Professional Documents
Culture Documents
Frederik P Agterberg*
Geological Survey of Canada, 601 Booth Street, Ottawa K1A 0E8, Canada
ABSTRACT: The relationship between fractal point pattern modeling and statistical methods of pa-
rameter estimation in point-process modeling is reviewed. Statistical estimation of the cluster fractal
dimension by using Ripley’s K-function has advantages in comparison with the more commonly used
methods of box-counting and cluster fractal dimension estimation because it corrects for edge effects,
not only for rectangular study areas but also for study areas with curved boundaries determined by re-
gional geology. Application of box-counting to estimate the fractal dimension of point patterns has the
disadvantage that, in general, it is subject to relatively strong “roll-off” effects for smaller boxes. Point
patterns used for example in this paper are mainly for gold deposits in the Abitibi volcanic belt on the
Canadian Shield. Additionally, it is proposed that, worldwide, the local point patterns of podiform Cr,
volcanogenic massive sulphide and porphyry copper deposits, which are spatially distributed within ir-
regularly shaped favorable tracts, satisfy the fractal clustering model with similar fractal dimensions.
The problem of deposit size (metal tonnage) is also considered. Several examples are provided of cases
in which the Pareto distribution provides good results for the largest deposits in metal size-frequency
distribution modeling.
KEY WORDS: fractal point pattern, spatial statistics, roll-off effect, cluster dimension, Pareto distribu-
tion.
(Malamud et al., 1998) are non-linear and result in to be unfavorable for drilling but situated within the
fractals or multifractals. study area. These edge correction methods can equally
It is increasingly realized that various processes well be applied in fractal point process modeling. This
that took place millions of years ago during the geo- paper advocates the use of such methods, which ac-
logical past also were non-linear. Ore deposits result- count for edge effects, allow irregularly shaped study
ing from hydrothermal processes generally display areas and, generally, are more precise than various
geometric characteristics that have been known to be box-counting methods more commonly used for the
fractal or multifractal for a long time (Agterberg, 2012, statistical treatment of fractal point patterns.
1995; Cheng, 2008). The same applies to oil and gas A closely related topic not covered in this paper
pools (Barton and La Pointe, 1995). Non-linear proc- is that fractal point patterns can be multifractal. The
ess modeling in these fields has two practical applica- spatial distribution of gold mineral occurrences in the
tions. On one hand, it has resulted in new exploration Iskut River map sheet, northwestern British Columbia,
techniques for the discovery of new deposits; on the is multifractal instead of monofractal (Cheng and Ag-
other hand, it allows statistical modeling of the size- terberg, 1995; Cheng, 1994). Although theory of mul-
frequency distributions of populations of known tifractal point processes was developed in the latter
deposits. Local singularity analysis (Cheng and two publications, there have been relatively few other
Agterberg, 2009; Cheng, 2005) is an example of non- applications along these lines, although generaliza-
linear modeling of geochemical data in mineral explo- tions to multifractal modeling can have advantages
ration and environmental applications that produce similar to those documented in the relatively many
new types of maps, which are significantly different existing multifractal applications to measures or
from conventional contour maps that tend to smooth chemical element concentration methods (Cheng,
out neighborhoods with significant enrichment of ore- 2012; Mandelbrot, 1999).
forming and other minerals. The points in a point pattern can have different
Point patterns can be fractal (Korvin, 1992; Feder, values. This has led to the modeling of “marked”
1988; Stoyan and Kendall, 1987). In general, fractal point processes (Cressie, 2001). This topic is of im-
point patterns are characterized by the fact that aver- portance in resource potential modeling where the
age point density (=number of points per unit of area) points represent mineral deposits or oil pools with
decreases when size of study area is increased. In this positively skewed size-frequency distributions and
respect, fractal point patterns differ from commonly possess other features that differ greatly from point to
used statistical point-process models such as the ran- point. With respect to modeling the size-frequency
dom Poisson and various clustering models in which distributions of ore and hydrocarbon deposits, Man-
the mean point density is assumed to be the same delbrot (1983) had challenged geoscientists by assert-
within the entire study area regardless of its size. ing that oil and other natural resources have Pareto
Mathematical statisticians (Baddeley and Turner, distributions and this “finding disagrees with the
2012; Baddeley et al., 2008; Baddeley, 2007; Cressie, dominant opinion, that the quantities in question are
2001; Rowlingson and Diggle, 1993, 1991; Ripley, lognormally distributed. The difference is extremely
1988, 1981, 1976; Diggle, 1983) have developed very significant, the reserves being much higher under the
precise methods to estimate the parameters of con- hyperbolic than under the lognormal law”. Although it
stant-density point processes. In Agterberg (1994), is well known that lognormal frequency distributions
Ripley’s method of correcting estimates of point pat- can be seen as multifractals, the problem of whether
tern parameters for edge effects is applied to wildcats natural resources are best modeled as Pareto- or log-
and gas discoveries in exploration “plays” in Alberta normal-type remains important and both approaches
with study area boundaries that are approximated by are used extensively in oil and other natural resource
multi-edge polygons instead of being rectangular in size-frequency modeling.
