Characterization of Particle-Size Distribution in Soils with a Fragmentation Model
Marco Bittelli,* Gaylon S. Campbell and Markus Flury
ABSTRACT
Particle-size distributions (PSDs) of soils are often used to estimate
other soil properties, such as soil moisture characteristics and hydraulic
conductivities. Prediction of hydraulic properties from soil texture
requires an accurate characterization of PSDs. The objective of this
study was to test the validity of a mass-based fragmentation model
to describe PSDs in soils. Wet sieving, pipette, and light-diffraction
techniques were used to obtain PSDs of 19 soils in the range of 0.05
to 2000 fun. Light diffraction allows determination of smaller particle
sizes than the classical sedimentation methods, and provides a high
resolution of the PSD. The measured data were analyzed with a mass-
based model originating from fragmentation processes, which yields
a power-law relation between mass and size of soil particles. It was
found that a single power-law exponent could not characterize the
PSD across the whole range of the measurements. Three main power-
law domains were identified. The boundaries between the three do-
mains were located at particle diameters of 0.51 ± 0.15 and 85.3 ±
25.3 ixm. The exponent of the power law describing the domain be-
tween 0.51 and 85.3 JJLIII was correlated with the clay and sand contents
of the soil sample, indicating some relationship between power-law
exponent and textural class. Two simple equations are derived to
calculate the parameters of the fragmentation model of the domain
between 0.51 and 85.3 ji.ni from mass fractions of clay and silt.
P ARTICLE-SIZE DISTRIBUTION in soil is one of the more
fundamental soil physical properties. It is widely
used for the estimation of soil hydraulic properties such
as the water-retention curve and saturated as well as
unsaturated conductivities (Arya and Paris, 1981;
Campbell and Shiozawa, 1992). Generally, a conven-
tional particle-size analysis involves the measurement
of the mass fractions of clay, silt, and sand. These frac-
tions may be used to find the textural class using a
textural diagram, commonly in form of a textural trian-
gle (e.g., Gee and Bauder, 1986). However, soil samples
that fall into a certain textural class may have consider-
ably different PSDs. For example, the textural class of
"clay" in the USDA classification scheme (Gee and
Bauder, 1986) contains soil samples that vary in clay
content between 40 and 100%. The size definitions of
the three main particle fractions of clay, silt, and sand,
used as diagnostic characteristics in most classification
schemes, are rather arbitrary, and they do not provide
complete information on the soil PSD.
A more accurate description of texture is obtained
by defining a PSD function. Commonly, PSDs are re-
ported as cumulative distributions, and different func-
tions have been proposed to fit experimental data. Bu-
chan et al. (1993) fitted several of these models to
experimental data and found the bimodal lognormal
distribution proposed by Shiozawa and Campbell (1991)
M. Bittelli, G.S. Campbell, and M. Flury, Department of Crop and
Soil Sciences, Washington State University, Pullman, WA 99164. Re-
ceived 26 Aug. 1998. *Corresponding author (bittelli@mail.wsu.edu).
Published in Soil Sci. Soc. Am. J. 63:782-788 (1999).
782
to be best suited. The Shiozawa and Campbell model
divides the particle distribution into two parts domi-
nated by primary (sand and silt) and secondary (clay)
minerals, respectively. However, as pointed out by Bu-
chan et al. (1993), the assumption of a lognormal distri-
bution in the clay fraction cannot be justified because
Shiozawa and Campbell (1991) had no data available
in that range.
One of the latest developments in the study of PSDs
in soils has focused on the use of fractal mathematics
to characterize particle sizes in soil (Turcotte, 1986;
Tyler and Wheatcraft, 1992; Wu et al., 1993). However,
questions remain about the validity and applicability of
fractal concepts to PSDs. There has been some discus-
sion about the proper use and definition of the term
"fractal" in the literature (Young et al., 1997; Pachepsky
et al., 1997; Baveye and Boast, 1998). Different concepts
of fractals are used, and these concepts lead to different
interpretations of fractal dimensions obtained. There-
fore it is essential to clearly specify the type of fractal
model used.
Particle- and aggregate-size distributions are often
rendered as cumulative functions, either as number of
particles larger than a certain diameter, or as mass
smaller than a certain diameter. These cumulative distri-
bution functions have been analyzed with power-law
relations and the exponents interpreted as fractal di-
mensions. Tyler and Wheatcraft (1989, 1992) analyzed
particle-size data ranging from 0.5- to 5000-u,m radii,
and observed that the fractal power law was not valid
across the entire extent of particle sizes. It is expected
that there are lower and upper limits to the validity
of fractal relations (Turcotte, 1986). Wu et al. (1993)
measured PSDs down to 0.02-jjun radius by using light-
scattering techniques, and found a power-law relation
between number of particles and particle radius valid
across a range of particle radii with a lower cutoff be-
tween 0.05 and 0.1 (am and an upper cutoff between 10
and 5000 |xm. Assuming that the exponent of a power-
law relation is a fractal dimension, Wu et al. (1993)
found a dimension of D = 2.8 ± 0.1 for the four soils
studied and suggested that this might be a universal
value of an underlying structure. Kozak et al. (1996)
analyzed PSDs of 2600 soil samples and found that for
50% of the samples power-law scaling of particle num-
bers vs. size was not applicable across the whole range
of particle sizes between 2 and 1000 ixm. The authors
indicate that power-law scaling might be applicable for
a narrower range of particle sizes, although this was not
analyzed in their study.
Most applications of fractal concepts to particle- and
aggregate-size distributions are based on the fragmenta-
tion model of Matsushita (1985) and Turcotte (1986).
Abbreviations: PSD, particle-size distribution; RMSE, root mean
square error.
 BITTELLI ET AL.: FRAGMENTATION MODEL TO CHARACTERIZE PARTICLE-SIZE DISTRIBUTION 783

