Professional Documents
Culture Documents
JIAPING WU
Zhejiang University, China
JUNYU HE
Zhejiang University, China
GEORGE CHRISTAKOS
Zhejiang University, China
San Diego State University, United States
Elsevier
Radarweg 29, PO Box 211, 1000 AE Amsterdam, Netherlands
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
Copyright © 2022 Elsevier Inc. All rights reserved.
No part of this publication may be reproduced or transmitted in any form or by any means, electronic or
mechanical, including photocopying, recording, or any information storage and retrieval system, without
permission in writing from the publisher. Details on how to seek permission, further information about the
Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance
Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.
This book and the individual contributions contained in it are protected under copyright by the Publisher (other
than as may be noted herein).
Notices
Knowledge and best practice in this field are constantly changing. As new research and experience broaden our
understanding, changes in research methods, professional practices, or medical treatment may become
necessary.
Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any
information, methods, compounds, or experiments described herein. In using such information or methods they
should be mindful of their own safety and the safety of others, including parties for whom they have a professional
responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability
for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or
from any use or operation of any methods, products, instructions, or ideas contained in the material herein.
vii
viii Contents
As its title dictates, the subject of this book is environmental science, agronomy, ecology,
the chronotopologic analysis and modeling of public health, epidemiology, economics, public
natural phenomena, that is, phenomena that policy, and risk management.
vary as functions of both their spatial (topos)a The book views the difference between train-
and the temporal (chronos)b coordinates, in real- ing and education in terms of the following met-
istic conditions of in situ uncertainty. The need to aphor: The way to get people build a ship is not
study such phenomena, in particular, has led to to teach them carpentry, but to inspire them to
significant developments in chronotopologic long for the infinite immensity of the sea.
data analysis (CTDA), which is an important Accordingly, the book’s methodology has four
part of the book. stages: The first stage is diagnosis (knowing
The notions, techniques, and thinking modes through), that is, identifying the basic features
discussed in the book aim at improving the of the phenomenon of interest, the second stage
understanding of the chronotopologic laws of is cardiognosis (knowing the heart of the phe-
change underlying the available numerical data- nomenon), that is, appreciating what is at stake
sets, while taking into consideration all relevant and choosing the appropriate mode of thinking,
core and site-specific knowledge bases, which the third stage is prognosis (foreknowledge), that
are subject to multi-sourced uncertainties and is, accurate prediction of certain important
associated measurement or observational errors aspects of the phenomenon, and the fourth stage
(conceptual, technical, computational). The core is epignosis (improved knowledge), that is, draw-
knowledge bases include scientific theories, ing important conclusions, inferences, and
physical laws, and models, whereas site-specific interpretations.
knowledge bases may have various forms and This methodology has some interesting
sources, including hard measurements, soft implications that become evident throughout
observations, secondary information, and auxil- the book, including the following: new routes
iary variables (ground-level measurements, sat- open when one type of data crosses another;
ellite observations, scientific instrument records, potentially difficult problems are considered
protocols and surveys, empirical graphs and from alternative standpoints, assessing how
charts). Understanding the spatial distribution well different methods can handle them, and
and temporal dynamics of knowledge bases how an improved problem solution is obtained
such as the above is a great challenge to the by a synthesis of methods acting in synergy; the
elucidation of crucial questions in many physi- presentation of worth noticing personal conclu-
cal, health, and social disciplines, including sions based on experience and insight is
geology, hydrology, geophysics, geography, facilitated.
a
The term “topos” (see topography, topology etc.) was the ancient Greek term for space.
b
The term “chronos” (see chronology etc.) was the ancient Greek term for time.
ix
x Preface
To achieve its particular goals, the book’s Individual topics addressed in the book may
focus is the rigorous presentation of the support- be probably found in separate publications, but,
ing theory, followed by a comprehensive and as far as we know, there is not any particular
balanced fusion of theory-driven and data- book that covers all these topics in a complete,
driven techniques. Accordingly, the assertions systematic, and integrative manner as the pre-
presented in the book have theory-based content sent book does. Arguably, then, readers with sci-
(the expression of a proposition about the real entific background or engineering training in
phenomenon being in a certain way) and are the aforementioned disciplines would appreci-
made in a data-based context (including the ate a systematic presentation in one volume of
available evidential support and real world the most important quantitative concepts,
conditions). In melioribus annis, this approach models and techniques studying the chronoto-
would have seemed natural. In praesenti annis, pologic behavior (combined spatial distribution
however, it is necessary to stress the importance and temporal dynamics) of natural phenomena
of this approach, since the excessive use of under conditions of uncertainty and site-specific
technological “black-box” nowadays runs the measurement errors (i.e., the vast majority of
risk to eventually create impoverished human phenomena encountered in the various scien-
“black-boxes.” Several numerical examples tific and engineering disciplines).
and actual case studies are included so that We would like to thank our families for their
the readers gain a hands-on experience and patience, and Ms Lindsay C. Lawrence of Else-
valuable insight concerning the implementation vier for her continuing encouragement during
of the presented notions, models, and tech- the book-writing project.
niques in the real world. Practice exercises at
the end of each chapter can help the readers Jiaping Wu
learn more from the text and hone their critical Junyu He
and practical skills. George Christakos
C H A P T E R
1
Chronotopologic data analysis
a
The term “topos” (see topography, topology, etc.) was the ancient Greek term for space.
b
The term “chronos” (see chronology, etc.) was the ancient Greek term for time.
c
The word παραδειγμα (paradeigma) has been used in famous texts, such as Plato’s Timaeus, as the model or the pattern
that God used to create the cosmos.
Quantitative Analysis and Modeling of Earth and Environmental Data 1 Copyright # 2022 Elsevier Inc. All rights reserved.
https://doi.org/10.1016/B978-0-12-816341-2.00005-8
2 1. Chronotopologic data analysis
emerged that is characterized by the This is a serious matter to anyone who takes a
addition of a significant computational moment to examine it with an attentive mind.
component (since theories of complex Therefore, we will revisit it in various parts of
phenomena became too complicated to the book.
solve analytically, numerical simulations
needed to be generated).
1.2 Interplay of science and mathematics
Furthermore, it has been argued by many
No readers of this book, most of whom are
investigators (scientists and nonscientists alike)
presumably quants (i.e., quantitative analysts),
that we currently enter the era of yet another
need to be reminded of the all-important rela-
paradigm:
tionship between two key components of
④ The big data-driven paradigm that seeks to human inquiry (see Fig. 1.1):
apply computational models to breathtaking
amounts of data obtained by instruments or
generated by simulators, then processed by Mathematics, which focuses on how to develop
the software, resulting in information stored abstract representations of the real world, and
in computers. science, which is concerned with the inverse
process, i.e., with how to use these abstract repre-
According to this paradigm, insight is gained sentations to obtain useful knowledge about the
through a self-reinforcing loop between experi- real world.d
mental data and statistical analysis (Succi and
Coveney, 2019). As should be expected,
serious objections to the purely data-driven Admittedly, this book uses a considerable
paradigm exist. Among them, Lewis H. Lapham amount of mathematics in the form of analysis,
has suggested that: modeling, and estimatione notions and tech-
@ Data mining engineers have no use for the niques. Compared to the complexity of the real
meaning and value of words. They come to
bury civilization, not to praise it.
In response to these and similar concerns, a
challenge to the purely data-driven perspective
has emerged, as follows:
• The big data perspective outlined above
makes extravagant claims, which, once
properly discarded, a synergistic synthesis
of large datasets with a sound scientific
theory could plausibly lead to a sensical FIG. 1.1 The opposite directions of mathematics and
paradigm-shift. science.
d
Yet, the two have some interesting sociological differences, including the fact that in science each generation of
investigators undoes the existing structure (replaces the theories of the previous generation with new ones etc.),
whereas in mathematics each generation adds a new piece to the existing structure.
e
The term “estimation” generally includes interpolation, extrapolation and prediction. The distinction between these
three kinds of chronotopologic estimation is critical: in interpolation and extrapolation attribute estimates are sought at
points, respectively, within and outside the chronotopologic sampling domain, and in prediction attribute estimates
are sought at points within the spatial sampling area but outside the temporal sampling period. The issue is discussed
in various parts of the book (e.g., Chapter 7).
1 From topos to chronotopos 3
world, these notions and techniques are not as Example 1.1 Assume one is given the equation
complex as some investigators may seem to E ¼ Fq, relating the electric field E with force F and a
q
believe. Those who do not believe so, it is rather point charge q. If q is divided by 2, i.e., 2, what is the
due to the fact that they may not realize how corresponding E value? Focusing on the mathe-
complicated real life actually is. matical grammar of the equation, the obvious
The links of mathematics with ordinary life answer is E ¼ 2F q . This answer is physically incor-
are rather obvious. Many mathematical ideas rect, though, since it neglects the essential knowl-
are ways of mathematizing ordinary ideas, edge that F is proportional to q, in which case the
e.g., the idea of a derivative that is a straightfor- correct answer is that there is no change in the
ward mathematization of the ordinary idea of value of E. This error happens when one (who
instantaneous change. Beyond this rather simple probably lacks the necessary background knowl-
notion, many complex mathematical concepts edge) looks at the equation and treats it purely as a
eventually come down to direct perceptual matter of mathematical grammar, neglecting to
experience. enrich its symbols with physical meaning.
Since science, a natura eius, is about finding
Since this book’s perspective is that an equa-
the analogy and identity in the most remote
tion’s mathematical grammar should be inte-
parts, it should be noticed at the outset that this
grated with physical meaning, our focus will
book’s own interpretation of Fig. 1.1 is as
not be only on technical deductions, but also
follows:
on the rich variety of inductive inferences asso-
ciated with our readers’ ordinary experience.
