or character-
year (a lower pitch represented a lower percentage
istic of a sound
increase). Krygier (1994) noted, “The pitch can be heard
• Duration: The length of time a sound is (or is not)
‘settling down’ as the percentage increase drops and
heard
steadies in the late 1980s. An anomaly can be heard in
• Rate of change: The relation between the durations
1991 where the animation switches from actual AIDS
of sound and silence over time
cases to model predicted AIDS cases” (p. 157). We sus-
• Order: The sequence of sounds over time
pect that viewers would readily hear the “settling
• Attack/decay: The time it takes a sound to reach its
down,” but would find it more difficult to detect the
maximum or minimum
anomaly; this might be a case in which experience
As Figure 2 shows, Krygier argued that the bulk of these would be advantageous (e.g., a musician would have an
variables would be effective for ordinal-level data. advantage over a nonmusician).
12:39:03.
In another application, Krygier created a multivariate results enhanced “the understanding of quantification
display by combining traditional visual variables (light- of uncertainty considerably” (p. 192).
ness and size) with sound variables. For the traditional A more recent example of the use of sound is the
visual variables, Krygier superimposed proportional iSonic software developed by Haixia Zhao and her
circles onto a choropleth map. The proportional circles colleagues (Zhao et al. 2005; Zhao 2006; Zhao et al. in
depicted “median income,” whereas the choropleth map press). Zhao and her colleagues argue that traditional
displayed the “percentage of population not in the labor approaches for presenting spatial data to the visually
force.” For the sound variables, Krygier used a pitch impaired have often utilized text readers that speak
within three different octaves (a register) to display geographic names and values in alphabetical order,
a “drive to work index,” and three levels of pitch within thus preventing a full understanding of the spatial pat-
each of the three octaves to represent the “percentage tern of a data set. As an alternative, iSonic provides a
poor.” He obtained values for the sound variables variety of novel approaches to assist in interpreting
by pointing at enumeration units. For example, for spatial data. For example, a choropleth map can be
an enumeration unit with a long drive to work and a low interpreted by associating map classes with violin
rate of poverty, a high-octave pitch followed by a low pitches and then playing these pitches in a fixed spatial
pitch within that octave would be heard. Krygier (1994) sequence, from left to right and from top to bottom.
indicated that “After a short period of [use] . . . Alternatively, it is possible to divide a map into regions
it becomes relatively easy to extract the four data and play the values for enumeration units in each
variables” (p. 158). region. The current version of iSonic (http://www.cs.
Several other researchers performed early experi- umd.edu/hcil/audiomap/) was developed as a result of a
ments with sound. Peter Fisher (1994a) developed soft- series of user studies done by Zhao and her colleagues
ware that utilized sound to portray the uncertainty in (e.g., Zhao et al. in press).
remotely sensed images. With Fisher’s software, a cur- As you can see in these examples, most of the
sor could be moved across a remotely sensed image research on sound has focused on how abstract sounds
under either user or automatic control (Color Plate 3). can be used to represent spatial data. Researchers also
As the cursor moved, the user heard a sound represent- have begun to explore how realistic sounds might be
ing the uncertainty associated with the current pixel utilized. An example is Glenn Brauen’s (2006) use of
location. For example, if duration were used as the sound to demonstrate the variability in Canadian fed-
sound variable, a long duration would indicate a pixel eral election results. In a traditional election map, each
with a low uncertainty (i.e., high reliability). In using electoral district would be shaded according to the
Fisher’s software, we found it relatively easy to deter- affiliation of the winning candidate. As an alternative,
mine the uncertainty associated with specific pixels, but Brauen created a sound map (http://gcrc.carleton.ca/
difficult to determine the uncertainty of groups of pix- cne/proof_of_concepts/elect2004/) in which the volume
els. This problem might be obviated by also using visual of a speech given by each candidate was proportional
methods to depict uncertainty. For example, we might to the percentage of votes received by that candidate.
utilize both sound and a set of gray tones to depict Brauen argued that this “ . . . reintroduces some of the
uncertainty (say, with dark areas representing more complexity of the election results into the map” (p. 64).
uncertain information. Similarly, The Cybercartographic Atlas of Antarctica
In addition to the study mentioned earlier, Lodha (http://www.carleton.ca/gcrc/caap/) uses sound to
and his colleagues (1996b) also used sound to visual- illustrate contested territorial claims in Antarctica:
ize the uncertainty in interpolation on isarithmic When a user places the mouse over an uncontested
maps. Lodha and his colleagues compared one inter- territory, the user hears a news report only from the
polation approach with another (for instance, inverse country claiming this region, whereas over a contested
distance compared with kriging). In discussing their territory, the user hears multiple news reports simulta-
results, they clearly felt that sound provided informa- neously. Pulsifer et al. (2007, 209) argue that “This
tion not attainable from a visual display. For example, layering of voices and languages increases the sense
they created a 3-D display in which both the height of confusion and tension between the different
and the color of glyphs were mapped to uncertainty countries.” Another interesting use of sound in the
(the higher and whiter the glyph, the greater the Cybercartographic Atlas is the association of music
uncertainty), and noted that it was “hard to glean with historical periods. For instance, Mozart’s String
information from a glyph” due to “the projection of Quartet no. 3 is played when a user clicks on the
the graphic.” However, when they mapped the uncer- Cook expedition (1772–75), whereas Miles Davis’s
tainty to pitch and presented the glyphs in sequence music accompanies the Fuchs and Hillary expeditions
from low to high uncertainty, they found that the in 1958.
547
12:39:03.
2.2 Touch (or Haptics) In a fashion similar to Krygier’s work with sound,
Amy Griffin (2001) proposed a haptic variable syntax
Strictly speaking, touch is the sensation we experience (Figure 3). In examining Figure 3, you can see that
when our skin comes in contact with an object, such Griffin split the haptic variables into three types: tactile,
as when we touch a three-dimensional physical mod- visual analog derived, and kinesthetic. The tactile haptic
el. Haptics is a broader term that includes not only variables involve only touch; for instance, by placing a
the sensation of contact, but also the kinesthetic finger on a portion of a map, given the appropriate tech-
receptors activated in our muscles, tendons, and joints, nology one should be able to feel whether the map is
as in the action of “sweeping the hand across a three- vibrating and how cool or warm it is. The visual analog
dimensional surface” (Golledge et al. 2005, 340). derived variables are named after the traditional visual
In practice, however, these terms are often used variables and again involve only touch; for instance,
interchangeably. circular disks of different sizes could be placed on a
FIGURE 3 A haptic variable syntax

for touch. (Courtesy of North
American Cartographic Information
Society and Cartographic Perspectives.)
12:39:03.
map, and a user could be asked to judge the differences They note that in Western cultures, smell is often
in the disks’ sizes. The kinesthetic variables involve placed at the bottom of the sensory hierarchy, but
“a movement of, or a change in the user’s position, or in clearly, there are instances in which smell is extremely
the position of a stimulus relative to the user” (p. 17); important, such as when evaluating the effectiveness of
for instance, Griffin suggested that increased friction winegrowing regions in France (Press and Minta 2000).
could correspond to increased terrain roughness. Lauriault and Lindgaard also note that untrained
Griffin specified that haptic variables, similar to sound people can recognize only about 16 scents, but that
variables, should be used only for nominal or ordinal specialists (e.g., those in the perfume industry) can rec-
levels of measurement (Figure 3). Although Griffin did ognize hundreds of scents. As an example of the use of
not actually test her proposed haptic variables with peo- smell in cartography, they reference the map Twin
ple, she suggested numerous applications such as the Cities Odorama: A Smell Map of Minneodorous and
following: Scent Paul. Scents are not actually embedded in this
map, but quotes such as the following provide an idea
“the vibration variable might be used to visualize the output of the importance of smell to inhabitants of the
of a seismic model that predicts the intensity of shaking due Minneapolis–St. Paul region:
to an earthquake. . . . Temperature might be used to repre-
sent water temperature in a river, so that the user could
Minneapolis’ Ichiban Japanese Restaurant always smells
track cold currents through space and time. Resistance could
wonderful when you walk by. It’s that distinctive Japanese
be used to represent measures of pollution; the more
food smell of Tepanyaki, Sushi and Tempura. It has a very
polluted an area, the more difficult it might be to . . . move
comforting emotional feel to me.
your hand through the area.” (p. 19)
Other researchers have begun to actually test the effec- Actually generating this particular smell could be chal-
tiveness of haptics. Particularly notable is the work of lenging, but Lauriault and Lindgaard note that hardware
Reginald Golledge, Daniel Jacobson, and their col- and software have been developed for generating smells
leagues, who have experimented with both sound and (e.g., TriSenx’s at http://www.trisenx.com/products.html).
haptics in the Haptic Soundscapes project (Jacobson As technology evolves and as our willingness to consider
et al. 2002; Golledge et al. 2005; Rice et al. 2005). the importance of the sense of smell also evolves, it will be
Golledge and his colleagues have noted a number of interesting to see what role smell plays in cartography.
challenging problems faced by the blind and visually
impaired. For instance, it is difficult for a blind user to 3 COLLABORATIVE GEOVISUALIZATION
determine the relationship between a small local region
and the extent of the larger map. One solution that Although we often think of maps and mapping software
Jacobson has experimented with is a graticule-like hap- as being used by individuals, there clearly are instances
tic grating that can be “. . . felt by the user and allows a in which collaboration among individuals is important.
user to maintain a sense of distance traveled, scale, and For example, a grade school teacher might ask students
direction . . . similar in nature to the information to interpret the pattern of earthquake activity on a
imparted by a graticule on a standard map” (Rice et al. printed map. If the map is of sufficient size, several
2005, 385). Their Web site http://soundscapes.geog.ucsb. students might gather around the map and discuss the
edu/ contains numerous examples of some of their pattern they see. When such collaborative activities
efforts, and a link to many related publications can be involve visualization, we refer to them as collaborative
found at http://www.immerse.ucalgary.ca/publications/. geovisualization (or geocollaboration). If we consider
One of the difficulties of interpreting Golledge et al.’s the full realm of printed and computer-based maps,
work is that many of us do not have access to haptic then we can conceive of collaboration that takes place
interface devices; we hope that this research will lead to at the same or at a different time, and at the same or at
an increased proliferation of the hardware necessary a different place. This produces a two-by-two matrix of
for haptic interfaces. possible instances for collaboration (Figure 4). For
example, an urban planning meeting generally would
2.3 Smell involve people collaborating in the same time and
place, whereas a school project involving both class-
Although it may seem far-fetched to acquire informa- work and fieldwork could require collaboration at
tion from maps via the sense of smell, Tracy Lauriault different times and places. In this section, we consider
and Gitte Lindgaard (2006) have made a strong case examples of collaborative geovisualization being under-
for the potential of using smell to interpret spatial data. taken by cartographers and other geoscientists.
549
12:39:03.
Same Time Different Time
Urban Strategic
Same planning military
Place meeting planning
Scientists School project

Different collaborate with involving classwork
Place decision-makers and fieldwork
FIGURE 4 Collaborative geovisualization can take place at

the same or at a different time (the columns) or the same or a
different place (the rows). (After Slocum et al. 2001. First pub-
lished in Cartography and Geographic Information Science FIGURE 5 Hedley and his colleagues achieved collaborative
28(1), p. 67. Reprinted with permission from the American geovisualization by using HMDs containing a video camera
Congress on Surveying and Mapping.) and a specialized projection table known as HI-SPACE.
(Courtesy of Nicholas R. Hedley, Mark Billinghurst, Lori
Postner, Richard May, and Hirokazu Kato, “Explorations in
3.1 Hedley and His Colleagues’ Work with AR the Use of Augmented Reality for Geographic Visualization,”
Presence 11:2 (April 2002), pp. 119–133. © 2002 by the
Nick Hedley and his colleagues have worked in the Massachusetts Institute of Technology.)
realm of augmented reality (AR). Here we discuss their
related effort in collaborative geovisualization (Hedley
et al. 2002), which utilized their previous work with AR, Sony Glasstron viewer was used rather than the HMD,
but also permitted computer interpretation of natural it was possible to achieve both AR and immersive VE
gestures and allowed a range of views—from the real views (Figure 7). The Sony Glasstron contained an iner-
world to AR to an immersive virtual environment tial tracker that enabled head orientation to be inter-
(VE). Hedley and his colleagues created an AR through preted and a pressure pad that could be used to move
the use of a head-mounted display (HMD) that had an through the VE.
attached video camera and associated software; thus, Hedley and his colleagues found that users preferred
when users looked at a card containing a unique pat- the Sony Glasstron to the more traditional hands-free
tern of marks, a 3-D image appeared at that location. HMD “because it was easy to share views with others
Natural gestures were interpreted via a specialized [and] because it reduced . . . incidences of nausea” (2002,
projection table known as the human information 127). Hedley et al., however, noted that users of the
workspace (or HI-SPACE) developed by Richard Glasstron often wanted to move backward in the VE (as
May (1999; http://www.hitl.washington.edu/projects/ in computer games), which was a feature that Hedley et
hispace/ ). An infrared light source was placed beneath al. had not made possible. Hedley and his colleagues also
the table, and a camera above the table recorded the noted several limitations of their initial collaborative
infrared light. When a user gestured (e.g., by pointing effort (known as AR PRISM): Users could not annotate
with a finger) or placed an object on the table, the the data or place their own virtual models in the AR
infrared light was blocked, producing dark images that scene; there was no support for viewing two-dimensional
could be interpreted by computer software. Initially, imagery (e.g., remotely sensed images); virtual models
users viewed a paper map that was placed on a table would sometimes disappear if a marker card was
that didn’t have any specialized apparatus. If users then obscured; and using a paper map as a base prevented
donned HMDs and examined a marked card placed on zooming and panning (pp. 128–129).
the map, they would see a 3-D model of terrain at that
location (Figure 5). An individual could then pick up 3.2 MacEachren and His Colleagues’ Work with
the model (by picking up the card) and pass it around ImmersaDesks and HI-SPACE
to others, all of whom could see the 3-D AR view. Other
types of geographic information also could be seen by Alan MacEachren and his colleagues (2000; 2004)
placing the appropriately marked cards on the map next experimented with collaborative geovisualization by
to a particular terrain model (Figure 6). If a handheld using ImmersaDesks to examine precipitation and
12:39:03.
A B
FIGURE 6 In Hedley and his colleagues’ collaborative geovisualization approach, other

types of geographic information could be seen by placing the appropriately marked cards
on the map next to a terrain model that was being examined. Here we see a soil card
placed next to a terrain model (A); when the soil card is interpreted, we get the result in
(B). (Courtesy of Nicholas R. Hedley, Mark Billinghurst, Lori Postner, Richard May, and
Hirokazu Kato, “Explorations in the Use of Augmented Reality for Geographic
Visualization,” Presence 11:2 (April 2002), pp. 119–133. © 2002 by the Massachusetts
Institute of Technology.)
temperature relationships in association with communication between the groups was possible only
Hurricane Agnes, which was a major weather event in via a speaker telephone (Internet tools proved unreli-
1972. They attempted both same-place and different- able).
place collaborations (for the same period of time) by Given the limitations of the ImmersaDesks,
having one group of researchers work with an MacEachren and his colleagues (2006b) have experi-
ImmersaDesk at Pennsylvania State University, and mented with other technologies to allow collaborative
having another group work with an ImmersaDesk at geovisualization. One of these is the HI-SPACE tech-
Old Dominion University. Color Plate 4 provides one nology used by Hedley et al. In contrast to the sophisti-
example of the images utilized in the Hurricane Agnes cated AR and VE capabilities that Hedley et al. used,
study. The bottom portion of the color plate shows a 3- MacEachren and his colleagues have emphasized the
D portrayal of terrain; the middle portion represents ability of the HI-SPACE table to allow users to interpret
temporal changes in precipitation rather than in eleva- natural hand gestures (Figure 8). They stress that the HI-
tion, with olive green “clouds” depicting precipitation SPACE environment allows “groups of individuals to
isosurfaces above a certain threshold value (Hurricane work in a round-table fashion” and is thus appropriate
Agnes is depicted by a “blanket” of rain multiple days “for small team decision-making where participants are
thick); and the top portion portrays the temperature at relatively equal partners in decisions” (p. 25). In the next
the end of the precipitation event. By examining section, we will consider their work on the collaborative
images such as these, researchers were able to see the large-screen system known as DAVE_G.
“substantially reduced temperature across the basin
following Hurricane Agnes, as the huge quantities of
water dumped on the region slowly evaporated” 4 MULTIMODAL INTERFACES
(MacEachren et al. 2000, 39). Although MacEachren
and his colleagues (2004) felt that the researchers had At present, the bulk of our work with personal comput-
gleaned useful information using the ImmersaDesks, ers is accomplished via a windows-icons-menus-pointing
MacEachren et al. noted a number of problems, includ- device (WIMP) interface. Although WIMP interfaces
ing that: the use of stereo glasses limited the are likely to remain common for the next 5 to 10 years,
researchers’ ability to utilize gestures and facial expres- researchers are experimenting with a variety of inter-
sions; only one individual could have control of the dis- faces that eventually might replace the WIMP paradigm.
play, and that individual needed to hand a wand and Novel ways of interacting with computers include
control glasses to another person desiring control; opti- speech, lip movements, pen-based gestures, free-hand
mal viewing was best for the individual in control; and gestures, and head and body movements. Speech refers
551
12:39:03.
FIGURE 7 A handheld Sony Glasstron viewer that enables FIGURE 8 Natural gestures can be recognized via the
both AR and immersive VE views. (Courtesy of Nicholas R. HI-SPACE projection table. An infrared light source is placed
Hedley, Mark Billinghurst, Lori Postner, Richard May, and beneath the table, and a camera above the table records the
Hirokazu Kato, “Explorations in the Use of Augmented infrared light. When users gesture (e.g., by pointing with their
Reality for Geographic Visualization,” Presence 11:2 (April fingers), the infrared light is blocked, producing dark images
2002), pp. 119–133. © 2002 by the Massachusetts Institute of that can be interpreted by computer software. (Courtesy of
Technology.) North American Cartographic Information Society and
Cartographic Perspectives.)
to using spoken commands (e.g., “Draw a line from New

York City to Albany”), which requires speech recogni- composed of two or more novel modes of interaction.
tion hardware and software; because speech recognition Thus far, the most common multimodal interfaces have
is imperfect, it is often supplemented by computer-vision utilized a combination of either speech and pen-based
methods that can interpret lip movements. A pen-based gestures or speech and lip movements. For example,
gesture refers to using a pen-like device to specify loca- when using speech and a pen-based gesture, the user
tions or to provide written input by touching a computer might point to the state of New York and say, “Shade
display or input device. In contrast, a free-hand gesture this region in blue.”
uses just the arm and hand to specify a location (or Multimodal interfaces have numerous advantages.
region) without touching the display. Obviously, free- One is that the associated novel forms of interaction
hand gestures require considerably greater computer are arguably more desirable than WIMP interfaces
processing, given that computer-vision methods must be because the former are more natural—imagine being
used to determine the location intended by the user, as able to simply say, “Show me the location of houses
opposed to the x and y coordinates that result from costing between 75,000 and 125,000” as opposed to
touching a display in a pen-based gesture. Sharon Oviatt having to learn and traverse the menus of a WIMP
(2002; 2007) is one researcher who has championed interface. Second, multimodal interfaces allow a wider
novel interaction methods. Oviatt has promoted the spectrum of users to utilize computers; for example,
notion of multimodal interfaces, which are interfaces a visually impaired user might prefer speech, whereas
12:39:03.
a child might prefer a pen-based gesture. A third et al. 2006a; Cai et al. 2006) argue that decision
advantage is the ability to use certain interaction modes makers need to interact with disaster-related software
in specialized situations; for example, in the field you in a natural manner, as opposed to utilizing GIS
might want your hands free, and thus speech is an obvi- specialists who are familiar with the menus and soft-
ous choice, whereas interaction during a meeting might ware of traditional GIS interfaces. Their system,
require a quiet, pen-based gesture. A fourth advantage known as the Dialogue-Assisted Visual Environment
is the need to reduce the task difficulty and errors for Geoinformation (DAVE_G), builds on XISM by
resulting from using only one mode. For example, in using both speech and free-hand gesture, but in a col-
speech-based input, users often have difficulty express- laborative framework (Figure 9). Interpreting the
ing themselves, particularly for spatial tasks. Oviatt input of multiple users is handled, in part, by assign-
(1997, 120) illustrated this point in a classic early study ing separate microphones and cameras to each collab-
involving interactive maps. To specify the location of orator. The greater complexity of the system also is
open space with just speech, a user stated, “Add an open dealt with by carefully analyzing the actions of deci-
space on the north lake to . . . include the north lake sion makers in real-world emergency operations
part of the road and north.” When both pen-based ges- centers before developing the software.
tures and speech were permitted, the user utilized a pen One key characteristic of ongoing work in multi-
to circle the area and stated, “Open space.” Overall, modal interfaces is that it is multidisciplinary. For
Oviatt found that when using multimodal interfaces, example, work at Penn State involves those in the
tasks were simplified, they were completed more departments of Computer Science, Geography, and
quickly, and there were fewer errors. Moreover, people Information Sciences and Technology. Clearly, the com-
preferred multimodal interaction. plexity of multimodal interfaces, particularly in the col-
Rajeev Sharma and his colleagues (2003) at laborative realm, necessitates a group research effort.
Pennsylvania State University and Advanced Interface
Technologies have done considerable work with multi-
modal interfaces involving speech and free-hand ges- 5 INFORMATION VISUALIZATION
tures. One of their early systems, known as XISM AND SPATIALIZATION
(Kettebekov et al. 2000), was used to simulate an urban
emergency response system. In this case, a single user Information visualization involves the visualization
played the role of an emergency center operator and analysis of nonnumeric abstract information. To
who dispatched vehicles to particular locations. The create a visualization of nonnumeric information,
user’s gestures were recorded by a camera located at graphic designers frequently utilize a spatial or geo-
the top of the display, and the user’s speech was graphic metaphor—a process known as spatialization.
recorded by a microphone dome hanging from the ceil- The notion is that people should feel comfortable in
ing. Sophisticated computer hardware and software interpreting this surface because of their familiarity
were then used to interpret the resulting speech and ges- with topographic surfaces.
tures. Emergencies were indicated in the display by the André Skupin and Sara Fabrikant (2003; 2008) are
animation of symbols and associated audible alarm sig- two geographers who have done extensive work in
nals. The user would then respond with an appropriate information visualization. An example of Skupin’s
verbal command (e.g., “Acknowledge this”) and then work is his attempt to visualize the relationships
use speech and gestures to dispatch the vehicles (e.g., among more than 2,000 abstracts submitted to the
the user might say, “Dispatch an ambulance from this 1999 annual meeting of the Association of American
station to that location” while gesturing to the station Geographers (Skupin 2002, 2004). Given that the dis-
and location). Because free-hand gestures were too cipline of geography normally is divided into three
coarse at some scales, the system permitted zooming via broad areas (i.e., human geography, physical geogra-
speech and gesture (e.g., the user would say, “Zoom phy, and techniques), Skupin wondered how these
here” and make the appropriate gesture). three areas, and their relations, might be expressed
More recently, Alan MacEachren and his col- in the relations among abstracts, and also whether
leagues in the GeoVISTA Center at Penn State have other divisions of the discipline might be appropriate.
teamed with Sharma and his colleagues to develop a One visualization that Skupin created appears
collaborative geovisualization system for emergency in Figure 10, where we see two-dimensional “maps” of
management situations commonly dealt with in a GIS the relationships among topics found in the abstracts.
framework (e.g., disaster relief associated with a hur- The leftmost map portrays the most generalized ver-
ricane). MacEachren and his colleagues (MacEachren sion, in which 10 regions arise, whereas the other two
553
12:39:03.
FIGURE 9 Two people working with

