Professional Documents
Culture Documents
net/publication/232149999
CITATIONS READS
3 63
4 authors:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Silvia Stanchi on 19 October 2015.
283
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
284 STANCHI ET AL. SOIL SCIENCE
(e.g., Bittelli et al., 1999; who proposed a Should such dependencies exist, opportunities
multidomain scaling) or to other model assump- to predict fractal scaling would appear.
tions (e.g., constant density and failure proba- Because the list of properties affecting the
bility, as suggested by Perfect, 1997; Anderson presence of scaling can be large, and it is not a
et al., 1998). priori clear which of them are the most efficient
Both hypotheses suggest an effect of soil predictors of the fractal behavior, statistical meth-
chemical and physical characteristics affecting ods have to be used to select the most influential
aggregation on the scaling properties in this size predictors. Two such methods have been exten-
range. Moreover, the first hypothesis might be sively applied in soil studies. Discriminant analysis
related to the PSD determination method, which was used first by Webster and Burrough (1974) to
is not unique (Gee and Bauder, 1986) because it describe soil classes and then applied in the study of
depends on the aim of the analysis (Richie, 1963). soil genesis (e.g., Berg, 1980), in the evaluation of
In fact, without the complete removal of organic the fertility characteristics (e.g., Frevert, 1986;
and inorganic cementing agents, sand-sized, silt- Bonifacio et al., 2000) and in the assessment of
sized, and clay-sized aggregates are actually the effect of agricultural practices on soil physical
determined, instead of primary particles. The properties (e.g., Quiroga et al., 1998). Classifica-
standard United States Department of Agricul- tion and regression trees (Breiman et al., 1993) are
ture procedure for PSD determination involves another technique that has been used to discover
the removal of soluble salts and organic matter structure in data, mainly when there are many
(Burt, 2004), but the choice of whether to per- variables that can mask the effect of each other
form additional pretreatments to remove inorganic (Rawls and Pachepsky, 2002; Scull et al., 2005).
cementing agents, such as Fe oxides, carbonates, or The objective of this work was twofold: (i) to
silica, is left to the operator. Depending on this investigate the fractal behavior of soil as inf lu-
choice, different size distributions will be obtained. enced by the presence of the fragmentation-
In contrast, water-dispersible PSD may also be enhancing treatment before PSD determination
determined for special purposes, such as the assess- and (ii) to predict the presence/absence of frac-
ment of soil erodibility (Burt, 2004). As a tal behavior from soil chemical and physical
consequence, the presence or the absence of the properties by applying discriminant analysis and
fractal scaling, too, might be related to soil classification trees and to compare the efficiency
chemical and physical properties, because aggre- of the two methods.
gating agents may be unevenly distributed in the
size classes (Tisdall and Oades, 1982). This may MATERIALS AND METHODS
affect the susceptibility of soil particles to fragmen-
tation and dispersion during sample preparation; Study Area and Soils
in fact, the dispersion procedures used before The study area is located in Southern Italy
textural analysis are actually fragmentation pro- (40-22–40-35 N; 15-42–16-01 E). It covers the
cesses (Bittelli et al., 1999). As a consequence of a Camastra river basin, with a total surface of 358
weak fragmentation process, after water or Na- km2 and a mean altitude of 950 m above sea
hexametaphosphate dispersion, but without the level. Vegetation is represented by the beech
removal of organic matter and Fe oxides, particles cover in the upper part of the slopes, grasslands,
are still aggregates, and therefore, they can under- fallow, or oak woods in the lower section. Bare
go further fragmentation when pretreated with soil and pioneer plant communities dominate
H2O2 and Na-dithionite-citrate-bicarbonate. the f lat alluvial areas. Geology is highly variable.