The same argument is obviously valid in terms
The interworkings of science and mathemat-
of data, i.e., data should not be treated purely
ics should be carefully considered in both direc-
as numbers, neglecting to enrich it with the
tions: not only mathematics should be used in
due physical meaning. In this respect, the fol-
scientific investigations, but science should be
lowing general conclusion is reached:
used in mathematical investigations too.
• Readers are encouraged to avoid falling into
the trap of reasoning purely in terms of
Accordingly, investigators should closely mathematical grammar and numerics, thus
fuse mathematical symbology with physical missing clues to a different interpretation
meaning, resulting in a powerful and productive given by physical meaning.
structure. That is, scientific applications require
The payoff of this approach, integrating
us to integrate our understanding of the real
mathematical grammar and numerics with
world with symbolic relations of mathematics,
physical meaning, is comprehensible, since it
thus adding meaning and structure to both. Sci-
can produce new knowledge, add to the existing
ence puts meaning to mathematics, adding
body of knowledge in a particular scientific
additional levels of structure, interpretation,
field, and, in many cases, scientific findings
and even tools. This book, then, is more inter-
can be transferred into useful technology.
ested in the scientifically meaningful employ-
ment of mathematical tools than in the purely
formal features of the tools. A direct consequence 1.3 Natural attributes
of the above considerations is the warning that
looking at an equation and treating it purely as Quantitative analysis, as practiced in sci-
a matter of mathematical grammar, while ences, relies on the tools of the discipline of
neglecting to enrich its symbols with physical applied mathematics, which studies the varia-
meaning, leaves one vulnerable to errors. tional characteristics of empirical observations
4 1. Chronotopologic data analysis
or measurementsf associated with natural attri- unsampled points (discussed in Section 4) may
butes. Before we proceed, we need a working decrease considerably with increasing data vari-
definition of the latter: ation. Accordingly, understanding the attribute’s
variational characteristics based on the available
data is a great challenge to the elucidation of cen-
A natural attribute is a measurable or observ- tral questions in many physical, social, and health
able entity characterizing selected aspects of a disciplines, including environmentology, geol-
natural phenomenon (physical, biological, social) ogy, agronomy, ecology, geography, public
that vary over a region during a time period. health, epidemiology, economics, social policy,
and risk management.
Simply put, everything that varies across
space and/or time in nature and can be mea- 1.4 Kinds of scientific data analysis
sured (using simple devices or sophisticated As regards the practice of data analysis in sci-
equipment of various kinds) or observed (via ences, where there is a strong physical depen-
perceptual experience) can be considered a nat- dence of data values on location and time,
ural attribute. In this setting, known (measured three kinds of studies have historically emerged
or observed) attribute values are characterized (in ascending order of modeling sophistication):
as attribute data.
Example 1.2 There is a wide variety of such ❶ Studies concentrating solely on the variation
natural attributes, including pollutant concen- of data across space, known as spatial data
tration, topographic elevation, ocean surface analysis (SDA), or solely on temporal data
temperature, rainfall intensity, human expo- variation, known as time data analysis (TDA).
sure, disease incidence, population mortality, ❷ Studies focusing on some kind of separable
land-use variables, commodity prices, and space-time data analysis (hereafter denoted
regional poverty indicators. as S-TDA), in the sense that they consider
space and time in isolation.
In cases such as the above, attribute character- ❸ Studies favoring a composite spacetime
istics of prime importance that need to be ade- data analysis (denoted as STDA), in the
quately defined and understood are the sense that space and time are considered
following: as an integrated whole, meaning that the
• The variational characteristics describe integration obeys the physics of the
attribute dynamics,g regional changes, and phenomenon.
interrelationships due to the underlying
In the above settings, space and time individ-
mechanisms of the phenomenon under
ually become three- and one-dimensional
study (Section 2).
projections of the four-dimensional domain,
Indeed, data variation across space and time is respectively. Conversely, the classical three-
a dominant characteristic that has tremendous dimensional geometry of SDA becomes a
impacts on the performance of certain quantita- four-dimensional geometry in space-time analy-
tive analysis and modeling aspects. For example, sis. Due to its wide popularity, let us comment a
the accuracy of attribute interpolation at little further on SDA. Although it may be seen,
f
The distinction between measurement and observation is discussed later in the book.
g
The term dynamics here refers to the temporal change of the attribute.
1 From topos to chronotopos 5
methodologically, as a special case of S-TDA,
there are considerable differences, as will become Time series analysis (TSA) is a collection of
clear in the following. Moreover, SDA was concepts and techniques for studying the statisti-
historically developed many decades before cal characteristics of time series data and making
S-TDA. Hence, one can find a plethora of SDA attribute forecasts for future times.
techniques—including inverse distance, splines,
trend surface, statistical regression, and geostatis- Historically, the earliest theoretical TSA
tics Kriging (Chapters 7 and 8)—that have been developments can be traced back to the pio-
successfully implemented in many disciplines. neering work of Yule (1927) and Walker
(1931). The next major development was the
1.4.1 Geostatistics and time-series statistics work of Box and Jenkins (1976) presenting the
SDA techniques did not develop uno ictu tem- complete time series modeling process (specifi-
poris. A case in point is the following well- cation, estimation, diagnostics, and forecast-
known SDA field: ing). One can find several TSA models in the
relevant literature, including moving averages,
exponential smoothing of various degrees,
Geostatistics consists of a collection of concepts analysis of variance, autoregressive integrated
and techniques that study spatially distributed moving average, among many other models.
attributes based on a space-dependent statistical There have been numerous TSA applications in
theory that was built on pre-existing results formu- almost any scientific discipline, including econo-
lated in different scientific disciplines with varying metrics, finance, geophysics, seismology, signal
objectives in mind. processing, pattern recognition, and meteorol-
ogy. The interested readers are referred to the
Indeed, the roots of geostatistics go back to very rich literature on the subject.
the 1940s–60s as a group of spatial correlation,
regression, and mapping techniques used in Example 1.3 Both SDA and TSA capitalize on
forestry and earth sciences with considerable spatial and temporal correlations, respectively.
success, see the pioneering work of Matern To most scientists, the fact that closely spaced
(1947) and Matheron (1965). Remarkably, sev- or temporally separated samples tend to be sim-
eral decades later the field was rediscovered ilar is not surprising, since such samples are
and renamed spatial statistics.h In any case, many likely to be influenced by similar physical
important advances have been made since the processes.
1960s, and the range of geostatistics or spatial An obvious disadvantage of SDA and TSA is
statistics applications nowadays is vast (the that they both neglect any cross space-time data
interested readers are referred to the relevant associations, and the reason of this neglect may
literature). be a physical justification, an apparently plausi-
TDA, on the other hand, generally focuses on ble approximation, or simply a matter of conve-
the study of sequences of attribute data ordered nience. Yet, reality often does not strictly satisfy
in time (time is often considered the indepen- the SDA or TSA assumptions. Instead, strong
dent variable). In particular, the systematic sta- correlations may occur simultaneously across
tistical study of this kind of time data was the space and time:
focus of a well-known TDA field:
h
A well-known spatial statistics book is that of Noel Cressie (Statistics for Spatial Data. J Wiley, NY, 1991).
6 1. Chronotopologic data analysis
• In reality, the links between attribute data Example 1.4 Space-time adaptive processing
values can be critically affected by space- extends adaptive antenna techniques to proces-
time cross-effects,i and the corresponding sors that simultaneously consider spatial
data distributions can be highly skewed. domain information (in the form of signals
received on multiple antenna array elements)
In a human sense, after all, to represent mat-
and temporal domain information (in the form
ters in a purely spatial manner is to freeze time
of multiple pulse repetition periods) of a coher-
and to deny the dynamism implicit in living.
ent processing interval.
A similar criticism holds when representing
matters in a purely temporal manner that denies Although S-TDA is very useful in many prac-
spatial dependency. By way of a summary, then, tical applications, its key limitation is rather pro-
the following conclusion is drawn based on the found: S-TDA essentially suggests seeing the
above considerations: world in slices. I.e., the S-TDA models, by defi-
nition, consider space and time as separate argu-
• Both SDA and TSA encounter profound
ments. In layman terms, then, the need for the
challenges to their methodologies that have
composite spacetime perspective proposed by
been based on notions and models that are,
STDA is based on the following rather straight-
respectively, timeless-space (SDA) and
forward argument:
space-independent (TSA).
It is reasonably anticipated that the adequate In many cases, it is hard to pin down the truth
responses to these challenges could lie in richer
when one is obliged to see the world in slices,
approaches that account for both space and time
because snapshots may conceal as much as they
in a systematic way. Such approaches are intro-
make plain.
duced in the remaining of this chapter and they
are, in fact, the subject matter of this book.
Otherwise said, STDA also studies the distri-
bution of natural attributes, but, while S-TDA
1.4.2 S-TDA and STDA views space and time as separate arguments,
For the above reasons, the advent of S-TDA STDA considers a composite spacetime argu-
has been seen as a significant development over ment. Classical statistics assumes nonspatiality
SDA and TSA because it offers a more accurate and that the samples are independent from
quantitative representation of the real world, one another, whereas S-TDA and STDA account
including a more realistic study of the space- for spatiality and temporality, although in dif-
time distribution of data from natural attributes. ferent ways. STDA is obviously a more recent
Indeed, many useful S-TDA models have been development than S-TDA.
developed with emphasis on the quantitative Undoubtedly, S-TDA and STDA offer
expression of attribute properties and relation- improved representations of reality, which is
ships that take into account the space-time local- why one can find a considerable number of
ization of these properties and relationships in a interesting real-world publications of S-TDA
direct way. The notions and methods discussed and STDA in a variety of scientific journals
in the book address these and similar issues. representing different and often unconnected
i
This means that, due to space-time cross effects, data values that are further apart in space but closer in time may turn
out to be more similar than those that are closer in space but further apart in time, and vice versa.