the DAVE_G collaborative
geovisualization system for emergency
management situations. The users
provide input by speech and free-
hand gestures (note the two cameras
placed above the display). (From
Rauschert, I., et al. (2002) “Designing a
human-centered, multimodal GIS inter-
face to support emergency manage-
ment.” Proceedings of the 10th ACM
International Symposium on Advances
in Geographic Information Systems,
McLean, VA, 119–124. (c) 2002 ACM,
Inc. Reprinted by permission.)
maps show the detail that is possible when maps of management dominate the top of the map. This is a
100 and 800 regions are used, respectively. In dis- heterogeneous area at the intersection of human and
cussing Figure 10A, Skupin stated: natural environments. (p. 56)
It would be interesting to see how visualizations such as
. . . human geography occupies the right half. . . .
these would evolve over time: Imagine comparing the
[A]spects of physical geography dominate the upper
left quadrant. The processing and modeling of geo-
visualization shown in Figure 10 with one developed 5
graphic data dominate the lower left quadrant. The or 10 years later.
cluster labeled geogr/new/research . . . contains many An example of Fabrikant’s work is the research she
abstracts that deal with the teaching of geography, undertook with Barbara Buttenfield to develop spa-
such as research into the development of new teach- tializations of documents found in GEOREF, an
ing tools and techniques. Issues surrounding resource online set of documents pertaining to literature in the
FIGURE 10 A spatialization of topics found in abstracts submitted to the 1999 meeting of

the Association of American Geographers: Map A represents the most generalized view,
whereas maps B and C show the detail that is possible when zooming to various levels.
(After Skupin 2002, “A cartographic approach to visualizing conference abstracts,” IEEE
Computer Graphics and Applications 22, no. 1, p. 56; © 2002 IEEE.)
12:39:03.
geological sciences (Fabrikant and Buttenfield 2001).

A
Figure 11 illustrates two different spatializations of
100 documents found in GEOREF developed by
Fabrikant and Buttenfield. In Figure 11A we see a
two-dimensional map of point symbols for the 100
documents, where each symbol is labeled according to
a first-level keyword found in GEOREF, and each
symbol is shaded according to the number of other
documents in GEOREF that contain the same key-
word. Documents that are near one another in the
two-dimensional map are presumably more similar to
one another in content than those that are farther
Topic Headings
away. In Figure 11B, we see a three-dimensional view (number of hits)
of the data portrayed in Figure 11A: There, the height
of the surface represents the frequency with which
various keywords are used in other documents (in this
case, the labels are topical index terms taken from the
GEOREF thesaurus). Fabrikant and Buttenfield B
argued that spatializations such as these serve as
“tools to provide an overview of available informa-
tion, to discover relationships between items in a data
archive, and to filter nonrelevant pieces of informa-
tion” (p. 263). Although these spatializations might
seem abstract, Fabrikant and Buttenfield found that
users gleaned considerable information from them
because they were able to take advantage of the geo-
graphic metaphor (i.e., that nearby things are more
similar to one another).
You might wonder how we convert nonnumeric
abstract information into the spatializations depicted
here. One key step is selecting an appropriate spatial
layout technique, such as multidimensional scaling
(MDS) or the self-organizing map (SOM) (Skupin and
Fabrikant 2008, 72). Unfortunately, space does not
permit us to delve into such techniques here. These
techniques are commonly discussed in intermediate
or advanced courses in quantitative methods, and so
we encourage you to develop a solid background in
quantitative methods.
6 SPATIAL DATA MINING
In recent years, computer scientists and other IT

professionals have struggled with extracting knowl-
edge from very large data sets. For example, a health- FIGURE 11 Spatializations of 100 documents found in
related database might have millions of patient GEOREF, an online bibliographic database: (A) a 2-D
records, with tens or even hundreds of attributes map depicting the number of other documents that share
for each patient. Traditional statistical approaches keywords with those documents plotted in x and y; (B)
such as correlation and regression can be utilized to 3-D maps of the same information. (From Fabrikant, S. I.,
examine such data, but these approaches have a high and Buttenfield, B. P. 2001. “Formalizing semantic spaces
computational burden and normally are utilized only for information access.” Annals of the Association of
for theory-driven hypothesis testing. As a result, American Geographers 91, no. 2, pp. 272, 273. Courtesy of
researchers have developed the technique known as Blackwell Publishing.)
555
12:40:28.
knowledge discovery in databases (KDD), which 0.95, respectively. Obviously, distance plays a greater
refers to exploratory approaches for finding interest- role in Color Plate 5B, with the result being more
ing patterns in databases and, ultimately, uncovering homogeneous regions. Murray and Shyy suggested
knowledge about those databases. Given that the that their method ultimately be used in a data explo-
intention is to mine useful information (and, ulti- ration framework in which the weights for crime rate
mately, knowledge) from these databases, the term and distance can be experimented with, along with the
data mining is often used instead of KDD; moreover, number of classes (note that six classes are used in
when the database is inherently spatial, the term Color Plate 5). In examining the maps in Color Plate
spatial data mining is often used. However, you 5, also note that particular crime rates are not assigned
should keep in mind that technically, data mining is to each class (that is, the classes are simply numbered
considered a subset of KDD. In general, KDD 1 to 6), and that the classes do not appear to have a
involves four steps: developing background knowl- logical visual order. The reason for these characteris-
edge, data preprocessing, data mining, and knowledge tics is that class limits under the bicriterion median
construction (Miller 2008). clustering can overlap one another, which is normally
We now consider two examples that illustrate the not the case on choropleth maps. Although overlap-
notion of spatial data mining. Due to space limitations ping classes are likely to complicate interpretation,
and the fact that we have not delved deeply into KDD, they do permit a more thorough integration of spatial
we have selected examples that involve few observa- relationships into the clustering process.†
tions and do not require completing all of the steps As a second example of spatial data mining, we
involved in KDD—thus, our focus is on data mining consider the efforts of Natalia Andrienko and her col-
and knowledge construction.* For our first example, leagues (2001b) to study regional groupings of adminis-
we consider the work of Alan Murray and Tung-Kai trative districts in Bonn, Germany. The five colored
Shyy (2000), who wished to examine property crime regions in Color Plate 6 represent one possible way in
rates in the 178 suburbs of Brisbane, Australia. Murray which a geographer might group the districts of Bonn.
and Shyy initially mapped the suburban crime rates To examine the characteristics of these regions,
using a traditional choropleth map (they used the Andrienko and her colleagues submitted the district
Fisher–Jenks optimal classification method). They identifiers (i.e., which region the district is a member
argued, however, that this approach was inappropriate of) and numerous thematic attributes associated with
because traditional classification methods are aspatial the districts to the spatial data mining algorithm C4.5
and consider only the values of crime rates, not the (Quinlan 1993). The objective of C4.5 is to determine
spatial relationships of the associated suburbs. Murray which attributes (and subranges thereof) are most
and Shyy then argued that both crime rates and the closely associated with previously identified classes (in
spatial relationships among suburbs could be consid- our case, the five regions).
ered by using bicriterion median clustering, a variant of As an illustration of the nature of the C4.5
a median cluster analysis approach commonly used in algorithm, consider the resulting decision tree shown
spatial data mining. They accounted for spatial rela- below the map of Bonn in Color Plate 6. At the top of
tionships between suburbs by including the distance the decision tree, you can see that “Percentage
between neighboring suburbs as an attribute in the (Age groups = 18 - 30) ,” or the percentage of the
clustering method (in addition to the attribute prop- population 18 to 30 years old, is the attribute that ini-
erty crime rates). tially best discriminates among the five regions, with
The key point about Murray and Shyy’s use of bicri- the break point being 19.03. Looking at the left por-
terion median clustering is that they were able to tion of the second level of the decision tree, you can
experiment with assigning different weights to each of see that a total of 44 districts have less than or equal
the attributes. To illustrate, Color Plate 5 portrays two to 19.03 percent of the population 18 to 30 years old,
different runs of their cluster method: In Color Plate whereas the right portion of the second level reveals
5A, the crime rate and the distance attributes are that 18 districts have greater than 19.03 percent of
weighted 1.0 and 0.25, respectively, whereas in Color the population 18 to 30 years old. Most important,
Plate 5B, crime rate and distance are weighted 1.0 and the right portion consists largely of the “Center”
† Although the same symbology is used for both maps in Color Plate 5,
note that the range of values falling in a particular class would not nec-
* More complex examples of data mining include those by Bertolotto essarily be the same on the two maps. For another cluster analysis
et al. (2007), Guo (2007), and Turdukulov et al. (2007b). approach that resulted in overlapping classes, see Rowles (1991).
12:40:28.
region (i.e., 83 percent of the 18 districts are in essential for understanding and solving complex
this region). Again, looking at the left portion of spatiotemporal problems. He states:
the second level, you can see that “Percentage
(Population groups = Foreigners),” or the percentage Geovisualization offers interactive access to the data
of foreigners in the population, breaks the 44-district behind the map that is realized by combining graphics
group into two subgroups. In the third level you can with geo-computational tools and database techniques. . . .
see that one of the subgroups has a percentage of for- However, to be able to deal with global challenges, we
need more. Visualization has to be combined with
eigners greater than 12.1, and is largely associated
analytics. (p. 115)
with the “South-East” region. Finally, again looking
at the right portion of the second level, you can see The key is that large multivariate spatiotemporal data
that “Percentage (Age groups = 0 - 18) ,” or per- sets cannot be interpreted using just visualization;
centage of the population 0 to 18 years old, splits the rather, interpreting them requires a synthesis of visual-
18-district group into two groups: There, the two ization and analytical methods.
North-West districts have a higher percentage of peo- NVAC was formed in response to the September 11,
ple who are 0 to 18 years old. To summarize, the C4.5 2001, terrorist attacks on the United States. Today,
method has revealed that the Center and South-East NVAC’s mission is to:
portions of Bonn appear to have different population
structures than other areas: The Center has a high • Detect, prevent, and reduce the threat of terrorist
percentage of people in the 18- to 30-year-old range, attacks
whereas the South-East has a high percentage of for- • Identify and assess threats and vulnerabilities to our
eigners. homeland
Andrienko and her colleagues argue that spatial • Recover and minimize damage from terrorist attacks,
data mining is best used in conjunction with interac- should they occur (http://nvac.pnl.gov/about.stm)
tive visualization techniques. Thus, each portion of the To get input from the academic community, NVAC
decision tree can be depicted on the map by clicking has created five Regional Visualization and Analytics
on the desired portion. For instance, in Color Plate 6 Centers (RVACS; http://nvac.pnl.gov/centers.stm).
note that the right portion of the second level of the One of these centers, the North-East Visualization
tree has a thick black line surrounding it, indicating and Analytics Center (NEVAC), is coordinated with
that it has been clicked on. Looking at the map, the GeoVISTA Center at Penn State, which we men-
you can see that small square boxes identify the 18 tioned earlier in this chapter. NEVAC’s Web site
associated districts. Admittedly, novel spatial data- (http://www.geovista.psu.edu/NEVAC/index.html)
mining methods are more complicated than tradi- indicates that although homeland security is a focal
tional thematic maps, but they have the potential point, advancements in visual analytics “will be rele-
to uncover patterns in the data that we might not vant for all domains in which an ability to make sense
normally detect (imagine performing a similar task out of complex information and apply knowledge to
for 100 attributes within the 3,000 + counties of the real-world decisions is needed,” as in disease epidemi-
United States). ology, environmental science and management, and
regional planning.
Examples of research involving visual analytics can
7 VISUAL ANALYTICS be found in the January 2007 issue of Cartographica
(as already mentioned), in the International Journal of
The term visual analytics was developed by the U.S. Geographical Information Science (Vol. 21, Issue 8,
National Visualization and Analytics Center (NVAC; 2007), in the Spring 2007 issue of Information
http://nvac.pnl.gov/ ) and was described in depth in Visualization, and as individual papers published else-
the book documenting NVAC’s research agenda: where (e.g., Guo et al. 2006). Many visual analytics
Illuminating the Path: The Research and Development papers have a strong data-mining emphasis; for
Agenda for Visual Analytics (Thomas and Cook instance, Guo (2007) found that spatial data mining
2005). Visual analytics is defined as “the science of was essential to analyzing the pandemic spread of a
analytical reasoning facilitated by interactive visual population of 1.6 million people associated with more
interfaces” that should “detect the expected and than 180,000 locations. Since visual analytics research
discover the unexpected” (Thomas and Cook 2005, often involves large multivariate spatiotemporal data
p. 4). In the Summer 2007 issue of the journal sets, it can be challenging for a reader without a strong
Cartographica, which focuses on visual analytics, background in geovisualization and data mining to
Menno-Jan Kraak argues that visual analytics is understand. As a starting point, we suggest reading the
557
12:40:28.
application papers by Turdukulov et al. (2007a) and Gartner and his colleagues (2007). This book is based
Weaver et al. (2007). Turdukulov et al. discuss the on a series of symposiums that were held at the Vienna
problem of exploring rip channels, which create dan- University of Technology in Austria in 2002, 2004, and
gerous currents at swimming sites in Australia. 2005. The location of the symposiums was partly a func-
Traditionally, researchers have tried to use animation tion of the extensive use of smartphones and related
for this purpose, but these small features are difficult technologies in Western Europe (Peterson 2007). The
to detect, due to change blindness. As an alternative, term TeleCartography refers to the “distribution of
Turdukulov et al. developed a sophisticated analytical cartographic presentation forms via wireless data
approach for detecting rip channels, as well as novel transfer interfaces and mobile devices” (Gartner 2007,
methods for visualizing rip channels’ change over 3). Another term that is sometimes associated with
time. Weaver et al. describe the development of an mobile mapping is ubiquitous cartography, which
interactive visual tool for exploring the visitation pat- refers to the “ability for users to create and use maps in
terns of guests at Pennsylvanian hotels in the late any place and at any time to resolve geospatial prob-
1800s. Their paper provides an interesting example of lems” (p. 3). An eight-chapter section of Gartner et al.’s
interdisciplinary work, in this case conducted by visual book that is relevant to this chapter is entitled
analytics experts and historical geographers, and is a “Visualization and Cartographic Communication.”
good illustration of the power of the Improvise soft- Some of the papers included there are particularly rel-
ware developed by Chris Weaver (2004). Improvise evant to mobile mapping (e.g., Döllner et al. 2007
permits those without extensive programming exper- describe a novel approach for navigating 3-D city mod-
tise to “. . . rapidly build and browse multiple coordi- els on mobile devices), while others are more relevant
nated views of their data” (p. 89) and is arguably a to related topics (e.g., Cartwright 2007 describes an
good example of “new methods for constructing visu- evaluation of a virtual environment). It is noteworthy
ally based systems” that are critical for visual analytics that two papers deal with augmented reality (those by
(Thomas and Cook 2005, p. 8). Schmalstieg and Reitmayr 2007 and Höllerer et al.
2007), as augmented reality can be considered part of
mobile mapping.
8 MOBILE MAPPING AND LOCATION-BASED It will be interesting to see how the area of mobile
SERVICES mapping and LBS evolves. Peterson (2007, 75) notes
that “worldwide cellular subscribers are forecasted to
Mobile mapping refers to the notion that we can get reach 3.2 billion by the end of 2010.” Assuming that
access to maps and related geospatial information as we many of these subscribers will be utilizing smartphones
move through our environment. This has been made and related location-based services, the impact on
possible through the use of personal data assistants cartography could be comparable to the impact the
(PDAs), smartphones, and wearable computers. By hav- Web has had. Certainly, this new technology has
ing access to the Web and determining the geographic the potential to dramatically change the extent to and
location of the device (e.g., via the global positioning the manner in which people utilize maps.
system (GPS)), such technology enables users to obtain
a host of information related to the local environment,
or what are known as location-based services (LBS). 9 KEEPING PACE WITH RECENT DEVELOPMENTS
For example, if you are visiting a city, you might use a
smartphone to determine the shortest walking route to One problem we faced in writing this text was keeping
a museum or the locations of restaurants that serve Thai pace with ongoing developments: The field of cartogra-
food. Mobile mapping and LBS provide a unique map- phy is changing so fast that any text becomes dated as
ping environment because of the small screen size of soon as it is written. We have handled this problem to
the associated technology, the lack of standard controls some degree by providing updated information on the
for interacting with the display (e.g., you wouldn’t use a home page for the text. Those wishing detailed infor-
mouse to interact with a smartphone), the need for mation on current research topics should also consider
rapid response (you may be on the move, and thus a reviewing recent issues of the journals and proceedings
slow response may not apply to where you are a few listed in Table 1. Note that the journals in the table are
seconds later), and the fact that your attention may not split into primary and secondary categories. Most of the
be completely focused on the display. articles within the publications in the primary list have
An overview of research issues related to mobile a cartographic component, whereas only some of the
mapping and LBS can be found in the book Location articles within the publications in the secondary list
Based Services and TeleCartography, edited by Georg focus on cartography. In addition to reviewing recent
12:40:28.
TABLE 1 Journals and conference proceedings TABLE 2 Conferences having topics of interest
containing articles dealing with thematic cartography to cartographers
and geovisualization
AAG (Association of American Geographers)
Primary journals ACM SIGGRAPH (Association for Computing Machinists, Special
Interest Group on Graphics and Interactive Techniques)
Cartographic Perspectives ACSM/ASPRS Annual Convention (American Congress
Cartographica on Surveying and Mapping)
Cartography ESRI International User Conference (ESRI)
Cartography and Geographic Information Science GIScience (University Consortium for Geographic
The Cartographic Journal Information Science)
Secondary journals ICA (International Cartographic Association)
International Symposium on Spatial Data Handling
Annals, Association of American Geographers (International Geographical Union)
Byte NACIS (North American Cartographic Information Society)
Computer Graphics Visualization (IEEE)
Computers & Geosciences
Environment and Planning A
Environment and Planning B: Planning and Design for Geographic Information and Analysis (NCGIA;
Geographical Analysis http://www.ncgia.ucsb.edu/ ) has produced numerous
Geospatial Solutions
publications of interest to cartographers. More
GEOWorld
IEEE Computer Graphics & Applications recently, the University Consortium for Geographic
IEEE Transactions on Visualization and Computer Graphics Information Science (UCGIS; http://www.ucgis.org/),
Information Visualization a group of more than 70 universities, professional
International Journal of Geographical Information Science organizations, and affiliates, has developed a research
International Journal of Human-Computer Studies agenda (McMaster and Usery 2004), much of which
Journal of Geography
is relevant to cartographers. The International Carto-
Journal of the American Statistical Association
Landscape and Urban Planning graphic Association (ICA; http://www.icaci.org/)
Photogrammetric Engineering and Remote Sensing has 22 ongoing commissions and working groups of
Presence relevance to cartography; some of these that are
Progress in Human Geography relevant to this text include Generalization and
Statistical Computing & Statistical Graphics Newsletter
Multiple Representation, Map Projections, Maps
The American Statistician
The Professional Geographer
and the Internet, Mountain Cartography, Spatial
Transactions in GIS Data Uncertainty and Map Quality, and Visualization
Transactions, Institute of British Geographers and Virtual Environments. Finally, you might wish
Proceedings
to examine the Carto Project, an undertaking by
ACM SIGGRAPH (Association for Computing
ACM SIGGRAPH (Association for Computing Machinists, Special Machinery, Special Interest Group on Computer
Interest Group on Graphics and Interactive Techniques)
ICA (International Cartographic Association)
Graphics) to explore how “viewpoints and tech-
Innovations in GIS (Proceedings of the U.K. National niques from the computer graphics community can
Conference on GIS Research: GISRUK) be effectively applied to cartographic and spatial
International Symposium on Spatial Data Handling data sets” (http://www.siggraph.org/~rhyne/carto/).
(International Geographical Union)
Visualization (IEEE)
Note: For proceedings, the name of the sponsoring organization SUMMARY
is provided in parentheses.
In this chapter we have examined a number of ongoing