Table 1. Soil classification, geological parent material, percentage of sand, silt and clay by weight, and organic C content for the 19 soils
used. Particle-size data were obtained by sieving and light-diffraction methods. Textural classes are according to the USD A classification.
Soils Soil classification! Geological parent material Sand Silt Clay OC§
o/
—————————— /o ——————————
Affoltern Typic Haplndalf moraine 47.4 48.5 4.1 2.2
Aeiigst Typic Hydraquent fluvial deposits 38.7 55.6 5.7 1.9
Buelach Typic Hapludalf gravel deposits 57.2 40.4 2.4 2.2
Les Barges Mollic/Aquic Udifluvent fluvial deposits 74.2 25.5 0.3 1.0
Mettmenstctten Lithic Ustorthent moraine 55.6 40 4.4 0.4
Murimoos Lithic Medihemist fluvial deposits 69.9 29.6 0.5 1.0
Obermumpf Lithic Rendoll limestone 25.6 69.2 5.2 0.5
Obfelden Typic Hydraquent fluvial deposits 36.3 59.6 4.1 0.6
Palousei Ultic Haploxeroll loess 13.2 68.6 18.2 nafl
Reckenholz Vertic/Typic Eutrochrept moraine 23.8 70.7 5.5 1.3
Red Bluffi Ultic Palcxeralfs fluvial deposits 17.9 36.5 45.6 na
Rheinau Arenic Eutrochrept fluvial gravels 68.1 29.2 2.7 0.8
Royalt Ultic Haploxeroll glaciofluvial sediments 30.7 63.1 6.2 na
Salkunit Xeric Palehumults glacial drift 11.9 59.7 28.4 na
Walla Wallal Typic Haploxeroll loess 8.3 78.4 13.3 na
Wetzikon 1 Lithic Ruptic-Alfic Eutrochrept moraine 48.9 46.7 4.4 0.9
Wetzikon 2 Rendollic Eutrochrept moraine 59.7 37.2 3.1 0.9
Wuelflingen Vertic/Typic Eutrochrept anthropogenic deposits 32.7 60.5 6.8 0.7
Zeiningen Ultic Hapludalf floess 40.5 55.4 4.1 0.4
t U.S. soil taxonomy.
j Soils from USA.
§ OC, organic C percentage by weight, determined with Walkley-Black method (Nelson and Sommers, 1982).
H na, not available.