1 From topos to chronotopos 7
disciplines. Yet, there is a very limited number each attribute of interest physically meaningful
of books on S-TDA, and even fewer on STDA chronotopologic features. In particular, while
theory and applications.j This is somewhat sur- classical statistics assumes that the samples are
prising, since there is a great potential and need independent from one another, CTDA considers
for comprehensive and systematic presentations sample dependence governed by physics, it is
of S-TDA and STDA in practice, due to the not tied to a distribution model that assumes
obvious fact that the vast majority of real data that all samples of a population are normally
vary across both space and time. Moreover, distributed, and it is devoted to the interpreta-
improved technology is becoming available that tion of uncertainties caused by various in-situ
facilitates the in-situ implementation of S-TDA conditions (Section 2 below).
and STDA, e.g., low-cost Temporal Geographic Due to the increasing availability of datasets
Information Systems, TGIS, with user-friendly that are simultaneously location and time
interfaces (Christakos et al., 2002). dependent, their systematic quantitative analy-
sis and modeling provided by CTDA are of
great importance to a variety of scientific and
1.5 The unifying chronotopologic data engineering disciplines, including earth, ocean
analysis notion and atmospheric sciences, environmental and
ecological engineering, health and epidemio-
It is now time to return to the notion of logical studies, risk assessment, social policy
chronotopology briefly introduced at the begin- and financial management.
ning of this chapter. It should be brought to the
reader’s attention that, for methodological and
presentation reasons, in this book a unifying 1.5.1 CTDA conditions
notion will be favored, on occasion, depending As a matter of fact, it should be emphasized at
on the context (e.g., when the properties of the outset that CTDA can be used in the study of
interest are shared by both S-TDA and STDA). natural attributes that satisfy these conditions:
① They are measurable or observable.
It is methodologically appropriate and occa- ② They vary within a well-defined chrono-
sionally convenient to use the general term chron- topologic domain.
otopologick data analysis (CTDA) that includes ③ They are characterized by chronotopologic
both cases of S-TDA and STDA. dependence.
④ They occur in conditions of in-situ
uncertainty.
Generally, CTDA aims at extracting implicit
knowledge such as chronotopologic relations In other words, the CTDA concepts and tech-
and patterns that may not be explicitly stored niques should be used in the study of natural
in the relevant databases. This extraction relies phenomena when meaningful values of the attri-
on the merging of scientific knowledge and log- bute may occur at every point in the chronotopo-
ical principles. CTDA distinguishes itself from logic domain of interest that can be potentially
classical data analysis in that it associates with measured or observed.
j
Spatiotemporal data analysis books include Chistakos G (Spatiotemporal Random Fields, Elsevier, the Netherlands,
2017; Modern Spatiotemporal Geostatistics. Oxford UP, NY, 2000); and Christakos G and Hristopulos DT (Spatiotemporal
Environmental Health Modeling, Kluwer Acad, MA, 1998).
k
The term “chronotopologic” includes the cases of space-time and spacetime data variability.
8 1. Chronotopologic data analysis
Example 1.5 While CTDA concepts and tech- information exists individually on each of these
niques can be implemented to study a large items, which must be brought together to assure
variety of natural attributes (Example 1.2 ear- consistency in concepts and methods. CTDA pro-
lier), they may not be necessarily useful when vides some initial principles and notions, as
points represent merely the presence of events follows.
(e.g., earthquake occurrence), people, or some
• The chronotopologic unit is the basic
physical object (e.g., volcanoes, buildings).
building block for the CTDA of space- and
CTDA methods are based on a rigorous prob- time-specific natural attributes.
lem definition, which is then used to formalize a
The choice of the chronotopologic unit should
CTDA protocol in order to answer the questions
be based on a relevant set of criteria (physical,
of interest. In this respect, CTDA is a systematic
ecological, social, etc.) and it may affect the
inquiry to produce new knowledge, refine or
CTDA results.
validate the existing knowledge regarding the
natural attributes considered, and generate attri- Example 1.7 A chronotopologic unit may be
bute estimates. associated with observable characteristics of ter-
restrial, marine and atmospheric domains. It
• The interdisciplinary mantra of CTDA is to
may correspond to the smallest or elementary
not work in silos and to value that which is
working unit that is representative of these
cross-cutting.
characteristics.
In this setting, CTDA surely benefits from
Information on natural attributes may be
developments in TGIS technology. This tech-
available in several different scales (say, city,
nology allows the chronotopologic visualiza-
county and state), and an adequate CTDA may
tion of a variety of attributes such as
require conveying information from one scale
individual populations, quality of life indices,
to another (see, also, Section 4 of Chapter 3). This
pollutant distribution, disease spread, and
also includes the methods of transferring infor-
company sales over time in the region of inter-
mation from one chronotopologic point to
est. To achieve that, it is enough to have a
another.
chronotopologic database, and the TGIS is
capable of presenting an animated set of col- • Chronotopologic scaling is the process of
ored maps that provide informative visualiza- conveying information from one chrono-
tions of the chronotopologic pattern of the topologic scale to another.
attributes of interest.
Scaling may take different forms: Down-
Example 1.6 CTDA applications currently scaling (conveying information from larger to
span many disciplines, with their methods vary- smaller chronotopologic domains), upscaling
ing in relation to the specific questions being (conveying information from smaller to larger
addressed, whether predicting air pollution, domains), and transferring (conveying informa-
examining suspiciously high frequencies of dis- tion from one point to another).
ease events, or handling the vast data volumes
Example 1.8 A general existing ecosystem
being generated by the Global Positioning System
data analysis framework operates at the national
(GPS) and Satellite Remote Sensing (SRS).
level, whereas a site-specific study may require a
much finer chronotopologic scale. Investigators
1.5.2 Generic CTDA notions need to understand when and how to convey
CTDA is a broad framework for studying natu- information from the point (e.g., pollution mon-
ral attribute distributions, linking attribute coordi- itoring) to the chronoregion (e.g., city popula-
nates, in-situ conditions and values. Considerable tion exposure) scale.
2 Chronotopologic variability, dependency and uncertainty 9
Adequate information gathering is important be kept in mind that this is an instructional book
in understanding CTDA results as well as for that relies on sound theory, scientific reasoning,
communication purposes. and substantive knowledge, rather than on mere
data massaging techniques and popular quick
• Chronotopologic aggregation is the process
fixes in challenging times.
of reducing certain measures characterizing
the chronotopologic attribute distribution
into simpler ones.
2 Chronotopologic variability,
Aggregation may refer to chronotopologic units dependency and uncertainty
or the corresponding natural attribute values. Ade-
quate aggregation relies on a satisfactory under- As CTDA techniques become powerful tools
standing of the underlying mechanisms, inter- in many fields (earth sciences, oceanography,
relationships, and their relative importance. These geography, natural resource management,
measures may be the same across different scales, biological conservation, and societal processes),
or they may differ between scales, in which case adequate quantitative characterizations of good
the appropriate links between them need to be quality chronotopologic datasets of natural attri-
established. Also, different CTDA models and butes are increasingly needed.
techniques may have their own chronotopologic
units, scaling notions, and aggregation tools.
2.1 Conceptual chronotopology
Example 1.9 Aggregation may consist in com-
bining several data variability measures into a Accordingly, some further comments con-
simpler set of measures. Aggregation may take cerning the notions of chronotopology and its
the form of indicators, indices, summary statis- geometric dynamics are in order at this point.
tics, and graphs. Disparate measures may be We start with the following two key notions:
aggregated using a reference state or a common
factor, such as unit of measurement.
Chronotopologic variability refers to the
degree of the attribute’s joint change (fluctuation)
1.5.3 Fourfold CTDA objectives
across space and time, whereas chronotopologic
In light of the above deliberations, we chose dependency refers to the degree of the attribute’s
to conclude this first section of the book by mak- joint smoothness (connectivity) across space
ing some suggestions regarding the main objec- and time.
tives of an adequate CTDA study.
Example 2.1 When studying air temperature that the attribute (e.g., chemical concentrations)
data all over the globe during 150 years in order can assume any value within a specified range
to describe global climate change (e.g., Jones to be distinguished from discrete attribute values
et al., 1986), one could calculate temperature that can be counted (e.g., number of days an indi-
averages at a given moment (say, day), but these vidual has been exposed to a pollutant) and can
numbers would not take into account the natural take a finite number of possible values. Although
chronotopologic variability of these data (at any chronotopologically continuous datasets play a
given moment, the range of temperatures pivotal role in investigations in the fields of earth
around the world is huge due to different local monitoring, ocean research, environmental
climates, seasons, topography, whereas no tem- risk assessment and planning, public health
perature data exist over large portions of the management and decision-making, they are
oceans). In addition to the standard temperature usually not readily available and are often
datasets, recent studies of global warming relied difficult and/or expensive to acquire (e.g., in
on many other data sources, including data on mountainous and deep marine regions).
the timing of the first appearance of tree buds
in spring, greenhouse gas concentrations in the
Example 2.2 The following real-world situa-
atmosphere, and measurements of isotopes of
tions illustrate the significance of continuous
oxygen and hydrogen from ice cores (Egger
datasets and the need for CTDA techniques that
and Carpi, 2011). The study of such diverse
can effectively compensate for their lack.
and multisourced datasets is essentially
• Environmental managers require chrono-
interdisciplinary, i.e., it requires the integrated
topologically continuous data over a region
efforts of scientists with a variety of backgrounds
to make effective and informed decisions
and expertise, see the report of the Intergovern-
concerning the prevention and control of
mental Panel on Climate Change (IPCC, 2007).
severe pollution, and scientists need
The profound inverse relationship that exists accurate data that are well-distributed in a
between the notion of variability and that of chronotopologic domain to make justified
dependency is described by the following interpretations.
relationship: • The marine environment in Australia is a
typical case where seabed mapping, habitat
• Complementarity relationship: the presence of
classification, and prediction of marine
higher (lower) attribute variability in a
biodiversity, essential for marine biodiversity
chronotopologic domain would mean weaker
conservation, need reliable continuous data
(stronger) attribute dependency in the domain.
of the marine environment. Yet, in most of
For reasons that will become evident later, the Australian marine regions such data are
complementarity assumes a key role in CTDA not available, and only sparsely and
practice. Another essential property of chrono- unevenly scattered point samples have been
topology is as follows: collected. Hence, chronotopologic
interpolation techniques (Section 4) are
• Dataset continuity: in principle, data can be
essential for computing biophysical variables
considered at any point of the chrono-
at the unsampled locations.
topologic domain.