publications, you might also find it useful to attend con- developments in the evolving discipline of cartography.
ferences on topics of interest to cartographers (Table One development is Daniel Carr and his colleagues’
2). Such conferences frequently include expositions work with linked micromap plots (LM plots) and
illustrating the latest technological advances in hard- conditioned choropleth maps (CCmaps). LM plots split
ware and software. Being aware of such advances is a spatial distribution into a series of micromaps that
important because the field of cartography is arguably enable map readers to focus on local patterns as
technology-driven. opposed to the overall spatial pattern. In addition, LM
It is also important to be aware of research groups plots enable graphical and statistical information (e.g.,
and organizations that can provide useful informa- confidence intervals) to be linked with spatial informa-
tion on ongoing developments. The National Center tion. Whereas LM plots involve placing an equal
559
12:40:28.
number of observations on each micromap, CCmaps using your hands in the field, then speech is an obvious
utilize other attributes to split (or condition) a thematic choice); and they reduce the errors resulting from using
attribute of interest into individual maps. For instance, only one mode of interaction. Particularly exciting work
we saw how the attributes “percent below the poverty in multimodal interfaces is being done at the
level” and “precipitation” could be used to explain the GeoVISTA Center at Penn State University, where
distribution of lung cancer. Like many novel techniques researchers are developing collaborative geovisualiza-
today, LM plots and CCmaps are most effectively tion systems for emergency management situations.
utilized in a data exploration framework. Normally, we think of using thematic maps to depict
A second ongoing development is the use of senses the distribution of some geographic phenomenon. In
other than vision to interpret spatial patterns. the realm of information visualization, however,
Specifically, we focused on the uses of sound, touch (or researchers are stretching our notion of what consti-
haptics), and smell. Abstract sound variables and a tutes a map by creating maps of nonnumeric abstract
haptic variable syntax provide a set of potentially useful information. Spatialization is the process of converting
variables for sound and touch, respectively, that are anal- such abstract information to a spatial framework. Sara
ogous to the visual variables. Numerous researchers Fabrikant and Barbara Buttenfield, two geographers
have experimented with sound, including John Krygier, who have developed information visualization applica-
Peter Fisher, and Suresh Lodha. More recently, Haixia tions, argue that users can glean useful information
et al. have developed iSonic, which enables sounds to be from such visualizations because these visualizations
played in distinctive spatial sequences, and Glenn Braun are able to take advantage of the geographic metaphor
has experimented with realistic sounds. Research on (i.e., that nearby things are more similar to one
touch has not been as extensive as the research on another).
sound, although Reginald Golledge and his colleagues Large databases consisting of millions of records and
are experimenting with sound and touch in the Haptic tens or hundreds of attributes often challenge our abil-
Soundscapes project. Very little research has been done ities to visualize information. As a result, researchers
with smell in cartography, but certainly, there is potential are utilizing knowledge discovery in databases (KDD)
for using smell, and hardware and software have even or, more simply, spatial data mining techniques, to
been developed for generating smells. discover interesting patterns in these databases. As
Collaborative geovisualization is another development examples of spatial data mining, we considered Alan
that stretches our traditional view of maps. Normally, we Murray and Tung-Kai Shyy’s clustering approach for
think of maps as being utilized by individuals, but clearly integrating the distance between enumeration units in a
there are many instances in which collaboration is essen- choropleth mapping problem and Natalia Andrienko
tial (think of schoolchildren who wish to visualize the and her colleagues’ use of the C4.5 algorithm to deter-
migration of the Monarch butterfly, or climatologists who mine which attributes are associated with subregions of
wish to visualize the impacts of different models of global a city. Similar to spatial data mining, visual analytics
warming). We looked at several applications of collabora- enthusiasts argue that massive spatiotemporal data
tive geovisualization. In one, Nick Hedley and his col- sets can be understood only through a combination of
leagues experimented with a variety of high-tech capabil- visualization and analytical methods. Although initially
ities including augmented reality (AR), an immersive developed for homeland security purposes, approaches
virtual environment (VE), and a specialized projection developed by visual analytics experts can also be used
table known as HI-SPACE. In another application, Alan in a broad range of applications; for instance, Chris
MacEachren and his colleagues experimented with Weaver and his colleagues described the development
ImmersaDesks and the HI-SPACE table. of an interactive tool for exploring the visitation
In addition to developing geovisualization software patterns of guests at Pennsylvanian hotels in the late
that will allow users to collaborate, researchers are 1800s, something that would be of interest to historical
experimenting with interfaces that utilize speech, lip geographers.
movements, pen-based gestures, free-hand gestures, and One of the more recent research thrusts involves
head and body movements. Multimodal interfaces are mobile mapping and location-based services (LBS).
novel interfaces that involve two or more of these tech- Because this technology is rapidly evolving, it is difficult
niques. Multimodal interfaces are desirable because to anticipate exactly what its impact will be on the field
they are more natural (contrast simply saying, “Find all of cartography (and what impact cartographic research
sites within 30 meters of the river” with learning and might have on the technology), but given the likelihood
traversing a set of menus); they allow a wider spectrum of more than 3 billion cellular users by 2010, it is possi-
of computer users; they permit users to interact with a ble that the impacts could be comparable to those that
computer in whatever mode is convenient (if you are the Web has had on cartography.
12:40:28.
Because the discipline of cartography continues to cartographers publish, attend conferences of interest
evolve, any text such as this will be out of date as soon to cartographers, and be aware of research groups
as it is written. Thus, to keep up-to-date, you should and organizations that provide useful information on
read the major journals and proceedings in which ongoing developments.
FURTHER READING
Carr, D. B., White, D., and MacEachren, A. M. (2005) interfaces to large-screen displays.” Cartographic Perspectives,
“Conditioned choropleth maps and hypothesis generation.” no. 54:16–34.
Annals of the Association of American Geographers 95, Compares and contrasts the HI-SPACE and DAVE_G systems.
no. 1:32–53.
MacEachren, A. M., Wachowicz, M., Edsall, R., and Haug, D.
Provides an extensive discussion of how CCmaps can be used to (1999c) “Constructing knowledge from multivariate
formulate and test hypotheses concerning two attributes that spatiotemporal data: Integrating geographical visual-
might explain a third attribute.
ization with knowledge discovery in database methods.”
Cartography and Geographic Information Science 28, no. 1, International Journal of Geographical Information Science
2001 (entire issue). 13, no. 4:311–334.
This special issue focuses on research challenges in geovisualization. A classic paper on spatial data mining that links geovisualization
Communications of the ACM 45, no. 8, 2002 (entire issue). to knowledge discovery in databases.
This special issue focuses on data mining; pages 54–58 deal with Miller, H. J. (2008) “Geographic data mining and knowledge
scientific applications including geospatial data. discovery.” In The Handbook of Geographic Information
Gartner, G., Cartwright, W., and Peterson, M. P., eds. (2007) Science, ed. by J. P. Wilson and A. S. Fotheringham,
Location Based Services and TeleCartography. Berlin: Springer. pp. 352–366. Malden, MA: Blackwell.
Reviews recent work on spatial data mining.
Covers recent research related to LBS and TeleCartography; for
an overview of LBS in the realm of GIS, see Brimicombe (2008). Miller, H. J., and Han, J., eds. (2001) Geographic Data Mining
Golledge, R. G., Rice, M. T., and Jacobson, D. (2006) and Knowledge Discovery. London: Taylor & Francis.
“Multimodal interfaces for representing and accessing geo- A collection of early readings on spatial data mining and knowl-
spatial information.” In Frontiers of Geographic Information edge discovery in databases.
Technology, ed. by S. Rana and J. Sharma, pp. 181–208. Berlin: Oviatt, S. (2008) “Multimodal interfaces.” In The Human-
Springer-Verlag. Computer Interaction Handbook: Fundamentals, Evolving
Reviews the use and potential of multimodal interfaces for Technologies, and Emerging Applications, 2nd ed., ed. by
interpreting geospatial information. A. Sears and J. A. Jacko, pp. 413–432. New York: Lawrence
Griffin, A. L. (2001) “Feeling it out: The use of haptic visual- Erlbaum Associates.
ization for exploratory geographic analysis.” Cartographic Provides an overview of multimodal interfaces.
Perspectives, no. 39: 12–29. Skupin, A., and Fabrikant, S. I. (2008) “Spatialization.” In The
Provides a thoughtful discussion of the potential of touch for Handbook of Geographic Information Science, ed. by J. P.Wilson
interpreting spatial patterns. and A. S. Fotheringham, pp. 61–79. Malden, MA: Blackwell.
Lauriault, T. P., and Lindgaard, G. (2006) “Scented cybercartog- Provides an overview of the field of spatialization.
raphy: Exploring possibilities.” Cartographica 41, no. 1: 73–91. Thomas, J. J., and Cook, K., eds. (2005) Illuminating the Path:
Discusses the potential of using the sense of smell for cartographic The Research and Development Agenda for Visual Analytics.
applications. Los Alamitos, CA: IEEE Computer Society.
MacEachren, A. M., Cai, G., Brewer, I., and Chen, J. (2006b) Provides an introduction to visual analytics, as defined by the
“Supporting map-based geocollaboration through natural National Visualization and Analytics Center (NVAC).
GLOSSARY
abstract sound variables: analogous to visual variables, these collaborative geovisualization: geovisualization activities in
are variables used to depict abstract sounds (e.g., the location which several individuals are involved in the visualization
and loudness of sound). process; for instance, individuals located at different universi-
abstract sounds: sounds that have no obvious meaning and ties might manipulate and visualize a climate-change model
thus require a legend for interpretation (e.g., a mouse click on (ideally, the results of one individual’s manipulations can be
a census tract produces a loudness value as a function of seen by other individuals immediately).
income).
561
12:40:28.
conditioned choropleth maps (CCmaps): a spatial pattern mobile mapping: users are able to access maps and related
for a single attribute is split into a series of choropleth maps in geospatial information as they move through the Earth’s envi-
which the rows and columns of the maps correspond to two ronment.
attributes that might explain the attribute that is displayed in multimedia encyclopedias: like paper encyclopedias, these
the choropleth map series. provide information about places, people, and events, but they
geocollaboration: see collaborative geovisualization. are enhanced through the use of sound, video, and ease of
haptic variable syntax: refers to the variables involved in linking various topics.
sensing spatial information with touch (haptically); for exam- multimodal interfaces: interfaces that involve two or more
ple, the vibration variable refers to the vibration we might feel novel approaches for interacting with the computer; for
when moving a tactile mouse over a map. instance, you might point to a location without touching the
haptics: the use of touch in virtual environments (e.g., a user screen (a free-hand gesture) and say, “Show me all homes
feels different vibrations when a force-feedback mouse is within 100 meters of this location.”
placed over map locations that have differing crime rates). realistic sounds: sounds that have meaning based on our
information visualization: involves the visualization and past experience; contrast with abstract sounds.
analysis of nonnumerical abstract information such as the top- sonification: using sound to represent spatial data.
ics that are discussed on the front page of a newspaper over a spatial data mining: a simplified term used for spatial appli-
monthlong period. cations involving knowledge discovery in databases (KDD).
knowledge discovery in databases (KDD): refers to spatialization: the process of converting abstract informa-
exploratory approaches for finding interesting patterns in tion to a spatial framework in which visualization is possible
databases, and ultimately uncovering knowledge about those (e.g., converting the nature of topics that are discussed on the
databases. front page of a newspaper over a monthlong period to a spa-
linked micromap plots (LM plots): a series of small maps tial framework).
termed micromaps, which divide a single spatial distribution TeleCartography: the distribution of maps via wireless data
into pieces; graphical and statistical information is also transfer interfaces and mobile devices.
included, and the emphasis is on local pattern perception.
ubiquitous cartography: the ability of users to create and
location-based services (LBS): the delivery of data and use maps in any place and at any time to resolve geospatial
information services as a function of a user’s location or pro- problems.
jected location; for example, you might use a smartphone to
determine the location of the nearest Thai restaurant. visual analytics (VA): a term developed by the U.S. National
Visualization and Analytics Center (NVAC); refers to a
synthesis of visualization and analytical methods.
REFERENCES
Andrienko, N., Andrienko, G., Savinov, A., Voss, H., and American Cartographic Association. (1991) Matching the
Wettschereck, D. (2001b) “Exploratory analysis of spatial Map Projection to the Need. Bethesda, MD: American
data using interactive maps and data mining.” Cartography Congress on Surveying and Mapping.
and Geographic Information Science 28, no. 3:151–165. Andrienko, N., Andrienko, G., and Gatalsky, P. (2000)
Aerts, J. C. J. H., Clarke, K. C., and Keuper, A. D. (2003) “Supporting visual exploration of object movement.”
“Testing popular visualization techniques for representing Proceedings of the Working Conference on Advanced
model uncertainty.” Cartography and Geographic Visual Interfaces AVI, ACM Press, Palermo, Italy, pp.
Information Science 30, no. 3:249–261. 217–220, 315.
Aggett, G., and McColl, C. (2006) “Evaluating decision sup- Andrienko, N., Andrienko, G., and Gatalsky, P. (2001a)
port systems for PPGIS applications.” Cartography and “Exploring changes in census time series with interactive
Geographic Information Science 33, no. 1:77–92. dynamic maps and graphics.” Computational Statistics
Ambroziak, B. M., Ambroziak, J. R., and Bradbury, R. (1999) 16:417–433.
Infinite Perspectives: Two Thousand Years of Three- Andrienko, N., Andrienko, G., Savinov, A., Voss, H., and
Dimensional Mapmaking. New York: Princeton Wettschereck, D. (2001b) “Exploratory analysis of spatial
Architectural Press. data using interactive maps and data mining.” Cartography
American Cartographic Association. (1986) Which Map Is and Geographic Information Science 28, no. 3:151–165.
Best: Projections for World Maps. Bethesda, MD: Anselin, L. (1998) “Exploratory spatial data analysis in a
American Congress on Surveying and Mapping. geocomputational environment.” In Geocomputation:
American Cartographic Association. (1988) Choosing a A Primer, ed. by P. A. Longley, S. M. Brooks, R.
World Map: Attributes, Distortions, Classes, Aspects. McDonnell, and B. MacMillan, pp. 77–94. Chichester,
Bethesda, MD: American Congress on Surveying and England: Wiley.
Mapping.
12:40:28.
Bertolotto, M., Di Martino, S., Ferrucci, F., and Kechadi, T. Guo, D., Chen, J., MacEachren, A. M., and Liao, K. (2006)
(2007) “Towards a framework for mining and analyzing “A visual inquiry system for space-time and multivariate
spatio-temporal datasets.” International Journal of patterns (VIS-STAMP).” IEEE Transactions on
Geographical Information Science 21, no. 8:895–906. Visualization and Computer Graphics 12,
Bertolotto, M., Di Martino, S., Ferrucci, F., and Kechadi, T. no. 6:1461–1474.
(2007) “Towards a framework for mining and analyzing Harding, C., Kakadiaris, I. A., Casey, J. F., and Loftin, R. B.
spatio-temporal datasets.” International Journal of (2002) “A multi-sensory system for the investigation of
Geographical Information Science 21, no. 8:895–906. geoscientific data.” Computers & Graphics 26:259–269.
Brauen, G. (2006) “Designing interactive sounds maps: Using Hedley, N. R., Billinghurst, M., Postner, L., May, R., and
scalable vector graphics.” Cartographica 41, no. 1:59–71. Kato, H. (2002) “Explorations in the use of augmented
Carr, D. B., and Pierson, S. M. (1996) “Emphasizing statistical reality for geographic visualization.” Presence 11,
summaries and showing spatial context with micromaps.” no. 2:119–133.
Statistical Computing & Statistical Graphics Newsletter 7, Höllerer, T., Wither, J., and DiVerdi, S. (2007) “‘Anywhere
no. 3:16–23. Augmentation’: Towards mobile augmented reality in
Cai, G., Sharma, R., MacEachren, A. M., and Brewer, I. unprepared environments.” In Location Based Services
(2006) “Human-GIS interaction issues in crisis response.” and TeleCartography, ed. by G. Gartner, W. Cartwright,
International Journal of Risk Assessment and Management and M. P. Peterson, pp. 393–416. Berlin: Springer-Verlag.
6, no. 4/5/6:388–407. Jacobson, R. D., Kitchin, R., and Golledge, R. (2002) “Multi-
Cartwright, W. (2007) “Landmarks and the perception of a modal virtual reality for presenting geographic informa-
space in web-delivered 3-D worlds.” In Location Based tion.” In Virtual Reality in Geography, ed. by P. Fisher and
Services and TeleCartography, ed. by G. Gartner, W. D. Unwin, pp. 382–400. London: Taylor & Francis.
Cartwright, and M. P. Peterson, pp. 329–344. Berlin: Kettebekov, S., Krahnstoever, N., Leas, M., Polat, E., Raju,
Springer-Verlag. H., Schapira, E., and Sharma, R. (2000) “i2Map: Crisis
Carr, D. B., Olsen, A. R., Courbois, J.-Y. P., Pierson, S. M., and management using a multimodal interface.” Proceedings of
Carr, D. A. (1998) “Linked micromap plots: named and the ARL Federated Laboratory 4th Annual Symposium,
described.” Statistical Computing & Statistical Graphics College Park, MD, CD-ROM.
Newsletter 9, no. 1:24–32. Lauriault, T. P., and Lindgaard, G. (2006) “Scented cybercar-
Chen, J. X., Carr, D. B., Wechsler, H., and Pan, Z. (2006) tography: Exploring possibilities.” Cartographica 41,
“Interactive visualization of multivariate statistical data.” no. 1:73–91.
International Journal of Virtual Reality 5, Krygier, J. (1993) “Sound and cartographic design.”
no. 3:1–10. Videotape. University Park: Deasy GeoGraphics
Döllner, J., Hagedorn, B., and Schmidt, S. (2007) “An Laboratory, Department of Geography, Pennsylvania State
approach towards semantics-based navigation in 3D city University.
models on mobile devices.” In Location Based Services Krygier, J. B. (1994) “Sound and geographic visualization.” In
and TeleCartography, ed. by G. Gartner, W. Cartwright, and Visualization in Modern Cartography, ed. by A. M.
M. P. Peterson, pp. 357–368. Berlin: Springer-Verlag. MacEachren and D. R. F. Taylor, pp. 149–166. Oxford,
Fabrikant, S. I., and Buttenfield, B. P. (2001) “Formalizing England: Pergamon.
semantic spaces for information access.” Annals of the Lodha, S. K., Wilson, C. M., and Sheehan, R. E. (1996b)
Association of American Geographers 91, no. 2:263–280. “LISTEN: Sounding uncertainty visualization.” IEEE
Fisher, P. F. (1994a) “Hearing the reliability in classified Visualization ’96, San Francisco, CA, pp. 189–195.
remotely sensed images.” Cartography and Geographic Lodha, S. K., Joseph, A. J., and Renteria, J. C. (2000) “Audio-
Information Systems 21, no. 1:31–36. visual data mapping for GIS-based data: An experimental
Gartner, G., Cartwright, W., and Peterson, M. P. (eds.). (2007) evaluation.” In Workshop on New Paradigms in
Location Based Services and TeleCartography. Berlin: Information Visualization and Manipulation (NPIVM ’99),
Springer-Verlag. ed. by D. S. Ebert and C. D. Shaw, pp. 41–48. New York:
The Association for Computing Machinery.
Griffin, A. L. (2001) “Feeling it out: The use of haptic visual-
ization for exploratory geographic analysis.” Cartographic MacEachren, A. M., Cai, G., Brewer, I., and Chen, J. (2006b)
Perspectives, no. 39:12–29. “Supporting map-based geocollaboration through natural
interfaces to large-screen displays.” Cartographic
Golledge, R. G., Rice, M., and Jacobson, R. D. (2005) “A
Perspectives, no. 54: 16–34.
commentary on the use of touch for accessing on-screen
spatial representations: The process of experiencing haptic MacEachren, A. M., Edsall, R., Haug, D., Baxter, R., Otto, G.,
maps and graphics.” The Professional Geographer 57, Masters, R., Fuhrmann, S., and Qian, L. (2000) “Virtual
no. 3:339–349. environments for geographic visualization: Potential and
challenges.” In Workshop on New Paradigms in
Guo, D. (2007) “Visual analytics of spatial interaction pat-
Information Visualization and Manipulation (NPIVM ’99),
terns for pandemic decision support.” International Journal
ed. by D. S. Ebert and C. D. Shaw, pp. 35–40. New York:
of Geographical Information Science 21, no. 8:859–877.
The Association for Computing Machinery.
563
12:40:28.
May, R. A. (1999) “HI-SPACE: A next generation workspace Sharma, R., Yeasin, M., Krahnstoever, N., Rauschert, I., Cai,
environment.” Unpublished Masters thesis, Washington G., Brewer, I., MacEachren, A., and Sengupta, K. (2003)
State University, Pullman, WA. “Speech-gesture driven multimodal interfaces for crisis
McMaster, R. B., and Usery, E. L. (eds.). (2004) A Research management.” Proceedings of the IEEE 91, no. 9:1327–1354.
Agenda for Geographic Information Science. Boca Raton, Skupin, A. (2002) “A cartographic approach to visualizing
FL: CRC Press. conference abstracts.” IEEE Computer Graphics and
Miller, H. J. (2008) “Geographic data mining and knowledge Applications 22, no. 1:50–58.
discovery.” In The Handbook of Geographic Information Skupin, A. (2004) “The world of geography: Visualizing a
Science, ed. by J. P. Wilson and A. S. Fotheringham, pp. knowledge domain with cartographic means.” Proceedings
352–366. Malden, MA: Blackwell. of the National Academy of Sciences 101 (Suppl. 1):
Murray, A. T., and Shyy, T.-K. (2000) “Integrating attribute 5274–5278.
and space characteristics in choropleth display and spatial Skupin, A., and Fabrikant, S. I. (2003) “Spatialization meth-
data mining.” International Journal of Geographical ods: A cartographic research agenda for non-geographic
Information Science 14, no. 7:649–667. information visualization.” Cartography and Geographic
Oviatt, S. (1997) “Multimodal interactive maps: Designing Information Science 30, no. 2:99–119.
for human performance.” Human–Computer Interaction Skupin, A., and Fabrikant, S. I. (2008) “Spatialization.” In The
12:93–129. Handbook of Geographic Information Science, ed. by J. P.
Oviatt, S. (2002) “Breaking the robustness barrier: Recent Wilson and A. S. Fotheringham, pp. 61–79. Malden, MA:
progress on the design of robust multimodal systems.” Blackwell.
Advances in Computers 56:305–341. Slocum, T. A., Blok, C., Jiang, B., Koussoulakou, A., Montello,
Peterson, M. P. (2007) “The transition from Internet to D. R., Fuhrmann, S., and Hedley, N. R. (2001) “Cognitive
mobile mapping.” In Location Based Services and and usability issues in geovisualization.” Cartography and
TeleCartography, ed. by G. Gartner, W. Cartwright, and M. Geographic Information Science 28, no. 1:61–75.
P. Peterson, pp. 73–88. Berlin: Springer-Verlag. Thomas, J. J., and Cook, K. (eds.). (2005) Illuminating the
Press, D., and Minta, S. C. (2000) “The smell of nature: Path: The Research and Development Agenda for Visual
Olfaction, knowledge and the environment.” Ethics, Place Analytics. Los Alamitos, CA: IEEE Computer Society.
and Environment 3, no. 2:173–186. Turdukulov, U. D., Blok, C. A., Ruessink, B. G., and Turner,
Pulsifer, P. L., Caquard, S., and Taylor, D. R. F. (2007) I. L. (2007a) “Connecting users with their data: An envi-
“Toward a new generation of community atlases—The ronment to explore the morphodynamics of rip channels.”
Cybercartograhic Atlas of Antarctica.” In Multimedia Cartographica 42, no. 2:139–151.
Cartography, ed. by W. Cartwright, M. P. Peterson, and G. Turdukulov, U. D., Kraak, M.-J., and Blok, C. A. (2007b)
Gartner, pp. 195–216. Berlin: Springer-Verlag. “Designing a visual environment for exploration of time
Rauschert, I., Agrawal, P., Fuhrmann, S., Brewer, I., Wang, H., series of remote sensing data: In search for convective
Sharma, R., Cai, G., and MacEachren, A. M. (2002) clouds.” Computers & Graphics 31:370–379.
“Designing a human-centered, multimodal GIS interface Wang, X., Chen, J. X., Carr, D. B., Bell, B. S., and Pickle, L. W.
to support emergency management.” Proceedings of the (2002) “Geographic statistics visualization: Web-based
10th ACM International Symposium on Advances in linked micromap plots.” Computing in Science &
Geographic Information Systems, McLean, VA, pp. Engineering 4, no. 3:90–94.
119–124. http://doi.acm.org/10.1145/585147.585172. Weaver, C. (2004) “Building highly-coordinated visualiza-
Rice, M., Jacobson, R. D., Gollege, R. G., and Jones, D. (2005) tions in Improvise.” Proceedings of the IEEE Symposium
“Design considerations for haptic and auditory map on Information Visualization (INFOVIS’04), Austin, TX,
interfaces.” Cartography and Geographic Information pp. 159–166.
Science 32, no. 4:381–391. Weaver, C., Fyfe, D., Robinson, A., Holdsworth, D., Peuquet,
Rowles, R. A. (1991) “Regions and regional patterns on D., et al. (2007) “Visual exploration and analysis of historic
choropleth maps.” Unpublished PhD dissertation, hotel visits.” Information Visualization 6:89–103.
University of Kentucky, Lexington, KY. Zhao, H. (2006) “Interactive sonification of abstract data—
Schmalstieg, D., and Reitmayr, G. (2007) “The world as a Framework, design space, evaluation, and user tool.”
user interface: Augmented reality for ubiquitous comput- Unpublished PhD dissertation, University of Maryland,
ing.” In Location Based Services and TeleCartography, ed. College Park, MD.
by G. Gartner, W. Cartwright, and M. P. Peterson, pp. Zhao, H., Smith, B. K., Norman, K., Plaisant, C., and
369–391. Berlin: Springer-Verlag. Shneiderman, B. (2005) “Interactive sonification
of choropleth maps.” IEEE Multimedia 12, no. 2: 26–35.
12:40:28.
A B
Vote for Perot, 1992 Vote for Perot, 1992
Percentage of Votes Percentage of Votes
COLOR PLATE 1 The choropleth map: An example of a thematic map. Map A uses illogically ordered hues, whereas map B uses
logically ordered shades of a single hue. Although map A might allow the reader to discriminate easily between individual states, it
does not permit the reader to perceive the overall spatial pattern as readily as map B. (Data source: Famighetti 1993, p. 583.)
COLOR PLATE 2 A combined proportional symbol–dot map that attempts to represent what the population might look like
in the “real world.” (After U.S. Bureau of the Census 1970.)
COLOR PLATE 3 A change map created using
MapTime. Red and blue circles represent population
decreases and increases, respectively, between 1950 and
1990. Note the distinctive region of population losses in
the Northeast; this pattern was not revealed in an anima-
tion of the data over this time period. (Courtesy of
Stephen C. Yoder.)
565
12:40:28.
COLOR PLATE 4 A frame from an animated fly-
through of downtown Lawrence, KS. (Courtesy of
Jerome E. Dobson and his Geog 980 seminar students.)
COLOR PLATE 5 An example of mapping Internet