shape. Not only do these “plays” have irregular A simple method to check whether a set of re-
boundaries, they may contain islands of terrain judged source estimates satisfies a Pareto distribution consists
Fractals and Spatial Statistics of Point Patterns 3
of constructing a log-log plot of amount of metal per the latter are not fractal because their dimension re-
deposit versus size rank (in descending order) as mains equal to the Euclidian dimension (=2) for in-
shown in the example of Fig. 1 for gold deposits in the creasing r.
Abitibi area on the Canadian Shield. If the Pareto Two methods (box-counting, and cluster density
model is satisfied, the points on this plot are on a determination) are commonly used for the fractal
straight line, usually with a relatively sharp downward modeling of point clusters. The first method consists
bend at the end, because of an economic cut-off effect of superimposing square grids with different spacings
for smaller deposits. In the following sections of this (Є) on the study region. The number of boxes (NЄ)
paper, the fractal point pattern of Abitibi gold deposits containing one or more points is counted for every
including many smaller (sub-economic) gold occur- square grid. The number NЄ decreases with increasing
rences will be studied in detail. Additionally, world- value of Є, mainly because there are fewer larger
wide point patterns of podiform chromium, volcano- boxes. If the box-counting fractal model is satisfied,
genic massive sulphide and porphyry copper deposits this decrease satisfies a power-law relation and shows
(data from Singer and Menzie, 2010) will be subjected as a straight line with slope-Db on a log-log plot where
to fractal modeling and Pareto distributions will be Db represents the fractal dimension. Cluster density
fitted to their sizes (tons of metal). determination can be performed by centering circles
with different radii (r) on all points in the study area,
3.5
3.0 counting how many other points occur within these
2.5
circles, and averaging the results. The relation be-
Log size
2.0
1.5 tween cluster densities then is according to a power-
1.0
0.5
law that plots as a straight line with slope Dc-2 on log-
0.0 log paper if the fractal model with Dc as its fractal di-
0.0 0.5 1.0 1.5 2.0 2.5
Log rank mension is satisfied.
Figure 1. Production and reserves (in metric tons Carlson (1991) obtained the following results: for
Au) of 107 lode deposits in Abitibi volcanic belt, distances less than 15 km, cluster density determina-
Superior Province, Canadian Shield, ranked ac- tion as applied to 4 775 precious metal deposits give
cording to their sizes (largest deposit first). Pareto Dc=0.83 versus Db=0.50 for box-counting. For greater
distribution fitted as straight line (after Agterberg, distances between 15 and 1 000 km, the two methods
1995). give Dc=1.17 versus Db=1.51. Note that for short dis-
tances, Db is less than Dc but for larger distances, it is
FRACTAL MODELING OF POINT PATTERNS the other way around. For both kinds of fractal cluster,
Carlson (1991) applied a fractal cluster model to the dimension seems to be bifractal. Carlson com-
hydrothermal precious-metal mines in the Basin and mented that differences between Db and Dc are “trou-
Range Province, western United States. The underly- bling but common in measuring fractal cluster dimen-
ing theoretical model is clearly explained in Feder sions”. The fractal dimensions for short distances are
(1988, chapter 3 therein). For the spatial distribution probably too low because of “roll-off” effects that
of points in the plane this model requires that the affect Db more strongly than Dc (see later). Because of
number of points within distance r of an arbitrary possible edge effects, the Db=1.51 estimate is may be
point is proportional to rDc where Dc represents the so- most accurate. Carlson (1991) interpreted his results
called cluster fractal dimension which is less than the in terms of a bifractal model assuming that fractal
Euclidian dimension of the embedding space (=2 for hydrothermal and fracture systems are effective over
two-dimensional space). It implies that the cluster scales from about 15 to 1 000 km. Various examples
density decreases with increasing r. As mentioned of geological bifractals in other types of application
before, a basic difference between this fractal model can be found in Korvin (1992).