In this model, the fragmentation of an initially intact
particle into smaller particles leads to a power-law rela-
tion between (i) number or (ii) mass of particles as a
function of particle size. These two types of fragmenta-
tion relations are known as number-based and mass-
based approaches (Turcotte, 1992). The power-law ex-
ponent of the number-based approach can be inter-
preted as fractal dimension (Matsushita, 1985; Turcotte,
1986). It is worth noting that the fragmentation model
does not lead to a geometrical fractal with the fractal
dimensions confined between Euclidian dimensions.
The sorting of particles by size in the fragmentation
model results in fractal dimensions ranging theoretically
between the limits of 0 and 3 (Turcotte, 1986). Borkovec
et al. (1993) experimentally determined fractal dimen-
sions of fragmentation and surface areas of soil particles
and found the two dimensions to be 2.8 ± 0.1 and 2.4 ±
0.1, respectively.
The objective of this study was to test the mass-based
fragmentation approach proposed by Turcotte (1986)
for characterizing PSDs, and to determine the range of
particle diameters where power-law scaling is applica-
ble. To test the general validity and the extent of power-
law scaling it is of fundamental importance to have
data that span several orders of magnitude. Traditional
sedimentation and hydrometer techniques for the mea-
surement of PSDs yield only limited data in the clay
fraction smaller than 2 u,m. Light-scattering methods
overcome this problem and provide data between 0.05
to 1000 |xm.
MATERIALS AND METHODS
Particle-Size Analysis
Nineteen soils were used in this study, five of them were
from the USA and 14 from Switzerland. The soils were chosen
such that they represent a wide variety of parent materials,
weathering conditions, and textures. Characteristic properties
of these soils are summarized in Table 1. All soil samples were
dried at 105°C, gently crushed, and passed through a 2-mm
sieve. Each sample was tested for the presence of carbonates
using cold 1M HQ, and if carbonates were present, the sample
was treated with 0.5 M sodium acetate at 75°C for at least 1 h.
After acetate treatment, samples were washed with deionized
water. The five soil samples from the USA were further pre-
treated by destroying organic matter using H2O2 (30%, w/w)
at 65°C. The 14 Swiss soil samples were not pretreated for
organic matter. The absence of pretreatment for organic mat-
ter could in some cases have affected the dispersion of particles
for the Swiss soils, leading to incomplete segregation, and
therefore to an underestimation of small particle fractions.
Organic matter contents of the Swiss soils, determined with the
Walkley-Black method (Nelson and Sommers, 1982), ranged
from 0.4 to 2.2% by weight (Table 1).
After pretreatment, all samples were dried at 105°C for 24
h. Prior to particle-size analysis, all soil samples were dispersed
in 1 g Lr1 hexametaphosphate solution and shaken for 24 h
to destroy aggregates. For the pipette analysis, samples were
wet sieved with the hexametaphosphate solution at 1000-,
500-, 250-, 125-, and 53-jmi mesh sizes. The material smaller
than 53 |j,m was then analyzed by the pipette method (Gee
and Bauder, 1986). To obtain four size classes between 2 and
50 jxm, sedimentation techniques based on Stoke's law were
used to obtain the following diameters: <2, <5, <10, and <20
(jtm. For the light-scattering technique, the soil samples were
wet sieved down to a size of 250 (Jim for the Swiss soils and
500 n,m for the U.S. soils. The particles passing the smallest
sieve mesh were collected in a bucket, dried at 105°C, and
subsequently analyzed by light diffraction. A 3-g aliquot of
the dried material was introduced into an ultrasonic bath unit
of a small-angle light-scattering apparatus (Malvern Master
Sizer MS20, Malvern, England) equipped with a low-power
(2 mW) Helium-Neon laser with a wavelength of 633 nm as
the light source.1 Suspension concentrations were adjusted to
an obscuration of the primary beam of =0.1 to 0.2%. The
obscuration values were set to optimize between best signal/
noise ratio and negligible multiple scattering effects. If the
sample concentration is too low, the obscuration and the inten-
1
Reference to company name does not reflect endorsement of
particular products by Washington State University.
784 SOIL SCI. SOC. AM. J., VOL. 63, JULY-AUGUST 1999
sity of the scattered light are low, leading to noisy data. If the
sample concentration is too high, then the light scattered from
a particle may be scattered again by a second particle, causing
errors in the final particle-size analysis. Prior to measurement,
samples were dispersed by sonication in an ultrasonic bath