• Botanists collect information that is discrete at
It should be noticed that data continuity in the a particular set of sampling locations and
above sense refers to the attribute chronotopol- time instances, but has to be converted in
ogy rather than to the attribute value per se. In some way into useful dynamic map
the latter case, attribute value continuity means representations. To achieve this goal one
2 Chronotopologic variability, dependency and uncertainty 11
needs to manipulate attribute data in a way
that accurately estimates attribute values in
chronotopologic domains that have not
been sampled. In essence, one needs from
the information available to produce
continuous maps at unknown domains.
l
Usually, n ¼ 2 or 3.
m
In the same CTDA setting, “location” is a topologic entity, whereas “point” is a chronotopologic entity.
12 1. Chronotopologic data analysis
variability (or dependency) of the natural to the well-known 1st law of geographical sci-
attribute X(p) or X(s, t) in a rigorous and inter- ence, which consists of an intuitively
pretable manner. As it turns out, the following rather profound statement by Waldo R. Tobler
mathematical operator plays a pivotal role in Tobler (1970):
CTDA.
@ Everything is related to everything else, but
near things are more related than distant
Quantitative models of attribute variability things.
(dependency) represent dynamical (time-
dependent) processes across space described by As CTDA shows, this intuitively appealing
a mathematical operator S[X(pi), X(pj)] for any empirical statement does not apply to all
points pi and pj. natural attributes, including those attributes
exhibiting periodic or wave-like variability,
Based on the form of the S-operator,n these where attribute values at longer distances and
variability (dependency) models have different separations apart may be substantively more
properties that guide their implementation in related than values at close distances and
practice. Accordingly, some of them are data- separations.
depended whereas some others are data-free.
We postpone a detailed consideration of the Example 2.4 Two typical cases of this situation
attribute variability (dependency) models until are shown in Figs. 2.2 and 2.3.
Chapters 4–6.o • Fig. 2.2 (also, Example 3.21, Section 3 of
Chapter 5) shows the variogram γ X of sea
2.2.1 Revisiting Tobler’s law surface heights that exhibits a wave pattern
Remarkably, CTDA analysis may generate (the variogram is a function that measures
some interesting albeit sometimes unexpected attribute variability across space and time,
results. For illustration, one such result is related Chapter 5). As is determined by the
FIG. 2.2 The variogram of sea surface heights exhibiting a wave variation pattern.
n
For technical details, see Eqs. (2.5a–2.5b), Section 2 of Chapter 5.
o
The covariance and variogram are the mainstream chonotopologic variability models expressed by the S-operator.
Additional models include the trivariance, contingogram and sysketogram.
p
As will be discussed in Chapter 5, the lower the variogram value the higher the chronotopologic correlation
(dependency) between the attribute values at the corresponding pair of points.
2 Chronotopologic variability, dependency and uncertainty 13
FIG. 2.3 Data points pi (i ¼ 1, …, 6) and interpolation point p7. The attribute value is indicated by the color level of the dot
according to the scale legend on the right. The triplet within a parenthesis denotes coordinates pi ¼ (si1, si2; ti), and the italic
number is the contribution of datum at pi to interpolation at p7. The contribution of p6 is much larger than those of p1, p2, p4 and
p5 that are farther away from p7; but the contribution of p3 is larger than those of p4 and p6, which are both closer to p7 (this is
called the “screen effect”).
variogram γ X values,p the sea surface heights contribution, whereas a negative value
at a space-time lag represented by represents a negative contribution, and the
C (distance 5 m and separation 10 s) with interpolated value was obtained using the
γ X ¼ 0.9614 m2 are more closely related than space-time ordinary Kriging interpolation
heights at a lag B (distance 10 m and technique (to be discussed in Chapter 8). As
separation 5 s) with γ X ¼ 0.9945 m2, and, in indicated by these numbers, samples that
turn, heights at a space-time lag B are more are further away from the attribute
closely related than heights at a lag interpolation point may have a greater
A (distance 5 m and separation 5 s) with influence on the interpolated value than
γ X ¼ 1.6160 m2. points much closer to the interpolation
• Chronotopologic interpolation would assign point.
bigger weights to faraway points than to
near ones.q This is, in fact, the case of Fig. Rather not surprisingly, then, these examples
2.3 that represents the so-called screen effect, seem to have illustrated numerically the follow-
where pi (i ¼ 1, …, 6) denote data points ing assertion:
and p7 the interpolated point (the triplet of
numbers within the parentheses indicate • The first half of the empirical Tobler’s law
space-time coordinates (si1, si2; ti)). The italic (i.e., “Everything is related to everything
numbers represent the contribution of each else”) is valid more often in practice than
datum at pi to the interpolated value p7. the second half (i.e., “near things are more
A positive value represents a positive related than distant things”).
q
These violations of Tobler’s law actually may be the starting point of valuable discoveries. After all, the progress of
any scientific field is based on the derivation of unexpected and previously unknown results.
14 1. Chronotopologic data analysis
As it turns out, this assertion points out unique definition of uncertainty, which is a mul-
another significant issue that is worth revisiting tifaceted notion (ontic, epistemic, or technical),
in later chapters. which may emerge in a number of distinct ways
and may refer to different things.
2.3 Concerning real-world uncertainty Given, then, that there is considerable uncer-
and its probabilistic description tainty about the definition of uncertainty itself,
In addition to chronotopologic attribute vari- the associated probability notion may refer to
ability, another key characteristic of CTDA is the more than one consideration:
frequent lack of in-situ certainty. Every investi- ❶ Probability as a mathematical quantity: a
gator and practitioner of almost every scientific number belonging to the range between 0
field can testify to this fact. Otherwise said, and 1 (or 0 and 100%) expressing the
humans do not seem capable of the kind of last- frequency of occurrence of possible
ing immunity from doubt that the ideal of cer- attribute values.
tainty requires. ❷ Probability as a state of mind: a state of
Example 2.5 When studying changes in global doubt or lack of certainty about the true
mean surface air temperature from 1861 to 1984, attribute value, or a state of incomplete
Jones et al. (1986) used processing techniques to understanding of the phenomenon, or a
correct for uncertainties and inconsistencies in state of surprise regarding the outcomes of
the historical data not related to climate. As they experiments or future events.
notice, “Sea surface temperatures were mea- ❸ Probability as an irreducible (inherent)
sured using water collected in uninsulated, can- property of the natural phenomenon:
vas buckets, while more recent data come either representing, e.g., Heisenberg’s uncertainty
from insulated bucket or cooling water intake principle of phenomena in microscales.
measurements, with the latter considered to be
0.3-0.7 °C warmer than uninsulated bucket Example 2.6 Uncertainty in science is closely
measurements.” Accounting for these uncer- related to the fact that most data have a range
tainties is a complicated matter, since most sea of possible values as opposed to a precise single
surface temperature data do not include a value). It might be enlightening to view matters
description of what kind of bucket was used. from a layman’s perspective. The layman’s clas-
Moreover, the air temperature measurements sification of the laws of uncertainty rather distin-
over the ocean were obtained aboard ships of guishes between four kinds, as described in
unknown type and size. But ship type and size Table 2.1.
determined the height at which measurements
were obtained, which introduced additional TABLE 2.1 Layman’s laws of uncertainty.
measurement uncertainties (temperatures can Law Description
change rapidly with height above the ocean).
Inevitability Almost every investigation step contains a
What is required, then, is a comprehensible factor of uncertainty, it’s an inevitable part
and widely applicable notion of uncertainty. of doing science in the real world
As it turns out, however, this is not necessarily Invisibility The absence of evidence is not evidence of
the case in the real world. absence
Obviousness In many data collections, the figure most
2.3.1 Multifaceted notion
obviously correct could be the mistake
In fact, what makes matters of uncertainty so
Interpretation No matter what the result, there is always
complicated is that the following argument can someone who could misinterpret it
be made with actual certainty: There is not a
2 Chronotopologic variability, dependency and uncertainty 15
Yet, despite the considerable and multi-
faceted difficulties caused by the above aspects A notion of practical certainty would be
of uncertainty, it is widely acknowledged that appropriate (a) when the supporting argumenta-
to understand anything about any phenomenon tion is rational or justified enough to be relied
of nature, one must understand its uncertainty. upon in practical deliberation and in deciding
Accordingly, the next suggestion is a reasonable what to do, or (b) when the stakes are low,
one: whereas a more rigorous or stricter uncertainty
notion may be appropriate when the stakes are
higher.