The presence of the fractal scaling in PSD Carbonate rocks, typical of mountain massifs,
can be therefore evaluated for different frag- are associated to marls on ancient river terraces
mentation pathways defined by the treatment and lower positions in the relief, whereas clays
before dispersion. In the case of aggregate-size are often present on highlands, covered some-
distributions, it has been observed that fractal times by more recent sands and silt deposits and
dimensions decrease with the increase in the conglomerate formations. Depending on site
organic matter content and increase with the conditions, Entisols and Inceptisols prevail, but
increase in clay (Perfect and Kay, 1991; Rasiah Alfisols and Mollisols are also represented.
et al., 1993). However, no attempt has been
made to find relationships between the presence Soil Analysis and Treatments
of the fractal behavior and soil chemical proper- Soil samples were collected from 33 profiles
ties inf luencing soil aggregation-fragmentation. (85 horizons), then air-dried and sieved with
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
VOL. 171 ~ NO. 4 FRACTAL BEHAVIOR IN PARTICLE-SIZE DISTRIBUTIONS 285
2-mm sieves. Zanini et al. (1995) reported pH, the classification using the misclassification rates.
organic C content, CaCO3 content, cation This value was selected after several trial runs of
exchange capacity (CEC), base saturation (BS), the model in which we varied the minimum
exchangeable cations, Na-dithionite-citrate- number of observations from 10 to 3 and with the
bicarbonate–extractable Fe (FeDCB), and NH4- goal of achieving a good detail in the classification
oxalate–extractable Fe (FeO) in those samples. while avoiding overbranching (Breiman et al.,
The soil texture was determined by the pipet 1993). By changing the limit value from 4 to 3, we
method (Gee and Bauder, 1986) following two did not achieve any further improvement in detail.
different dispersion methods. We prepared two
replicates for each determination method. RESULTS
Method 1 included the pretreatment of the 2-
mm fraction of the samples with hydrogen Differences Between Soil Properties in Fractal
peroxide and Na-dithionite-citrate-bicarbonate and Nonfractal Samples
(Mehra and Jackson, 1960) and then dispersion Values of the fragmentation fractal dimension
with Na-hexametaphosphate. Method 2 con- DfP (Table 1) on pretreated samples ranged from
sisted of dispersion with Na-hexametaphosphate 2.627 to 2.938, with an average value of 2.837,
without any pretreatment. Symbols ‘‘P’’ and whereas on non-pretreated samples, they ranged
‘‘NP’’ are used for pretreated and non-pretreated from 2.334 to 2.896, with an average of 2.727.
samples, respectively. Values of DfP were significantly higher than
The fragmentation fractal dimension Df was DfNP (t test, P G 0.01), indicating a greater
estimated from PSDs according to Tyler and degree of fragmentation of soil after the destruc-
Wheatcraft (1992), using five size classes: coarse tion of organic and inorganic binding agents. A
sand (0.2–2 mm), fine sand (0.05–0.2 mm), poor, although significant, correlation (r = 0.325,
coarse silt (0.02–0.05 mm), fine silt (0.002–0.02 P G 0.01) was observed between DfNP and DfP.
mm), and clay (G0.002 mm). Values of Df were The determination coefficient varied from
calculated as slopes of the log-log plot of 0.80 and 0.99 for NP samples, and the 0.95
cumulative mass versus upper sieve radius. To threshold allowed us to identify 44 samples
evaluate the goodness of fit, two criteria were displaying strict fractal behavior (F), with an
used: the determination coefficient of the regres- average value of 0.97. The 41 NF samples had
sion line, following the example of Rasiah et al., an average r2 of 0.90 (Table 1).
(1995) and of Millan et al., (2003), and the eval- The Kolmogorov-Smirnov statistic was non-
uation of the residuals, to investigate the pres- significant (always P 9 0.30), indicating the
ence of systematic deviations from the model. A normality of distributions of residuals. No indi-
threshold value of 0.95 for the r2 was used to cations of bias in the model were found from the
discriminate between fractal (F) and nonfractal analysis of the probability plots. However, the
(NF) subsets. The Kolmogorov-Smirnov test presence of a trend was observed in the scatter-
(Sokal and Rohlf, 1969) was used to assess the plots of standardized residuals versus predicted
statistical significance of the deviation from values for 34 samples, all in the NF set.
normal distribution. Values of DfP and DfNP The P samples showed determination coef-
were compared using a paired-samples t test ficients varying from 0.79 to 0.99, but only 31
(Sokal and Rohlf, 1969). cases had values more than 0.95 and were thus
Discriminant analysis was performed on NP considered to belong to the F subset. The NF
and P data sets using the SPSS software (SPSS Inc., subset showed average r2 values of 0.88, whereas
Chicago, Ill). Presence of fractal behavior was used r2 in the F group was 0.97, on the average
as the categorical grouping variable. Initially, all (Table 1). The residual analysis showed results
the variables were included in the analysis, then similar to those obtained with NP samples.