for 25 min. A focal length of 300 mm was used with an ordinary
Fourier Optics configuration, and a focal length of 45 mm was
used for the inverse Fourier Optics configuration. The inverse
configuration allows the accurate measurement of scattering
at high angles in order to correctly measure the very fine
particles (sizes down to 0.01 (jun). Particle-size distribution
was obtained by fitting full Mie scattering functions for spheres
(Kerker, 1969).
Data Analysis
Soils are formed by weathering of geological parent mate-
rial. The weathering results in a fragmentation of the initial
solid rock or sediment. It has been recognized that the prod-
ucts of fragmentation in nature can often be described with
fractal concepts. For different types of objects, a power-law
relation between the number and size of objects has been
proposed (Mandelbrot, 1982; Matsushita, 1985; Turcotte,
1986)
N(r > R) = [1]
where N(r > R) is the number of objects per unit volume
having a radius r larger than R, C is a constant of proportional-
ity, and D is the fractal dimension. For soil particles, Turcotte
(1986) and Tyler and Wheatcraft (1992) pointed out that it is
generally more convenient to express the number-based power
law (Eq. [1]) as a mass-based form. The mass-based approach
is compatible with data obtained from experimentation, where
usually mass fractions rather than number fractions are mea-
sured. The mass-based form of Eq. [1] is expressed as (Tur-
cotte, 1986; Tyler and Wheatcraft, 1992)
M(r < R) _ ( R
[2]
R L, upper
where M(r < R) is the mass of soil particles with a radius
smaller than R, Mr is the total mass of particles with radius
less than ^L.uPPer, /?L,upper is the upper size limit for fractal behav-
ior, and v is a constant exponent. This power law can be
related to the fractal number relation by taking incremental
values as shown by Matsushita (1985) and Turcotte (1992).
Taking the derivatives of Eq. [1] and [2] with respect to the
radius R yields, respectively,
dN [3]
and
dM <* R"'1 dR [4]
Assuming a constant density of soil particles, the volume of
a particle with radius r is proportional to its mass m, hence
r 3 oc m; therefore, for incremental particle numbers and masses
we have
R'dN °c dM [5]
Substituting Eq. [3] and [4] into [5] gives (Turcotte, 1992)
R-D-I a fl-sflv-i [6]
from which it follows that
D = 3- v [7]
Equation [7] relates the exponent v of the mass-based ap-
proach to the exponent D of the number-based approach. As
100 1000
Particle Diameter (urn) Particle Diameter (urn)
Fig. 1. Cumulative particle-size distributions for four soils obtained
by two different experimental methods.
shown below by our experimental data and discussed in the
literature (Turcotte, 1992), the power-law relation given in
Eq. [2] has also a lower limit of validity. The radius R of
particles satisfying Eq. [2] is confined between RLJO^U <
D ^ D
** ^ ^*L,upper-
The mass-based fragmentation approach was used to ana-
lyze experimentally determined PSD data. The lower and up-
per limits /?Ljower and .Rapper as well as the power-law exponent
D = 3 - v were determined by the following procedure. A
linear regression was used to fit Eq. [2] on a log-log plot to
the experimental data. The entire range of experimental data
was used first and the residuals were calculated. Subsequently,
the upper- and lower-range data points were eliminated and
new residuals and root mean square errors (RMSE) were
calculated. In an iterative procedure, the RMSE error was
minimized by eliminating data points at the upper and
lower boundaries.
RESULTS AND DISCUSSION
Comparison between Pipette and
Light-Diffraction Methods
Most of the textural data reported in the literature
have been measured by sedimentation techniques, such
as hydrometer or pipette. It is therefore illustrative to
briefly compare experimental results obtained by pi-
pette and light-scattering methods. The results obtained
by the two techniques were in excellent agreement in
our study. Figure 1 shows a qualitative comparison be-
tween pipette and light-scattering methods for four soils.
Experimental differences in the cumulative fraction at
a given particle size obtained by the two methods were
in the order of 0.3 to 11.7%. Similar results were ob-
tained by Wu et al. (1993), who found that sedimenta-
tion and light-scattering techniques were in good agree-
ment for the majority of the soil samples used in their
experiment.
Characterization of Particle-Size Distribution
In Fig. 2, cumulative mass fractions are plotted as a
function of particle diameter on double logarithmic
scale for four soils. The plots clearly show that a single
 BITTELLI ET AL.: FRAGMENTATION MODEL TO CHARACTERIZE PARTICLE-SIZE DISTRIBUTION 785