and commonly used statistical point-process models Differences in fractal point pattern dimension can
such as the Poisson and Neyman-Scott models is that be due to different reasons. One explanation is that Db
4 Frederik P Agterberg
and Dc are not necessarily the same; e.g., they are been criticized by many authors but Raines (2008) in
expected to be different for a multifractal cluster. a comprehensive review concluded (a) in several other
With respect to the differences between Db and Dc: applications, differences between Db and Dc are less
due to limited resolution of maps, the measurements than those found by Carlson, and (b) in general, the
on fractals normally become biased for small dis- bifractal cluster model would be satisfied because di-
tances (Є or r) and this can cause measured fractal mensions for shorter distances are indeed less than
dimensions to be biased as well. Theoretically, a frac- those for larger distances. Raines (2008) was able to
tal cluster model has the property of self-similarity in reduce bias in fractal cluster estimation to some extent
that point patterns for any enlarged subarea would be by using GIS-based methods. However, it is possible
similar even if the zooming-in is repeated indefinitely. to improve fractal estimation procedures more signifi-
In practice, graphical representations become increas- cantly by adopting the methods developed by mathe-
ingly incomplete with increased enlargement. This matical statisticians as will be shown in the next sec-
type of bias does not only apply to fractal point pat- tion on the basis of a practical example.
terns but to other types of fractals as well. It is known
as the “roll-off” effect and has been studied in detail CLUSTER DENSITY DETERMINATION OF
by structural geologists (e.g., Blenkinsop, 1995; GOLD DEPOSITS IN THE KIRKLAND LAKE
Pickering et al., 1995; Walsh et al., 1991). When sam- AREA ON THE CANADIAN SHIELD
ple size is large, the “roll-off” effect can be modeled Figure 2 (from Agterberg et al., 1993) shows the
by using a continuous curve that asymptotically ap- locations of 295 gold deposits (mines and occurrences)
proaches the fractal straight line on log-log paper. in a rectangular area of 4 185 km2 in the vicinity of
However, for smaller samples, the observed “roll-off” Timmins and Kirkland Lake in the Abitibi area on the
effect may create an artificial sequence resembling Precambrian Canadian Shield. It is part of a much
two or more separate straight-line segments. Down- larger study area (=89 600 km2) containing 1 306 gold
ward bias in Db toward the origin is stronger than that deposits with Dc=1.514 to be discussed in more detail
in Dc for short distances. in the next section. Cluster density estimation applied
On the other hand, estimates of Dc for large dis- to Fig. 2 gives D c =1.493, which is close to the
tances may become less precise and biased downward
more strongly than those of Db if edge effects are not
taken into account. Unless Ripley’s K(r) function (see
later) is used, it is difficult to account for edge effects
in Dc because: (1) much useful information is dis-
carded if all largest circles around the points (along
with all smaller circles within the largest circles) are
required to be fully contained within the study area,
and (2) the boundary of the study area may not have a
simple (rectangular or circular) shape but can be
highly irregular and then would have to be approxi-
mated by a polygon instead of a rectangle, thus creat-
ing additional complications. These two edge effect
problems can be avoided by using Ripley’s K(r) func-
tion. Although estimates of Db based on the largest Figure 2. Example of point pattern for 295 gold
boxes are relatively imprecise, because there are rela- deposits in the vicinity of Timmins and Kirkland
tively few of them, they can have the advantage of Lake, Abitibi volcanic belt on Canadian Shield.
remaining unbiased provided that exactly the same Circles are gold mines (mainly past producers);
study area is covered by all box-counting grids. triangles represent small, unmined deposits (ex-
The results obtained by Carlson (1991) have tracted from Agterberg et al., 1972).