space—a visualization of the backbone of NSFNET, an
early version of the Internet. Purple and white vertical
lines represent low and high Internet traffic, respectively.
(Courtesy of the National Center for Supercomputing
Applications (NCSA) and the Board of Trustees of the
University of Illinois.)
COLOR PLATE 6 An example of scientific visualiza-

tion involving the analysis of a dispersal of an airborne
contaminant in urban street canyons. The caption for the
image indicated that “The results shown here used approx-
imately 10–50 million finite volume computational cells,
and required hundreds of hours of CPU time to calculate.”
(From the U.S. Environmental Protection Agency’s
Environmental Modeling and Visualization Laboratory,
http://www.epa.gov/nesc/.)
12:40:28.
A B
Greenness Values Vegetation Difference Values

Difference in Biomass Amounts Decreased Little or no Increased
Low Biomass High Biomass Water Greenness Change Greenness
C D
Vegetation Difference Values Vegetation Difference Values

Difference in Biomass Amounts Decreased Little or no Increased Difference in Biomass Amounts Decreased Little or no Increased
Greenness Change Greenness Greenness Change Greenness
COLOR PLATE 7 Potential images to be used with the GreenReport: (A) a basic greenness map used to represent current
vegetation conditions; (B) a change map in which the “Little or no Change” category might be confused with some of the cate-
gories on the basic greenness map; (C) a change map in which a gray shade is used for the “Little or no Change” category to avoid
confusion; (D) a change map in which gray is used for the “Little or no Change” category, and colors not used on the greenness
map are used to depict changes in greenness. (Courtesy of Kansas Applied Remote Sensing Program, University of Kansas.)
Display: MPA.faces COLOR PLATE 1 Three-

7 LF overburden dimensional model of an open-pit
6 LF coal
5 Shale 1 Z exaggeration: 7.81 coal mining site. (Data Source:
4 Limestone Azimuth: 295.7
3 Shale Inclination: 13.4
Pennsylvania Department of Environ-
2 UK coal ment Protection; courtesy of Dynamic
1 Underclay
Graphics, Inc., Alameda, CA.)
567
12:40:28.
Color Visual Variables
Hue
Lightness
Saturation
COLOR PLATE 2 Color visual variables for quantitative phenomena. For visual vari-
ables for black-and-white maps, see Figure 3.

None
Hue Recommended
COLOR PLATE 3 Hue, a color visual variable for qualitative phenomena. For visual vari-

(saturation) (hue)
COLOR PLATE 4 Representing the percentage of wheat harvested in Kansas counties using different visual variables: (A) satura-
tion, (B) hue. For black-and-white visual variables, see Figure 11.
12:40:28.
A B
Vote for Perot, 1992 Vote for Perot, 1992
Percentage of Votes Percentage of Votes
COLOR PLATE 1 The choropleth map: An example of a thematic map. Map A uses illogically ordered hues, whereas map B uses
logically ordered shades of a single hue. Although map A might allow the reader to discriminate easily between individual states, it
does not permit the reader to perceive the overall spatial pattern as readily as map B. (Data source: Famighetti 1993, p. 583.)
COLOR PLATE 2 A combined proportional symbol–dot map that attempts to represent what the population might look like
in the “real world.” (After U.S. Bureau of the Census 1970.)
COLOR PLATE 3 A change map created using
MapTime. Red and blue circles represent population
decreases and increases, respectively, between 1950 and
1990. Note the distinctive region of population losses in
the Northeast; this pattern was not revealed in an anima-
tion of the data over this time period. (Courtesy of
Stephen C. Yoder.)
565
16:03:18.
COLOR PLATE 4 A frame from an animated fly-
through of downtown Lawrence, KS. (Courtesy of
Jerome E. Dobson and his Geog 980 seminar students.)
COLOR PLATE 5 An example of mapping Internet

space—a visualization of the backbone of NSFNET, an
early version of the Internet. Purple and white vertical
lines represent low and high Internet traffic, respectively.
(Courtesy of the National Center for Supercomputing
Applications (NCSA) and the Board of Trustees of the
University of Illinois.)
COLOR PLATE 6 An example of scientific visualiza-

tion involving the analysis of a dispersal of an airborne
contaminant in urban street canyons. The caption for the
image indicated that “The results shown here used approx-
imately 10–50 million finite volume computational cells,
and required hundreds of hours of CPU time to calculate.”
(From the U.S. Environmental Protection Agency’s
Environmental Modeling and Visualization Laboratory,
http://www.epa.gov/nesc/.)
16:03:18.
A B
Greenness Values Vegetation Difference Values

Difference in Biomass Amounts Decreased Little or no Increased
Low Biomass High Biomass Water Greenness Change Greenness
C D
Vegetation Difference Values Vegetation Difference Values

Difference in Biomass Amounts Decreased Little or no Increased Difference in Biomass Amounts Decreased Little or no Increased
Greenness Change Greenness Greenness Change Greenness
COLOR PLATE 7 Potential images to be used with the GreenReport: (A) a basic greenness map used to represent current
vegetation conditions; (B) a change map in which the “Little or no Change” category might be confused with some of the cate-
gories on the basic greenness map; (C) a change map in which a gray shade is used for the “Little or no Change” category to avoid
confusion; (D) a change map in which gray is used for the “Little or no Change” category, and colors not used on the greenness
map are used to depict changes in greenness. (Courtesy of Kansas Applied Remote Sensing Program, University of Kansas.)
Display: MPA.faces COLOR PLATE 1 Three-

7 LF overburden dimensional model of an open-pit
6 LF coal
5 Shale 1 Z exaggeration: 7.81 coal mining site. (Data Source:
4 Limestone Azimuth: 295.7
3 Shale Inclination: 13.4
Pennsylvania Department of Environ-
2 UK coal ment Protection; courtesy of Dynamic
1 Underclay
Graphics, Inc., Alameda, CA.)
567
16:03:18.
Hue
Lightness
Saturation
COLOR PLATE 2 Color visual variables for quantitative phenomena. For visual vari-

None
Hue Recommended
COLOR PLATE 3 Hue, a color visual variable for qualitative phenomena. For visual vari-

(saturation) (hue)
COLOR PLATE 4 Representing the percentage of wheat harvested in Kansas counties using different visual variables: (A) satura-
tion, (B) hue. For black-and-white visual variables, see Figure 11.
16:03:18.
Geoid 2003
50
Latitude (degrees North)
40
30
130 120 110 100 90 80 70 60

Longitude (degrees West)
COLOR PLATE 1 GEOID03 is a model of the geoid in the United States for the conterminous United States. GEOID03
heights range from a low of -50.01 meters (164.07 feet), represented by magenta in the Atlantic Ocean, to a high of 3.43 meters
(11.25 feet), shown as red in the Rocky Mountains and the Labrador Strait. (From the National Geodetic Survey; http://www.ngs.
noaa.gov/GEOID/GEOID03/.)
A B
COLOR PLATE 1 The orthographic projection showing (A) North America and its proximity to Europe and (B) a view of
South America and Antarctica. (From pp. 52–53 of Harrison, R. E. (1944) Look at the World: The Fortune Atlas for World Strategy.
New York, NY: Alfred A. Knopf.)
569
16:03:18.
Additive Colors Subtractive Colors
COLOR PLATE 1 Principles of additive and subtractive color. For additive color, overlapping red, green, and blue lights reveal
how cyan, magenta, yellow, and white can be created. For subtractive color, the reverse is the case: cyan, magenta, and yellow combine
to produce red, green, blue, and black. To obtain a true black with subtractive colors, it is often necessary to add a black layer.
COLOR PLATE 2 The Munsell color model. (Courtesy of

GretagMacbeth.)
16:03:18.
100
80
1.0 y
60
Y%
40
0.4
20
0.2
0.2
0.4
0.6
0.8
1.0 x
COLOR PLATE 3 A three-dimensional view of the Yxy CIE system. (Courtesy of A. Jon Kimerling.)
CURRENCY
Bolivar
Boliviano
Dollar
Franc
Guilder
Guarani
Nuevo Sol
Peso
Real
Sucre
Source: U.S. Department of Commerce.

National Trade Data Bank, May 6, 1999.
COLOR PLATE 1 Unique hues are preferred to line and

dot patterns when symbolizing areal features with qualitative
data. (Compare with Figure 12.)
571
16:22:37.
Ethnolinguistic Families, Iran
C A S P I A N
S E A
Indo-European
Altaic
Tehran
Hamito-Semitic
Indo-European and
Hamito-Semitic
Provincial Border
Areas shown in white are sparsely

populated or uninhabited.
Source: United States Central Intelligence

Agency. Map File #505103 (547149) 2-82.
P
E
R
S
IA
N
G
U
L
F
0 100 200 300 400
Miles
COLOR PLATE 1 Map resulting from completion of the design process, and based on the sketch
map illustrated in Figure 4.
A Grounds and Buildings

Department of Public Works
Carport 5
Building A
Carport 2
Building B
Carport 4
Building C Source:
Department of
Public Works,
Carport 1 Saint Jerome,
NJ.
Carport 3
Owned by DPW
Leased from AC&T
Permanent Structure
Portable Structure
0 200
Feet
B Grounds and Buildings

Department of Public Works
Carport 5
Building A
Carport 2
Building B
Carport 4
Building C
Carport 1
Carport 3
0 200 Owned by DPW

Feet Leased from AC&T
Source: Department of Public Works, Permanent Structure COLOR PLATE 2 (A) A poorly balanced design in which
Saint Jerome, NJ.
Portable Structure map elements compete for space. (B) A well-balanced design in
which map elements exist in relative harmony and equilibrium.
16:22:37.
Suitability of Residential Parcels
Land Park and Curtis Park Neighborhoods, 2007
50
Bro
adw
ay
2nd
Ave
.
99
d.
24th St.
Blv
ide
r.
ers
D
rk
Riv
Pa
d
Lan
Franklin Blvd.
5 C.K.
M
HIG cCLATC
H SC
CUR
H
HOO Y
Blvd
L
SA
lvlvdd. .
TIS
CR
ortrt BB
RIV
AM
reeeppo
PAR
ER
EN
Fre N
T
CURTIS
K V
O
LAN
D
PAR
K
IL
City of Sacramento
LAG
E (pro
5
PARK
80
Ranking
pose
Least Suitable
)d
Area
Are a Mildly Suitable Locational Criteria:
of 1. Zoned for single-family residence
Detail
50 Moderately Suitable
2. Not within 500 feet of freeway or ramp
Highly Suitable 3. Not within 250 feet of active railroad
Most Suitable 0 0.25 0.5 Mile
5 99 Ranking Criteria (equally weighted):
Unranked Parcels Shown in White 1. Lower crime density
2. Higher proportion of college graduates 2006 Crime Report Database, City of Sacramento Police Department;
Crime Hot Spot 3. Higher elevation (reduced flood risk) County of Sacramento GIS.
COLOR PLATE 3 The final real estate site suitability map, resulting from the case study in section 2.
Cyan: 105º
Black: 45º
COLOR PLATE 1 Different halftone screen angles allow tints of black and
cyan to mix on the page.
573
16:22:37.
COLOR PLATE 2 Rosette pattern produced from
halftone screens of the process colors, each at a differ-
ent screen angle.
Cyan: 105º
Magenta: 75º
Yellow: 90º
Black: 45º
COLOR PLATE 3 An overlay

proof of a map produced by the
California State Automobile Associa-
tion, San Francisco, California. © 2002
California State Automobile Associa-
tion. Used by permission.
A B
(lightness) (lightness and saturation)
COLOR PLATE 1 A comparison of two sequential color schemes: (A) only lightness varies; (B) both lightness and saturation
vary, with saturation increasing from the first to the third class and then decreasing for the latter two classes. (Colors based on
CMYK values shown in Brewer 1989.)
16:22:37.
A B
(hues span portion of color circle) (hues span entire color circle)
COLOR PLATE 2 A comparison of two sequential color schemes in which differing hues are used: (A) the hues span a portion
of the color circle; (B) the hues span the entire color circle. (CMYK values for maps provided by Cynthia Brewer.)
Adults Not Graduating from High School, 1990
14.9 to 19.2 19.9 to 22.5 22.9 to 26.1 27.9 to 30.0 31.7 to 35.7
COLOR PLATE 3 An example of a diverging scheme in

which two sequential schemes converge on a neutral gray.
575
16:22:37.
Color Schemes Used by Brewer and Colleagues
A Purple/Green Diverging B Brown/Blue-Green Diverging
C Purple/Blue Sequential D Spectral
COLOR PLATE 4 Some of the effective color schemes tested in Brewer et al.’s (1997) work: (A) a purple/green diverging scheme,
(B) a brown/blue–green diverging scheme, (C) a purple–blue sequential scheme, and (D) a spectral scheme. (CMYK values for maps
provided by Cynthia Brewer.)
COLOR PLATE 5 ColorBrewer, a program

that enables novices to easily select color
schemes for choropleth maps. (Courtesy of
Cynthia Brewer.)
16:22:37.
Color Schemes Used by Mersey
A Unordered Hues B Two-Hue Progression
C Spectral Hues D Hue-Value Series
E PMS-Value Series F Black-and-White Value Series
COLOR PLATE 6 Color schemes used in a study by Mersey (1990): (A) unordered hues, (B) a two-hue progression, (C) spectral
hues, (D) a hue-value series, (E) a PMS-value series, and (F) a black-and-white-value series. Colors shown are based on Mersey’s
specification of CMY and PMS colors. The PMS value series involved overprinting an orange-brown color with a black ink to
increase the value range of the orange-brown color. Mersey found that the hue-value series worked best overall. (After Mersey 1990.)
577
16:22:37.
COLOR PLATE 7 Using data
exploration software to visualize the
data in different ways, in this case, as
four methods of classification (from
the program ExploreMap).
A B C
Population Density
(persons per km 2)
Percent Error Count Error
50 to 37,233 20,000 to 38,852
0
25 to 50 10,000 to 20,000
0 to 25
10 to 25 5,000 to 10,000
25 to 75
-10 to 10 -5,000 to 5,000
75 to 175
-25 to -10 -10,000 to -5,000
175 to 375 -50 to -25 -20,000 to -10,000
375 to 13,551 -100 to -50 -395,877 to -20,000
COLOR PLATE 1 Example of dasymetric and associated error maps resulting from Eicher and Brewer’s work: (A) dasymetric
map based on the binary method; (B) percent error map; (C) count error map. (After Eicher and Brewer 2001. First published in
Cartography and Geographic Information Science 28(2), p. 134. Reprinted with permission from the American Congress on Surveying
and Mapping. Courtesy of Cory Eicher and Cynthia Brewer.)
16:22:37.
COLOR PLATE 10 Using geo-icons to depict
A B multiple attributes associated with the ability to
traverse off-road terrain. Note how the view
varies as the scale changes from small (A) to
large (C). (D) depicts the five attributes being
mapped: (1) road proximity, (2) stream proximity,
(3) terrain roughness, (4) slope steepness, and (5)
vegetation resistance. The color scheme extends
from blue to red. (After Zhang and Pazner 2004.
First published in Cartography and Geographic
Information Science 31(1), p. 37. Reprinted with
permission from the American Congress on
C
D
1 2
5
4 3
COLOR PLATE 11 Results of clustering the