1- ary was 0.51 jjum and of the silt-sand domain boundary

Log-Log Log-Log
was at 85.3 u,m, with a coefficient of variation of 15 and
0.1-
25%, respectively. The consistent occurrence of three
0.01 - power-law domains in all 19 soils and the close agree-
ment of the domain scales indicate similarity between
0.001 -
Salkum
the different soils, particularly when considering the
Aeugst
0.0001 - wide textural variability of the samples ranging from
0.3 to 46% clay. In a similar study on four soils, Wu et
1- al. (1993) also found three domains where a power law
Log-Log Log-Log
was applicable, but the limits between the domains were
located at 0.05 to 0.1 and 10 to 5000 u.m. The consistency
of the limits for the three domains needs therefore to
be investigated across a greater number of soils.
Walla Walla Wuelflingen We denote the three fractal dimensions determined
o
in our study as Dclay, £>sill, and Dsand. The fitted values of
0.1 1 10 100 1000 0.1 1 10 100 1000 the fractal dimensions obey in all cases the relation: Dday
Particle Diameter (urn) Particle Diameter (urn) < DSji, < Aand- In the clay domain, the fractal dimension
Fig. 2. Log-log plots of particle-size distributions for four soil samples. ranged from 0.118 to 1.21, in the silt domain from 1.728
Symbols denote experimental data, solid lines denote model fits. and 2.792, and in the sand domain from 2.839 to 2.998
(Table 2). These data are consistent with the limits of the
power law cannot describe the data across the entire fragmentation approach given as 0 < D < 3 (Turcotte,
range of measured particle sizes. There is evidence that 1986). The generally high value of the coefficient of
different power laws apply for three domains in all of determination R2 shows that the fragmentation models
the 19 soils. The solid lines in Fig. 2 are the curves of are good descriptions of the PSDs in the three domains.
Eq. [2] fitted to the different domains of particle sizes Some soils showed poor power-law agreement in the
on the log-log plots. sand domain (e.g., Obfelden and Reckenholz). Experi-
Optimized parameters of the fragmentation model mental data in the sand as well as in the clay domain
together with the median particle diameter for all the are limited by the experimental procedures, namely the
19 soils are shown in Table 2. The median diameter has maximum particle size as allowed by the 2-mm sieve
been calculated from the measured PSDs by linearly mesh and the minimum particle size determined by the
interpolating the 50% quantile (e.g., Sokal and Rohlfs, light-scattering technique.
1995). The identified power-law domains separate the The model used in the derivation of Eq. [2] is based
particle sizes in three classes, which we denote as clay, on the fragmentation of an initially intact particle into
silt, and sand domains. The diameter boundaries be- smaller particles (Matsushita, 1985; Turcotte, 1986). An
tween the clay and silt domains ranged from 0.33 to intact cubical particle of size h is fragmented into eight
0.99 |xm, and between silt and sand domains from 45.3 identical cubes of size h/2. Each of these smaller cubes
to 126.7 (Jim. The average of the clay-silt domain bound- is further divided in cubes with size h/4, and so forth.