Fractals and Spatial Statistics of Point Patterns 5
estimate based on the larger area. Figure 2 shows erful way to characterize the second-order properties
large subareas without any gold deposits (lacunarity). of a point process is provided by Ripley’s K(r) func-
This is either because the bedrock containing gold tion, which is proportional to expected number of fur-
deposits in these subareas is covered by relatively ther events within distance r from an arbitrary event
thick Quaternary deposits or it consists of barren rock point. The relation between λ2(r) and K(r) is λ2(r)∝
types such as Archean gneiss belts without gold de- K’(r)/r where ∝ denotes proportionality and K’(r) is
posits. The pattern of Fig. 2 is anisotropic because of the first derivative of K(r) with respect to r. Applica-
strong east-west structural trends. By means of addi- tion of this statistical point process technique to fractal
tional experiments to be summarized below it can be cluster theory yields λ2(r)∝rDc-2 and K(r)∝rDc where
shown that neither lacunarity nor anisotropy signifi- ∝ denotes proportionality.
cantly alters our estimate of Dc (≈1.5). For rectangular areas such as that of Fig. 2, the
Most problems of lack of precision and accuracy K(r) function can be estimated readily by the method
outlined in the previous section can be avoided by originally developed by Ripley (1976) according to
adopting methods of statistical point-process model- which edge effects due to the boundaries of the study
ling (Agterberg, 1994; Ripley, 1988; Diggle, 1983). In area are avoided. Diggle (1983) has published explicit
this type of modeling, the first-order and second-order formulae for unbiased estimation of K(r) for event
properties of a spatial point process for events are de- points in a rectangular study area. Application of this
scribed by its intensity function λ1(x) and second-order method to the pattern of Fig. 2 resulted in the pattern
intensity function λ2(x, y) where x and y represent the of solid circles in Fig. 3a. The slope of the straight
locations of two event points in the plane. For an iso- line fitted to these points gives the estimate Dc=1.493
tropic, stationary point process, λ1(x)=λ and λ2(x, mentioned before.
y)=λ2(r) where r is distance between x and y. A pow- Results of two experiments to study the effects of
2.5 2.0
(a) (b)
2.0 1.5
1.5
1.0
LogK (r)
LogK (r)
1.0
0.5
0.5
0.0 0.0
-0.5 -0.5
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
Logr Logr
Figure 3. Log-Log plots of K(r)-function versus distance r (in km) to estimate cluster fractal dimension Dc
(estimated from slopes of best-fitting straight lines). (a) Point pattern for gold deposits in Fig. 2; (b) south-
ern half of point pattern shown in Fig. 2. Solid circles show results for original (anisotropic) point patterns;
open circles show results for transformed (approximately isotropic) point pattern. Estimates of Dc for solid
circles are (a) 1.493 and (b) 1.498; estimates of Dc for open circles are (a) 1.506 and (b) 1.524. Solid circles in
Fig. 3b are based on fewer points because edge correction formula used requires r<h/2 where h is height of
map area (for north-south direction) in km (after Agterberg et al., 1993)
lacunarity and anisotropy are also shown in Fig. 3. tion of the previous method to the point pattern for the
The largest holes in the point pattern of Fig. 2 occur in southern half of Fig. 2 resulted in the pattern of solid
the northern half of this study area. For this reason, a circles in Fig. 3b with Dc=1.467 instead of Dc=1.493.
new study area consisting only of the southern half of The fact that these two estimates are nearly equal to
Fig. 2 was defined. It contains most of the gold depos- one another illustrates that lacunarity does not signifi-
its in Fig. 2 and has three holes (lacunae) that are cantly affect our estimation of Dc.
smaller than the two largest holes in Fig. 2. Applica- The effect of anisotropy was studied as follows.
6 Frederik P Agterberg
12.0
terberg (1981) had used the following computer algo- 10.0
rithm to estimate the reduced second moment measure 7.0
Second-order intensity
L 2c( r )/ l c=4.452 r -0.486
λ2(r)/λ (cf., Section Fractal modeling of point patterns). 5.0
A grid with (5 km×5 km cells) was successively cen- 4.0
mineral deposit density control areas for mainly (1) one would expect that the three best-fitting straight
podiform chromite; (2) volcanogenic (Cyprus+Kuroko) lines for point density in Fig. 6 would be approxi-
massive sulphide and (3) porphyry copper deposits mately horizontal because number of deposits per tract
(Singer and Menzie, 2010, Table 4.1 therein). For de- then would be proportional to tract area for each de-
posit type per permissive area, these authors listed (a) posit type. Instead of this, the three best-fitting lines in
area (km2), (b) number of deposits, and (c) median the log density versus log permissive plots have nega-
and total tons of metal. Figure 6 (based on Singer and tive slopes equal to 0.53, 0.62 and 0.61 for chromite,
Menzie, 2010, Table 4.1 therein) shows log-log plots volcanogenic massive sulphide and porphyry copper,
for the relations between deposit density and permis- respectively. From the relatively large R2 values of
sive area for the three selected deposit types. If the these least squares fits it can be derived that the three
deposits would be randomly distributed across each slopes are significantly greater than zero.