• Uncertainty-related assumptions about a
phenomenon should be made on the basis
of rational reasons, i.e., they should be In CTDA, one can distinguish between
justified assumptions. temporal vs. intemporal practical uncertainty
(i.e., certainty at a specific time vs. lasting cer-
Rational reasoning shows that data uncer- tainty), as well as, between context-dependent
tainty does not necessarily imply that the data vs. context-independent practical certainty.
are wrong and useless. Quite the contrary,
Example 2.8 While in certain applications it is
rigorous uncertainty quantification becomes
often sufficient to focus on the conditions
very important in this respect, since the magni-
required for a phenomenon to be certain at a
tude of uncertainty offers an assessment of how
time or within a certain context, in other applica-
confident the investigators would be on the
tions, it is recognized that a phenomenon might
accuracy of their data (i.e., higher uncertainty
be certain at one time (when a clear and distinct
implies lower confidence to the data).
perception of the phenomenon has been estab-
Example 2.7 While the uncertainty level of lished) or within a particular context but not at
historical air temperature data increases going another time (when one no longer has or is not
further back in time, it has been reduced consid- attending to this clear and distinct perception)
erably, for all practical purposes, in more recent or within a different context.
years starting around 1900 (Mann et al., 1998). In
another case, although one cannot specify the 2.3.3 Sources of uncertainty
exact amount of precipitation during next
Real-world case studies encounter consider-
month, nevertheless, one can assume, as practi-
able uncertainties from different sources, which
cally certain, that there will be no major weather
is why CTDA studies put great emphasis on
pattern changes. Hence, the known weather
the reliability and quality of the data used.
conditions of previous years provide good
A list of key uncertainty sources is given in
grounds for reasonably assuming a certain prob-
Fig. 2.4, and a brief review of them follows
ability rainfall distribution.
(see below items ❶ through ❹).
❶ One of the most common sources is in-situ
2.3.2 The limited notion of practical data uncertainty, which is linked to
certainty experimental, observation and interpretation
Surely, the level of rigor associated with errors, heterogeneity and scale variation, as
uncertainty determination may also depend on well as to limited or sparse data, see
the study conditions and goals. In this respect, Example 2.1 of Section 2.
the following notion could be useful in ❷ One of the most important sources is model
applications: (or conceptual) uncertainty, i.e., whether or
16 1. Chronotopologic data analysis
not a mathematical model used by CTDA is satellite field models do not fully capture the
an adequate representation of the real ionospheric processes, particularly in the auro-
phenomenon. ral region; Finlay et al., 2017).
Model uncertainty may be linked to the ontic ❸ Computational uncertainty could be the result
condition, which refers to an inherent feature of of imperfection of a numerical technique and
the natural phenomenon, and the epistemicr the finiteness of machine precision.
condition, which refers to incomplete under-
Example 2.11 Computational uncertainty may
standing of the phenomenon.
characterize the sensitivity of a numerical prob-
Example 2.9 This kind of uncertainty may be lem solution taking into account the errors and
due to imperfections and idealizations (due to inaccuracies of inputs, algorithms, truncations,
ontic or epistemic conditions) made in the con- and roundings.
struction of a model representing a physical
phenomenon, and to the choices of the relevant ❹ Parameter uncertainty quantifies the lack of
probability functions to describe statistically the knowledge about the actual value of a
model parameters. parameter, and is linked to data
erraticness, physical properties, process
Altera pars, perhaps not surprisingly, there are optimization, design variables, rate
several CTDA studies in which uncertainty is coefficients, and unknown boundary and
linked to the opposite situation, as follows: initial conditions (BIC).
• Uncertainty may be linked to an unmodeled Example 2.12 Parameters that are inputs to
source that often can cause errors and biases computer models are uncertain when their exact
that are chronotopologically correlated. values can neither be computed in physical
experiments nor be inferred by statistical tech-
Example 2.10 Geomagnetic field studies con- niques. When more than one parameter is
sider both measurement errors, i.e., instrument involved, their uncertainties are correlated and
accuracy (see data uncertainties from the Swarm their proper merging is needed to fully assess
mission; Tøffner-Clausen et al., 2016) and errors uncertainty in an attribute derived from more
from unmodeled sources (even the most modern than one parameter.
r
The branch of philosophy that concerns itself with knowledge is called epistemology, or methodology, or philosophy
of science. It is the focus of some important developments in CTDA (see, also, subsequent chapters).
2 Chronotopologic variability, dependency and uncertainty 17
s
We will revisit each one of these three CTDA components, at various parts of the book.
18 1. Chronotopologic data analysis
TABLE 2.2 Probability interpretations. The above definitions emphasize both the clear
Probability
distinction between theory and evidence about the
interpretation Contextual description natural attribute of interest and their strong com-
plementarity. Another interesting distinction
Classical The ratio of cases in which an attribute
should be made between theory, which is about
occurs over all possible cases
predicting what has not been observed yet, and
Frequentist The value of classical ratio when inference, which is based on evidence alone and
number of cases is very large (tends to
cannot exceed the domain of the data it uses.
infinity)
As regards scientific theories, they typically
Geometric The ratio of the domain in which an develop gradually over time into increasingly com-
attribute occurs over the entire domain
plex models representing concentrated human
of interest
knowledge that often spans several centuries.
Subjective The result of personal experience,
preferences and intuition Example 3.1 In physics, the original Newton’s
theory was advanced considerably during the
Scientific The result of rational assessment based
on logical reasoning and scientific following three centuries, through the works
knowledge of Bohr, Einstein, Heisenberg, Higgs, and many
others to form today’s Standard Model.
more detailed discussion of the quantitative This gradual increase in complexity may be
uncertainty models employed in scientific due to various reasons, i.e., as the understanding
applications. of the phenomenon improves, new variables
need to be included in the theory, thus increas-
ing its complexity, or, a phenomenon may man-
3 Theory and evidence ifest only at a higher level of complexity and not
at lower levels, or, a theory that is valid in one
A chief objective of CTDA is to adequately use
scale maybe not so in a different scale. On the
the existing body of knowledge in a specified
other hand, a deeper physical understanding
area of science in an efficient and systematic
can lead to theory simplification, i.e., by omit-
way in order to better understand a phenome-
ting variables that play a minor role in the phe-
non and make predictions. Let us start with a
nomenon, or certain correlations are found to
brief yet concise description of what constitutes
have negligible effects.
a “theory” and what constitutes “evidence”.
t
Classical examples of theories are Coulomb’s law of electrical attraction and repulsion, Maxwell’s electromagnetic
equations, and the ideal gas equations describing relations between temperatures and pressures of enclosed gases.
3 Theory and evidence 19
In the words of Auguste Comte: conclusions drawn. Yet, reality is more compli-
cated and lies beyond the convenient slogan
@ If it is true that every theory must be based
above.
upon observed facts, it is equally true that
facts cannot be observed without the
guidance of some theory. Without such A dataset has a natural “soul” that needs to be
guidance, our facts would be desultory and explored strictly within the specified phenome-
fruitless; we could not retain them: for the nal context so that theory-driven (content-
most part, we could not even perceive them. dependent) conclusions can be drawn.
Indeed, it is a theory that can take data anal- absent. Similarly, when one attempts to forecast
ysis beyond mere numerics and assign the nec- future developments in financial affairs or politics,
essary meaning to the data and avoid common future evidence is, indeed, a contradiction in
numerical traps like spurious data correlations. terms, and one has no other option than to theorize
Also, theoretical developments may be more or hypothesize about possible future scenarios.
captivating than empirical evidence in cases
where randomization and replication are diffi- 3.4 Synergy
cult and several confounders exist.
The crux of the matter is that to understand
the world, one needs both theory and evidence.
One should take advantage of the available evi-
3.3 Evidence unavailability
dence in all its forms (traditional and emerging)
It is not surprising that the necessity of theory but without ignoring the treasure of core knowl-
in many real-world investigations is also estab- edge accumulated in scientific theories. And, the
lished on the basis of practical evidence-related deeper meaning of the distinct yet complemen-
reasons. Indeed, people need to understand the tary features of theory and evidence discussed
world, and when sufficient empirical evidence above can be summarized as follows:
is not available, they naturally tend to rely on
theory.
Beyond the distinct yet crucial roles theory and
evidence individually play in scientific inves-
tigation, perhaps even more important is the real-
Theoretical developments may be necessary ization that they act in synergy, i.e., there is an
because of unavailable evidence, i.e., in many intrinsic and essential relationship between
real-world cases, evidence is not available at the the two.
current state.