Table 2. Fragmentation fractal dimensions, median particle diameter, and cutoff boundaries, estimated from particle-size distribution
data obtained by the light-diffraction method for the 19 soils.
Silt domain
(Jlay domain silt domain sand domain
Median Lower Upper
Soils DA, R1 A,,, R2 A™, R2 diameter ds. boundary boundary
,un |xm M-m
A ff oiler n 0.808 0.96 2.239 0.99 2.930 0.95 46.19 0.42 94.35
Aeugst 0.606 0.97 2.294 0.99 2.979 0.94 36.51 0.38 93.93
Buelach 0.596 0.96 2.122 0.99 2.898 0.96 60.01 0.99 98.41
Les Barges 0.808 0.99 1.768 0.99 2.839 0.98 74.96 0.53 124.58
Mettmenstetten 0.792 0.96 2.297 0.98 2.858 0.98 109.79 0.58 74.99
Murimoos 0.255 0.99 1.728 0.99 2.948 0.91 74.44 0.56 112.92
Obermumpf 0.701 0.96 1.801 0.99 2.974 0.98 30.96 0.40 56.93
Obfelden 1.210 0.97 2.152 0.98 2.969 0.85 24.86 0.40 69.98
Palouse 0.118 0.96 2.504 0.99 2.996 0.91 14.34 0.44 54.21
Reckenholz 0.789 0.96 2.238 0.99 2.998 0.81 29.22 0.40 71.58
Red Bluff 0.174 0.96 2.792 0.99 2.921 0.99 2.88 0.51 77.92
Rheinau 0.799 0.96 2.251 0.99 2.815 0.97 77.58 0.42 126.73
Royal 0.987 0.95 2.269 0.99 2.981 0.94 35.56 0.56 90.46
Salkum 0.214 0.97 2.618 0.98 2.953 0.99 8.18 0.63 45.31
Walla Walla 0.896 0.94 2.384 0.98 2.973 0.91 16.57 0.61 50.86
Wetzikon 1 0.796 0.99 2.249 0.99 2.931 0.98 50.04 0.38 98.57
Wetzikon 2 0.808 0.95 2.201 0.98 2.901 0.95 67.71 0.33 122.84
Wuelflingen 0.896 0.96 2.279 0.99 2.994 0.99 36.44 0.55 71.28
Zeiningen 0.795 0.96 2.182 0.99 2.991 0.96 30.85 0.34 101.71
Average 0.51 85.3
Standard deviation 0.15 25.3
Coefficient of variation, ''/o 15 25
786 SOIL SCI. SOC. AM. J., VOL. 63, JULY-AUGUST 1999