permissive tract, e.g., according to a Poisson model,
-0.5
Log(density)
Log(density)
R 2=0.642 1
10
10
-3.0
-2.5 -3.5 -5.0
-3.0 -4.0 -5.5
1.0 1.5 2.0 2.5 3.0 3.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
10 10 10
Log(area) Log(area) Log(area)
Figure 6. Log-log plots of original density (red squares) and size-weighted density (blue rectangles) versus
permissive tract area for three types of deposits. (a) Podiform Cr deposits; (b) volcanogenic massive sul-
phide deposits; (c) porphyry copper deposits (original data from Singer and Menzie, 2010, Table 4.1
therein). Note that slopes of best-fitting straight lines for original density plots are nearly equal to one an-
other and have relatively large R2 values. Slopes of lines for size-weighted density plots are not significantly
different from zero for types (a) and (b) but similar to the density plot slope for porphyry coppers. In order
to facilitate comparison, the size-weighted density scale was augmented by 3 in Fig. 6a and diminished by 3
in Fig. 6c.
Although the permissive tract approach differs tracts are fairly large.
from cluster density estimation, it also can be used for Singer and Menzie (2010, p. 60–64 therein) also
fractal cluster estimation. From the slopes of the lines have considered the relation between total tonnage of
of best fit it follows that all three deposit types satisfy deposits in their Table 4.1 with area of permissive
a fractal cluster model with Dc equal to 1.47, 1.38 and tract and deposit density simultaneously. In our Fig. 6
1.39, respectively. These results are not only remarka- a second set of points is shown for deposits of each
bly similar to one another but also to those obtained type weighted according to their total metal tonnage.
previously for gold deposits in the Abitibi volcanic For podiform chromite and volcanogenic sulphide
belt on the Canadian Shield with Dc≈1.5. Singer and deposits, the slopes of the lines fitted to the second set
Menzie (2010) had already established empirically of points are not statistically significant but for the
that deposit density for the three types of deposits sig- porphyry coppers there is an indication that size-
nificantly decreases with size of permissive tract but weighted density also decreases with tract area. The
they did not offer a fractal explanation. A factor lack of clear-cut relationships between size-weighted
probably contributing to the fact that deposit density deposit density and tract size is probably due to the
clearly decreases linearly with tract size in this type of extremely large positive skewness exhibited by the
application is that the boundaries of the permissive metal tonnages for all three types of deposits. How-
areas have curved shapes determined by local geology. ever, the size-frequency distributions can be studied
Roll-off effects should be negligibly small because the independently of geographic locations of the deposits.
Fractals and Spatial Statistics of Point Patterns 9
Figure 7 shows that for each of the three deposit mates. The slopes in the log rank versus log tonnage
types (podiform chromite, volcanogenic massive plots are -10.66, -0.406 and -0.801 for podiform Cr,
sulphide and porphyry copper) the largest deposits volcanogenic massive sulphide and porphyry copper
approximately satisfy a Pareto size frequency distribu- deposits, respectively. For other applications of the
tion. Because the three data sets are not large, Pareto frequency distribution to mineral resource data,
Quandt’s (1966) method was used for parameter esti- see Mandelbrot (1995), Turcotte (1997) and Agterberg
mation, because it yields unbiased (consistent) esti- (2011, 1995).