Some key features of theory and evidence,
together with some comments concerning their
Example 3.3 Evidence concerning the future interplay, are presented in Table 3.1.
occurrence of a phenomenon is usually unavail-
Example 3.5 Interesting cases of the synergy
able, and empirically inaccessible attribute
between theory and evidence in two different
domains may exist.
scientific disciplines are as follows:
As a matter of fact, logically the notion of • In modern health sciences, there is an urgent
“future evidence” may be a contradiction in need to combine evidence bases with
terms, since the “evidence” expresses a current theoretical and computational models in
state (i.e., one definitely talks about “current evi- order to adequately investigate the huge
dence”), whereas the future refers to an unrea- amounts of data from healthcare and
lizable yet state (which may or may not be medical research.
realized in the future, but this is not determined • In earth sciences, many applications merge
with certainty). the information contained into geophysical
observations with that coming from
Example 3.4 When investigators study the
dynamical models, a process often known
weather conditions at a future time, they need to
as assimilation.
rely heavily on theoretical developments in the
form of complicated weather models and hypoth- Although theory develops before any evi-
eses, surely supported by past evidence, whereas dence becomes available in the form of data,
direct evidence of future events is profoundly yet it can be properly modified (updated or
3 Theory and evidence 21
Theory is a scientific and systematic description of Evidence is obtained by observing natural and
knowledge that usually relies on a set of assumptions about experimentally generated events, objects, and effects
the phenomenon of interest
Theory is customarily represented as a collection of Evidence sources are theory-laden observations and research
sentences, propositions, beliefs and formulas, and their findings. If evidence is separated from theory, it is
logical consequences disconnected from its origin and it may become devoid of
scientific meaning
Theory uses evidence (real data) to calibrate its parameters Evidence-based practice applies the most credible and
reliable evidence available in order to understand and
describe a phenomenon
Theory and theorizing are cornerstones of a scientific Evidence is a conceptual term, and, thus, it is linked to
discipline. Research findings and conclusions are usually knowledge fields that are conceptual. As such, the evidence
presented in the form of theories and hypotheses does not look the same to observers with different conceptual
resourcesa
A sound theory helps avoid both unanchored abstract Evidence depends on the relevant context. By changing the
thought and a narrow empiricism that concerns itself only context, the notion of evidence may change and even become
with observation or measurement and unidirected data problematicb
gathering
A theory can be challenged and defended in different ways Very few investigators would be convinced by a study
and at different times as new empirical evidence and drawing an apparently interesting conclusion without a
investigative techniques are introduced plausible, at least, theory adequately fitting the relevant data
a
For example, to measure soil porosity one associates it with an appropriate concept; and proponents of a caloric account of heat do not understand heat
experiment results in the same way as those who view heat in terms of mean kinetic energy or radiation.
b
For example, if dark clouds are observed in the sky, there is a high likelihood of rain, but if dark clouds are observed when the temperature is far below freezing,
there may be no rain.
calibrated) when new and reliable evidence is constructed and the way the evidence is
emerges in the process of scientific inquiry. obtained, including the type of evidence
Actually, this seems to be an appropriate proce- that counts in scientific reasoning.
dure for combining theory with evidence. In
If a theory suggests that the current evidence is
practice, theory and evidence should rest com-
inadequate, it implies that more research on that
fortably with each other and they should also
specific subject is necessary, whereas when hard
reinforce each other (detecting data correlations
data contradict a theory or the theory cannot
facilitates theory testing and refinement; and
explain it, the theory must be revised accordingly.
combined with the increasing computing
Naturally, theory and practice have not been
power, the data may suggest different ways of
developed to the same degree in all disciplines.
developing a theory).
Example 3.6 Weather forecasting is a disci-
• When significant differences are found pline that is both theory- and practice-strong,
between theory and evidence, they can lead whereas public health is considered a theory-
to improvements in both the way a theory weak and practice-strong discipline.
22 1. Chronotopologic data analysis
• That the law of gravity is not perfectly Remarkably, the kind of “quality control”
reliable, because of wind currents and introduced by the accuracy indicators is not part
other intervening factors, is not an excuse of deterministic interpolation (see, e.g., the inverse
for one to ignore its predictions and decide distance techniques to be discussed in Chapter 7).
to jump from a ravine. Any common- Moreover, some techniques do not take into
sensical human being understands that account intrinsic properties of the interpolated
despite its imperfection, the law remains phenomena (e.g., rapid changes in the data with
reasonably valid for all practical purposes. distance vs. more gradual changes, or law-
• Similarly, to decline to make critical climate obeying variations), as they only take account of
change estimates on the grounds that the the location/instance of the data points. The
data is imperfect (as many investigators CTDA techniques to be discussed in the book will
have argued) would probably not be a good try to address the above and similar issues.
decision, especially given the gravity of the
matter. As it turns out, investigators can
indeed draw sufficiently informative 4.3 Interpretation
inferences with the evidence we have. We conclude with the following comment
concerning the interpretation of the CTDA out-
4.2 Confidence and accuracy comes, like the chronotopologic attribute
maps.
On a relevant note, since chronotopologic esti-
mation is subject to error, it is necessary to assess
how much uncertainty can be tolerated and how In general, when investigators interpret the
much confidence can be placed in the generated attribute maps, they attempt to explain the uncov-
attribute estimates. Arthur Rudolph, the scientist ered patterns and trends, bringing all of their
who developed the Saturn V moon rocket, background knowledge, experience and skills to
expressed this critical issue very succinctly: bear on the question and relating their data to
existing scientific ideas.
@ You want a valve that doesn’t leak, and you
try everything possible to develop one. But
the real world provides you with a leaky Example 4.2 Weather forecasting is an
valve. Then, you have to determine how interpretation made by a meteorologist of data
much leaking you can tolerate. collected by satellites using climate models.
On practical grounds, then, the need for toler- Given the personal nature of the knowledge
ance and confidence assessment leads to the the investigators rely upon, interpretation is
development of several estimation accuracy indica- bound to be subjective to some extent, in the
tors, which are absolutely necessary given the data sense that by bringing a different background,
heterogeneity and uncertainty (Section 2). After thinking mode and experience to bear on the
all, a general motto of scientific investigations is: same dataset, an investigator can come to very
different conclusions. A widely publicized case
Being aware of how much one does not know in point is the global climate debate concerning
is often more important than being cognizant of the presence or absence of a temperature trend
how much one knows. in time, which is the focus of the following
example.
24 1. Chronotopologic data analysis
Example 4.3 The study of global warming ❶ Pure chronogeographic statistics techniques
turned out to be an arena of strong disagree- (point center, diffusion, inverse metrics, etc.;
ments concerning data analysis and interpreta- Chapters 4 and 7). They are concerned only
tion. Jones et al. (1986) interpreted the changes with the chronotopological characteristics
in global mean surface air temperature from (coordinates, distances, instances) of the
1861 to 1984 to imply a long-term warming trend phenomenon.
(e.g., they emphasized that 1980, 1981, and 1983 ❷ Attribute chronogeographic statistics
were the three warmest years in the entire techniques (Moran’s index, Geary’s index,
dataset). On the other hand, Lindzen (1990) etc.; Chapters 4, 7, and 12). They are
opposed this interpretation using a number of concerned both with the chronotopological
arguments, like the inadequacy of the existing characteristics of the phenomenon and its
data collection network to correct for data uncer- physical values (mass, weight, temperature,
tainty, substantial gaps in temperature coverage concentration, etc.). They include multiple
(especially over the oceans), errors in data anal- linear regression (MLR), land-use regression
ysis, and inappropriate interpretation of the (LUR), and linear mixed-effect models.
global mean temperatures, thus, reaching the ❸ Standard geostatistics techniques (ordinary
opposite conclusion that there is no trend in and simple Kriging, statistical regression,
the data. This is a classical case where different etc.; Chapters 5 and 8). They are popular
investigators made very different interpreta- techniques that focus on hard (uncertainty-
tions and drew very different conclusions from free) data, and they usually rely on rather
the same datasets, as a result of their different restrictive modeling assumptions (linear
backgrounds, experiences, and thinking modes. estimators, Gaussian data distribution,
low-order statistics).
Ut aequum, disagreement is not uncommon in
❹ Modern geostatistics techniques (Bayesian
scientific investigations. When it occurs, it rou-
maximum entropy, stochastic logic,
tinely leads to more data collection and analysis,
factoras, etc.; Chapters 6 and 9). They are
but perhaps more importantly it should also
very versatile, incorporate core knowledge,
lead to the consideration and appreciation of
assimilate both hard and soft (uncertain)
the distinct thinking modes used by the different
data, and avoid the restrictive modeling
investigators.
assumptions of the standard techniques
(nonlinear estimators and non-Gaussian
data distribution are automatically
5 A review of CTDA techniques considered).
❺ Machine learning techniques (artificial
As already noted, there exist several CTDA
neural networks, random forest, gradient
techniques (both S-TDA and STDA) available
boost machine, etc.; Chapter 12). They are
in the relevant literature. Naturally, each tech-
often characterized as “black-box,”
nique has its own set of specific features and
because they do not involve confirmed
assumptions.
mathematical expressions representing the
substance of the natural attribute of
5.1 Classification interest. Yet, due to their high
The CTDA techniques to be discussed in this nonlinearity, self-adaptation, self-learning
book, in varying degrees of detail, belong to the and other features, they can fit very well
following distinct groups: complex real world processes.