New York State data shown in Table 3 using the
Cluster Analysis Results UPGMA method. Numbers within the legend
UPGMA Method represent cluster numbers specified by the SPSS
software used to generate the clusters.
Moderately Low % Unemployment

1.
High % African-American
2. Near State Averages
3. High % Unemployed
4. High Infant Mortality
5. Moderately Increasing Population

Rapidly Increasing Population
6. High % Unemployment
Moderately High Infant Mortality
High % Unemployment
7. Moderately Low % African-American
Low Infant Mortality
16:26:13.
A Area Population Density, 1991 B Population Concentration, 1991
Average for Ward Areas Average for Ward Populations
Residents per Residents per

Square Kilometer Square Kilometer
up to 1 3 up to 1
1 to 2 4 1 to 2
2 to 5 7 2 to 5
5 to 10 10 5 to 10
10 to 25 17 10 to 25 1
25 to 50 19 25 to 50 3
50 to 100 16 50 to 100 5
100 to 300 12 100 to 300 8
300 to 1,000 6 300 to 1,000 14
1,000 to 2,000 3 1,000 to 2,000 17
2,000 to 3,000 1 2,000 to 3,000 14
3,000 to 4,000 1 3,000 to 4,000 13
4,000 to 6,000 1 4,000 to 6,000 15
6,000 to 10,000 6,000 to 10,000 7
10,000 & above 10,000 & above 2
Percent of Percent of
Land Area All Residents
Scale Scale
= 1,000km2 = 250,000 people
COLOR PLATE 1 Maps of population density in Britain using (A) a traditional equivalent projection and (B) a cartogram. The
equivalent projection suggests that most of Britain is dominated by relatively lower population densities (the blues and greens),
whereas the cartogram provides a detailed picture of the variation in population densities within urban areas. (From Dorling
1995a, p. xxxiii. Courtesy of Daniel Dorling.)
COLOR PLATE 2 Minard’s (1865) classic flow map illustrating the export of French wine by sea in 1864. This is an example of
a distributive flow map in which there is an attempt to depict the actual route of flow. (From Robinson, Arthur H. (1982) Early
Thematic Mapping in the History of Cartography. University of Chicago Press; original from l’Ecole Nationale des Ponts et
Chaussées, Paris.)
585
16:26:13.
COLOR PLATE 3 A radial flow map that depicts the quantity of migrants from various regions of the United States to each
state; choropleth shading depicts the total number of migrants to each state. (Source: U.S. Geological Survey (1970).)
COLOR PLATE 4 A map depicting wind speed and direction over Australia—an example of a continuous flow map. Note that
wind speed is redundantly represented by larger, more opaque arrows. (From Turk, G., and Banks, D. (1996) “Image-guided
streamline placement.” SIGGRAPH 96 Conference Proceedings, New Orleans, LA, pp. 453–460. © 1996 ACM, Inc. Reprinted by
permission.)
16:26:13.
COLOR PLATE 5 A visualization of flows associ-
ated with Internet traffic—an example of a telecommu-
nications flow map. (From http://www.caida.org/tools/
visualization/walrus/gallery1/, Copyright © 2003 The
Regents of the University of California. All rights
reserved. Used by permission.)
A
COLOR PLATE 6 Screens from Yadav’s MigMap, a
program for depicting migration flows: (A) a frame from
an animation; (B) a small multiple based on choropleth
symbols. Note the use of redundancy in (A) (migration is
a function of both the width and the color of the symbol).
(Courtesy of Sunita Yadav-Pauletti.)
587
16:26:13.
COLOR PLATE 1 Tanaka’s
approach is modified by
adding colors between the con-
tours to give the map a more
realistic appearance. (From
Kennelly and Kimerling 2001.
First published in Cartography
and Geographic Information
Science 28(2), p. 119. Reprinted
with permission from the
American Congress on
COLOR PLATE 2 A comparison of three methods of depict-

A ing topography: (A) Raisz’s physiographic method; (B)
Harrison’s manual shaded relief method; (C) a USGS digital
shaded relief portrayal developed by Thelin and Pike. (Map A
copyright by Erwin Raisz, 1967. Reprinted with permission by
RAISZ LANDFORM MAPS, 800-277-0047.)
COLOR PLATE 3 An example of a shaded relief map that

Imhof created by hand. (From Imhof, E., Cartographic Relief
Presentation, Berlin, New York: Walter de Gruyter 1982.)
16:26:13.
COLOR PLATE 4 A portrayal of the Earth’s topography
developed by Treinish (1994) using the IBM Visualization COLOR PLATE 5 Moellering and Kimerling’s MKS-
Data Explorer (DX). The technique uses Gouraud shading to ASPECT™ approach for symbolizing aspect based on the
represent aspect and slope, and hypsometric tints are used to opponent-process theory of color. The image was created using
represent elevation above or below sea level. (Courtesy of 128 colors and was scanned from a slide provided by Moellering.
Lloyd Treinish, IBM T. J. Watson Research Center.) (Courtesy of Harold Moellering and A. Jon Kimerling.)
A B
C D
COLOR PLATE 6 Brewer and Marlow’s

method for portraying aspect and slope. The
image is for the San Francisco North 7 and one-
half minute USGS quad. Caution should be taken
in interpreting these colors because the RGB val-
ues recommended by Brewer and Marlow have
been converted to CMYK values for printing.
COLOR PLATE 7 Some of the visualizations available in LandSerf.

Shaded relief maps are overlaid with: (A) a green–brown–purple–white
color scheme depicting elevation ranges; (B) a white–yellow–red color
scheme depicting slope; (C) features found in the region, with channels
shown in blue, ridges in yellow, and planar regions in gray; and (D) a gen-
eralization of (A) using a 65 × 65 cell window. (Courtesy of Jo Wood.)
589
16:26:13.
COLOR PLATE 8 A classic
A
bird’s-eye view of Anniston, AL, in
1888: (A) the full image; (B) an
enlargement of a portion of the
image. (Source: Glover, E. S. Bird’s
eye view of Anniston, Ala. 1888.
Map. Shober & Carqueville Litho.
Co., Chicago, c1888. Library of
Congress, Geography and Map
Division. Map Collections: Cities
and Towns, Panoramic Map
Collection 1847–1929. http://memory.
loc.gov/ammem/pmhtml/panhome.
html (accessed July 20, 2007).)
COLOR PLATE 9 Berann’s

classic hand-drawn panorama of
Denali National Park, Alaska.
(Courtesy of National Park
Service; artist, Heinrich Berann.)
16:26:13.
COLOR PLATE 10 A map of Crater Lake National Park that utilizes Patterson’s rules
for creating more realistic views of vertically viewed topographic maps. (Courtesy of
National Park Service.)
COLOR PLATE 11 Examples of physical models created using technology developed by Solid Terrain Modeling: Mt. Everest
is on the left, and Hurricane Floyd is on the right. (Courtesy of Solid Terrain Modeling.)
COLOR PLATE 1 Solutions for disorientation in fly-

bys: spotlight path (the light blue-shaded corridor);
landmarks (the flags with the landmark labels); labels
(e.g., the cities of Oakland and San Francisco); floating
compass (the tic marks on the horizon); and heads-up
display text (the text in the lower right). (Courtesy of
Mark Harrower.)
591
16:26:13.
A
COLOR PLATE 2 Treinish’s portrayal of the ozone hole over Antarctica: (A) the ozone symbolization is shown as an ortho-
graphic projection (this symbolization was used in an animation of the data); (B) small multiples depicting monthly averages of
ozone during the spring warming for the Southern and Northern Hemispheres. (Courtesy of Lloyd Treinish, IBM T. J. Watson
Research Center.)
1850 1900 1950 1990
COLOR PLATE 3 Four frames from the USGS’s animation of urban growth in the San Francisco–Sacramento region described in
Buchanan and Acevedo (1996). Note the relatively smooth transitions between the frames. (Courtesy of William Acevedo.)
16:26:13.
COLOR PLATE 4 An illustration of Harrower’s
(2002) use of visual benchmarks (the yellow and red
circles). (Courtesy of Mark Harrower.)
COLOR PLATE 5 Andrienko et al.’s (2001) interactive

software that is intended to improve on “fixed” animations:
(A) a frame from an animation of the raw data; (B) a frame
from a “comparison” option, in which each year is com-
pared against 1992. (Courtesy of Knowledge Discovery
Department Team, Fraunhofer Institute IIS, http://www.iais.
fraunhofer.de.)
593
16:26:13.
A Cleveland OH pop: 1940.0: 878336.0
B
Salem OR pop: 1880.0: 2538.0
COLOR PLATE 7 Using Brownrigg’s notion of a spacetime

COLOR PLATE 6 Maps taken from animations in Deep map to discover some facts that are unlikely to be discovered in
Thunder, a research project for developing sophisticated an animation: (A) that Cleveland, OH, experienced two
weather prediction models and associated visualizations. declines in population, a brief one in the 1930s and then a
(Courtesy of Lloyd Treinish, IBM T. J. Watson Research longer one beginning in 1950; and (B) that the population of
Center.) Salem, OR, reduced to 0 shortly after it was founded, and then
began to increase again. (Courtesy of Richard Brownrigg.)
A B
COLOR PLATE 1 The Subset option for ExploreMap: (A) A dispersion graph appears, illustrating the distribution of the data.
Note that the current class breaks, mean, median, and standard deviation can be toggled on; in this case, only the median is shown.
(B) The user selects all values less than the median, and these are displayed in blue.
16:26:13.
COLOR PLATE 2 A frame of an animation from Aspens. COLOR PLATE 3 A frame of an animation obtained
The size of a leaf represents total cumulative tree growth when a leaf in Color Plate 2 is clicked on. The size of trees
for a study site, whereas leaf color represents incremental represents total cumulative growth, whereas tree color rep-
growth for each year. Clicking on a leaf reveals an anima- resents incremental growth. Clicking on an icon in the upper
tion for clones associated with that study site (Color Plate right can lead to animations of clones for other study sites.
3). (Courtesy of Barbara P. Buttenfield and Christopher R. (Courtesy of Barbara P. Buttenfield and Christopher R.
Weber.) Weber.)
COLOR PLATE 4 A screen from the “enumerated” software within Project Argus. Three variables are shown at the top of the
screen: percent black, percent of population 25 years and older with any college education, and percent below the poverty level.
In the top right is a bivariate choropleth map of the first two variables; high values on these two variables are represented by
increasing amounts of yellow and blue, respectively. In the central portion are point graphs and a scatterplot of the first two vari-
ables. The lower and lower right views show a parallel coordinate plot and Dorling’s cartogram method, respectively. (Jason Dykes,
Department of Geography, University of Leicester.)
595
16:26:13.
COLOR PLATE 2 A frame of an animation from Aspens. COLOR PLATE 3 A frame of an animation obtained
The size of a leaf represents total cumulative tree growth when a leaf in Color Plate 2 is clicked on. The size of trees
for a study site, whereas leaf color represents incremental represents total cumulative growth, whereas tree color rep-
growth for each year. Clicking on a leaf reveals an anima- resents incremental growth. Clicking on an icon in the upper
tion for clones associated with that study site (Color Plate right can lead to animations of clones for other study sites.
3). (Courtesy of Barbara P. Buttenfield and Christopher R. (Courtesy of Barbara P. Buttenfield and Christopher R.
Weber.) Weber.)
COLOR PLATE 4 A screen from the “enumerated” software within Project Argus. Three variables are shown at the top of the
screen: percent black, percent of population 25 years and older with any college education, and percent below the poverty level.
In the top right is a bivariate choropleth map of the first two variables; high values on these two variables are represented by
increasing amounts of yellow and blue, respectively. In the central portion are point graphs and a scatterplot of the first two vari-
ables. The lower and lower right views show a parallel coordinate plot and Dorling’s cartogram method, respectively. (Jason Dykes,
Department of Geography, University of Leicester.)
595
16:29:28.
COLOR PLATE 5 A map created in CommonGIS using the
“Qualitative colouring” option. (Courtesy of Knowledge Discovery
Department Team, Fraunhofer Institute IAIS, http://www.iais.
fraunhofer.de.)
A B
Iceland
Moldova
Macedonia
Albania
COLOR PLATE 6 Data exploration in CommonGIS involving the removal of an outlier: (A) an unclassed choropleth map of
all data for birth rates in Europe; (B) the Albanian outlier is removed. (Courtesy of Knowledge Discovery Department Team,
Fraunhofer Institute IAIS, http://www.iais.fraunhofer.de.)
A B
Porto
Lisbon
COLOR PLATE 7 Data exploration in CommonGIS using a dynamic, diverging color scheme: (A) an unclassed choropleth
map of the percent of the population greater than or equal to 65 in divisions of Portugal; (B) a diverging color scheme is applied
to a portion of the data range, illustrating that the lowest values occur around the biggest cities of Portugal such as Lisbon and
Porto. (Courtesy of Knowledge Discovery Department Team, Fraunhofer Institute IAIS, http://www.iais.fraunhofer.de.)
16:29:28.
North
Dakota
Nevada
COLOR PLATE 8 Value flow diagrams utilized in CommonGIS. The diagrams depict
the burglary rates in the United States from 1960 to 2000. Values above the median are
shown in red, and values below the median are shown in blue. Note that value flow dia-
grams having a similar structure in Figure 8 (i.e., those enclosed by the solid line) are red
in this color plate. (Courtesy of Knowledge Discovery Department Team, Fraunhofer
Institute IAIS, http://www.iais.fraunhofer.de.)
COLOR PLATE 9 Using GeoDa to display a percentile

map and a boxmap of the percent foreign born in Florida
counties; note that Dade County (in extreme southern COLOR PLATE 10 Using GeoDa to analyze the relation-
Florida) is highlighted with cross-hatching on the maps and is ship between percent foreign born and percent Catholic for
a distinct outlier on the histogram (indicated by the yellow counties in Florida; scatterplots of the raw and transformed
color). (Courtesy of Luc Anselin.) data are shown in the lower left and right, respectively, and
the histogram shows the distribution of percent Catholic (see
Color Plate 9 for a histogram of percent foreign born).
(Courtesy of Luc Anselin.)
597
16:29:28.
COLOR PLATE 11 The interface for
Health VisPCP, using heart disease for
white men as an example. Major ele-
ments include a bivariate choropleth
map (upper left), a scatterplot of the
variables depicted on the map (middle
right), and a parallel coordinate plot
(lower left). (Courtesy of Robert
Edsall.)
COLOR PLATE 12 Using ESTAT to

analyze the relationship between per-
cent of votes for George Bush and per-
cent Catholic. Note the correlation of
-0.209 in the scatterplot and that there
are few dark gray areas on the bivariate
map, indicating that few areas are high
in both percent Catholic and percent
who voted for George Bush. In the par-
allel coordinate plot, note that Bush
votes are positively correlated with
Baptist adherents (r = 0.20). (Courtesy
of Anthony Robinson and Jin Chen.)
COLOR PLATE 13 Using ESTAT to

analyze the relationship between percent
of Kerry votes and percent Baptist. The
bivariate map shows those counties that
are highest in both variables (the top
quantile in each case); note that these
counties are all in the South, particularly
along the southern Mississippi River.
(Courtesy of Anthony Robinson and Jin
Chen.)
16:29:28.
COLOR PLATE 14 Using
ESTAT to further analyze the rela-
tionship between percent of Kerry
votes and percent Baptist. In the
parallel coordinate plot (PCP), the
percent votes for the Kerry axis has
been brushed from its highest value
to just slightly above 50 percent.
Note that in the PCP, the winning
Kerry counties had a low percent-
age of Baptists and a wide range of
both percent Catholics and income,
and that there is a positive correla-
tion (0.46) between the winning
Kerry counties and female head of
household. (Courtesy of Anthony
Robinson and Jin Chen.)
COLOR PLATE 16 The Globe Display option within IDV.

Here, two attributes are displayed: wind speeds in excess of 55
meters per second (as an isosurface) and air temperature near
the Earth’s surface (as a Color-Filled Contour). (Courtesy of
the Unidata Program Center.)
COLOR PLATE 15 Examples of using IDV to depict the

wind speed at an elevation where we typically find the jet
stream: (A) a Color-Filled Contour; (B) a Contour Plan View.
(Courtesy of the Unidata Program Center.)
599
16:29:28.
COLOR PLATE 1 The default display for
RVIS. On the left is an isarithmic map result-
ing from interpolating between the 49 point
locations for which dissolved inorganic
nitrogen (DIN) values were collected. On
the right is an uncertainty map based on a 95
percent confidence interval computed using
kriging. Note that different hues are used for
each map, and that lightness and saturation
vary within each hue. (Courtesy of David
Howard and Alan MacEachren.)
A
Rate
Reliable? per
Y N 100,000
6.5 to 34.8
5.1 to 6.4
3.8 to 5.0
3.0 to 3.7
2.2 to 2.9
1.6 to 2.1
0.0 to 1.5
COLOR PLATE 2 MacEachren et al.’s (1998) approach for

depicting uncertainty, in which parallel lines are split down
their length, with one side black and the other white. (From
“Visualizing georeferenced data: Representing reliability B
of health statistics.’’ Environment and Planning A 30:
pp. 1547–1561. Pion Limited, London. Courtesy of Alan
MacEachren.)
COLOR PLATE 3 Visualizations developed by Cliburn et

al. (2002): (A) height of the surface shows water surpluses and
deficits based on historical data, whereas the surface is col-
ored to reflect changes anticipated based on average GCM
model data; (B) transparency (an intrinsic approach) is used
to depict areas that are uncertain. (Courtesy of Daniel C.
Cliburn.)
16:29:28.
COLOR PLATE 4 Extrinsic visu-
alization developed by Cliburn et al.
(2002). Orange and purple bars rep-
resent the range of GCM predictions
at a particular location, and small
pyramid-like symbols at the end of
bars denote which GCMs are associ-
ated with the extreme low or high
point on a bar. (Courtesy of Daniel
C. Cliburn.)
COLOR PLATE 1 A map of Health Care Assistance

Receipts by County for 2002 created through American
Factfinder (http://factfinder.census.gov/). The map was down-
loaded in .pdf format. COLOR PLATE 2 A screen capture of The National Atlas
of the United States (http://www.nationalatlas.gov/) showing
major hurricane paths in the western Atlantic Ocean during
the 2000s.
601
16:29:28.
COLOR PLATE 3 Images from Tapestry of Time and Terrain, part of
A
The National Atlas of the United States: (A) the “Color Legend” depict-
ing the geological time scale; (B) an image that appears when
Precambrain is clicked in (A).
COLOR PLATE 4 A screen capture from World Wind (http://

worldwind.arc.nasa.gov/) showing the Grand Coulee dam (located at the
white cross-hairs) along the Columbia River in Washington State. Note
the high-resolution satellite imagery as the backdrop for the scene.
COLOR PLATE 5 A screen cap-

ture of the National Map (http://
nmviewogc.cr.usgs.gov/) showing the
railroad network across the Front
Range of the Rocky Mountains of the
United States.
16:29:28.
COLOR PLATE 6 A screen capture of the New York City Smell Map (http://www.gawker.com/
maps/smell/) of the subway stations around lower Manhattan Island. In this case, someone reported
that the Spring Street/Lafayette Street station had a distinctive body order and food smell. (Image
courtesy of Gawker.com; reproduced with permission.)
COLOR PLATE 7 A screen capture of PennDOT’s iVidlog (http://www.dot7.state.pa.us/ividlog/

video_locate.asp) entering the town of Glencoe, PA. iVidlog provides a virtual driving experience
across portions of Pennsylvania’s road network. (Image courtesy of Pennsylvania Department of
General Services, Bureau of Publications; reproduced with permission.)
603
16:29:28.
COLOR PLATE 8 A map of Iraqi religious and ethnic
groups as they existed in 1978. The map was downloaded from COLOR PLATE 9 A portion of William Henry Holmes’s
the Perry-Castañeda Map Library (http://www.lib.utexas.edu/ 1882 rendition of the Grand Canyon—an example of the more
maps/) as a .jpg image. than 8,800 maps available in the David Rumsey Map
Collection. (Courtesy of the David Rumsey Map Collections,
www.davidrumsey.com.)
COLOR PLATE 10 Four historical maps of Boston created

using the QuadViewer option within a GIS browser available in
the David Rumsey Map Collection. Note that the modern street
network is overlaid on each map. (Courtesy of the David Rumsey
Map Collections, www.davidrumsey.com.)
16:29:28.
COLOR PLATE 11 A screen
capture of the Sterling, VA, NWS
Doppler Radar (http://www.erh.noaa.
gov/lwx/). Notice that the bottom of
the map includes options to turn
“on” and “off” various layers.
COLOR PLATE 12 A screen cap-

ture of the United States showing
various weather data in three-hour
increments. By moving the mouse
over each time frame in the left-hand
column, the data on the map reflect
the conditions for each time frame.
(From the National Weather Service
at http://www.weather.gov/forecasts/
graphical/sectors/.)
605
16:29:28.
COLOR PLATE 2 An example of a wall-sized display for
creating a VE. The image portrays decision makers examining
the uncertainty associated with a water balance model.
COLOR PLATE 1 An example of Ian Bishop’s early work (Courtesy of James R. Miller.)
on visual realism—a simulation of a proposed lake created as
a result of filling an existing open-cut coal mine with water.
(Reprinted from Visualization in Modern Cartography, Ian
Bishop, “Using Wavefront Technology’s Advanced Visualizer
Software to visualize environmental change and other data,”
pp. 99, 101–103, Copyright 1994, with permission from
Elsevier.)
COLOR PLATE 3 The user interface for Fuhrmann and MacEachren’s software that utilized a
flying saucer metaphor. On the left is the egocentric view; on the right is an overview map with a
directional cone indicating the user’s current position and orientation. (Courtesy of Sven
Fuhrmann.)
16:29:28.
A COLOR PLATE 4 Examples of the Bath virtual city:
(A) one block of the more than 150 blocks included in the
city, (B) an aerial view of the Abby in Bath. (Courtesy of
CASA, University of Bath.)
COLOR PLATE 5 A scene from Virtual Los Angeles. (Courtesy of Bill Jepson, Director
UCLA Urban Simulation Laboratory.)
607
16:32:34.
COLOR PLATE 6 A scene generated by Community Viz. (Copyright © 2001 ESRI. All rights reserved.)
COLOR PLATE 7 How panoramic images created by panoraMap

are utilized in the Virtual Field Course project. The orange arrows on
the left correspond to the three views shown on the right. (Courtesy of
Jason Dykes, City University; Reproduced by permission of Ordnance
Survey on behalf of HMSO. © Crown copyright 2007. All rights
reserved. Ordnance Survey Licence number 100047826.)
16:32:34.
COLOR PLATE 8 An illustration of the power of panoraMap to link
other kinds of information with panoramic images and to explore this infor-
mation in a spatial context. The lower left views correspond to the orange
arrows shown above. In the lower right is a georeferenced digital movie con-
taining audio and video collected in the field, whereas the upper right is a
hypertext document containing images, links, and other information.
(Courtesy of Jason Dykes, City University.)
COLOR PLATE 9 Alternative approaches for visualizing forested landscapes: (A) a

traditional approach in which different land use/land covers are symbolized by polygons
of differing colors; (B) Dunbar et al.’s use of photorealistic virtual forests. (From Dunbar,
M. D., Moskal, L. M., and Jakubauskas, M. E. (2004) “3D visualization for the analysis of
forest cover change.” Geocarto International 19, no. 2:103–112; courtesy of Taylor &
Francis Ltd., http://www.informaworld.com.)
609
16:32:34.
COLOR PLATE 10 Views of a forested landscape at the stand
A level: (A) Dunbar et al.’s approach based on Visual Nature
Studio; (B) a photograph of the same area. (From Dunbar, M. D.,
Moskal, L. M., and Jakubauskas, M. E. (2004) “3D visualization
for the analysis of forest cover change.” Geocarto International
19, no. 2:103–112; courtesy of Taylor & Francis Ltd., http://www.
informaworld.com.)
COLOR PLATE 1 Example of a linked micromap plot