1.2 1.4
(a) (b) 1.6
1.0 1.2 (c)
1.4
0.8 1.0 1.2
Log (rank)
Log (rank)
Log (rank)
0.6 0.8 1.0
y =-10.656 x +1.292 7 y =-0.800 5 x +4.134 2
0.8
0.4 0.6
y =-0.406 2 x +1.512 3 0.6
10
10
0.2 0.4
10
0.4
0.0 0.2 0.2
-0.2 0.0 0.0
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 1.0 1.5 2.0 2.5 3.0 3.5 3.0 3.5 4.0 4.5 5.0 5.5
10 10 10
Log (t) Log (t) Log (t)
Figure 7. Log-log plots of size rank versus metal tonnage for the three types of deposits (the straight lines
were fitted to largest deposits only and represent Pareto distributions).
search in the Mineral Industries, Canadian Institute of Ore-Mineral and Toxic Trace Elements in Stream Sedi-
Mining, Metallurgy and Petroleum, Montreal. 1: 43–53 ments. Computers & Geosciences, 35(2): 234–244,
Agterberg, F. P., Chung, C. F., Fabbri, A. G., et al., 1972. doi:org/10.1016/j.cageo.2008.02.034
Geomathematical Evaluation of Copper and Zinc Potential Cressie, N. A. C., 2001. Statistics for Spatial Data. Wiley, New
of the Abitibi Area, Ontario and Quebec. Geological Sur- York
vey of Canada Paper. Dept. of Energy, Mines and Re- Diggle, P. J., 1983. Statistical Analysis of Spatial Point Patterns.
sources, Yukon Academic Press, London. 72
Baddeley, A., Turner, R., 2012. Package ‘Spatstat’, Manual, Falconer, K. J., 2003. Fractal Geometry: Mathematical Founda-
Version 1.30.0 (Released 2012-12-23) [PDF]. http://140. tions and Applications. Wiley, Chichester
247.115.171 Feder, J., 1988. Fractals. Plenum, New York
Baddeley, A., Møller, J., Pakes, A. G., 2008. Properties of Re- Gupta, V. K., Troutman, B. M., Dawdy, D. R., 2007. Towards
siduals for Spatial Point Processes. Annals of the Institute a Nonlinear Geophysical Theory of Floods in River Net-
of Statistical Mathematics, 60(3): 627–649, doi:10.1007/ works: An Overview of 20 Years of Progress. In: Tsonis,
s10463-007-0116-6 A. A., Elsner, J. B., eds., Nonlinear Dynamics in Geo-
Baddeley, A., 2007. Spatial Point Processes and Their Applica- sciences. Springer, New York. 121–150, doi:10.1007/978-
tions. Lecture Notes in Mathematics: 1–75, doi:10.1007/ 34918-3_8
978-3-540-38175-4_1 Korvin, G., 1992. Fractal Models in the Earth Sciences. El-
Barton, C. C., La Pointe, P. R., 1995. Fractals in Petroleum sevier, Amsterdam
Geology and Earth Processes. Plenum, New York. 317 Lovejoy, S., Schertzer, D., 2007. Scaling and Multifractal
Blenkinsop, T. G., 1995. Fractal Measures for Size and Spatial Fields in the Solid Earth and Topography. Nonlinear
Distributions of Gold Mines: Economic Implications. In: Processes in Geophysics, 14(4): 465–502
Blenkinsop, T. G., Tromp, P. L., eds., Sub-Saharan Eco- Malamud, B. D., Morein, G., Turcotte, D. L., 1998. Forest
nomic Geology. Special Publication Geological Society of Fires: An Example of Self-Organized Critical Behavior.
Zimbabwe, 3: 177–186 Science, 281(5384): 1840–1842, doi:10.1126/science.281.
Carlson, C. A., 1991. Spatial Distribution of Ore Deposits. 5384.1840
Geology, 19(2): 111–114, doi:10.1130/0091-7613(1991) Mandelbrot, B. B., 1975. Les Objects Fractals: Forme, Hazard
Cheng, Q. M., 1994. Multifractal Modeling and Spatial Analy- et Dimension. Flammarion, Paris. 1–19
sis with GIS: Gold Mineral Potential Estimation in the Mandelbrot, B. B., 1983. The Fractal Geometry of Nature.