^3.0-
Q^
O
.2 2 - 5 " O

c
CD
£ 2.0-

1.5-

1.0 I ' I ' I '

20 40 60 80
Sand Content (%)
Probability of Fragmentation p
Fig. 3. Fragmentation fractal dimensions D and probabilities p of Q* "
^— A A
o
fragmentation. The solid line represents Eq. [8], symbols are calcu- c A ***
A A
O

lated with Eq. [8] from experimentally determined fractal di- 0 2.5-
~^ O
80) -
A
E 2.0-
b
The fragmentation of a cube has a certain probability i

p, which is assumed to be constant for all orders of

fragmentation. A cube can maximally disintegrate into
eight smaller cubes (p = 1) and minimally into one
smaller cube (p = 1/8). As shown by Turcotte (1986),
the fragmentation probability p is related to the fractal
dimension D by
D = log(8p)
[8]
log 2
where the range of possible fractal dimensions is 0 <
D < 3. Figure 3 shows a plot of Eq. [8] along with values
calculated from the analysis of our experimental data.
A scale-independent fragmentation process would have
a constant fragmentation probability. Evidently, frag-
mentation probabilities varied across almost the entire
range of 1/8 < p < 1. It is interesting that the fractal
dimensions for the three domains are typically Z)c,ay <
Am < Aand- It appears that for the 19 soils studied, the
probability of fragmentation is scale dependent, and
in particular it decreases with decreasing size of the
particles. There is experimental evidence that fragmen-
tation of soil and sediment aggregates is scale dependent
(Perfect et al., 1993; Rasiah et al., 1993). Larger aggre-
gates tend to fracture more easily than smaller aggre-
gates (Perfect, 1997). Considering soil particles as prod-
ucts of a fragmentation process, our results are in
qualitative agreement with observations from aggregate
failure studies.
Particle-size distribution measurements are strongly
influenced by the experimental methods of dispersion
of the soil particles. The dispersion itself can be regarded
as a fragmentation process. Organic matter increases
aggregate stability and hence leads to less fragmentation
(Rasiah et al., 1993). Therefore the omission of organic
matter removal in the Swiss soils probably leads to less
dispersion of smaller particles. This explains the smaller
clay fractions determined in the Swiss soils compared
with the U.S. soils (Table 1). Based on the fragmentation
model, we would also expect smaller fractal dimensions
for the clay fraction of the Swiss soils compared with
the U.S. soils; however, there is no evidence that this
is the case (Table 2).
Following Tyler and Wheatcraft (1992), Dsi,t vs. clay
and sand fraction was plotted in Fig. 4 to demonstrate
the relation between fractal dimension and soil texture.
!"•
2
o Present study
A Tyler and Wheatcraft (1 992)
"" 1.0- , I , , .
0 10 20 30 40 50 60
Clay Content (%)
Fig. 4. Fragmentation fractal dimension of the silt domain Dai vs.
clay and sand percentage. Data from Tyler and Wheatcraft (1992)
were obtained from the entire range of the particle-size distribution
used in their study.
The fractal dimensions reported by Tyler and Wheatcraft
(1992), plotted in this figure, were obtained by applying
Eq. [2] to the entire range of the PSD data, which ranged
from 1 to 50 |xm, 0.5 |jim to 5 mm, and 16 u,m to 1 mm
for different data sets. Fractal dimensions given by Tyler
and Wheatcraft are therefore not directly comparable
with our DSji,, but nevertheless, Fig. 4 shows a trend
between the D value and the clay and sand contents.
The fractal dimension increases with clay content, and
decreases with sand content. These results suggest that
the power-law relation of Eq. [2] can be used to charac-
terize PSD in soils, and may be an alternative to the
conventionally used approaches, such as the lognor-
mal distribution.
Calculation of Parameters of the Fragmentation
Model from Mass Fractions of Clay and Silt
It is evident that no single power law can characterize
the PSD of a soil across the entire scale usually measured
in a particle-size analysis. For the majority of the sam-
ples, 46 to 86% (with an average of 71%) of the total
mass is carried by particles with diameters between 0.51
and 85.3 jxm, the silt domain of the distribution. On a
log-log scale, the PSD of the silt domain is a straight
line and is therefore characterized by two parameters,
the intercept and the slope of the power-law distribu-
tion. If we know any two points on this line, we can
calculate the model parameters of the silt domain. As
Table 2 shows, the USDA boundaries between clay and
silt (2 u,m), and between silt and sand (50 |xm) are within
the silt domains for all 19 soils. Therefore we can use
these standard particle-class fractions to calculate the
two parameters of a power-law particle-size distribution
in the silt domain. As the second parameter besides
the fractal dimension D we choose the median particle
diameter dso of the PSD. The median particle diameter
 BITTELLI ET AL.: FRAGMENTATION MODEL TO CHARACTERIZE PARTICLE-SIZE DISTRIBUTION 787