some were excluded by using the stepwise method The 31 soils showing fractal behavior after
and determining the Wilks lambda index, the pretreatment mainly belonged to the sandy clay
eigenvalue, and the misclassification rate to assess loam, sandy loam, and clay texture classes. Clays
the goodness of the model (Norusis, 1985). We and clay loams dominated instead in the P non-
also carried out classification tree analysis (Breiman fractal subset. Most of the non-pretreated fractal
et al., 1993) using the Splus software (MathS.ft samples were sandy loam, sandy clay loam, and
Inc. Cambridge, MA). We set 4 as the minimum clay, whereas the majority of the NP nonfractal
number of observations in a branch that can be were loam, sandy loam, and clay loam (Fig. 1).
further split. We then evaluated the goodness of In addition, soil physical and chemical properties
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
286 STANCHI ET AL. SOIL SCIENCE
TABLE 1
Average values of significant chemical and physical properties for different
PSD determination methods, considering F and NF subsets
P (n = 85) NP (n = 85)
Variable
F. (n = 31) NF. (n = 54) P F (n = 44) NF (n = 41) P
DfP 2.800 2.859 G0.001 2.718 2.737 0.001
DfNP 2.724 2.730 0.843 2.817 2.859 0.502
r2 0.97 0.88 G0.01 0.97 0.90 G0.01
Coarse sand P (%) 26.3 13.3 G0.001 23.3 12.4 G0.001
Coarse sand NP (%) 34.6 28.4 0.055 36.3 24.6 G0.001
Fine sand P (%) 24.2 17.1 G0.001 22.3 16.9 0.007
Fine sand NP (%) 24.3 19.6 0.027 23.7 18.7 0.015
Coarse silt P (%) 5.8 5.3 0.588 5.1 5.9 0.357
Coarse silt NP (%) 7.0 7.4 0.707 6.0 8.6 0.002
Fine silt P (%) 15.4 25.4 G0.001 17.8 26.0 G0.001
Fine silt NP (%) 16.4 23.5 G0.001 16.1 26.1 G0.001
Clay P (%) 28.4 38.9 G0.001 31.6 38.8 0.005
Clay NP (%) 17.6 21.2 0.219 17.8 22.0 0.126
Coarse sand P/coarse 0.77 0.49 G0.001 0.67 0.51 0.006
sand NP (%)
pH 7.8 7.2 0.002 7.5 7.4 0.681
CaCO3 (g/kg) 85.7 62.1 0.323 67.6 73.9 0.785
Corg (g/kg) 11.2 21.0 0.020 14.6 20.6 0.144
N (g/kg) 1.25 2.14 0.027 1.51 2.14 0.102
CEC (cmolc/kg) 19.1 27.8 0.001 21.3 28.2 0.005
Caexch (cmolc/kg) 16.29 20.46 0.042 15.80 22.31 0.001
Mgexch (cmolc/kg) 1.24 1.77 0.023 1.39 1.78 0.088
Kexch (cmolc/kg) 0.36 0.46 0.207 0.36 0.49 0.084
BS (%) 95.1 83.7 0.004 86.9 88.9 0.608
FeDCB (g/kg)- 14.8 19.1 0.017 17.2 17.9 0.692
FeO (g/kg)‘ 1.9 3.5 0.012 2.3 3.6 0.039
P indicates texture after pretreatment with H O and Na-dithionite-citrate-bicarbonate; NP, no pretreatment.
2 2
.
F indicates strict fractal behavior; NF, nonstrictly fractal behavior.
showed significant differences between methods, the sand content and the NP coarse silt, the fine
as shown in Table 1. The F and NF groups had silt, and the P clay contents showed significant
different coarse sand (P) and fine silt contents, CEC, differences among the four groups described in
and exchangeable Ca content and FeO content. Table 2. The soils with ‘‘always fractal’’ PSD had
Only 26 samples showed fractal behavior in lower DfP, a coarse sand ratio (coarse sand
both cases, and 36 in none. Five samples were contents with pretreatment/coarse sand contents
fractal only when pretreated. Eighteen samples without pretreatment) closer to 1, relatively low
showed the opposite effect; that is, fractal behavior C and Fe contents, and lower exchangeable Ca
without pretreatment changed to the nonfractal content together with high BS. The ‘‘always
one with pretreatment. According to the changes nonfractal’’ soils had higher DfP, a lower coarse
in fractal behavior, four groups were identified sand ratio, and higher C and Fe contents
(Table 2). Analysis of variance showed that some compared with the ‘‘always fractal’’ ones.
