You are on page 1of 9

Multifractal Characterization of Soil Pore Systems

Adolfo N. D. Posadas, Daniel Gimenez,* Roberto Quiroz, and Richard Protz


ABSTRACT

and biological activity (Bouma et al., 1977; RingroseVoase, 1987; Pagliai and De Nobili, 1993). The selection
and interpretation of the morphological parameters of
soil pores that best characterize soil structure, however,
is still a subject of research (Droogers et al., 1998;
Holden, 2001). Statistical methods of characterization
of pore-solid arrangement from images emphasize the
spatial structure of pores with the advantage of being
more amenable to modeling soil structure and processes
(Dexter 1976, 1977; Moran and McBratney, 1997; Garboczi et al., 1999; Horgan, 1999; Vogel and Roth, 2001).
Among the statistical methods used to characterize
soil structure, fractal techniques are relatively common
in soil science (Anderson et al., 1996; Pachepsky et al.,
1996; Gimenez et al., 1997). Fractal geometry assumes
that the dependence of the properties of a system with
scale (scaling) can be represented with a power law,
with the exponent being typically a function of a fractal
dimension. However, it has become increasingly evident
that knowledge on the fractal dimension of a set is
insufficient to characterize its geometry (Loehle and
Li, 1996). The fractal dimension, D, characterizes the
average properties of a set and cannot provide information on deviations from the average behavior of a power
law. For example, the box-counting technique is used
to estimate D from images of pore systems by covering
them with a grid of boxes of various sizes. The technique ignores the variations in pore density within a
box other than categorizing boxes as empty or occupied
(Vicsek, 1992). As a result, sets with different appearance or textures may have similar fractal dimensions
(Mandelbrot, 1982; Voss, 1988). On the other hand, a
multifractal analysis captures the inner variations in a
system by resolving local densities and expresses them
in the shape of a multifractal spectrum (Hentschel and
Procaccia, 1983; Frisch and Parisi, 1985; Halsey et al.,
1986; Chhabra et al., 1989; Chhabra and Jensen, 1989).
The multifractal concept has been useful in studies of
spatial arrangement of physical and chemical quantities
(Stanley and Meakin, 1988; Feder, 1988; Evertsz and
Mandelbrot, 1992; Cheng and Agterberg, 1996), turbulence (Meneveau and Sreenivasan, 1991), and geology
(Muller and McCauley, 1992; Cheng, 1999). In soil science, multifractal techniques have been applied to the
characterization of particle- and pore-size distributions
(Grout et al., 1998; Caniego et al., 2001; Posadas et
al., 2001; Martn and Montero, 2002), surface strength
(Folorunso et al., 1994), and spatial variability of soil
properties (Kravchenko et al., 1999). The only published
report on the application of a multifractal method to
soil pores is an analysis of the histogram of pore area

Spatial arrangement of soil pores determines soil structure and is


important to model soil processes. Geometric properties of individual
pores can be estimated from thin sections, but there is no satisfactory
method to quantify the complexity of their spatial arrangement. The
objective of this work was to apply a multifractal technique to quantify
properties of ten contrasting soil pore systems. Binary images (500
by 750 pixels, 74.2 m pixel1) were obtained from thin sections and
analyzed to obtain f() spectra. Pore area and pore perimeter were
measured from each image and used to estimate a shape factor for
pores with area larger than 0.27 106 m2. Mean area of the lower
(MAL) and upper (MAU) one-half of cumulative pore area distributions were calculated. Pore structures with large (MAU 10
106 m2) and elongated pores exhibited flat f()-spectra typical of
homogenous systems (three soils). Massive type structure with small
(MAU 1 106 m2) rounded and irregular pores resulted in asymmetric f()-spectra (two soils). Well defined and symmetric f()spectra were obtained with soil structures having elongated pores of
intermediate size (1 106 MAU 10 106 m2) clustered around
relatively small structural units (five soils). Multifractal parameters
defining the maximum of the f()-spectra were correlated to total
porosity (P 0.001), and silt content (P 0.05). This study demonstrates that the spatial arrangement of contrasting soil structures can
be quantified and separated by the properties of their f()-spectra.
Multifractal parameters quantifying spatial arrangement of soil pores
could be used to improve classifications of soil structure.

arly attempts to classify the arrangement of pores


and solids from thin sections led to the development of semi-quantitative classification systems based
mainly in the relationship between fine (i.e., inorganic
and organic soil colloid) and coarse material (Jim, 1988).
Application of these systems to the characterization of
soil thin sections have shown that arrangement of soil
material is correlated to particle-size distribution, organic matter content, and pedogenic processes (Eswaran and Banos, 1976; Brewer, 1979; Goenadi and
Tan, 1989). The development of the general field of
mathematical morphology allowed quantifying the
shape, size, and connectivity of soil pores (RingroseVoase, 1987; Horgan, 1998; Holden, 2001). Size, shape,
and spatial arrangement of pores have been used to
classify soil structure (Ringrose-Voase and Bullock,
1984; Pagliai and De Nobili, 1993). Pore shape and size
have been related to water flow, pedogenetic processes,