120- CONCLUSIONS
RMSE =12.58
100- 1:1
There is evidence that cumulative PSDs in soils follow
I BOH
-O
a power-law distribution, consistent with a fractal frag-
"" 60H mentation model. The mass-based approach suggested
55 by Matsushita (1985) and Turcotte (1986) showed good
T3 40-

20-
agreement between the fractal model and our experi-
mental data. Three main domains—a clay, silt, and sand
^ I '
40
I '
60
I '
80 100
I
120
domain—were identified where power-law scaling was
applicable. The limits between the domains were rela-
dso (calculated) tively constant for different soil types, but do not coin-
3.0-
RMSE = 0.09 cide with the traditional boundaries between clay, silt,
2.5- 1:1 and sand. Fragmentation fractal dimensions of the three
domains increased in the order: clay < silt < sand do-
¥
-§2.0-] main. A method is imposed to estimate the parameters
'»
of the fragmentation model of the PSD in the silt domain
o i.sH from standard textural data of clay, silt, and sand
fractions.
1.0- I I I
1.5 2.0 2.5 3.0
ACKNOWLEDGMENTS
Dsiit (calculated) We thank Alan Busacca and Sandra Lilligren for assistance
Fig. 5. Experimental and calculated values of median diameter ds during the laboratory analyses. The manuscript benefitted
and of fragmentation fractal dimension of the silt domain DlU, for from fruitful discussions with Claudio O. Stockle, Sally D.
all 19 soils used in this study. Calculated values are from Eq. [9]
and [10]. RMSE is the root mean square error.
Logsdon, and Philippe Baveye.

is chosen because it is often used in empirical relations

to predict other soil properties, and as such is a useful
parameter to know. The median particle diameter dso
and the fractal dimension of the silt domain Dsi|t can be
calculated from standard textural data as follows
(2 - Am) log 2 - log(mciay)
dsn = ex [9]
3 - DM
and
n _ J
, _ log(msii, + mday) - log(mday)
^-'silt ~~ , _„ , _.
log 50 - log 2
_ log(l + msili/mciay)
= 3
[10]
log 25
where mday and msilt are the mass fractions of clay and silt,
log is the natural logarithm, and the median diameter d50
is given in micrometers. Note that the median di0 in Eq.
[9] is a non-log-transformed parameter. Mass fractions
of clay, silt, and sand are readily available for many
soils, and from these data the median diameter and
the fractal dimension of the silt domain can then be
computed with Eq. [9] and [10]. Note that in the deriva-
tion of Eq. [9] and [10] we assumed the USDA textural
definition; for other classification schemes, the equa-
tions have to be adapted accordingly.
We tested the validity of the two equations with our
19-soil data set. In Fig. 5, experimental dx and jDsill values
(Table 2) are plotted vs. values calculated based on
standard particle-size fractions using Eq. [9] and [10].
There is a good agreement between experimental and
calculated values, particularly for £>silt. The RMSE in
Fig. 5 shows that the dso is not estimated as well as the
Am, and generally the model tends to underestimate
the experimental value. The estimation of the fractal
dimension is well performed by Eq. [10].
788 SOIL SCI. SOC. AM. J., VOL. 63, JULY-AUGUST 1999