(LM plot). (Courtesy of Daniel Carr.)
16:32:34.
COLOR PLATE 2 Example of conditioned choropleth maps (CCmaps). The attribute being studied is the lung cancer mortality rate
for white men ages 65 to 75. Potential explanatory attributes—annual precipitation and percent below the poverty line—are depicted
from left to right, and from bottom to top, respectively. (Courtesy of Daniel Carr.)
COLOR PLATE 3 A screen from Fisher’s software for portraying

the uncertainty in remotely sensed images via sound. As the cursor
moves across the image, the user hears a sound representing the
uncertainty associated with the current pixel location. In the case
depicted here, a long duration would indicate a pixel with low uncer-
tainty (i.e., high reliability). (Courtesy of Peter F. Fisher.)
COLOR PLATE 4 An image viewed in MacEachren and his

colleagues’ collaborative work with ImmersaDesks. The bottom por-
tion shows a 3-D portrayal of terrain; the middle portion represents
temporal changes in precipitation (time increases with height), with
olive green “clouds” depicting precipitation isosurfaces above a certain
threshold value (Hurricane Agnes is depicted by a “blanket” of rain
multiple-days thick); and the top portion portrays temperature at the
end of the hurricane event. (From Figure 11 (p. 24) of MacEachren and
Brewer (2004) “Developing a conceptual framework for visually-
enabled geocollaboration,” International Journal of Geographical
Information Science 18(1), pp. 1–34; courtesy of Taylor and Francis Ltd.,
www.tandf.co.uk/journals.)
611
16:32:34.
COLOR PLATE 5 Two maps resulting from applying Murray
and Shyy’s median clustering method: (A) weights of 1.0 and 0.25
are assigned to crime rates and distance, respectively; (B) weights of
1.0 and 0.95 are assigned to crime rates and distance, respectively.
A (An adaptation of Figures 4 and 5 from Murray and Shyy (2000)
“Integrating attribute and space characteristics in choropleth dis-
play and spatial data mining.” International Journal of Geographical
Information Science 14(7), pp. 649–667; courtesy of Taylor & Francis
Ltd., http://www.tandf.co.uk/journals.)
Class Groups
1
2
3
4
5
6
COLOR PLATE 6 An illustration of Natalia