chemical and physical properties were significantly
different among these four groups (Table 2). Discriminant Analysis and Classification Trees
Among those properties were DfP (P G 0.001), The discriminant function for the P set was:
the ratio of coarse sand contents with pretreat- 0:676 coarse sand þ 0:490 base saturation
ment and without pretreatment (P G 0.001), the j0:412 f ine silt ð1Þ
organic C (P = 0.035) content, pH (P G 0.001),
the CaCO3 content (P = 0.048), the CEC (P = for P G 0.001, with a misclassification rate of 15
0.002) and the exchangeable Ca content (P = of 85 (Table 3), whereas the one describing the
0.007), the BS (P G 0.001), FeDCB (0.049), and NP set was:
FeO (P = 0.019). Considering textural variables, 0:883 f ine silt þ 0:503 Ca þ 0:482 coarse silt ð2Þ
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
VOL. 171 ~ NO. 4 FRACTAL BEHAVIOR IN PARTICLE-SIZE DISTRIBUTIONS 287
for P G 0.001, with misclassification rate of 9 the misclassification rate was lower in the NP
of 85. data set.
Eigenvalues, indicating the ratio of the sum Classification trees were developed separately
of squares between groups to that within groups for P and NP samples (Fig. 3). The presence/
and therefore commonly used to evaluate the absence fractal scaling was the categorical variable
goodness of the model (Norusis, 1985) ranged to classify.
from 0.71 (P set) to 0.93 (NP set), which can still In the P tree, the fine silt content higher than
be considered acceptable. The Wilks lambda 26.3% determined the first partition, leading to
statistic, another index expressing the goodness the NF subset. The second partition was deter-
of the classification, was 0.59 for P and 0.52 for mined by the coarse sand content: if the coarse
NP, which is satisfactory according to Webster sand percentage was less than 9.0%, then the soil
(1971). The P data set discriminant function was was classified as NF. If not, the fine silt content
dominated by a textural variable (coarse sand), entered again, leading to the NF subset when
followed by BS and fine silt. In the NP data set, lower than 8.7%. The misclassification rate was
another textural variable, that is, fine silt, showed 3 of 85 (Table 3).
the highest coefficient, followed by exchangeable In the NP tree, the fine silt content also
Ca content and coarse silt. The subset centroids determined the first partition. If the fine silt
in NP (j0.98 for NF and 0.92 for F) were better content was more than 25.0%, the soil was
separated than the centroids in P (j0.63 for NF immediately classified as NF with no misclassifi-
and 1.10 for F). This suggested a larger overlap cation. The soil was also classified as NF with fine
between subsets in P. The absolute frequencies silt less than 25.0% and coarse silt more than
of discriminant scores clearly showed an over- 10.3%. If, instead, coarse silt was less than 10.3%
lapping of the F and NF samples, when the and the Ca content was less than 21.8 cmolc/kg,
pretreatment is used (Fig. 2). As a consequence, the subset was F, as well as if the Ca content
Fig. 1. Texture of pretreated and non pretreated samples. The - line separates regions of fractal and nonfractal
behavior as suggested by Tyler and Whetcraft (1992).