5 A review of CTDA techniques 25
Each group of CTDA techniques has its pros attribute-specific. In fact, there exist several
and cons. Although many factors affect a tech- chronotopologic estimation techniques available
nique (including sample size, scale, sampling in the relevant literature (mainly interpolation
design, data properties, and law of change), rather than extrapolation techniques, Section 5
there are not always consistent findings about above).
how exactly these factors affect the technique’s Interestingly, of the several changes in the
performance (accuracy, informativeness, sub- meaning of the term CTDA over the years, a cer-
stantiveness). Therefore, it may not be always tain proportion is due to a tendency of accompa-
an easy task to select an appropriate chronoto- nying features to become gradually defining ones
pologic modeling and interpolation technique (in some cases in addition to the original defining
for a given input dataset. It is noteworthy that properties, and some others displacing the origi-
different data interpolation techniques can gen- nal defining properties). It is also possible that the
erate different maps that are interpreted as confusion between defining and accompanying
representing very different phenomena. properties could produce misconceptions that
In view of the classification introduced, it is stand in the way of a realistic view of the phe-
often methodologically useful to make a key nomenon. Misconceptions about the physical
distinction: world often originate in semantic shortsighted-
ness, but are usually liable to correction through
CTDA distinguishes between defining and observation.
accompanying properties of a class of techniques,
where the former refer to properties that hold for
any member of the class and the latter to proper- 5.2 Computer technology
ties that hold only for a certain subset of the class. Naturally, one could not underestimate the
important contribution of technology in
The test whether a property is a defining one computer-assisted CTDA. During the last few
or not consists of investigating whether its decades, CTDA underwent a profound transfor-
absence would exclude a technique from inclu- mation from being manual (and tedious) to pre-
sion in the specific CTDA class. Chronotopology dominantly computer-based (this was made
is a defining property of the CTDA class, possible mainly due to advances in computer
whereas uncertainty is an accompanying prop- hardware during the 1990s).
erty, since it holds for some members of the
• Computers play a major role in the
CTDA class (e.g., geostatistics, Chapter 8) but
development and implementation of CTDA
not for others (e.g., deterministic chronogeo-
techniques.
metric, Chapter 7). Depending on their defining
and accompanying properties, one may distin- Among the advantages of computer software
guish between techniques that are global vs. is that it makes CTDA much more efficient and
local, exact vs. inexact, and formal vs. substanti- effective, allowing improved numerical accu-
ve.u CTDA techniques that have been developed racy, higher speed, time-savings, and handling
for and applied to various disciplines (physical, complicated computations (see, also,
biological, and social) are often data-specific or Section 8). Technology advances of various
u
The distinction between a formal and a substantive technique is that the former considers data as numbers assigning
no physical meaning to them, whereas the latter views data as physical quantities assigning different interpretations to
them, depending on the phenomenon.
26 1. Chronotopologic data analysis
forms have made it easier for investigators to As such, visualization is enabling the act of
collect and organize evidence in ways that allow mapping in a very effective and efficient man-
systematic analysis and encourage worldwide ner. This brings to mind what Albert Einstein
collaboration among groups of CTDA experts once said about scientists’ ultimate goal:
and institutes.
We conclude our very brief review with a @ The natural scientists try to develop, in the
word of caution. Although the continuous fashion that suits him best, a simplified and
development of more powerful computers in intelligible picture of the world; he then
the past has facilitated improvements in real- tries to some extent to substitute this
world CTDA that would not have been possible cosmos of his for the world of experience,
otherwise, in the view of many scholars, this and thus to overcome it. This is also what
does not seem to necessarily be the case any- the painter, the poet, or the speculative
more. Instead, it has been argued that processors philosopher try to do.
do not continue getting faster at the previous In the present case, the “intelligible picture
pace; instead, it is hoped that the continuation of the world” is represented by the properly
of the steady increase in power will come from visualized attribute map. Surely, achieving
massively parallel machines. intelligibility in mapping depends on choosing
the kind of visualization tool that works best
6 Chronotopologic visualization given the study objectives. In this respect, mod-
ern visualization is a very powerful technology
technology
that provides a large collection of tools that
can generate chronotopologic maps of varying
After CTDA has been used to derive quantita-
levels of sophistication (contour maps, domain
tive data dependency measures and generate
coloring maps, animated maps, etc.). The idea
numerical attribute estimates from isolated data,
is not so much to follow a formula, but rather
the next task is usually to portray these results
to be aware of the many different aspects of
on a continuous chronotopologic domain. In
such a context, the following objective is of par- visualizing a chronotopologic map or other
CTDA results.
ticular interest:
Generally, experience has shown that visual-
ization is an incredibly effective medium for
Chronotopologic mapping is the act of repre- informing or conveying a viewpoint or a mes-
senting the numerical CTDA results, involving sage across. Indeed, it is widely acknowledged
data and attribute estimates, in a continuous that the informativeness and usefulness of
and visually informative way. CTDA results can be greatly improved by
means of powerful visualization techniques.
Given its objective above, mapping is closely More specifically, based on the attribute data
linked to the following form of information at the sampled points and the estimated attri-
communication: bute values at the unsampled points a detailed
and, hopefully, highly enlightening map can be
Visualization in the form of maps, graphs, visualized. The emphasis is on the visualization
charts, diagrams, and animation is communica- of attribute properties and relationships that
tion of the available CTDA information to the takes into account the chronotopologic localiza-
interested individuals or groups of individuals tion/instantanization of these properties and
accurately and efficiently. relationships.
7 The range of CTDA applications 27
6.1 The emergence of visual thinking clearly and also sets them in the appropriate
context that can potentially influence people’s
Humans are inherently visual. The human views and decisions. Therefore, a good visual-
brain has the ability to literally process and store ization technique should add to data’s meaning
visual information faster than text, symbols, and/or clarity.
numerics, and statistics. Visualization takes More about this interesting topic will be pre-
complex information that most people don’t sented in the following chapters. In the mean-
have the time, skills, or attention span to compre- time, let us note that increasingly sophisticated
hend, and transforms it into easily understood, and high-quality visualization technology
pleasant and visually focused presentations. In products become available nowadays that are
this setting, then, another important human often both affordable and easy to use. The inter-
cognition issue emerges that transcends mere ested readers are referred to the relevant visual-
technological considerations: ization literature (e.g., Tufte, 2001; Knaflic,
2015; Healy, 2018; Kirk, 2016; and references
The visualization of CTDA results requires the therein).
adaptation of a new way of thinking, viz. visual
thinking, and, accordingly, to develop a deep
7 The range of CTDA applications
understanding of what it means to think visually.
CTDA (in its S-TDA and STDA formulations)
The modern scientists are expected to be able
provides useful quantitative tools of chrono-
to think visually and use visualization technol-
topologic dependency characterization and
ogy to its full potential so that they can generate
mapping. Besides the characterization and
plots that allow them and others (the public,
visual perception of attribute distributions, it is
stakeholders, decision-makers) to grasp informa-
very important to translate the emerging chron-
tion in an efficient and physically meaningful
otopologic patterns into useful objective and
manner. Given that CTDA involves a variety of
measurable entities, which is why CTDA can
knowledge bases, core and specificatory, shared
help the investigators answer key questions
by many different individuals with different
linked to real-world problems of numerous
backgrounds (scientific and cultural), this has
disciplines, such as those listed in Table 7.1.
made visualization a very important common lan-
The list in Table 7.1 is by no means exhaustive –
guage. One then comes to appreciate the claim
the readers may propose many other CTDA
that if data is information (of various kinds) about
issues that are clearly relevant to their fields of
the world, then visualization is the creative com-
interest. On the other hand, it may come as no sur-
munication of this information. Therefore:
prise to our readers that many fields of inquiry
have contributed to CTDA’s rise in its modern
• A good visualization technique does not only
forms, including cartography and surveying,
need to be technically adequate and
botanical studies of global plant distributions
aesthetically pleasant, but contextually rich
and local plant locations, forestry, ethnological
and functional too.
surveys of population movement, human expo-
The first two requirements focus on the eyes sure investigations, landscape ecological studies
and follow certain presentation rules of visual of vegetation blocks, environmental studies of
grammar, whereas the last two requirements population dynamics, biogeography research,
focus on the mind and respond to context epidemiology, and disease spread assessment
(rather than set it). According to this perspec- combined with locational healthcare delivery
tive, visualization conveys facts and ideas data, and space-time econometrics.
28 1. Chronotopologic data analysis
Environmental science: Investigators need to know the • Does pollution exceed certain thresholds
chronotopologic distribution of pollutant concentrations • In what parts of a region, and when and how frequently
during peak days does threshold exceedance happen
• Which are the most anticipated public health effects
• Which are the most efficient prevention and control
measures
Earth sciences: Investigators need to derive estimates of a • What is the fluctuation of water table elevation in a region
wide range of physical attributes in large domains and during the summer months
during long time periods • How weather variables—such as temperature, air pressure,
relative humidity, precipitation, cloud cover, wind speed,
and direction—change from a region to the next, or from
current to future
Mining: Data analysts need to estimate from a usually • Can those samples be used to estimate the mineral
limited set of samples of different periods the extension of a distribution in that site or to forecast its deposit in the
mineral deposit in a site near future
• What is the accuracy of the estimates to be used in decision-
making and management
Agronomy: Scientists often must analyze a region for • How to choose the independent variables—soil, vegetation,
agricultural zoning purposes and geomorphology
• What is the dependency structure of such variables within
the domain
• How to determine what the contribution of each variable is
in order to define where the growth of each type of crop is
more adequate
Epidemiology: Health professionals collect data about • Does the distribution of disease cases develop a specific
disease occurrence chronotopologic pattern and which are its major features
• Is there any association with any pollution sources
• Do geographical disease patterns vary with time
• Should a state of emergency be declared, how long should
it last, and which regional hospitals should participate
Social studies: People would like to know whether there is • Are there any “hot spots” in this distribution
any chronotopologic concentration in regional theft • Are thefts that occur at certain regions or during specific
distribution seasons correlated to the socio-economic characteristics of
these regions
Example 8.2 The interface of the STAR-BME conceptualizations are activated. Thus,
software library is shown in Fig. 8.2. Briefly, the changes in symbolic structure or arrangement
process introduced by STAR-BME is as follows: may cause changes in conceptualization.
set the coordinate reference system, import hard A difference in the arrangement of the
and soft data, data detrending with Gaussian ker- symbols in an equation could imply a
nel or spatiotemporal mean (optional), spatiotem- different organization of the body of
poral covariance modeling and estimation, knowledge, which could, in turn, lead to
prediction points, BME prediction and different interpretations. Newton’s second law
visualizations. can be represented by two mathematically
equivalent equations, F ¼ mγ and γ ¼ mF , where
Ergo, a reasonable suggestion would be that
F is the total force on an object, m is its mass
the readers visit the above sites and decide
and γ its acceleration. Explain why although
which software library best fits their needs.
their mathematical formulations are the same,
Moreover, the users of the software tools
the conceptualization of the two equations
should make sure that the underlying assump-
could be quite different.
tions and limitations are well understood and
2. One can think of many natural quantities that
not hesitate to question their applicability in
exhibit chronotopologic variation, such as
the specific case study, if necessary. It is hoped
rainfall, vegetation, population density,
that the readers will be sufficiently encouraged
human mortality, economic wealth,
to apply one or more of these software tools into
unemployment and crime rates. Suggest
their CTDA data to make sense of theory
attributes in natural sciences that
and data.