Andrienko and her colleagues’ use of the C4.5
algorithm for spatial data mining. The portion of
the decision tree highlighted with the thick black
line is depicted on the map by small squares.
(After Andrienko et al. 2001. First published in
Cartography and Geographic Information Science
28(3), p. 158. Reprinted with permission from the
American Congress on Surveying and Mapping.)
17:34:09.
Index
Page references followed by "f" indicate illustrated 249-256, 262, 265-266, 268, 271, 273-275, maps, 181
figures or photographs; followed by "t" indicates a 279-282, 369, 405, 407, 409, 428, 433, 451, Centrifugal force, 129, 138, 141, 143, 147-148
table. 457, 466, 478, 504, 506, 511, 513, 516, 550 Channel, 197-199, 208, 210, 407, 439
Basin, 369, 468, 527, 531, 538, 551 Chesapeake Bay, 485, 530, 539
Basins, 432 Chicago, 18, 23, 25-29, 33, 37, 42, 52, 62, 64, 145,
A Bathymetry, 531 148, 172, 194, 236-238, 259, 280, 305, 327,
A horizon, 134, 442 Beach, 5, 41, 66 329, 375, 397, 417, 517, 523-524, 585, 590
Aa, 546, 590 shoreline, 41 Chile, 36, 181, 218
Advanced treatment, 18, 327, 436 Bedrock, 3, 307-308, 319 China, 280, 283, 406, 436-437, 456, 458, 507, 538
Africa, 2-3, 91, 126, 181, 183, 199, 247, 360, 379, Beijing, China, 280, 283, 436-437, 456, 458, 538 television, 280
400, 451, 512, 538, 580-581 Bering Sea, 183 Cholera, 26-28
ethnicities, 91 Berry, Brian, 35 Cities, 5, 9, 11, 20, 40-43, 46-50, 54, 57, 61, 89, 93,
population concentrations, 183 Biodiversity, 527 134, 136, 230, 340, 344-345, 347, 369, 378,
African Americans, 388, 390, 394 Biomass, 15, 567 398, 433, 441, 447, 451-453, 457, 468-469,
Age, 21, 24, 135, 145, 196, 219, 230, 279, 282, 293, Birds, 433, 451 472, 506, 508, 515, 520, 526-529, 536, 549,
335, 381, 403, 434, 444, 506, 538, 545, Boston, 29, 236, 258-259, 327, 367, 369, 415-416, 590-591, 596
556-557, 582 438, 458, 514, 604 census tracts, 41, 398, 457
Aggregate, 107, 488 Boundaries, 19, 26, 41, 57, 59, 73, 86, 88-89, 94-95, central, 40, 57, 398, 472, 508
Aggregation, 41, 52, 61, 110, 113, 115-117, 125-126 97, 99, 101, 105, 111-112, 121-122, 126, edge, 61
Agnes, Hurricane, 551, 611 217, 221, 223, 242-243, 245-246, 250, 264, government, 5, 508
Agriculture, 1, 20-21, 107, 225, 305, 362, 379, 381, 285-286, 296-297, 302, 309, 332, 337, 339, Classification, 32, 64, 65-84, 91, 105-107, 112-113,
506 341, 361, 364-366, 397, 404-405, 409, 440, 115, 117-118, 127, 239, 242, 250, 258, 264,
AIDS, 88-89, 151-152, 171, 416, 441, 451, 457, 546 442, 451, 483, 500, 507, 510, 515, 520, 529, 285, 287-289, 299, 301-305, 352, 357, 362,
Alaska, 89, 132, 184, 219, 384, 433, 515, 590 537 371-374, 388, 390, 396, 398, 417, 444-445,
Albania, 472 cultural, 223, 500 447, 463-464, 466, 469, 473, 478, 482, 489,
Alberta, 36, 467, 479 Europe, 26, 246, 507 496-497, 499, 502-504, 516-517, 556, 578,
Alexander the Great, 435-436 geometric, 94, 112, 223 580
Alfisols, 104 mountains, 26, 510 Clay, 23-24, 58, 66, 96, 320, 322, 379
Altitude, 319, 476, 508, 511, 529 physical, 405, 442, 529 Climate, 88, 134, 192, 220, 311, 387, 506, 508, 515,
Aluminum, 271, 282 plate, 94, 101, 223, 243, 250, 365-366, 405, 409, 530, 539, 542, 561
American Geographical Society, 127, 348, 423-424, 442, 451, 507, 510, 515, 529 global, 508, 530
426 religious, 89, 217 latitude and, 134
Amplitude, 267 transform, 520 system, 134, 515, 530, 539
Andes, 411 water, 86, 89, 246, 332, 341, 361, 366, 520 types of, 387, 506
Angles, 24, 50, 116, 120, 131, 133, 137, 148, 151, Branch, 37, 527 world, 134, 192, 311, 508, 515, 539
156, 164-166, 169, 171, 186, 188, 190, 200, Brazil, 3, 199 Climate change, 508, 530
268, 282, 328, 383-384, 436, 438, 522, 573 Breaks, 65-67, 69, 71-73, 75, 77, 82-83, 374, 377, Climate changes, 530
Angola, 3 393, 444-445, 466, 474, 502, 504, 557, 594 Climate regions, 134
Angularity, 116, 120, 309, 313 British Columbia, 384, 435, 479 Climatologists, 110, 125, 319, 560
Animals, 200 Climatology, 21, 384, 443, 459
Antarctica, 14, 21, 159, 177, 179, 192, 443, 509, 547, C Clouds, 551, 564, 611
564, 569, 592 California, 5, 24, 41, 54, 82, 227, 230, 245, 286, 288, Coal, 510, 521, 567, 606
Arctic Ocean, 183-184 318, 371, 399, 407-408, 441, 458, 462, 533, Coalescence, 113-114, 118, 125-126, 364
Argentina, 181 544, 574, 587 Coast, 115, 121-122, 124, 172, 194, 384, 387, 447,
Aristotle, 134, 136 Canada, 11, 63, 179, 181, 378, 384, 412, 417, 447, 472
Arizona, 60, 288, 516, 518, 532 467, 479, 507, 517, 538-539 Coastal region, 232
Asia, 175, 181, 183, 189, 193, 512 Cancer, 58-59, 99, 134, 136, 286, 332, 402-403, 445, Coastline, 115, 118, 121-123, 126, 232
migration from, 175 475, 488, 497, 544-545, 560, 611 Cod, 340-341
population distribution, 175, 183, 193 Capacity, 280, 384, 583 Collision, 113
Atlanta, 42, 61, 417, 451-452 Capture, 19, 35, 101, 117, 183, 450, 505, 508-509, Color, 2, 6, 8-10, 12, 15, 53, 64, 87, 91-95, 101-102,
Atlantic Ocean, 25, 160, 569, 601 601-603, 605 104-107, 142, 192, 195-211, 220, 223-224,
Atmosphere, 87, 107, 416, 443, 521, 545 Carbon, 87 230-231, 241, 243, 246-247, 249-250, 252,
carbon dioxide in, 87 Carbon dioxide, 87 257, 262-263, 265-276, 279-283, 285,
density, 107, 416, 521 Caribbean, 3, 447 288-294, 296, 299, 302-305, 308, 324-329,
early, 416, 443, 521 Cartography, 1-21, 23-38, 39, 48, 64, 65, 82-83, 85, 334, 358, 365-366, 368, 375-377, 379, 381,
movement, 416, 521 105, 107, 109-110, 112, 117-118, 125-127, 383-384, 395-400, 405, 407-409, 412, 421,
ozone, 107, 443, 521 129, 143-145, 148, 149-151, 172-173, 424-426, 429-437, 441-444, 448, 450-452,
structure of, 521 175-176, 194, 195, 197, 206, 209-210, 457, 462-463, 466-467, 469-477, 479, 486,
Atmospheric circulation, 325, 579 213-214, 220, 228, 230, 236, 238, 239, 241, 488-489, 491-493, 499, 502, 504-516,
Australia, 138-139, 181, 402, 414, 507, 556, 558, 586 258-259, 261, 280, 282, 285, 292, 303-305, 521-522, 526-529, 531, 543-545, 547, 551,
mining, 556 307, 327, 329, 331, 333-334, 340-341, 343, 556-557, 565-612
size, 138, 414, 558 349, 367-370, 371, 375, 379, 397, 399-400, of objects, 527
Average annual precipitation, 220, 506 401, 407, 415-417, 419-420, 427, 436-438, Columbia River, 432, 508, 602
439-440, 442-443, 445, 449, 456-459, 461, Combines, 126, 144, 197, 231, 388, 506, 511, 520,
476-477, 479, 481, 487, 495-497, 499-503, 537
B Communications, 561
Baltimore, 30, 42, 58, 515, 528 516-518, 519, 525, 536-539, 541-542,
549-550, 558-564, 578, 584-585, 588, 606, development, 561
Bar, 111, 124, 214, 225-228, 235-237, 243, 245, Communities, 500, 527
247-248, 250, 256, 349, 400, 449, 451, 467, 612
Cascade Mountains, 422 Community, 21, 25, 36, 138, 221-222, 500-501,
474, 492, 601 508-511, 527-528, 557, 559, 564, 608
Bar scale, 111, 214, 225-228, 235-237, 243, 247-248, Caves, 436
Celsius scale, 89 Compaction, 56, 61-62
250, 256 Compass, 25, 115, 178, 228, 349, 442, 591
Barriers, 502 Census tracts, 6, 15, 19, 41, 105-106, 122, 307, 325,
328, 335, 339, 397-398, 444, 455, 457, Competence, 25
Bars, 44, 62, 200, 345-346, 349-350, 382-384, 396, Composition, 29, 394, 400
398-399, 449, 483, 492, 506, 510, 601 462-464, 478, 504, 546
Central America, 187-188 Compression, 190, 277
Base, 8, 23-24, 29, 34, 36, 59, 61, 103, 111-112, 130, Computer hardware and software, 553
135, 142, 148, 177, 194, 204, 214, 217-218, Central Asia, 181, 183
map, 181, 183 Concentration, 33, 87, 192, 197, 299, 353, 358, 402,
223, 228, 232, 235, 237, 239, 242-243, 245,
613
16:01:20.
469, 486, 521, 585 302, 366-367, 374, 398, 464, 542, 547, 562 Erosion, 14, 34, 529
Connections, 3 Dome, 553 soil, 14
Consumers, 521 Doppler radar, 512, 515, 605 Estuaries, 530
Continents, 141-142, 144, 147, 177-178, 186, 192, images, 515 Estuary, 530
529 Drainage, 2, 318, 325, 432, 468 Ethiopia, 3
Convergence, 133, 140, 159, 183, 187, 197, 502, Drainage basins, 432 Ethnicities, 91
517-518 Drift, 315, 328-329 Ethnicity, 26, 91, 185
Core, 23-24, 30, 32, 36, 277, 476 Dust Bowl, 408, 416 Europe, 25-27, 175, 178, 189, 192-193, 199, 208, 246,
of Earth, 24 412, 507, 512, 558, 569, 596
Cores, 58 E cities, 596
Corn Belt, 3 Earth, 10, 12, 14, 18, 24, 36, 87, 106-107, 110-111, Evaporation, 387
Correlation, 40-41, 48, 50-52, 57-58, 61-64, 345, 360, 126, 129-148, 150-154, 159-163, 165, Evapotranspiration, 384, 583
372-375, 377, 387, 395-398, 400, 445, 464, 167-168, 171-173, 175-177, 180-181, Evolution, 30, 36, 84, 172, 500
469, 475, 545, 555, 598-599 186-188, 193-194, 214, 216, 224-225, 237, Eye, 15-16, 19, 33, 186, 195-197, 200-201, 208-211,
Costa Rica, 493 250, 279, 307, 322, 327, 336, 397, 399, 407, 229, 241, 266-268, 298, 307-308, 324, 328,
Crater, 434, 591 416, 419, 424, 426, 433-438, 443, 448, 456, 346-348, 355, 367, 379, 382, 412, 415, 419,
Crests, 195-196, 210 458-459, 476, 490, 496, 499, 501, 507-509, 433, 437-438, 442, 454-456, 522, 524-525,
Crops, 14-15, 217, 286, 361-362, 381 516-517, 520, 523, 526, 528-531, 536-540, 527, 534, 537, 590
Crust, 141-142, 545 562, 589, 599
Crystal, 195, 201-202, 210, 435 core, 24, 36, 476 F
Crystals, 202, 210 core of, 476 Fahrenheit scale, 89
Culture, 3, 17, 20-21, 37 crust, 141-142 Farm size, 88
Currents, 141, 384, 408, 416, 549, 558 history of, 24, 36, 126, 129, 134-135, 145, 148, Fatalities, 366, 500, 512-514, 516
surface, 141, 408, 416 172, 416, 436, 499, 517 Fault, 308, 319
Cycling, 72-73, 83, 441, 457 interior, 135, 322, 437 Faults, 307-308, 318-319
Cyclones, 506 motions, 137 Feedback, 6, 236, 238, 257, 351, 562
tropical, 506 orbit, 141 Fires, 14, 531-532
orientation, 24, 106, 214, 216, 237, 250, 416, 520, Fish, 530
D 526 Fishing, 17
Dams, 508 radius of, 152-153, 163, 173, 177 Flat maps, 434
Dating, 24-25, 307 resources of, 443 Flooding, 249
Death rate, 14, 286 structure, 12, 106, 126, 416, 419, 426, 433, 435, Floodplain, 487-488, 510
Death rates, 286, 488 437, 443, 459, 496, 528, 531, 538 Florida, 29, 56, 66, 68-69, 71, 80-81, 121-124, 288,
Deaths, 14-15, 59, 286, 375, 389-390, 416, 445, 482 Earth system, 397, 399, 448 384, 473-474, 597
Declination, 228 science, 397, 399 Floyd, Hurricane, 435, 591
Delta, 201-202 Earth system science, 397, 399 Focus, 3, 5, 13, 15, 17, 23-24, 26-27, 37, 40, 48, 80,
Denali National Park, 433, 590 Earthquakes, 442-443 87, 95, 101, 126, 138, 143-144, 167,
Density, 2, 5, 7, 13, 15, 19, 21, 26-27, 40-42, 44-45, location, 442-443 175-176, 181, 183, 185, 187-188, 193, 201,
48-49, 51, 54, 98-99, 105-107, 113-115, 117, Ecology, 125 209, 219, 234, 249, 251-252, 270, 276, 286,
184, 219-220, 248-249, 257, 269, 274, Economic development, 467 288, 292, 303, 315, 327, 344-345, 359, 372,
286-287, 321-322, 328, 331-341, 363-366, Ecosystem, 530-531 375, 383, 387, 390, 392, 398, 401, 407-408,
368-369, 383-384, 399, 401-403, 405, Ecuador, 138 415, 419, 429, 439, 442, 444-445, 450, 452,
409-413, 415-417, 420-421, 467, 471-472, Education, 1, 3, 16, 23, 29, 34-37, 39, 44-45, 48-52, 455, 457, 459, 464-467, 469, 473-475, 478,
479, 487, 504-506, 521, 573, 578, 583, 585 54, 65, 85, 109, 129, 149, 175, 195, 201, 482, 493-494, 500-501, 510, 530-531, 534,
ice, 411 213, 218, 239, 261, 285, 289, 295, 300-301, 542-543, 556, 558-559
Denver, 144, 148 307, 331, 343, 371-372, 374, 378, 401, 419, Fog, 484-485, 496
Department of Agriculture, 20-21, 225 439, 444, 448, 456-457, 459, 461, 464, 469, Fold, 263
Department of Commerce, 58, 64, 144, 172, 194, 481, 499, 502, 519, 535-536, 538-539, 541, Folding, 263, 274
223-224, 497, 571 595 Forcing, 73
Deposition, 48 Egypt, 136 Forest fires, 14, 532
Development, 2, 8, 15, 23-25, 27-29, 31, 34, 36-38, Electrical charge, 202 Forests, 19, 360, 440, 520, 531, 537, 609
41, 52, 59, 127, 134-135, 144, 172, 194, Electromagnetic spectrum, 15, 195, 209-210, 289, Fraction, 43, 91, 94, 110-111, 124-126, 214, 224-225,
224, 228, 234, 299, 308, 327, 331, 337, 340, 304, 324, 328, 507 237, 500
345, 370, 396, 401, 436, 440, 442-443, Electron, 198, 201, 209-210, 280 France, 24, 26, 87, 137, 508, 539, 549
455-456, 458-459, 465, 467, 470, 477-478, Electrons, 201, 209-210 Freezing point, 89
480, 482, 496, 513, 522, 526-528, 538, Elements, 6, 14, 29, 31, 40-41, 59, 61-64, 144, Freshwater, 530
541-564 149-173, 194, 201, 210, 213-238, 239-243, Friction, 549
future prospects, 456 246-252, 254-259, 264, 272, 305, 370, 383, Fronts, 9
Diamonds, 241 429, 435, 438, 444, 446-447, 450, 481-482,
Diffuse reflection, 426 497, 505, 515, 523, 543, 572, 598 G
Diffusion, 401-403, 417, 465 Elk, 96 Galaxy, 545
Digital elevation models (DEMs), 327, 502, 514 Emissions, 508 Gamma rays, 196
Discharge, 468, 483, 530 Energy, 90, 195-196, 206, 209-210, 412, 416 Gases, 457
Disease, 88, 445, 456, 474-475, 493, 497, 543, 557, kinds of, 196 Gender, 17
598 kinetic, 90 Gentrification, 110
chronic, 445 levels, 90 Geocentric, 139-140, 147
heart, 445, 474-475, 598 England, 18, 20-21, 26, 38, 61, 63-64, 84, 107, 127, Geographic information systems (GIS), 2, 14, 19, 501
Distance, 14, 57-58, 61, 63, 86, 93, 107, 110, 115, 129, 131, 135, 147-148, 167, 194, 209, 259, Geography, 1, 3, 10-12, 20-21, 23-29, 31-36, 38, 54,
117-119, 130-132, 135-138, 140, 143, 148, 280, 305, 327, 367, 369, 399-400, 404-405, 59, 62, 64, 84, 107, 110, 120, 125-127, 134,
151, 155, 163, 168, 170, 172, 175, 179, 184, 416, 458-459, 469, 479, 495, 497, 517-518, 183, 236-237, 278, 304, 306, 327, 370, 390,
187, 190-191, 195-196, 210, 224-226, 256, 527, 537-538, 562-563 399, 401, 405, 409-410, 414, 416-417, 433,
275, 307, 309-319, 325-328, 363, 365, Entisols, 104 437, 440, 448, 450-451, 456, 458-459, 479,
392-396, 398, 401-402, 416, 420, 424, 428, Environment, 1-2, 10, 12, 14, 18-20, 34-35, 61, 63, 75, 483, 520-521, 523, 525-528, 531, 536-539,
431, 444-447, 485, 490, 500, 539, 547, 549, 101, 103, 113, 125, 185, 214, 236-237, 239, 553-554, 559, 563-564, 590, 595
556, 560, 612 277-278, 281, 291, 299, 340-341, 352, 360, approaches to, 35, 64, 107, 458
Distribution, 1, 3, 5-9, 14-15, 20, 26, 40-41, 43-44, 48, 367, 369, 371, 381-382, 385, 405, 409, content of, 126, 526
53, 57, 61-63, 70-71, 80, 82, 86, 88-89, 95, 415-416, 439, 443, 456, 458, 462-463, 475, human, 3, 12, 20, 23, 26, 62, 64, 84, 110, 126, 401,
97, 100-101, 103, 105, 107, 118, 141, 147, 477, 480, 497, 499-500, 502, 505-506, 508, 405, 414, 416, 459, 536-537, 539,
161, 175, 181, 183, 185-186, 192-193, 197, 510, 513-514, 520-522, 524-533, 535-539, 553-554, 559, 563-564
222, 239-240, 249, 263, 279, 281-282, 292, 545, 550-551, 553, 558-560, 562, 564, 567, regions, 26, 54, 134, 401, 410, 416, 553-554, 564
295, 301-303, 317, 325, 332, 340-341, 344, 600 scale and, 110, 120, 125-127
360-363, 372, 374, 376-377, 385, 396, 400, Environmental factors, 403 Geoid, 129, 141-142, 144, 147-148, 569
415, 439, 441-442, 448, 455, 465-466, 472, Equator, 24, 36, 129-135, 137-143, 147-148, 150, 155, Geologic time, 507
474, 488, 499, 501-503, 514, 517, 529, 542, 158-159, 162-163, 171, 173, 176-178, time scale, 507
558-560, 562, 581, 594, 597 180-184, 186, 188, 193, 491 Geologic time scale, 507
density and, 26, 332, 340, 415 region, 138, 176-178, 180-181, 183, 188, 193, 491 Geology, 29, 62, 64, 327, 329, 369, 506, 523
patterns and, 466 Equilibrium, 246, 249, 256, 258-259, 572 environmental, 327
Divergence, 410 Era, 23-24, 26, 30, 36, 151, 420, 433 Geomorphology, 424
Divide, 20, 43, 72, 91, 98, 105, 286, 288, 294-295, Eratosthenes, 24, 129, 136, 143
16:01:20.
Georgia, 36, 61, 288, 310, 417, 451, 504-505 Isolines, 26, 99, 106, 308, 328, 371, 387, 396, 399, Magnetite, 141
Geoscientists, 241, 319, 520, 526, 549 486 Magnitude, 52, 58, 63, 71, 77, 98, 104, 106, 113, 221,
Germany, 36-37, 138, 496, 539-540, 556 250, 286-287, 291, 295, 304, 322, 344-345,
Gerrymandering, 56, 62 J 350-351, 364, 367-369, 382-383, 398,
Gibraltar, 134 Japan, 132, 220, 406, 508, 540 407-409, 411, 415-416, 429, 440-441,
Glass, 31, 122, 202, 210 time, 406 446-447, 452, 455, 457, 469, 484, 491, 497,
Global Circulation Models (GCMs), 491 Jason, 469, 525, 595, 608-609 546
Global Positioning System (GPS), 17, 543, 558 Jet stream, 441, 457, 476, 478, 599 Map projections, 23, 31, 35, 37, 129-130, 144-145,
Global warming, 491, 560 Jets, 281 148, 149-173, 175, 177-178, 182, 187,
Globe, 23-24, 36, 111, 144, 150-153, 155-159, 161, Joints, 548 192-194, 224-225, 417, 483, 491, 495-496,
163-165, 167-168, 171-173, 178, 181, 187, 530, 559
304, 399, 404, 443, 476, 491, 500, 507-508, classes of, 151, 156
530, 537, 599 K Maps, 1-21, 23-27, 29, 31-38, 40, 54, 57-59, 61-62,
Gradient, 115, 322, 420 Kelvin scale, 90 64, 65, 68-70, 73-74, 77-84, 89-95, 97-101,
Grand Canyon, 435, 514, 520, 529, 579, 604 Kenyatta, Jomo, 218 103-105, 107, 111-112, 121, 125-127, 132,
Grand Coulee Dam, 602 Kinetic energy, 90 134-135, 139, 143-145, 147-148, 151, 159,
Gravity, 56, 137, 141-142, 147, 247, 545 161, 166, 169, 171-172, 175-178, 180-181,
specific, 141-142, 147 L 184-187, 192-194, 195, 197, 199, 201,
Great circle, 132-134, 147, 168, 177 Labor, 449, 547 203-204, 208, 210-211, 214-216, 219-220,
determining, 132, 134 Lag, 451 223-230, 235-238, 239-242, 248, 250-251,
routes, 168 Land use, 17, 45, 89, 91, 101, 106, 113, 331-332, 255, 257-259, 262-265, 269-270, 272-273,
Great circles, 132-133, 170, 172, 176 335-336, 339-340, 361-362, 365-367, 370, 276-282, 285-288, 291-306, 307-313,
Great Lakes, 299, 383, 410 415, 420, 434-435, 437, 482, 489-490, 317-320, 322-329, 332, 334-337, 339-341,
Greenland Ice Sheet, 411 492-493, 496, 510, 527, 580, 609 344-345, 347, 349, 351-353, 355-358,
Groundwater, 533 multiple, 17, 91, 331, 339-340, 493 360-362, 364, 366-370, 371-377, 379-381,
Gulf coast, 121-122 sequential, 489 385, 388, 390, 396-400, 401-417, 419-421,
Landforms, 13, 21, 424-425, 436, 438 425-426, 433-438, 439-442, 444-449,
H Language, 26, 235, 265, 272, 276-279, 281-282, 415, 451-459, 462-470, 472-475, 477-480,
Halos, 231, 256 477, 501, 507, 509, 513, 522 481-483, 486-490, 492-497, 499-503,
Harrison, John, 129, 135, 144 standard, 265, 279, 281-282, 522 506-507, 510-512, 514-518, 520, 525-526,
Head, 19, 24, 106, 368, 475, 495, 520, 522-524, 529, Latin America, 29 528-529, 542-543, 545, 547, 549, 553-564,
535, 537, 542, 550-551, 560, 599 Latitude, 11, 35, 87, 107, 129-132, 134-140, 142-148, 568, 575-576, 578, 580-581, 585, 588-589,
Health care, 504, 601 151-159, 162-163, 165, 167-168, 173, 591, 594, 597, 603-604, 611-612
Heart disease, 445, 474-475, 598 175-176, 183-185, 190-191, 193, 225-227, flat, 13, 83, 159, 172, 241, 336, 425-426, 434
Heat, 281, 457, 533 237, 312, 476, 508, 511, 569 projection, 143, 151, 159, 161, 166, 169, 171-172,
latent, 281 determination of, 143, 227 175-178, 180-181, 184-187, 192-194,
Highlands, 66 Lead, 14, 33, 52, 89, 99, 105, 201, 215, 287, 312, 210, 224, 239, 242, 250, 258, 312, 324,
Hispanics, 457, 462-464, 478 364-365, 372, 410, 463, 487, 495, 525, 532, 401-406, 408, 415-416, 437, 482, 492,
Horn, 421, 425, 436, 438 545, 549, 595 542-543, 547, 560, 562, 585
Hot spot, 255, 257, 573 Lebanon, 56 scale, 5, 13, 21, 24, 89-90, 111-112, 121, 125-127,
Hotspots, 508 Library of Alexandria, 24 134, 139, 143-144, 151, 159, 161, 166,
Housing, 41, 338-339, 382, 388, 448, 462-464, 482, Light, 16, 35, 65, 92, 94, 104, 106, 110, 120, 134, 152, 171-172, 176-177, 185-187, 192-194,
504, 512 158, 165, 173, 195-198, 201-204, 206-210, 214, 216, 219, 224-228, 230, 235-238,
forms, 388, 462, 464 223, 228, 238, 266, 268-269, 271, 275, 281, 239, 242, 248, 250-251, 258, 279-280,
public, 464, 512 288-291, 294, 304, 324-325, 336-337, 327, 336, 339, 360, 370, 400, 404, 420,
U.S., 41, 388, 482, 504, 512 420-421, 423-426, 429-431, 435-437, 453, 435-438, 442, 452, 455, 463, 468, 482,
Humidity, 545 469, 472, 485-486, 488, 513, 515, 550, 552, 488, 503, 506-507, 510, 512, 514-515,
Hurricanes, 506 591 526, 528, 549, 585
category, 506 Lightning, 415 scale of, 24, 111, 121, 125, 151, 219, 224, 242,
tracking, 506 strikes, 415 250, 512
United States, 506 Limbs, 383 topographic, 1-2, 15, 19, 21, 32, 37, 111-112, 139,
Hydrology, 479, 504, 510 Limestone, 3, 567 144, 151, 172, 177, 180, 203, 280, 319,
Hypotheses, 52, 200, 475, 545, 561 Location, 1-2, 14, 19, 40-41, 45-46, 54-57, 61-62, 322, 327, 411, 419, 426, 434-438, 482,
Hypothesis, 390, 399, 555, 561 87-89, 95, 101, 105-106, 113, 129-132, 487, 510, 553, 591
134-135, 139, 142-143, 148, 150, 152, Marble, 34-35, 415, 508
160-167, 173, 176, 179-180, 190-191, Markets, 17, 26
I 201-203, 210, 219, 225, 227-228, 232, 252, Matter, 9, 32, 82, 111, 433, 500
Ice, 89, 411, 509, 539 256, 277, 306, 307-308, 313, 317-318, Meanders, 404
density of, 411 326-329, 336, 350, 360-362, 369, 381, 383, Mercator projection, 160-162, 178, 181, 192, 491-492
freezing point, 89 408, 415, 420, 424, 435, 439, 441-443, 451, Meridian, 129, 131-133, 135-136, 140-143, 145,
Ice sheets, 411 455, 463, 466, 468-469, 482, 484, 486-487, 147-148, 152-153, 158-159, 164-165,
Ice shelf, 509 491-492, 496-497, 500, 502, 506-508, 172-173, 175, 177-178, 180-182, 184-185,
Iceberg, 509 511-512, 515-516, 520-521, 526-527, 529, 190, 194, 227
Iceland, 226, 472, 596 531, 533, 535, 542-543, 546-547, 550, Mesoscale, 459
Igneous rocks, 141 552-553, 558, 560-564, 601, 611 Meteorology, 479
Impression, 68, 99, 169, 183, 193, 274, 350, 363, 372, mathematical, 62, 95, 150, 152, 164, 173, 307, Mexico, 288, 384, 440-441, 480
374, 420, 424, 433, 440, 454-455, 490, 495 313, 327-328, 350, 362, 369 Microsoft, 270, 501, 522
India, 3, 139 site, 89, 256, 277, 469, 500, 502, 506, 511-512, Migration, 56, 175, 185-187, 189, 193, 401, 406-409,
Industrialization, 433 515-516, 527 415-417, 560, 587
Infant mortality, 375, 379, 385, 388-390, 395, 584 situation, 176, 190, 420, 515, 535 distance, 175, 187, 401, 416, 560
Inselberg, 62, 64 London, 18, 20-21, 37-38, 62, 64, 82, 84, 127, 135, net, 407-408
Insolation, 86, 308 167, 172, 194, 198, 209-210, 238, 280, 282, Mine reclamation, 510
Institute of British Geographers, 82, 559 326, 329, 340, 382, 399, 405, 416, 436, 448, Mining, 313, 457, 462, 475, 477, 510, 542-543,
Intensity, 19, 73, 94, 105, 141, 187, 202-203, 287, 339, 508, 517-518, 528, 536-539, 561, 563, 600 555-557, 560-564, 567, 612
381, 452, 515, 520, 525-526, 535, 537, 549 Longitude, 11, 35, 87, 107, 129-132, 134-135, Mississippi River, 120, 475, 531, 598
Interface, 35, 278, 312, 437, 450, 457-459, 474, 495, 140-147, 151-154, 156-157, 163, 172-173, Model, 5-6, 9-10, 12-13, 15, 18-20, 52, 112-113, 115,
497, 501, 505, 507, 509-510, 512-513, 517, 175, 184, 190, 227, 237, 312, 476, 508, 511, 125, 140-143, 148, 150-151, 173, 177, 195,
522, 526, 529, 532, 534-536, 538, 549, 569 198-199, 203-204, 206-208, 210, 239-242,
551-554, 563-564, 598, 606 search for, 508 249, 257-258, 263-264, 282, 290, 315-317,
International Date Line, 144, 304, 399 Los Angeles, 42, 53, 348, 399, 441, 462, 527, 536, 319-320, 322, 326, 336, 340-341, 368, 387,
Internet, 10-11, 18-19, 21, 35, 262-263, 276-282, 408, 607 417, 419, 426-427, 430-431, 435-438, 440,
416, 499-503, 507, 510, 515-518, 529, 533, Los Angeles Basin, 527 443-444, 449, 455, 458, 465, 468, 487, 491,
538, 543, 551, 559, 564, 566, 587 Los Angeles, California, 399, 441, 462 495-496, 501, 528, 530, 535, 538-539, 542,
Iowa, 42, 286, 288, 525, 533, 538 Loxodrome, 133 546, 548-551, 561-562, 567, 569-570, 600,
Iran, 572 Lunar eclipse, 137 606
Iraq, 500, 512-513, 516 Lung cancer, 402-403, 545, 560, 611 Moldova, 472, 596
Ireland, 26 Molecules, 90
Iron, 13, 141, 425 Mollisols, 104
Islands, 117-118, 121, 126, 383, 506 M
Macedonia, 472, 596 Monitoring, 455-456, 511
615
16:01:20.
Monsoon, 451 Photons, 195 Ratio, 6, 28, 52, 56, 62, 86, 89-91, 104-106, 110, 121,
Monsoons, 23, 25 Physical environment, 14, 529 126, 171, 173, 224, 237, 285-287, 306, 322,
Moscow, Russia, 127 Pittsburgh, 42 335, 340, 345, 349, 351, 369, 400, 406, 428,
Mouth, 118, 125-126, 382 Place, 2, 14, 17, 20-21, 23, 33, 36, 53, 70, 72, 74, 77, 465-466
Music, 547 82, 87, 89, 112-113, 135, 140-141, 143, 178, Reflection, 234, 426, 429
184, 186, 193, 195, 197, 200-201, 210, 214, Reformation, 29
N 216-217, 223-224, 229, 231-233, 235, 237, Regions, 26, 54, 56, 61, 97-99, 134, 138, 143,
National Academy of Sciences, 34, 36, 403, 417, 564 242-243, 247, 250-251, 255-256, 267, 289, 176-177, 185, 194, 217-218, 224, 241, 285,
National Geographic Society, 169, 172 295, 309, 313, 325, 328, 337, 356, 360-362, 302, 319, 329, 333-334, 345, 361, 364,
National Ocean Service, 144 364-365, 407, 465, 468-469, 472, 476, 367-368, 376, 381, 387, 398, 401-402, 404,
National Oceanic and Atmospheric Administration 499-500, 507-509, 511, 513, 518, 525, 406, 410, 412, 415-416, 431-432, 435, 447,
(NOAA), 530 531-532, 537, 549-551, 558, 562, 564 467, 487, 490-491, 507, 515, 529-530, 543,
National Park Service, 434, 436, 438, 499, 503-504, Planar projection, 157-158, 173, 178, 180, 186-187 545, 547, 549, 553-554, 556, 564, 586, 589
516, 590-591 Plankton, 383-384, 583 formal, 224
National Science Foundation, 33 Plants, 26, 104 geographic, 54, 56, 61, 143, 176-177, 194,
Natural disasters, 508 Plasma, 201, 531 217-218, 224, 329, 333-334, 401-402,
Natural gas, 223 Plates, 91, 202, 210, 262, 270-271, 273-276, 279, 406, 415-416, 435, 487, 490-491, 507,
Net migration, 407-408 282, 292, 325, 334, 376, 416, 432, 475, 479 515, 529, 543, 547, 553-554, 564
Neutral, 223, 281, 289, 304, 488, 502, 575 PM, 447, 468 Relative location, 546
New Brunswick, 21, 127 Polar regions, 185 Relief, 13, 21, 26, 30, 328, 419-421, 424-428, 432,
New England, 497 Pollution, 11, 549 435-438, 446, 506-507, 531, 553, 588-589
New Mexico, 288, 440-441 Population, 2, 5-7, 9, 11, 14-15, 17, 19-20, 23, 25-27, Religion, 26, 89, 475
New York City, 183, 286, 388, 402, 499, 511, 516, 552, 30, 40-49, 51, 54, 56-58, 61-64, 66, 68, 71, remote sensing., 280
603 77, 82, 89, 91, 99, 105-107, 111, 113, 132, Renaissance, 23-24, 29, 36, 144, 456
New Zealand, 241 175, 183, 185-186, 189, 192-193, 208, Reservoirs, 361
Newton, Isaac, 129, 137 219-220, 230, 248, 286, 288, 296, 300, 308, Resources, 21, 281, 443, 448, 456, 459, 462, 464,
Nitrogen, 485-486, 600 322, 327-328, 331-341, 345-347, 361, 369, 501, 518, 530, 533
Nodes, 11, 141, 310 372, 374-375, 381, 388-390, 395, 400, Response, 197, 206, 210, 235, 269, 301, 303, 305,
North America, 2, 139, 143, 147-148, 192, 434, 507, 401-406, 414, 416, 440-441, 444-445, 337, 350, 375, 493, 511, 524, 528, 532, 543,
535, 569 450-453, 455, 457, 462-465, 468-472, 475, 553, 557-558, 563
North Atlantic Treaty Organization (NATO), 471 481-482, 488, 499, 504, 529, 547, 556-557, Revolution, 35, 134, 138, 280, 282
Nucleus, 200 565, 574-578, 584-585, 594-596 Rice, 97, 368, 549, 561, 563-564
Population concentrations, 183 Ridges, 431-432, 435, 589
Population density, 2, 5, 7, 19, 26-27, 40-42, 44-45, Risk, 20, 257, 331, 336, 341, 488, 563, 573
O 48-49, 51, 54, 105-107, 113, 219-220, 248, Rivers, 1, 112, 120, 230, 234, 468, 506-507, 515
Observations, 41, 43-44, 46, 48, 50, 54, 62-63, 68, 286, 328, 331-334, 336-341, 402, 405, Robinson projection, 178, 194
70-74, 77, 82-83, 173, 371, 374, 377, 388, 471-472, 504, 578, 585 Rock, 141, 433
390, 392-396, 398-399, 470, 474, 556, 560, arithmetic, 41 igneous, 141
581 Population growth, 440-441, 450 sedimentary, 433
Occlusion, 442 Portugal, 472, 596 Rocks, 141-142, 379, 381, 507, 529
Ocean, 25, 122, 144, 160, 175, 177-178, 180-181, Poverty, 40-42, 45, 47-52, 54, 224, 469, 545, 547, 560, metamorphic, 141
183-184, 188, 193, 241, 312, 383-384, 398, 595, 611 source, 507
407-408, 416, 530, 569, 583, 601 Power, 16-18, 44, 104, 202, 263, 311, 319, 336, Rocky Mountains, 410, 420, 510, 569, 602
color, 241, 383-384, 398, 407-408, 569, 583, 601 349-351, 367-369, 374, 414, 440, 442, 445, Ross Ice Shelf, 509
geography, 25, 183, 416 463, 502, 507, 516, 529, 534-536, 558, 609 Rotation, 103, 130-131, 134-135, 138, 141, 147-148,
living in, 530 Prairies, 13, 425 201, 324
margins, 407 Precession, 141 Russia, 104, 127, 175, 183-185, 193, 451
Ocean currents, 384, 408, 416 Precipitation, 9, 87-88, 217, 220, 310-312, 317, immigration, 193
Oceanic crust, 545 319-320, 326-327, 362, 368, 371, 384, 387,
Oceanography, 479 410, 441, 451, 467-468, 491, 494, 506, 512,
Oceans, 129-130, 135, 141-142, 144, 147, 177-178, S
515, 537, 545, 550-551, 560, 583, 611 Sahel, 451
181, 193 Prediction, 451, 594
Offshore, 118 Salinity, 398, 476, 531-532
Preservation, 165-166, 169, 320 Salt, 89
Oil, 26, 87, 89, 93, 99, 106, 274, 344-345, 367, 401 Pressure areas, 410
Oman, 3 San Francisco, 11, 20, 216, 441, 457-458, 477-480,
Prime meridian, 129, 131-132, 135, 142-143, 147-148, 497, 539, 563, 574, 589, 591-592
Ontario, 383-384, 389 153, 178, 182
Opaque, 268, 358-360, 369, 414, 443, 586 San Francisco Bay, 11, 216, 458
Process, 9, 15, 19-20, 47, 52, 56, 62-63, 73, 78, 82, Sand, 362, 379, 581
Overburden, 567 107, 110-113, 115, 117-120, 122, 125-126,
Overfishing, 530 Sandstone, 308
134, 150, 152, 156, 171-172, 175-176, 181, Satellite imagery, 434, 507-508, 529, 602
Ozone, 21, 107, 325, 443-444, 455, 459, 521, 579, 183, 193, 195-197, 200, 204, 206, 208-211,
592 Satellite images, 383, 435, 514-515
214, 228, 235, 237, 239-242, 249-252, Satellites, 14, 106, 141, 144, 148, 176, 337, 420
depletion, 325, 444, 579 256-259, 262-264, 266-272, 274-275,
Ozone depletion, 325, 444, 579 Saturation, 86, 92, 94, 103-106, 195, 203-204,
279-282, 285, 287, 292, 294-295, 302-303, 206-208, 210, 288-289, 325, 430-431, 483,
307-309, 315-316, 318-319, 325-329, 350, 486, 489, 496, 568, 574, 579, 600
P 364, 366, 371, 374, 376, 388, 390, 392-393, light, 92, 94, 104, 106, 195, 203-204, 206-208, 210,
Pandemic, 557, 563 396-397, 400, 402, 415, 429-430, 433, 435, 288-289, 325, 430-431, 486
Paradigms, 23, 35, 37, 536, 563 438, 444, 450, 465, 473, 488, 493, 500-503, Saudi Arabia, 401
Parallel, 25, 32, 40, 48-50, 62-64, 91, 117, 122, 126, 522, 525, 533, 537, 539, 542-543, 546, 553, Savannas, 360
130, 132-133, 136, 141, 145-148, 152, 556, 560-563, 572, 574, 589 Scale, 5, 13, 21, 24, 87, 89-90, 106, 109-127, 134,
155-158, 163, 173, 184, 227, 317, 420-421, Projections, 23-24, 29, 31-32, 35-37, 129-130, 132, 137, 139, 142-144, 150-151, 153, 158-159,
435-437, 464, 470, 474-479, 482, 488-489, 144-145, 148, 149-173, 175-178, 180-188, 161-168, 170-173, 176-177, 185-188,
595, 598-600 190-194, 224-225, 404, 415, 417, 457, 483, 190-194, 214, 216, 219, 224-228, 230,
Parallels, 130-134, 144, 155-159, 173, 175, 184-186, 491, 495-496, 530, 559, 562 234-238, 239, 242-243, 247-248, 250-252,
190, 193, 225, 231, 304, 377, 399 Ptolemy, 23-24 256, 258, 279-280, 327, 336, 339, 360, 363,
Parcel, 255 Ptolemy, Claudius, 23-24 370, 384, 400, 404, 420, 422, 431-432,
Paris, 26-27, 135, 137-138, 172, 529, 585 Pythagoras, 136 435-438, 442, 452, 455, 463, 468, 482, 488,
Perception, 8, 15-16, 19, 21, 107, 195, 197, 200, 491, 503, 506-507, 510, 512, 514-515, 526,
208-211, 215, 241, 259, 303, 305, 351, R 528, 535, 549, 584-585, 602
367-369, 381, 399, 459, 480, 496-497, 539, Radar, 196, 510, 512, 515, 605 Scales, 32, 89-91, 110-111, 113-114, 121, 143, 173,
542-543, 562-563 Radar images, 515 183, 225-227, 245, 294, 296, 303, 305-306,
Period, 9, 17, 19-20, 23-26, 28-30, 33-36, 53, 93, 134, Radial pattern, 407, 416 345, 420, 510, 553
137, 186, 192, 203, 287, 310, 317, 399, Radiation, 106, 336 bar, 111, 225-227, 245
443-444, 446, 450-451, 457, 468, 472-473, gamma, 106 graphic, 305, 420, 553
478, 512, 542, 547, 551, 562, 565 Radio waves, 196 Scanner, 111, 200-201, 270
Periods, 14, 23-24, 36, 137, 189, 362, 372, 396, 440, Railroads, 217-218, 406, 510 Scattering, 436
469, 474, 500, 547 Rain, 318, 451, 515, 551, 611 Scenarios, 10, 302, 527
Petroleum, 329 Rain gauge, 318 Scientific inquiry, 135
Phoenix, 399 Rainbow, 404, 443 Sea level, 87, 90, 107, 141-142, 148, 419, 426, 589
Photogrammetry, 337, 489 Rapids, 42 Seasons, 91
16:01:20.
Seattle, 83, 127, 400, 406, 538 Sweden, 138, 459 440, 506-507, 510-511, 516, 518
Sections, 29, 40, 118-119, 126, 224, 242, 264, 273, Sweep, 449 networks, 112, 407, 416, 516
332, 344, 350, 371, 435, 440, 449, 462, 489, System, 10, 15, 17, 19, 21, 25, 36, 53, 75, 107, Triangulation, 24, 137, 148, 307, 309-311, 313,
512, 531 110-111, 120-121, 124-127, 129-148, 153, 318-319, 325-328, 485
Sedimentary rock, 433 172-173, 184, 194, 195, 197, 200, 202-204, Tropic of Cancer, 134, 136
Sedimentary rocks, 142 206-211, 224, 226, 237, 250, 258, 269-270, Tropic of Capricorn, 134
Services, 31, 107, 270, 282, 382, 400, 443, 502, 507, 276, 280, 282, 295, 305, 326, 329, 340-341, Tropical cyclones, 506
516-518, 533, 542-543, 558, 560-564, 603 367, 383, 397, 399-400, 437-438, 448, Troposphere, 325, 579
business, 270, 282 450-451, 456, 467-468, 476-477, 479, Troughs, 440-441
Settlements, 336 496-497, 499, 503-504, 510-511, 515-516, Tsunami, 336
Sextant, 134, 148 518, 522, 525, 527, 530, 533-534, 536, Tundra, 221
Shale, 567 538-539, 543, 551, 553-554, 558, 563, 571 Turkey, 539
Shields, 19, 524, 535, 537 Systems, 2, 8, 14, 18-21, 24, 28-31, 37-38, 48, 52, 64,
Shoreline, 41 82, 87, 105, 107, 121, 126-127, 144-145, U
Silicon, 443 148, 172, 194, 202, 209, 211, 214-215, United Kingdom, 338, 528-529
Sill, 315 234-238, 240, 258-259, 262, 265, 269-270, United Nations, 185
Silt, 379 276, 278-279, 281-282, 304-305, 326-327, United Nations Environment Program, 185
Sites, 10, 48, 86-87, 89, 438, 467-468, 484, 499-500, 329, 340, 366, 370, 379, 383, 399-400, United States, 2, 4-8, 10, 13-15, 20-21, 27-29, 31, 34,
502-504, 510-512, 515-517, 543, 558, 560, 415-416, 436-438, 443, 445, 449, 456, 37, 41, 48, 63, 70, 77, 88, 91, 105, 110, 124,
595 458-459, 476-477, 479, 487, 495, 497, 501, 127, 135, 142-144, 151, 160-161, 175-176,
Slide, 117, 393, 589 517, 525, 527-529, 531, 533, 535, 537-539, 185-187, 189, 192-194, 199, 208, 215, 220,
Slides, 31 542, 553-554, 558, 560-564 222, 224-225, 245, 292, 312, 322, 331, 336,
Slope, 51-52, 172, 318, 324-325, 331, 336-337, 339, closed, 64 352, 358, 378, 380, 384, 387, 400, 407-408,
361-362, 384-385, 410, 419-422, 425-432, Earth as, 529 410, 420, 424-425, 433, 436-438, 446-447,
436-438, 545, 584, 589 equilibrium, 258-259 449, 465, 470, 473-475, 479, 481, 484, 486,
Slopes, 361, 420, 425-427, 429, 432, 437 feedback, 236, 238, 562 488, 497, 499-500, 506-507, 510-511,
Small circles, 132-133, 158, 368 open, 278-279, 282, 539, 553 514-517, 530, 543-546, 557, 569, 572, 582,
Snow, 27 586, 597, 601-602, 605
Snowflakes, 399 T concentration, 192, 358, 486
Soil, 14, 58, 86, 104, 234-236, 318, 361-362, 379, Taxonomy, 538 Universe, 137
384, 399-400, 497, 551, 583 Temperature, 1-2, 19, 89-90, 203, 270, 293, 304-305, Upslope, 421
moisture, 384, 583 312, 317, 328, 345, 371, 383-384, 387, 398, Urban areas, 6, 120, 183, 361-362, 405, 438, 440,
Soils, 15, 86, 362, 467-468 441, 444, 450, 476, 491, 515, 530, 537, 549, 455, 585
Sorting, 53, 376 551, 583, 599, 611 urbanized areas, 6
Source regions, 412 apparent, 2 Urban heat island, 533
South Africa, 3, 126, 400, 538 controls, 444 Urbanization, 458, 530
Soviet Union, 139, 185, 500 data, 1-2, 19, 89-90, 293, 304, 312, 317, 328, 345, Urbanized areas, 6, 112, 332, 528
Space, 11, 35, 44, 47, 58, 61, 63, 87-88, 105, 107, 371, 383-384, 387, 398, 441, 444, 450,
110-111, 114-115, 121, 127, 141, 145, 160, 476, 491, 515, 537, 549, 583
162, 168, 176, 186, 203-204, 206, 209, 214, V
latitude, 312, 476 Vault, 518
216-218, 221, 223-224, 226, 229-230, 233, scales, 89-90, 305, 345
236-237, 239, 246-248, 251-252, 255-256, Vegetation, 2, 10, 15, 17, 26, 115, 227, 384-385, 398,
sea surface, 383, 583 456, 512, 567, 584
258-259, 297, 303, 315, 325, 328-329, 344, world distribution of, 1
346, 350, 356-357, 364, 367, 379, 382-384, Vienna, 211, 280, 558
Temperature maps, 304-305 Visible light, 195-196, 210
396, 401, 408, 415-416, 431-432, 434-435, Temperature scales, 89
439, 448-450, 452-453, 455-456, 458-459, Volume, 3, 18, 20, 24, 26, 31, 73, 83-84, 115, 125-127,
Tension, 547 262, 269-270, 272-274, 276, 279-281, 304,
465, 468-469, 476, 500-501, 506, 508-509, Test, 73, 78, 120, 206-208, 321, 328, 372, 390, 493,
511, 521, 528-530, 532, 542, 546, 549-553, 307-308, 320-322, 326-329, 344, 347, 355,
549, 561 367-370, 400, 407, 479-480, 491, 497, 512,
555-556, 560-561, 563-564, 566, 572, 612 Testing, 399, 456, 495, 555, 562
Spatial analysis, 2, 14, 36, 62, 64, 249, 456, 475, 479, 531, 538, 547, 566
Texture, 93, 95, 223, 340-341, 379, 381, 383, 399,
528 441, 493, 582
Species, 507, 529-530 The Netherlands, 36, 517-518, 525 W
invasive, 507 Theories, 195, 197, 240 Wales, 404-405
Springs, 58 Theory, 15, 18, 25, 32, 35, 37, 59, 64, 77, 107, 126, Water, 14, 24, 31, 86, 89, 106, 129, 141, 147, 164,
Stars, 385 135, 195, 197, 203, 206, 208-211, 238, 241, 178, 220, 230, 232, 246-247, 274, 317, 322,
States, 2, 4-10, 13-15, 20-21, 27-29, 31, 34, 37, 41, 268, 274, 291, 299, 303, 327, 329, 429, 435, 331-332, 335, 341, 345, 361-362, 366-368,
48, 56, 63, 70, 73, 77, 86, 88, 91, 95, 452, 517, 522, 533, 555, 589 383-384, 408, 468, 489, 491-492, 495-497,
105-106, 110, 113, 124, 127, 135, 142-144, Thermometers, 483, 496 506, 520-521, 530-532, 549, 551, 567, 600,
151, 160-161, 175-176, 185-187, 189, Thunder, 451, 454, 594 606
192-194, 197, 199, 208, 215, 220, 222, Thunderstorm, 442-443, 455, 515 in soil, 497
224-225, 245, 288, 292, 299, 301, 310, 312, Tides, 141, 530, 532 states of, 506
322, 331-332, 336, 345, 350, 352, 358, Time, 6, 9, 14, 17, 19-20, 23-24, 28-32, 34-36, 41, 43, Water balance, 384, 491, 495, 497, 606
377-378, 380, 384, 387, 400, 402, 407-410, 66, 77, 80, 83, 87, 91, 93, 98, 104, 111, 127, Waterfalls, 508
416, 420, 424-425, 433, 436-438, 441, 130, 134-137, 144, 155, 158, 189, 192, 228, Watershed, 116, 531
446-447, 449, 451, 455, 457, 465, 470, 231, 235, 262-264, 270, 273-274, 277, 280, Waveform, 195-196, 209
472-475, 479, 481, 484, 486, 488, 497, 282, 285, 292, 298, 302-303, 307, 311, Wavelength, 94, 106, 195-197, 206, 208, 210, 308,
499-500, 506-507, 510-511, 514-517, 530, 318-319, 326, 329, 333, 336, 362, 372, 324, 326, 328
543-546, 557, 565, 569, 572, 582, 586, 597, 374-375, 384-385, 387, 396, 399, 401, 406, Waves, 141, 196, 532
601-602, 605 408-409, 415-416, 419-420, 430, 435-436, height, 141
compact, 56 439-441, 443-444, 446, 448, 450-453, length, 141
elongated, 484 455-458, 462-466, 468-469, 472-479, 482, orbital, 141
federal, 497, 499, 510, 516 485-486, 493, 499-500, 502-503, 506-507, wavelength, 196
Strabo, 134 511-512, 515-516, 520-521, 527, 529-534, Weather, 9, 19, 87, 99, 187, 222, 307-308, 310, 324,
Stratosphere, 325, 579 545-546, 549-551, 554, 558, 562-564, 327-329, 345, 412, 439, 441, 448, 450-452,
Streamlines, 412 565-566, 602, 605, 611 457, 459, 494-495, 500, 514-517, 529, 539,
Streams, 2, 116, 120, 406, 420, 468, 506-507 Tin, 420 551, 594, 605
discharge, 468 Tokyo, Japan, 132, 540 fronts, 9
gradient, 420 Topographic maps, 2, 32, 112, 139, 180, 434, 436, 591 jet stream, 441, 457
valleys, 420 Topography, 107, 307, 322, 410-411, 413, 421, 425, Wells, 87, 222, 264, 308, 344, 366-367
Stress, 2, 89, 124, 191, 351, 374, 473, 551 433, 437-438, 443, 476, 510, 537, 588-589 West Indies, 135
Strike, 151, 201 Tornadoes, 175, 187, 192, 387-388 Wetlands, 493
Strikes, 168, 173, 195, 415 damage, 187 Wind, 106, 237, 279, 401, 408-412, 414-416, 451,
lightning, 415 Township, 101, 319 476, 478, 484-485, 495, 499, 506-509,
Strokes, 229, 237 Trade winds, 23, 25 515-516, 546, 586, 599, 602
Sublimation, 280, 457 Translucent, 268, 451 measurement, 106, 484
Suburbs, 556 Transmission, 89, 216, 240, 500 monsoon, 451
Summer solstice, 136 Transparency, 481, 483-485, 491, 495-496, 600 Winds, 23, 25, 410, 412, 448, 530
Sun, 24-25, 129, 134, 136, 429, 539 Transparent, 273-274, 281, 358-360, 369, 443, 491 Women, 16-17, 488, 497
Surf, 459, 497 Transportation, 17, 26, 112, 217, 401, 406-407, 416, World population, 192, 327, 441
617
16:01:20.
World War I, 29
X
X-rays, 202
Y
Yellowstone National Park, 531
Z
Zebra mussels, 507
Zenith, 429
Zimbabwe, 377
16:01:20.