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
288 STANCHI ET AL. SOIL SCIENCE
TABLE 2
Selected chemical and physical properties of soil samples for different determination methods (average values)
Always fractal Never fractal Only P fractal Only NP fractal
P
(n = 26) (n = 36) (n = 5 ) (n = 18)
DfP G0.001 2.794b 2.863a 2.831ab 2.850a
DfNP 0.386 2.734 2.746 2.670 2.696
Coarse sand P (%) G0.001 26.6a 10.7b 24.4a 18.4b
Coarse sand NP (%) G0.001 34.2a 22.9b 36.9ab 39.5a
Fine sand P (%) 0.004 24.9a 16.4b 20.8ab 18.6ab
Fine sand NP (%) 0.033 25.7 18.9 17.3 20.9
Coarse silt P (%) 0.054 6.1 6.1 3.9 3.7
Coarse silt NP (%) 0.010 6.4 8.3 10.5 5.5
Fine silt P (%) G0.001 14.8b 27.1a 18.1ab 22.0a
Fine silt NP (%) G0.001 15.3b 26.6a 22.4ab 17.2b
Clay P (%) 0.001 27.6b 39.7a 32.7ab 37.3a
Clay NP (%) 0.150 18.5 23.3 13.2 16.9
Coarse sand P/coarse G0.001 0.79a 0.49b 0.65ab 0.48b
sand NP (%)
pH G0.001 7.9a 7.4a 7.1ab 6.8b
CaCO3 (g/kg) 0.048 99.5 82.3 13.6 21.6
Corg (g/kg) 0.035 8.7b 20.0a 24.3a 23.0a
CEC (cmolc/kg) 0.002 17.6b 28.5a 26.6ab 26.5ab
Caexch (cmolc/kg) 0.007 15.77b 22.77a 18.96ab 15.83ab
Mgexch (cmolc/kg) 0.133 1.21 1.84 1.39 1.65
Kexch (cmolc/kg) 0.275 0.36 0.51 0.35 0.36
BS (%) G0.001 98.04a 90.14a 80.05ab 70.88b
FeDCB (g/kg) 0.049 14.5 18.1 16.8 21.2
FeO (g/kg) 0.019 1.4b 3.5a 4.1ab 3.5ab
P indicates texture after pretreatment with H2O2 and Na-dithionite-citrate-bicarbonate; NP, no pretreatment.
exceeded this threshold value with DfNP less results, both in terms of explained variance and
than 2.816. In this case, the misclassification rate goodness of classification in this case. The
was 4 of 85 (Table 3). classification tree is reported in Fig. 4. The fine
silt content again entered as the first retained
Transitions Between F and NF Groups variable, with a threshold value of 25%, whereas
The presence/absence of conservation of the second branching was determined by fine silt
fractal behavior was predicted using as target and CEC. High fine silt contents immediately
variables the four groups described in Table 2. led to always NF behavior. Coarse and fine silt
The discriminant analysis led to unsatisfactory contents appeared several times as intermediate
split variables. The exchangeable Ca and Mg
TABLE 3 contents appeared after, leading to the terminal
Number of misclassified cases in discriminant analysis and branches. Lower silt contents, associated with
classification trees low CEC and low Mg contents, always led to
Data- F (observed) but NF (observed) the fractal group. A final branching was deter-
Method
set classified as NF but classified as F mined by pH, too, where highest values were
Discriminant P. 8 7 associated with always fractal samples. In this
analysis case, the misclassification error was 9 of 85.
Discriminant NP. 3 6
analysis
Classification P. 0 3 DISCUSSION
tree
Classification NP. 1 3 The P data set was obtained after destruction
tree of the organic and inorganic binding agents that
F indicates strict fractal behavior; NF, nonstrictly fractal affect the formation of aggregates from primary
behavior. particles (Tisdall and Oades, 1982), the NP
.
P indicates texture after pretreatment with H2O2 and ref lects a lower level of fragmentation, and ag-
Na-dithionite-citrate-bicarbonate; NP, no pretreatment. gregates are still present.
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
VOL. 171 ~ NO. 4 FRACTAL BEHAVIOR IN PARTICLE-SIZE DISTRIBUTIONS 289
Fig. 2. Absolute frequencies and discriminant scores for the F and NF subsets, for pretreated and non pretreated
samples.
In both cases, we detected deviations from and in several chemical properties (Table 1).