A.N.D. Posadas, International Potato Center (CIP), P.O. Box 1558,


Lima 12 Peru and Universidad Mayor de San Marcos, FCF-DAFI,
P.O. Box 10584, Lima 1- Peru; D. Gimenez, Dep. of Environmental
Sciences, Rutgers Univ., 14 College Farm Road, New Brunswick, NJ
08901; R. Quiroz, International Potato Center (CIP), P.O. Box 1558,
Lima 12 Peru,; R. Protz (deceased), Dep. of Land Resource Science,
Univ. of Guelph, Guelph, ON, Canada. Received 27 Aug. 2003. *Corresponding author (gimenez@envsci.rutgers.edu).
Published in Soil Sci. Soc. Am. J. 67:13611369 (2003).
Soil Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA

Abbreviations: MAL, mean area of the lower one-half of cumulative


pore-area distributions; MAU, mean area of the upper one-half of
cumulative pore-area distributions.

1361

1362

SOIL SCI. SOC. AM. J., VOL. 67, SEPTEMBEROCTOBER 2003

Table 1. Soil classification and selected properties of the studied soil horizons.
Horizon
Group
1

2
3

Soil
1
2
3
4
5
6
7
8
9
10

Soil classification
Spodosol
Argiaquic Argialboll
Typic Haplorthox
Typic Haplorthox
Inceptisol
Orthoxic Tropudult
Mollic Albaqualf
Petric Plinthudult
Entisol
Entisol

Type

Depth

C
B
B
B
C
B
B
A
BC
B

cm
6573
3440
3038
7078
2025
3340
4752
18
2032
614

Organic
matter

Sand

Silt

Clay

15.0
66.2
22.0
6.0
4.9
21.0
67.1
51.0
63.5
65.2

31.0
25.1
58.0
72.0
0.4
31.0
30.2
37.0
33.8
33.4

%
0.1
0.4
1.2
0.7
0.3
0.3
0.4
1.7
1.3
1.1

54.0
3.1
20.0
22.0
94.6
48.0
2.7
12.0
2.7
1.4

According to the soil taxonomy.

that did not consider the spatial arrangement of pores


(Caniego et al., 2001).
A multifractal analysis done on binary images has
shown that spatial patterns of pores in rocks are multifractal (Muller and McCauley, 1992). This conclusion
cannot be extrapolated to soil pore systems because of
the differences in genesis between rocks and soils. Thus,
the objective of this paper was to apply a multifractal
method to binary images of soil pores representing contrasting soil structures. This approach has the advantage
of integrating the spatial properties of pore systems.

Fractal dimensions offer a systematic approach to quantifying irregular patterns that contain an internal structure repeated over a range of scales (Meakin, 1991). For a fractal
object the number of features of a certain size , N(), varies as:

[1]

where D is the fractal dimension. Equation [1] is a scaling (or


power) law that has been shown to describe the size distribution of many objects in nature. The box-counting technique
is used to obtain the scaling properties of two-dimensional
fractal objects by covering a measure with boxes of size L and
counting the number of boxes containing at least one pixel
representing the object under study, N(L ):

logN(L)
D0 lim
L0 log(1/L)

[2]

Using Eq. [2], the box-counting dimension D0 can be determined as the negative slope of log N(L ) versus log(L ) measured over a range of box sizes. The disadvantage of the boxcounting technique is that the process does not consider the
amount of mass inside a box Ni (L ) and is, therefore, not able
to resolve regions with high or low density of mass. Multifractal
methods are suited for characterizing complex spatial arrangement of mass because they can resolve local densities (Vicsek,
1992). In practice, a way to quantify local densities is by estimating the mass probability in the ith box as:

Pi(L) Ni(L)/NT

[4]

where i is the Lipschitz-Holder exponent characterizing scaling in the ith region or spatial location (Halsey et al., 1986).
In our case, these exponents reflect the local behavior of the
measure Pi (L ) around the center of a box with diameter
L, and can be estimated from Eq. [4] as i log Pi (L )/
log(L )(Fig. 1d and 1e). Note that similar i values are found
at different positions in an image. The number of boxes N()
where the probability Pi has exponent values between and
d is found to scale as (Chhabra et al., 1989; Halsey et
al., 1986):

N() Lf()

THEORY

N() D

Pi(L) Li

[3]

where Ni(L ) is the number of pixels containing mass in the


ith box and NT is the total mass of the system. Examples of
spatial patterns of Pi (L 10) and Pi (L 50) for Soil 7
(Fig. 4, Table 1) are shown in Fig. 1b and 1c, respectively (L
is expressed in pixels). Also important is to quantify the scaling
(or dependence) of Pi with box size L. For heterogeneous
or non-uniform systems the probability in the ith box Pi(L )
varies as:

[5]

where f() can be defined as the fractal dimension of the set


of boxes with exponent . Equation [5] generalizes Eq. [1] by
including several indices to quantify the scaling of a system.
Multifractal measures can also be characterized through
the scaling of the qth moments of Pi distributions in the form
(Chhabra et al., 1989; Korvin, 1992):
N(L)

i1

P iq(L) L(q1)Dq

[6]

where Dq are the generalized fractal dimensions defined from


Eq. [6] as:
N(L)

Dq lim
L0

1
q1

log P iq(L)
i1

logL

[7]

The exponent in Eq. [6] is known as the mass exponent of


the qth order moment, (q ) (Halsey et al., 1986; Vicsek, 1992):

(q) (q 1) Dq

[8]

From Eq. [7] we see than when q 0 all the boxes have
a weight of unity, the numerator becomes N(L ), and Dq becomes the capacity dimension, D0 (Eq. [2]). Similarly, when
all the boxes have the same probability, that is, Pi 1/N, Dq
D0 for all values of q and (q ) becomes a linear function of
q (homogeneous fractal). Two other special cases are for q 1
and q 2. The values D1 and D2 are known as the entropy
dimension and the correlation dimension, respectively. The
entropy dimension is related to the information entropy of
Shannon and Weaver (1949), which quantifies the decrease in
information as the size of the boxes increases. The correlation
dimension D2 is mathematically associated with the correlation
function and computes the correlation of measures contained
in a box of size L.
The connection between the power exponents f() (Eq. [5])
and (q ) (Eq. [8]) is made via the Legendre transformation
(Callen, 1985; Halsey et al., 1986; Chhabra and Jensen, 1989):

POSADAS ET AL.: MULTIFRACTAL CHARACTERIZATION OF SOIL PORE SYSTEMS

1363

Fig. 1. Illustration of multifractal theory applied to a binary image. (a) binary image of Soil 7 (500 750 pixels), spatial pattern of probabilities
Pi(L ) calculated with Eq. [3] using (b) L 10 and (c) L 50 pixels, and spatial pattern of the exponent i estimated with Eq. [4] using (d)
L 10 and (e) L 50 pixels.

f[(q)] q(q) (q)

[9a]

and

(q)

d(q)
dq

[9b]

The function f() is concave downward with a maximum at


q 0. In natural systems, and f() are not evaluated at the
limit L 0, but rather in the scaling region in which and
f() can be described as powers of L, which also restricts the
range of q values that can be used.

MATERIALS AND METHODS


Ten thin sections of soils with contrasting soil structure and
soil texture were selected for this study (Table 1). Binary
images of soil thin sections were produced following the proce-

dure outlined in VandenBygaart and Protz (1999). Images


had a size of 500 750 pixels, with a pixel size of 74.2 m.
Distributions of pore area, A, and perimeter, P, were calculated from each image using NIH-Image software (Rasband,
1993). Pore area was used to calculate cumulative distributions. Mean pore areas of the lower one-half of a distribution,
MAL, and upper one-half of a distribution, MAU, were calculated. Pore shapes were estimated by calculating a shape index,
F, for pores with pore area larger than 0.27 106 m2 or
50 pixels2:

4A
P2

[10]

Three basic pore shapes were defined based on the values


F: planar when F 0.2, irregular when 0.2 F 0.5, and
rounded when F 0.5 (Bouma et al., 1977). For each image,

1364

SOIL SCI. SOC. AM. J., VOL. 67, SEPTEMBEROCTOBER 2003

Fig. 3. Example of f(q ) and (q ) functions estimated in the range of


q values in which the numerators of Eq. [12] and [13] were linear
with log L (see Fig. 2).

N(L)

(q) lim
L0

Fig. 2. Examples of application of (a) Eq. [12], and (b) Eq. [13] to
the binary image of Soil 7 for selected q values. Values of f(q )
and (q ) were obtained from the slope of plots similar to those
in (a) and in (b), respectively. The plot of (q 1.0) illustrates
data that resulted in R2 0.88, one of the lowest found in this
study (R2 for the rest of the plots can be found in Table 3).

the percentage of total porosity falling in each one of the pore


shape categories was determined.
The method developed by Chhabra and Jensen (1989) was
implemented in MatLab 6.0 (The Math Works Inc., Natick,
MA) and used to calculate the f()-spectra. Images were partitioned in boxes of size L, for L 2, 3, 5, 10, 25, 50, 100, 125,
and 250 pixels. A family of normalized measures i(q,L ) was
constructed for positive and negative values of q covering
variable ranges in steps of 0.1:

i(q,L)

P iq(L)
N(L)

[11]

P iq(L)

i1

where Pi(L ) is the fraction (or probability) of pores contained


in each ith box of size L (Eq. [3]). Note that for any value of
q, the normalized measures take values in the interval [0,1].
The direct computation of f(q ) values is (Chhabra et al., 1989;
Chhabra and Jensen, 1989):
N(L)

f(q) lim
L0

i1

i(q,L)log[i(q,L)]
logL

[12]