Terry A. Slocum, Robert B. McMaster, Fritz C. Kessler, Hugh H. Howard - Thematic Cartography and Geovisualization - Pearson New International Edition-Pearson Education Limited (2013)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Terry A. Slocum, Robert B. McMaster, Fritz C. Kessler, Hugh H. Howard - Thematic Cartography and Geovisualization - Pearson New International Edition-Pearson Education Limited (2013)

Uploaded by

Copyright:

Available Formats

Slocum, Terry A., et al.

Thematic Cartography and Geovisualization:

Visit us on the World Wide Web at: www.pearsoned.co.uk

© Pearson Education Limited 2014

ISBN 10: 1-292-04067-X

British Library Cataloguing-in-Publication Data

Printed in the United States of America

1. Thematic Cartography and Geovisualization

Slocum, Terry A., et al. Thematic Cartography and Geovisualization:

Volume of Slave Trade, 1700–1870

Corn for Grain or Seed, 1992

1 Dot: 10,000 Acres

Wheat for Grain, 1992

1 Dot: 10,000 Acres

wish to map the distribution of total population in the

Slocum, Terry A., et al. Thematic Cartography and Geovisualization:

Population Density, 1990

Persons per Square Mile

0 to 11 12 to 28 29 to 51 52 to 119 120 to 63,231

Millions of People Millions of People

0.5 15.1 29.8

A Forested Land, 1992 B Forested Land, 1992

Millions of Acres Percentage of Total Land Area

Slocum, Terry A., et al. Thematic Cartography and Geovisualization:

A Foreign-Owned Agricultural Land, 1992 B Foreign-Owned Agricultural Land, 1992

Percentage of Privately Owned Land

Slocum, Terry A., et al. Thematic Cartography and Geovisualization:

TABLE 1 Internet uses for cartography

FIGURE 7 An example of mapping

An excellent recent example of the capability pro-

concrete visual representations—whether on paper or to avoid confusion with the superscript 3.

Slocum, Terry A., et al. Thematic Cartography and Geovisualization:

Slocum, Terry A., et al. Thematic Cartography and Geovisualization:

Slocum, Terry A., et al. Thematic Cartography and Geovisualization:

Slocum, Terry A., et al. Thematic Cartography and Geovisualization:

FIGURE 1 Halley’s 1701 chart of

FIGURE 2 Frère de Montizon’s

FIGURE 3 D’Angeville’s 1836

FIGURE 4 Petermann’s 1848 con-

3.1 Period 1: The Incipient Period

John Paul Goode

themselves to cartography specifically, but some were

Guy-Harold Smith Richard Edes Harrison

3.2 Period 2: The Post-War Era and the Building

A seminal event in the evolution of American acade-

FIGURE 7 Guy-Harold Smith, who was an early thematic

Slocum, Terry A., et al. Thematic Cartography and Geovisualization:

Arthur Robinson established himself as the unofficial

Slocum, Terry A., et al. Thematic Cartography and Geovisualization:

University of Washington and their establishment in the

Moreover, several were going to have to teach without

Slocum, Terry A., et al. Thematic Cartography and Geovisualization:

most significant changes include: (1) closer integration

4 THE PARADIGMS OF AMERICAN CARTOGRAPHY

As academically oriented graduate programs emerged

4.1 Analytical Cartography

Slocum, Terry A., et al. Thematic Cartography and Geovisualization:

analytical cartography: a theoretical branch of cartography

Slocum, Terry A., et al. Thematic Cartography and Geovisualization:

McMaster, Robert B., et al. Thematic Cartography and Geovisualization:

TABLE 1 Sample data for 50 U.S. cities (sorted on murder rate)

Irvine, CA 0.0 2.6 95.1 780 2607 110

McMaster, Robert B., et al. Thematic Cartography and Geovisualization: