You are on page 1of 9

JURNAL INTERNASIONAL

ILMU UKUR TANAH

Disusun Oleh:

Renold Yanspri Pratama


21410013

DOSEN : DR. Ligal Sebastian,S.T.,M.SI / KH. Rizal, S.T

JURUSAN TEKNIK SIPIL


UNIVERSITAS PALEMBANG
TECHNICAL ARTICLE

Soil Survey: Pedotransfer Function of Linear Extensibility


Percent for Soils of the United States
Cathy A. Seybold and Zamir Libohova

ABSTRACT:
Soil survey is an ongoing process from initial soil mapping through soil survey updates. A national model of linear extensibility percent (LEP) is needed
to improve this process. The objective of this work was to develop and validate models for estimating LEP using general linear models and readily
available soil survey properties. Measured data from the Kellogg National Soil Survey Laboratory database (Lincoln, Nebraska) were used to develop
the prediction models. Twenty LEP prediction equations were developed based on pH and major mineralogy classes. Noncarbonate clay and, de-
pending on the soil pH, either cation-exchange capacity or effective cation-exchange capacity explained between 42% and 86% of the total variation
in LEP. Model equations using cation-exchange capacity as a predictive variable collectively produced a prediction root mean square error (RMSEP)
of 1.44% and mean error (ME) of −0.16%. For low pH soils, the model equations using effective cation-exchange capacity as a predictive variable
collectively produced an RMSEP of 1.29% and ME of −0.034%. The small negative MEs indicate an overall underestimation of LEP. Breaking down
the validation results further among the different mineralogy groups produced a range of RMSEp from 0.42% to 1.80%. The smectitic group had the
largest and the siliceous group had the lowest RMSEp. The prediction accuracy is considered adequate for soil survey purposes, and it is expected
that LEP estimates will ultimately enhance soil survey interpretations. The models will be added to the soil survey database for soil scientists to use
when measured data are not available.
Key Words: COLE, linear extensibility percent, prediction, shrink-swell, soil survey
(Soil Sci 2017;182: 1–8)

L inear extensibility percent (LEP) is a measure of the linear vol-


ume change in soils as a result of a change in water content
(Grossman et al., 1968; Burt, 2011). It is expressed on a whole-soil
Soil survey interpretations use LEP to assess potential shrinkage
and swelling impacts in a wide range of rural and urban land uses
and also for engineering and geotechnical purposes. Cracks created
basis with units in percentages. Linear extensibility percent charac- when soils shrink can affect soil infiltration rate and cause rapid
terizes the shrink-swell capacity of soils and was designed for use movement of water downward into the soil, which can carry
in soil genesis and characterization studies (Hallberg, 1977). chemicals to the groundwater (de Jong et al., 1992). Soil swelling,
Shrink-swell arises from the movement of water into and out of on the other hand, can close cracks and reduce infiltration rates, pos-
interlayer spaces of primarily 2:1 phyllosilicate clay minerals that sibly causing excessive soil surface runoff and erosion (Favre et al.,
causes the mineral to expand and contract (Bohn et al., 1979; 1997). Soils with high shrink-swell potentials can cause damage
Vaught et al., 2006). Linear extensibility percent is calculated to plant roots, cracking or buckling of pavements, cracking of inte-
from the coefficient of linear extensibility (COLE) as LEP = 100  rior walls in buildings, foundation damages, slope failures, break-
COLE. The COLE is measured in the laboratory as follows (Grossman age of underground pipes, and complete structural collapse (Mbagwu,
et al., 1968; Soil Survey Staff, 2014): 1992; Mandal et al., 2005; US Department of Agriculture Natural
Resources Conservation Service [USDA-NRCS], 2016). In engi-
  neering, LEP can be used for quantifying shrinkage factors such as
Dbd 1=3
COLE ¼ −1 the shrinkage limit, shrinkage ratio, and percent volume change
Dbm
(Hallberg, 1977). Shrink-swell is also important in the genesis of
soil structure (Bronswijk, 1989), and linear extensibility (COLE
Where Dbd is bulk density at oven dryness (g cm−3), and Dbm is multiplied by the layer thickness) is used as a criterion in soil taxon-
bulk density at -33 kPa (g cm−3) on natural fabric soil samples omy (Soil Survey Staff, 1999).
(clods). Bulk densities are corrected for any rock fragments to esti- Linear extensibility percent is a basic soil property that is captured
mate COLE on a whole-soil basis. in the National Soil Information System (NASIS) database of the
USDA-NRCS for soil map unit components (i.e., soils/series that
National Soil Survey Center, US Department of Agriculture Natural Resources make up the map unit). Field estimates of shrink-swell potential in-
Conservation Service, Lincoln, NE. clude observation of desiccation cracks, slickensides, gilgai, soil
Address for correspondence: Cathy A. Seybold, PhD, National Soil Survey Center,
creep, and leaning utility poles and are placed in one of three classes
USDA-NRCS, 100 Centennial Mall North, Rm 152, Lincoln, NE 68508. of shrink-swell potential (USDA-NRCS, 2016). In soil survey, clas-
E-mail: cathy.seybold@lin.usda.gov ses of shrink-swell are defined based on the LEP of the soil (USDA-
Financial Disclosures/Conflicts of Interest: None reported.
NRCS, 2016). However, more accurate estimates require measure-
ments, which are very time consuming and costly for an entire soil
Received September 7, 2016.
survey area. Therefore, estimation procedures or models are often
Accepted for publication December 20, 2016. used in soil survey, but there is no national model for estimating
Copyright © 2017 The Author(s). Published by Wolters Kluwer Health, Inc. This is an LEP. There are still areas where initial soil mapping is occurring,
open-access article distributed under the terms of the Creative Commons Attribution- and there are many areas going through updates, especially on a ma-
Non Commercial-No Derivatives License 4.0 (CCBY-NC-ND), where it is permissible to jor land resource area basis. To improve estimates of LEP, a model
download and share the work provided it is properly cited. The work cannot be changed in that is national in scope and uses available properties within the soil
any way or used commercially without permission from the journal.
survey database is needed.
ISSN: 0038-075X Several soil properties have already been related to the shrink-
DOI: 10.1097/SS.0000000000000191 swell capacity of soils. The type and amounts of clay in combination

Soil Science • January 2017 • Volume 182 • Number 1 www.soilsci.com 1


Seybold and Libohova Soil Science

are known to affect the shrink-swell capacity of soils, with clay min- MATERIALS AND METHODS
eralogy being the dominant factor (Seed et al., 1962; Franzmeier and
Ross, 1968; McCormack and Wilding, 1975; Kariuki and van der Data Selection and Analysis
Meer, 2004). Smectite clay minerals have a high capacity to shrink Soil horizons were selected from the KSSL characterization data-
and swell, because of their low layer charge (weak bonds) that al- base (Lincoln, NE), which contains more than 37,000 pedons with
lows water (a polar molecule) to enter between the mineral basal measured chemical and physical properties representing geographi-
planes, causing an expansion or swelling (Bohn et al., 1979). cally diverse soils from across the conterminous United States,
Schafer and Singer (1976) found COLE to be highly correlated to Hawaii, and Alaska. The measured data prior to year 2000 were used
the percent expandable clay, with a lower correlation to the ratio of for model development, whereas data from 2001 to 2010 were used
expandable clay to total clay or total clay content. Franzmeier and for model validation. The data set starts around 1959 when the saran-
Ross (1968) indicated that if clay is expressed on a volume basis coated clod method for determining bulk density was first used in
(an effect of the arrangement of soil particles in space) the rela- soil survey (Brasher et al., 1966), and methodology has not changed.
tionship between COLE and clay content could be improved. In The following measured data were used: total sand (2–0.05 mm),
addition to clay type, cations associated with the clay minerals total silt (0.05–0.002 mm), and total clay (<0.002 mm) (pipette
have also been shown to have an effect on the swelling potential method); carbonate clay (<0.002 mm); organic carbon content
of soils, such as Na-montmorillonite (Al-Rawas, 1999). Zhang (acid-dichromate digestion method, discontinued in 2000); total C
et al. (2016) showed Ca-montmorillonite to have greater swelling (dry combustion); CEC (1.0 N NH4OAc at pH 7); effective CEC
compared with Na-montmorillonite, which was attributed to a (ECEC); bulk density at −33 kPa matric potential (Db,33, saran-
stronger hydration of the Ca-montmorillonite. The plastic index coated clods); oven-dry bulk density (Db,o, saran-coated clods); wa-
(PI) of soils is known to be highly correlated to the expansive ter retention at −1,500 kPa (<2 mm sieved); gypsum content
property of soils (Seed et al., 1962; de Jong et al., 1992; (<2 mm); CaCO3 equivalent (electronic manometer method); pH
Puppala et al., 2013). Other properties have been related to the (in water and 0.01 M CaCl2 in 1:2 soil/solution suspensions); and
shrink-swell potential such as specific surface area (Ross, 1978; LEP (between oven dryness and −33 kPa matric potential measured
de Jong et al., 1992; Grey and Allbrook, 2002) and cation- on saran-coated clods, calculated from COLE). All determinations,
exchange capacity (CEC) (Gill and Reaves, 1957; Grey and except those for bulk density and LEP, are made on air-dried (30°
Allbrook, 2002; Kariuki and van der Meer, 2004). The latter C–35°C), crushed, and sieved (<2 mm) soil samples. All of the pre-
property integrates the total amount of clay and activity of the viously mentioned methods are described in Soil Survey Staff
clay (Yule and Ritchie, 1980). In summary, several properties that (2014) and are reported on an oven-dry basis. Information on the
are basic to soil survey such as total clay, CEC and PI would be soil taxonomic classifications from the soil profile descriptions
useful in predicting LEP. was also used.