In addition, values of (q ) were computed by evaluating:

i1

i(q,L) log[Pi(L)]
logL

[13]

For each q, values of f(q ) and (q ) were obtained from


the slope of plots of the numerators of Eq. [12] and [13] vs.
log L over the entire range of L values considered (2250
pixels). The range of q values over which both functions were
linear, q, was selected considering the coefficients of determination (R2) of both fits (Fig. 2). The f(q ) and (q ) functions
obtained over a given q (Fig. 3), were used to construct the
f()-spectra as an implicit function of q and L. In addition, we
tested the validity of the results by verifying that the tangent of
the graph f() vs. at (q 1) is the bisector defined by
df()/d q. The point of intersection corresponds to
f[(1)] [1] D1 (Evertsz and Mandelbrot, 1992). The
symmetry of multifractal spectra was evaluated by comparing
the width of the spectra from their center [(0)] to (|qi|).
Values of |qi| were the same in both the positive and negative
domains and equal to the smaller of the two defining a q interval.

RESULTS AND DISCUSSION


Thin sections for this study were primarily selected
to include contrasting soil structures (Fig. 4). Soil texture
among selected soils showed wide ranges in the percentages of sand (1.494.6%), silt (4.967.1%), and clay (0.4
72%). Organic matter content was relatively low (0.1
1.7%) because mainly subsurface horizons were sampled
(Table 1). In this study, distributions of cumulative pore
area ranged from relatively balanced distributions to
bimodal ones with few pores concentrating relatively
large percentage of the porosity (Fig. 5). Formation of
soil structure implies development of a bimodal poresize distribution with the largest pores (macropores)
being found between structural units. The largest variability in the distribution of pore area across soils occurred in the macropore size, as indicated by the wide
range of variation in the values of the MAU (Table 2).
Typically, macropores resulting from the formation of

POSADAS ET AL.: MULTIFRACTAL CHARACTERIZATION OF SOIL PORE SYSTEMS

1365

Fig. 5. Selected cumulative distributions of pore area measured on


binary images.

Fig. 4. Binary images of the studied soil thin sections separated in


groups of similar pore properties.

structural units are elongated in shape and interconnected (Bouma et al., 1977; Ringrose-Voase and Bullock, 1984). Except for Soil 9 and Soil 10, between 50
and 74% of the total porosity contributed by pores with
area larger than 0.27 106 m2 was formed by elongated
pores suggesting that pedogenetic processes were active
in these soils (Table 2).
Three groups of soils were distinguished using the
values of porosity and MAU. Soils in Group 1 (Soil 1

to Soil 5) have relatively higher porosities and smaller


MAU values than soils in Group 2 (Soil 6 to Soil 8), but
in both groups elongated pores concentrate the largest
proportion of soil porosity (Table 2). On the other hand,
soils in Group 3 (Soil 9 and Soil 10) have the lowest
porosity and smallest values of MAU. Pore shapes in
this group were predominately irregular and rounded
(Table 2). Rounded pores are typically the result of
random packing of particles or aggregates, whereas irregular voids may originate from the compaction of
rounded pores or from biological activity (RingroseVoase, 1987; Pagliai and De Nobili, 1993). A visual
assessment of soil structure shows that structural units
in Group 1 were smaller than in Group 2, but in both
cases more developed than the two samples in Group
3 that exhibited massive structure characterized by a
coherent mass and the absence of structural units
(Fig. 4).

Multifractal Analysis
A crucial step in multifractal analysis is to determine
the range of both L and (negative and positive) moments of order q over which a multifractal method is

Table 2. Total porosity, mean pore area of the lower (MAL) and upper (MAU) one-half cumulative distribution of this property, and
pore shape classes (expressed as percentage of total porosity) measured from binary images.
Pore-shape classes

Group
1

2
3

Soil

Total
porosity

MAL
106m2

MAU
106m2

1
2
3
4
5
6
7
8
9
10

0.17
0.15
0.20
0.16
0.21
0.09
0.13
0.13
0.06
0.06

0.051
0.046
0.068
0.037
0.066
0.053
0.054
0.083
0.025
0.034

5.049
3.832
3.204
0.899
1.770
8.608
77.646
12.699
0.345
0.692

Only pores with area 0.27 106 m2 are included.

F 0.2

0.2 F 0.5

F 0.5

% Total porosity
50.2
55.2
49.8
34.9
51.7
61.6
73.5
67.4
11.7
13.2

18.6
11.4
23.8
16.9
17.1
13.3
3.35
9.5
18.3
28.1

1.7
2.3
1.9
2.4
1.2
3.0
2.8
4.2
2.4
8.8

1366

SOIL SCI. SOC. AM. J., VOL. 67, SEPTEMBEROCTOBER 2003

Table 3. Selected multifractal parameters the residual error of the estimates from the analysis of binary images. Also shown are the
values of the coefficients of determination of the fits (R2).
q
Group
1

2
3

Soil

1
2
3
4
5
6
7
8
9
10

1.4
2.5
1.4
2.0
1.3
1.0
1.3
1.0
1.0
2.0

2.3
2.8
3.5
3.0
5.0
1.3
1.0
1.4
2.4
4.0

R2

D0
1.75
1.73
1.79
1.78
1.82
1.58
1.64
1.65
1.53
1.56

0.03
0.04
0.03
0.03
0.03
0.05
0.04
0.04
0.05
0.04

0.99
0.99
0.99
0.99
0.99
0.98
0.98
0.99
0.98
0.98

R2

D1
1.70
1.67
1.73
1.72
1.77
1.56
1.63
1.63
1.49
1.50

0.09
0.02
0.02
0.02
0.02
0.08
0.07
0.09
0.03
0.04

0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99

f [(1)]

R2

0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99

1.70
1.66
1.73
1.71
1.74
1.57
1.62
1.64
1.51
1.52

0.18
0.10
0.10
0.10
0.10
0.10
0.12
0.16
0.17
0.12

(0)

R2

0.98
0.98
0.98
0.98
0.99
0.95
0.96
0.97
0.97
0.96

1.81
1.80
1.85
1.84
1.88
1.61
1.67
1.68
1.59
1.63

0.04
0.05
0.04
0.04
0.04
0.06
0.06
0.05
0.06
0.06

max

R2

0.88
0.87
0.90
0.92
0.93
0.85
0.87
0.85
0.85
0.87

1.94
2.01
2.00
2.03
2.06
1.65
1.74
1.72
1.65
1.77

1.30
1.40
1.30
1.00
1.00
1.30
1.20
1.30
1.10
1.40

min

R2

0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99

1.67
1.60
1.67
1.65
1.66
1.56
1.63
1.63
1.44
1.36

0.05
0.09
0.08
0.06
0.10
0.05
0.07
0.02
0.05
0.16

D0, D1, and f [(1)] are fractal dimensions at q 0, q 1, and q 1, respectively; (0), max, and min are the values of the Lipschitz-holder exponent
at q 0, and at the most negative (q) and most positive (q) values that defined the range q over which the multifractal method was applied.

applicable (Saucier and Muller, 1999). In our case, this


meant determining the range of L and q in which the
numerators of Eq. [12] and [13] are linear functions of
log L. A constant L range (2250 pixels) was used in
this study to avoid representing multifractal properties
over a small interval of scales. The range of q values
was selected considering the coefficients of determination (R2) of the fits (see Fig. 2), which in general were
equal or larger than 0.95 (Table 3). In a few cases, lower
values of R2 were accepted when the relatively poorer
fit involved only one of the two functions (Fig. 2). The
largest variation in q was observed in the range of
positive q values (Table 3). Moments q 0 magnify
the contribution of boxes with high concentration of
pores to the estimates of either f(q) or (q). The opposite is true for q 0. Soils in Group 1 and Group 3
exhibited multifractal properties over a wider range of
moments q than soils in Group 2 (Table 3). Since soils
in Group 2 have larger pores than soils in Group 1 and
Group 3, it is possible that a complete characterization
of pore scaling in soils of Group 2 would require sampling an area larger than the provided by a thin section.
Analyses of local porosity and of the mean entropy of
a soil structure have shown that, in general, the size of
thin sections is enough for an appropriate representation of the lower moments of porosity distributions
(Dexter, 1977; VandenBygaart and Protz, 1999), but
there are no studies on the relationship between sampled area and the higher moments of a measure of porosity.
The f()-spectra among groups showed distinctively
different shape and symmetry (Fig. 6). The curvature
and the symmetry of the f()-spectra provide information on the heterogeneity of a system defined by the
diversity of scaling exponents needed to characterize it.
The f()-spectra of homogeneous systems with fractal
support is reduced to a single point corresponding to
the maximum value of f() found at q 0, that is,
f[(0)] D0. Heterogeneity can be assessed at q 0
by the magnitude of the differences in the values of D0
and (0), or more generally, by the magnitude of
changes around D0 in both the f() and axes. Values
to the right and left of D0 represent negative and positive
q values, respectively (see Fig. 3). In the f() axis, comparison is between D0 and the values of f[(1)] or (D1)

and f[(1)]. In the axis, the difference (max min)