(a) do not vary in space,
Concluding, the discussion in this chapter is a
(b) do not vary in time,
skeletal description of the various CTDA ele-
(c) vary neither in space nor in time.
ments and features, with the understanding that
3. Consider the chronotopologic coordinates
much of the ensuing discussion in Chapters 2–13
associated with:
will serve to flesh out the skeleton.
(a) the occurrence of volcano eruption,
(b) soil moisture samples,
9 Practice exercises (c) household density in the state of California,
(d) migration of swallows from Ireland to S.
1. How a physical law is presented by an Africa, and
equation matters in determining what (e) SARS disease incidences in East Asia.
9 Practice exercises 31
Which ones of the above can be studied by problems are more effectively addressed
the CTDA techniques? using __ techniques.”
4. In one sentence, what is CTDA all about and
what kinds of problems can it solve, in your 10. Compare the chronotopologic variability
view? and dependency of the temperature
5. Describe natural attributes in your field of distribution in the countryside vs. the
interest characterized by chronotopologic industrial city center.
dependence. 11. Design an investigation of the
6. Consider the following phenomena: chronotopologic variation of atmospheric
(a) the occurrence of your father’s birth, humidity in a geographical region during
(b) pollutant concentration samples, a time period of your choice. List sources
(c) poverty levels in North Carolina counties, that may generate uncertainty.
(d) the battle of Marathon, and 12. Provide some special situations or data that
(e) Plague mortality throughout India in the cannot be analyzed by CTDA techniques.
early 1900s. 13. List uncertainty sources associated with
Which ones of the above can be studied different types of sampling (random,
by CTDA techniques? regular, stratified, etc.).
7. With which ones of the following 14. Compare the variability of daily
descriptions do you agree, if any? temperature in May at a city of your
There are scientific fields that: choice vs. the variability of daily
(a) capitalize on chronotopologic depen- temperature during the entire
dence to estimate attribute values at calendar year.
unsampled points, 15. Compare the variability of daily humidity
(b) are concerned with the study of in May at a specific city of your country
phenomena that vary in a composite vs. the variability of daily humidity in
spacetime domain, May at several cities of your country.
(c) focus on numerical techniques that deal 16. Provide examples of attribute distributions
with chronotopology in conditions of that violate Tobler’s law of geographical
in-situ uncertainty, science.
(d) offer a way of describing the continuity 17. Write a brief review of CTDA techniques.
of natural phenomena and include 18. Which is the branch of philosophy that
adaptations of regression techniques concerns itself with knowledge, and why
to take advantage of this continuity. is it important for CTDA research
8. In what way do S-TDA and STDA purposes?
distinguish themselves from classical data 19. Do you consider implementing CTDA
analysis, and in what way are they techniques, and in what specific fields?
distinct from each other? 20. Should CTDA be concerned with the
9. Fill the empty spaces (__) in the following representation and analysis of common-
statement: sense knowledge or with solving
problems designed to test research-
“Compared to the __ statistics that merely motivated axiomatic theories, or with both?
examine the statistical distribution of a set 21. Like all types of human inquiry, CTDA too
of sampled data, __ incorporates both the
can be influenced by the underlying
statistical distribution of the samples and
the chronotopologic correlation among them. conceptual or methodological perspective.
Because of this difference, many real-world For illustration, Table 9.1 illustrates some
32 1. Chronotopologic data analysis
TABLE 9.1 Differences between the positivist and the interpretivist approach to human inquiry.
Issue Positivist perspective Interpretivist perspective
key differences between the positivist and defined as the practice of seeking
interpretivist approaches. What kind of knowledge, and following reason and
approach do you use in your research? evidence, even if it disagrees with a deep-
Justify. stated belief system. Justify.
22. Do you agree or not with the view that truth-
seeking is the critical thinking skill that is
C H A P T E R
2
Chronotopology theory
Whether the natural attribute of interest con- An operational chronotopology can achieve
cerns human populations or the physical envi- certain real-world key objectives: determining
ronment, chronotopology is of fundamental what kinds of natural attribute distributions exist,
importance. Historically, two distinct concep- assessing how they vary in-situ, describing the rel-
tions of chronotopology have been developed, evant body of knowledge, and classifying the
around which all the other CTDA notions attributes into different scientific disciplines.
revolve:
• The classical conception, in which the In this respect, some initial characterizations
chronotopologic domain serves as an arena are in order regarding the main constituents
for the natural phenomena to take place, of an operational chronotopology. These
and which is independent from the characterizations concern the correspondence
phenomena. between elements of the material (real or phys-
• The modern conception, according to which ical) world and these of the chronotopologic
the chronotopologic domain does not claim domain, as follows:
Quantitative Analysis and Modeling of Earth and Environmental Data 33 Copyright # 2022 Elsevier Inc. All rights reserved.
https://doi.org/10.1016/B978-0-12-816341-2.00012-5
34 2. Chronotopology theory
❶ A material pointa (or particle) in the domain that merely serves as an attribute
material (or real) world consisting of argument (location and instance of the
continuous matter and an immaterial occurrence of an attribute value).
(abstract or geometric) pointb representing
only position and time instance in the Example 1.1 Fig. 1.1A shows a point P and a
chronotopologic domain. body in the material world (consisting of a set
❷ A material body (or object) in the material of material points) that are mapped into a point
world and its abstract configuration in the p and a configuration, respectively, in the chron-
chronotopologic domain within which the otopologic domain, i.e., the configuration of the
configuration resides and moves. body is obtained by fixing the material points of
❸ A mobile point in the chronotopologic the body to abstract points on the chronotopolo-
domain representing the activity of a gic domain. The study of this representation (i.e.,
material particle in the material world and a material point or body residing and moving in
an immobile point in the chronotopologic the chronotopologic domain) is the focus of
FIG. 1.1 (A) Elements of the material (or real) world and their mapping in the chronotopologic domain; and (B) attribute
value and its mapping in the chronotopologic domain.
a
I.e., an entity having size (width, no length and no depth).
b
I.e., a sizeless entity in geometry having no size (no width, no length and no depth).
Another random document with
no related content on Scribd:
The Beggar.
By Jules Bastien-Lepage.
The more he become master of his brush, the more the rustic
work haunted him. He was still a thorough countryman. Although he
had now at intervals the refinements of elegance and little bursts of
worldliness; although he had exchanged the modest atelier in the
Impasse du Maine for a house in the Quartier Monceau, the world
soon wearied him, and he was glad to go back to his village.
This six weeks’ absence, of which he speaks in his letter to his
friend Baude, was spent in an excursion to Venice, and in
Switzerland. He came back only half delighted, and brought back
only a few unimportant sketches.
Italy and the splendours of Venetian art had left him cold. In this
world of history and mythology he was not at home. He sickened for
his meadows and his Meusian forests.
During his rapid visits to Paris in 1881 and 1882, the painting of
various portraits, notably that of Madame Juliette Drouet, and the
compulsory tax of visits and soirées occupied him almost entirely.
We saw but little of him. But these successes, and the adulation
lavished upon him in Parisian drawing-rooms, did not change him.
He was still the loyal, joyous comrade, faithful to old ties; very
good, very simple; happy as a child when he found himself in a circle
of intimate friends.
We were both members and even founders of an Alsace-Lorraine
dinner, the Dîner de l’Est, which was always given in summer in the
country. One of the last meetings at which he was present, took
place at the end of May, 1881.
A boat had been engaged, which was to take the diners to the
bridge at Suresnes, and to bring them back at night. When we
arrived at the landing-stage, a blind man was standing by the
footbridge, attended by a young girl, who held out her sebilla to the
passers-by.
“Come, gentlemen! all of you, put your hands in your pockets!”
gaily commanded Bastien, and he passed over first, preaching by
example. And the eighty, or a hundred guests of the Dîner de l’Est,
passed one after another over the footbridge, each one leaving in
the child’s sebilla a coin, large or small.
When we were on the deck, Bastien turned round to look at the
blind man and his girl, who were amazed at this unexpected windfall,
and were slowly counting their money.
“What a lovely group?” he said to me. “How I should like to paint
that child!”
While waiting for dinner we walked in the Bois de Boulogne. The
acacias and hawthorns were in flower. The lawns, newly shorn, gave
out a perfume of mown grass. Jules, joyfully drawing in this air
impregnated with country odours, laughed like a happy child.
At that moment all was going well with him. His Mendiant had had
a great success at the Salon; his last visit to England had been very
prosperous; his head was full of fine projects for pictures. “It is good
to be alive!” he exclaimed, as he played with a flower he had plucked
from the bushes…. On the way back he gave himself up to all sorts
of roguish fun. Mounted on the prow of the boat he sang, with his full
voice, the Chant du Départ.
The vibrating tones resounded powerfully between the two
sleeping river banks; the sky was splendid, twinkling with
innumerable stars. From time to time Bastien lighted a rocket and
sent it up overhead, shouting a loud hurrah!
The fusée mounted slowly into the night, showering down many-
coloured sparks, then fell suddenly and sank in the dark water. Alas!
it was the image of the short and brilliant years that remained for him
to live.
IV.