strict self-similarity for several samples. We When the samples are pretreated, so that the
realize that defining fractal behavior in terms of action of aggregating agents is not taken into
the threshold value of the determination coef- account, Df is expected to depend mainly on the
ficient introduces some arbitrariness in the type of parent material and therefore on the
separation of soil samples in this study. How- specific physical and fragmentation properties of
ever, the qualitative results on the role of basic the material (Turcotte, 1986, 1989). Soil devel-
soil properties in the presence of fractal scaling opment is expected to smooth this effect, but
would hold when we varied the threshold value textural differences related to parent material are
from 0.92 to 0.97 (data not shown). Moreover, still visible in Alfisols (Macias and Chesworth,
the residual analysis indicated that no systematic 1992). The differences found in PSDs of pre-
deviations from the model are present; there- treated soils originated from different geologic
fore, the determination coefficient can be used substrates in the study area (P G 0.01, data not
as an objective index to split the two groups of shown) supported these hypotheses. For exam-
behavior. Even if some authors (Bittelli et al., ple, all the soils developing on fine-textured
1999) related the lack of strict self-similarity to parent materials, such as clays and marls, had the
the presence of different fractal domains (sand, highest DfP, with an average of 2.862, whereas
silt, and clay) in the particle range, we could not those on coarse-textured sediments, such as
adopt this approach because of the limited sands, typically showed the lowest DfP, with an
number of classes. average value of 2.781, indicating a PSD
The F and NF soils after the P treatment dominated by coarser particles. In the study
showed significant differences both in the DfP area, therefore, NF soils are generally developed
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
290 STANCHI ET AL. SOIL SCIENCE
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
VOL. 171 ~ NO. 4 FRACTAL BEHAVIOR IN PARTICLE-SIZE DISTRIBUTIONS 291
Fig. 4. Prediction of variations in fractal behavior for different determination methods. In brackets, the number of
samples for each final branch. FF indicates always fractal; NFNF, always nonfractal; PYF, NPYNF, fractal when
pretreated but nonfractal when not pretreated; PYNF, NPYF, nonfractal when pretreated but fractal when
pretreated.
between the coarse sand contents determined successfully predicted using classification trees
after and before pretreatment (Table 2), indicat- and discriminant analysis. The textural variables,
ing scarce aggregation in the coarse sand class, primarily silt content, had the major discrimi-
which resulted in a minor disgregation due to nant role. Properties related to the cation
pretreatment. These samples mostly came from exchange complex also had some inf luence on
profiles evolving on sands. Aggregation was the fractal behavior. Better results were obtained
more pronounced in soils always displaying with classification trees, which appeared to be a
nonfractal behavior, as indicated by low coarse more versatile tool for prediction.
sand index. These soils were typically developed
on clays or other fine-textured parent materials
that can undergo greater fragmentation when
ACKNOWLEDGMENTS
the pretreatment is adopted. These relations
between fractal behavior and aggregation were This work was supported by SITAFspa
further confirmed by the higher values of funding.
organic C and iron oxides contents in the
always NF soils than in the always F (Table 2). REFERENCES
Ahmed, M., and J. M. Oades. 1984. Distribution of
CONCLUSIONS organic matter and adenosine triphosphate after
fractionation of soils by physical methods. Soil
The presence of fractal scaling in soil PSDs Biol. Biogeochem. 16:465–470.
for two different texture determination methods Anderson, A. N., A. B. McBratney, and J. W.
(i.e., with or without pretreatment for organic Crawford. 1998. Applications of Fractals to Soil
matter and iron oxides removal) could be Studies. Adv. Agron. 63:1–76.
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
292 STANCHI ET AL. SOIL SCIENCE
Berg, R. C. 1980. Use of Stepwise Discriminant plied to soil aggregation. Soil Sci. Soc. Am. J. 55:
analysis to assess soil genesis in a youthful sandy 1552–1558.
environment. Soi Sci. 129:353–365. Perfect, E. 1997. Fractal models for the fragmentation
Bittelli, M., G. S. Campbell, and M. Flury. 1999. of rocks and soils: A review. Eng. Geol. 48:185–198.
Characterization of particle-size distribution in Pulleman M. M., and Marinissen J. C. Y. 2004.
soils with a fragmentation model. Soil Sci. Soc. Physical protection of mineralizable C in aggre-
Am. J. 63:782–788. gates from long-term pasture and arable soil.
Bonifacio, E., S. Santoni, and E. Zanini. 2000. Soil Geoderma. 120:273–282.
properties required by some southern Africa fruit Quiroga, A. R., D. E. Buschiazzo, and N. Peinemann.
trees as assessed by discriminant analysis. Arid Soil 1998. Management discriminant properties in
Res. Rehab. 14:253–263. semiarid soils. Soil Sci. 163:591–597.