is used as an indication of the heterogeneity of a system.
In our analysis, porosity and D0 values were highly
correlated (Table 4). These results are similar to the
ones reported by Lipiec et al. (1998), who correlated
box-counting dimensions of pore volume with porosity
for a soil compacted at various levels (R 0.97). Correlation between D0 and porosity provides no information
on the scaling properties of a pore structure because
calculation of D0 assumes a homogeneous soil structure
(at every partition level, all boxes have the same probability). On the other hand, (q) can distinguish among
soil structures by quantifying the average scaling of mass
density (or probability) with L. Porosity and (0) were
also highly correlated (Table 4), but a comparison between D0 and (0) values reveal differences in pore
structure among soils. Soils in Group 2 exhibited the
most homogeneous pore system as demonstrated by the
proximity of their values to the 1:1 line (Fig. 7a). This
tendency to homogeneity (or simple scaling) in soils of
Group 2 is further confirmed by the notably smaller
values of the differences D0 D1, D0 f[(1)], and
(max min) for soils in this group (Table 3). The implication of this finding is that pore systems in Group 2 soils
can be characterized almost entirely by the capacity
dimension D0. This interpretation needs to be weighted
by considering that soils in this group scaled over a
narrow range of q values (Table 3), which might indicate
that in the presence of macropores sampling area should
be larger than the one used here. This hypothesis is
supported by the negative correlation between log of
MAU and the largest value of positive q over which the
multifractal method applies (Table 4).
In a symmetric spectrum, the widths from (0) to
(|qi|) are equal (or very similar) implying that regions
with high and low concentrations of mass scale similarly.
Soils in Group 1 and Group 2 had f()-spectra that were
more symmetrical than spectra from soils in Group 3
as indicated by the proximity of the -intervals to the
1:1 line (Fig. 7b). The left-hand side of the f()-spectra
of soils in Group 3 exhibited the lowest f() values
in this region of the spectrum. Fractal dimensions with
values similar to the observed f() values have been
obtained in analyses of dye-stained patterns, which are
typically disconnected in the plane perpendicular to flow

POSADAS ET AL.: MULTIFRACTAL CHARACTERIZATION OF SOIL PORE SYSTEMS

1367

(Table 3 and Fig. 6). On the other hand, several studies


have demonstrated relationships between textural composition and soil microstructure (Eswaran and Banos,
1976; Brewer, 1979). If general relationships between
multifractal parameters of soil structure and soil properties are established, it could serve as a first approximation to the characterization and modeling of soil pore
systems. In our study silt content was significantly correlated to D0, (0), and D1 (Table 4). These parameters
represent the behavior at or around the maximum point
of the f()-spectra, and the correlations could be the
result of silt (and to a lesser extent sand) content being
correlated to total porosity. Even though more research
is needed to understand the mechanisms underlying
these correlations, they illustrate the potential for developing characterization schemes of soil structure that
includes soil texture as a general framework (Eswaran
and Banos, 1976; Brewer, 1979). More accurate characterization of relationships between texture and soil
structure may need to involve the parameterization of
the particle-size distribution (Posadas et al., 2001).

CONCLUSIONS

Fig. 6. f()-spectra for soils in (a) Group 1, (b) Group 2, and (c)
Group 3, as shown in Fig. 4.

(Lipiec et al., 1998). The absence of structural units


(massive structure) and pore clustering is also evident
in the lowest values obtained for these two soils.
It is generally recognized that particle- and pore-size
distributions are not well correlated in structured soils.
For instance, Soil 7 and Soil 9 have very similar contents
of sand, silt, and clay (Table 1), but different f()-spectra

The multifractal f()-spectra successfully separated


contrasting soil structural types. Three soil groups, distinguished based on size and shape of pores, resulted
in f()-spectra that were distinctively different. Properties of the f()-spectra important to distinguish among
groups were: (i) the range of negative and positive q
values over which scaling was verified, and (ii) the shape
and symmetry of the spectra. Soil pore structure containing clusters of relatively large pores exhibited scaling
over a reduced range of moment q and relatively flat
f()-spectra, which might have resulted from a limited
sampling area in relation to the size of the largest pores.
Asymmetrical f()-spectra were obtained in samples of
massive structure characterized by low porosity dominated by small rounded and irregular pores and a matrix
without noticeable structural units. The lowest values
of and f() obtained in this group suggest a pore
system with disconnected pores and low spatial density.
Symmetric f()-spectra exhibiting the typical concave
downward shape were obtained with soil structures that
combined small structural units and relatively small
pores.
The advantage of multifractal techniques is that a
f()-spectrum integrates and quantifies spatial properties of a pore system. Conceivably, multifractal parameters could be used to improve classifications of soil
structure. Thus, an approximation to the shape and
properties of a spectrum could be useful as an approximation to pore structure. The relationship between silt
content and multifractal parameters found in this study
indicate the potential for such approach, but a systematic and more extensive analysis of images of soil sections representing various levels of the soil taxonomy
is needed to identify and interpret parameters from the
f()-spectrum and their relation to soil properties and
processes.

1368

SOIL SCI. SOC. AM. J., VOL. 67, SEPTEMBEROCTOBER 2003

Table 4. Correlation coefficients between selected multifractal parameters and pore properties measured in thin sections (total porosity,
log MAL, and log MAU) or on different soil samples (organic matter, sand, silt, and clay contents).
Multifractal
parameters
D0
D1
(0)
max
min
q
q

Total porosity

log MAL

log MAU

Organic matter

Sand

Silt

Clay

0.940***
0.987***
0.940***
0.844**
0.505
0.118
0.368

0.493
0.613
0.387
0.189
0.062
0.446
0.237

0.039
0.196
0.079
0.225
0.016
0.350
0.715*

0.335
0.364
0.321
0.323
0.666
0.139
0.022

0.520
0.547
0.492
0.403
0.351
0.281
0.407

0.669*
0.669*
0.641*
0.544
0.438
0.147
0.342

0.090
0.050
0.086
0.084
0.045
0.181
0.181

* Significant at the 0.05 probability level.