Mitchell-Sulphurets Area, Northwestern British Columbia: Freeman, San Francisco. 262
[Dissertation]. University of Ottawa, Ottawa Mandelbrot, B. B., 1995. The Statistics of Natural Resources
Cheng, Q. M., 2008. Non-Linear Theory and Power-Law Mod- and the Law of Pareto. In: Barton, C. C., La Pointe, P. R.,
els for Information Integration and Mineral Resources eds., Fractals in Petroleum Geology and the Earth Sci-
Quantitative Assessments. Mathematical Geoscience, ences. Plenum, New York. 1–12
40(5): 195–225, doi:10.1007/s11004-008-9172-6 Mandelbrot, B. B., 1999. Multifractals and 1/f Noise. Springer,
Cheng, Q. M., 2005. A New Model for Incorporating Spatial New York
Association and Singularity in Interpolation of Explora- Park, N. W., Chi, K. H., 2008. Quantitative Assessment of
tory Data. Quantitative Geology and Geostatistics, 14(5): Landslide Susceptibility Using High-Resolution Remote
1017–1025, doi:10.1007/978-1-4020-3610-1_106 Sensing Data and a Generalized Additive Model. Interna-
Cheng, Q. M., 2012. Multiplicative Cascade Processes and tional Journal of Remote Sensing, 29(1): 247–264, doi:10.
Information Integration for Predictive Mapping. Nonlinear 1080/01431160701227661
Processes in Geophysics, 19: 57–68, doi:10.5194/npg-19- Pickering, G., Bull, J. M., Sanderson, D. J., 1995. Sampling
57-2012 Power-Law Distributions. Tectonophysics, 248(1–2): 1–20,
Cheng, Q. M., Agterberg, F. P., 1995. Multifractal Modelling doi:org/10.1016/0040-1951(95)00030-Q
and Spatial Point Processes. Mathematical Geology, 27(7): Quandt, R. E., 1966. Old and New Methods of Estimation and
831–845 the Pareto Distribution. Metrica, 10(1): 55–82, doi:10.
Cheng, Q. M., Agterberg, F. P., 2009. Singularity Analysis of 1007/BF02613419
Fractals and Spatial Statistics of Point Patterns 11
Raines, G. L., 2008. Are Fractal Dimensions of the Spatial Statistical Physics Approach to Understanding the Multis-
Distribution of Mineral Deposits Meaningful? Natural Re- cale Dynamics of Earthquake Fault Systems. Reviews of
sources Research, 17(2): 87–97, doi:10.1007/s11053-008- Geophysics, 41: 1019, doi:10.1029/2003 RG000135
9067-8 Sharma, A. S., 1995. Assessing the Magnetosphere’s Nonlinear
Ripley, B. D., 1976. The Second-Order Analysis of Stationary Behavior: Its Dimension is Low, Its Predictability, High.
Point Processes. Journal of Applied Probability, 13(2): Reviews of Geophysics, 33(S1): 645, doi:10.1029/95RG
255–266 00495
Ripley, B. D., 1981. Spatial Statistics. Wiley-Interscience, New Singer, D., Menzie, W. D., 2010. Quantitative Mineral Re-
York source Assessments: An Integrated Approach. Oxford
Ripley, B. D., 1988. Statistical Inference for Spatial Processes. University Press, New York
Cambridge University Press, Cambridge. doi:10.1017/ Stoyan, D., Kendall, W. S., 1987. Stochastic Geometry and Its
CBO9780511624131 Applications. Wiley, Chichester. 125
Rowlingson, B. S., Diggle, P. J., 1991. Estimating the K- Turcotte, D. L., 1997. Fractals and Chaos in Geology and Geo-
Function for a Univariate Spatial Point Process on an Ar- physics. Cambridge University Press, Cambridge
bitrary Polygon. Lancaster University Mathematics De- Uritsky, V. M., Donovan, E., Klimas, A. J., 2008. Scale-Free
partment Technical Report MA91/58. Lancaster Univer- and Scale-Dependent Modes of Energy Release Dynamics
sity, Lancaster. 1–15 in the Night Time Magnetosphere. Geophysical Research
Rowlingson, B. S., Diggle, P. J., 1993. Spalncs: Spatial Point Letters, 35 (21): L21101, 1–5
Pattern Analysis Code in S-Plus. Computers & Geo- Walsh, J., Watterson, J., Yielding, G., 1991. The Importance of
sciences, 19(5): 627–655, doi:org/10.1016/0098-3004(93) Small-Scale Faulting in Regional Extension. Nature, 351:
90099-Q 391–393, doi:10.1038/351391a0
Rundle, J. B., Turcotte, D. L., Shcherbakov, R., et al., 2003.