Boix-Fayos, C., A. Calvo-Cases, A. C. Imeson, and Rasiah, V., B. D. Kay, and E. Perfect. 1993. New
M. D. Soriano-Soto. 2001. Inf luence of soil mass-based model for estimating the fractal dimen-
properties on the aggregation of some Mediterra- sion of soil aggregates. Soil Sci. Soc. Am. J. 57:
nean soils and the use of aggregate size and stability 891–895.
as land degradation factors. Catena. 44:46–67. Rasiah, V., E. Perfect, and B. D. Kay. 1995. Linear and
Breiman, L., J. H. Friedman, R. A. Olshen, and C. J. nonlinear estimates of fractal dimension for soil
Stone. 1993. Regression trees. Chapman and Hall aggregate fragmentation. Soil Sci. Soc. Am. J. 59:
and CRC Press, Boca Raton, FL. 83–87.
Burt R. 2004. Soil survey laboratory methods Rawls, W. J., and Y. A. Pachepsky. 2002. Soil
manual. USDA-NRCS, pp. 34–65. consistence and structure as predictors of water
Curtin, D. 2002. Possible role of aluminium in retention. Soil Sci. Soc. Am. J. 66:1115–1126.
stabilizing organic matter in particle size fractions Richie, J. A. 1963. Earthwork tunnelling and the
of Chernozemic and Solonetzic soils. Can. J. application of soil testing procedures. J. Soil
Soil Sci. 82:265–268. Conserv. NSW. 19:111–129.
Frevert, T. 1986. A statistical model on the fertility of Scull, P., J. Franklin, and O. A. Chadwick. 2005. The
the giant Taro cultivation pits at Tarawa Atoll application of classification tree analysis to soil type
(Central Pacific). Catena. 13:321–331. prediction in a desert landscape. Ecol. Model. 18:
Gee, G. W., and J. W. Bauder. 1986. Particle-size 1–15.
analysis. In: Methods of Soil Analysis. Klute A (ed.). Sokal R., and J. Rohlf. 1969. Biometry. W.H. Freeman
Part I—Physical and Mineralogical Methods, 2nd and Co, San Francisco, CA, pp. 166–171.
ed. ASA, SSA, Madison, WI, pp. 383–409. Tisdall, J. M., and J. M. Oades. 1982. Organic matter
Macias, F., and W. Chesworth. 1992. Weathering in and water-stable aggregates in soils. J. Soil Sci.
humid region, with emphasis on indigenous rocks 33:141–163.
and their metamorphic equivalent. In: Weathering Turcotte, D. L. 1986. Fractals and fragmentation.
soils and Paleosoils. Developments in Earth Sur- J. Geophys. Res. 91:1921–1926.
face Processes. n. 2. Martini I. P. and W. Chesworth Turcotte, D. L. 1989. Fractals in geology and
(eds.). Elsevier, Amsterdam, NL, pp. 283–306. geophysics. Pure Appl. Geophys. 131:171–196.
Mehra, O. P., and M. L. Jackson. 1960. Iron oxide Tyler, S. W., and S. W. Wheatcraft. 1992. Fractal
removal from soils and clays by a dithionite-citrate scaling of soil particle-size distributions: Analysis
system buffered with sodium bicarbonate. Clay. and limitations. Soil Sci. Soc. Am. J. 56:362–369.
Clay Miner. Proc. 7th National Conference on Webster, R. 1971. Wilks criterion: A measure for com-
Clays and Clay Minerals, Washington, DC, 1958, paring the value of general purpose soil classifica-
pp. 317–327. tions. J. Soil Sci. 22:254–260.
Millan, H., M. Gonzalez-Posada, M. Aguilar, J. Webster, R., and P. A. Burrough. 1974. Multiple
Dominguez, and L. Céspedes. 2003. On the discriminant analysis in soil survey. J. Soil Sci. 25:
fractal scaling of soil data. Particle-size distribu- 120–134.
tions. Geoderma. 117:117–128. Zanini, E., E. Bonifacio, N. Alliani, and V. Boero.
Norusis, M. J. 1985. SPSSx Advances Statistics Guide. Genesi e caratteri chimici, mineralogici e struttur-
Chicago, IL, SPSS Inc., pp. 213–217. ali dei suoli marginali dell’Appennino Lucano.
Perfect, E., and B. D. Kay. 1991. Fractal theory ap- Miner. Petrogr. Acta. 38:177–188.
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
View publication stats