** Significant at the 0.01 probability level.
*** Significant at the 0.001 probability level.
Significant at the 0.1 probability level.
D0 and D1 are fractal dimensions at q 0 and q 1, respectively; (0), max, and min are the values of the Lipschitz-holder exponent at q 0, and at
the most negative (q) and most positive (q) values that defined the range q over which the multifractal method was applied.

Fig. 7. Plots of (a) (0) vs. D0 and (b) widths of -intervals: (0) (qi,) vs. (qi,) (0) for soils in the three groups considered. The
numerical values of qi, and qi, were the same and equal to the smallest of the two values defining a q interval (see Table 3). Proximity
to the 1:1 line implies (a) more homogeneous distributions, or (b) more symmetric f()-spectra. In plot (b) two soils of Group 2 had identical
values and show as one point in the graph.

POSADAS ET AL.: MULTIFRACTAL CHARACTERIZATION OF SOIL PORE SYSTEMS

ACKNOWLEDGMENTS
Prof. Richard Protz passed away suddenly on 17 Nov. 2001
without being able to participate in the preparation of this
paper. We gratefully acknowledge his active and valuable contributions throughout the duration of this project. The authors
are grateful to Dr. Ana Mara Tarquis for useful discussions
on multifractal theory, and to the reviewers for their helpful
and constructive comments.

REFERENCES
Anderson, A.N., A.B. McBratney, and E.A. FitzPatrick. 1996. Soil
mass, surface, and spectral fractal dimensions estimated from thin
section photographs. Soil. Sci. Soc. Am. J. 60:962969.
Bouma, J., A. Jongerius, O. Boersma, A. Jager, and D. Schoonderbeek. 1977. The function of different types of macropores during
saturated flow through four swelling soil horizons. Soil Sci. Soc.
Am. J. 41:945950.
Brewer, R. 1979. Relationships between particle size, fabric and other
factors in some Australian soils. Aust. J. Soil Res. 17:2941.
Callen, H.B. 1985. Thermodynamics and an introduction to thermostatistics. 2nd ed. John Wiley & Sons, New York.
Caniego, F.J., M.A. Martn, and F. San Jose. 2001. Singularity features
of pore-size soil distribution: Singularity strength analysis and entropy spectrum. Fractals 9:305316.
Cheng, Q., and F.P. Agterberg. 1996. Multifractal modeling and spatial
statistics. Math. Geol. 28:116.
Cheng, Q. 1999. Multifractality and spatial statistics. Comput.
Geosci. 25:949961.
Chhabra, A.B., C. Meneveu, R.V. Jensen, and K.R. Sreenivasan.
1989. Direct determination of the f () singularity spectrum and
its application to fully developed turbulence. Phys. Rev. A 40:
52845294.
Chhabra, A.B., and R.V. Jensen. 1989. Direct determination of the
f() singularity spectrum. Phys. Rev. Lett. 62:13271330.
Dexter, A.R. 1976. Internal structure of tilled soil. J. Soil Sci. 27:
267278.
Dexter, A.R. 1977. A statistical measure of the structure of tilled soil.
J. Agric. Eng. Res. 22:101104.
Droogers, P., A. Stein, J. Bouma, and G. de Boer. 1998. Parameters
for describing soil macroporosity derived from staining patterns.
Geoderma 83:293308.
Eswaran, H., and C. Banos. 1976. Related distribution patterns in
soils and their significance. An. Edaf. Agriobiol. 35:3345.
Evertsz, C.J.G., and B.B. Mandelbrot. 1992. Multifractal measures.
p. 921953. In H.-O. Peitgen et al. (ed.). Chaos and Fractals. New
Frontiers of Science. Springer-Verlag, New York.
Feder, J. 1988. Fractals. Plenum Press, New York.
Folorunso, O.A., C.E. Puente, D.E. Rolston, and J.E. Pinzon. 1994.
Statistical and fractal evaluation of the spatial characteristics of
soil surface strength. Soil Sci. Soc. Am. J. 58:284294.
Frisch, U., and G. Parisi. 1985. Fully developed turbulence and intermittency. p. 84. In M. Ghil et al. (ed.). Turbulence and preditability
in geophysical flows and climate dynamics. North-Holland, NY.
Garboczi, E.J., D.P. Bentz, and N.S. Martys. 1999. Digital images and
computer modelling. p. 141. In P-Z Wong (ed.). Experimental
methods for porous media. Academic Press, San Diego, CA.
Gimenez, D., R.R. Allmaras, E.A. Nater, and D.R. Huggins. 1997.
Fractal dimensions for volume and surface of interaggregate
poresScale effects. Geoderma 77:1938.
Goenadi, D.H., and K.H. Tan. 1989. Relationship of soil fabric and
particle-size distribution in a Davison soil. Soil Sci. 147:264269.
Grout, H., A.M. Tarquis, and M.R. Wiesner. 1998. Multifractal analysis of particle size distributions in soil. Environ. Sci. Technol. 32:
11761182.
Halsey, T.C., M.H. Jensen, L.P. Kadanoff, I. Procaccia, and B.I. Shrai-

1369

man. 1986. Fractal measures and their singularities: The characterization of strange sets. Phys. Rev. A 33:11411151.
Hentschel, H.G.E., and I. Procaccia. 1983. The infinite number of
generalized dimensions of fractals and strange attractors.
Physica D 8:435444.
Holden, N.M. 2001. Description and classification of soil structure
using distance transform data. Eur. J. Soil Sci. 52:529545.
Horgan, G.W. 1998. Mathematical morphology for analysing soil
structure from images. Eur. J. Soil Sci. 49:161173.
Horgan, G.W. 1999. An investigation on the geometric influences on
pore space diffusion. Geoderma 88:5571.
Jim, C.Y. 1988. A classification of soil microfabrics. Geoderma 41:
315325.
Korvin, G. 1992. Fractal methods in the earth science. Elsevier, Amsterdam.
Kravchenko, A., C.W. Boast, and D.G. Bullock. 1999. Multifractal
analysis of soil spatial variability. Agron. J. 91:10331041.
Lipiec, J., R. Hatano, and A. Słowinska-Jurkiewicz. 1998. The
fractal dimension of pore distribution patterns in variously-compacted soils. Soil Tillage Res. 47:6166.
Loehle, C., and B.-L. Li. 1996. Statistical properties of ecological and
geological fractals. Ecol. Modell. 85:271284.
Mandelbrot, B.B. 1982. The fractal geometry of nature. 2nd ed. W.H.
Freeman and Co., New York.
Martn, M.A., and E. Montero. 2002. Laser diffraction and multifractal
analysis for the characterization of dry soil volume-size distribution.
Soil Tillage Res. 64:113123.
Meakin, P. 1991. Fractal aggregates in geophysics. Rev. Geophys.
29:317354.
Meneveau, C., and K.R. Sreenivasan. 1991. The multifractal nature
of turbulent energy dissipation. J. Fluid Mech. 224:429484.
Moran, C.J., and A.B. McBratney. 1997. A two-dimensional fuzzy
random model of soil pore structure. Math. Geol. 29:755777.
Muller, J., and J.L. McCauley. 1992. Implication of fractal geometry
for fluid flow properties of sedimentary rocks. Transp. Porous
Media 8:133147.
Pachepsky, Ya., V. Yakovchenko, M.C. Rabenhorst, C. Pooley, and
L.J. Sikora. 1996. Fractal parameters of pore surfaces as derived
from micromorphological data: Effect of long term management
practices. Geoderma 74:305319.
Pagliai, M., and M. De Nobili. 1993. Relationships between soil porosity, root development and soil enzyme activity in cultivated soils.
Geoderma 56:243256.
Posadas, A., D. Gimenez, M. Bittelli, C.M.P. Vaz, and M. Flury. 2001.
Multifractal characterization of soil particle-size distributions. Soil
Sci. Soc. Am. J. 65:13611367.
Rasband, W. 1993. NIH-Image v. 1.52. An image processing system
for the Macintosh. U.S. National Institute of Health, NTIS, Springfield, VA.
Ringrose-Voase, A.J. 1987. A scheme for the quantitative description
of soil macrostructure by image analysis. J. Soil Sci. 38:343356.
Ringrose-Voase, A.J., and P. Bullock. 1984. The automatic recognition
and measurement of soil pore types by image analysis and computer
programs. J. Soil Sci. 35:673684.
Saucier, A., and J. Muller. 1999. Textural analysis of disordered materials with multifractals. Physica A 267:221238.
Shannon, C.E., and W. Weaver. 1949. The Mathematical Theory of
Communication. University of Illinois Press. Urbana, IL.
Stanley, H.E., and P. Meakin. 1988. Multifractal phenomena in physics
and chemistry. Nature 335:405409.
VandenBygaart, A.J., and R. Protz. 1999. The representative elementary area (REA) in studies of quantitative soil micromorphology.
Geoderma 89:333346.
Vicsek, T. 1992. Fractal growth phenomena. 2nd ed. Word Scientific
Publishing Co., Singapore.
Vogel, H.-J., and K. Roth. 2001. Quantitative morphology and network representation of soil pore structure. Adv. Water Resour.
24:233242.
Voss, R.F. 1988. Fractals in nature: From characterization to simulation. p. 2169. In H.-O. Peitgen and D. Saupe (ed.). The Science
of Fractal Images. Springer-Verlag, New York.

You might also like