Several indexes or models that can predict the swelling or shrink-
age capacity of soils have been developed. An artificial neural net- Soil samples with −1,500 kPa water-to-clay ratios greater than 0.6
work model was developed by Ikizler et al. (2010) to predict were removed from the database, except those with the following
lateral and vertical swelling pressures of soils. Seed et al. (1962) de- mineralogy classes: ferritic, gibbsitic, allitic, sesquic, parasesquic,
veloped a swelling potential index for engineering purposes (on ferruginous, magnesic, halloysitic, amorphic, ferrihydritic, glassy,
compacted clays) that used the percent total clay, clay activity, and and isotic. When the ratios are greater than 0.6, incomplete disper-
an expansion coefficient depending on the clay type. Vertical shrink- sion during the particle size determinations introduces inaccuracies,
age in Texas Vertisols using small cores was estimated by the CEC especially in soils dominated by silicates (Stolpe and Lewis, 1990;
alone (R2 = 0.76) (Yule and Ritchie, 1980). Pruška and Šedivý Soil Survey Staff, 1999; Burt, 2011). Incomplete dispersion has less
(2015) predicted swelling pressures of expansive soils using a con- of an impact on the accuracy of low clay content soils. Soil samples
sistency index (PI) and a colloidal activity index on tuffaceous clays, with more than 3% organic carbon were also removed from the da-
mudstones, claystones, chalk marlites, and neogene clay soils in the tabase to reduce potential error associated with irreversible shrink-
Czech and Slovak Republic (R2 = 0.88). Schafer and Singer (1976) age from organic matter. Organic matter represents a nonreversible
developed a prediction model for COLE using the expandable clay shrinkage portion of the soil, and LEP characterizes the reversible
fraction and porosity for 16 California soils. Their validation results shrinkage of soils (Soil Survey Division Staff, 1993). Soils with
showed an R2 of 0.59 and very little bias for a wide variety of soils fragments were also removed from the database as COLE is calcu-
(using 160 soil samples from the Kellogg Soil Survey Laboratory lated on a whole-soil basis. If coarse fragments are present, they are
[KSSL] database in Lincoln, NE). An expansive index was devel- corrected for in the calculation of COLE and thus LEP (Holmgren,
oped by Thomas et al. (2000) using 12 soils of various parent mate- 1968). For this study, soils without coarse fragments were desired
rials from Virginia. They used the swelling 2:1 minerals, PI, liquid
limit, and a swell index and summed them into an index. However,
TABLE 1. Statistical Properties of the Development Data Set
none of the previously mentioned models are specific to the predic-
tion of LEP. In the case where COLE was predicted, it was confined Property n Min Max Median Mean S.D.
to only a few soils and used parameters that are not readily available
Total clay (%) 16,359 1.5 95.0 26.1 28.9 16.4
in soil survey. This makes the previously mentioned models not suit-
able for use in predicting LEP for the diverse range in soils of the Carbonate clay (%) 6,991 <1.0 38.4 2.0 3.4 4.3
United States.
Total sand (%) 16,359 <1.0 97.7 22.6 30.3 25.7
The objective of this study was to develop and validate a model
for estimating LEP using general linear models and soil properties CEC (cmol(+) kg) 12,758 0.1 100.6 17.3 18.9 11.4
readily available within the NASIS soil survey database. The model
ECEC (cmol(+) kg) 3,601 0.1 89.1 6.4 9.3 8.8
would be limited to the basic soil properties captured in soil survey
that are related to LEP such as total sand, silt, and clay; total organic CaCO3 equiv. (%) 14,466 <0.1 95 3.0 9.1 13.1
matter; bulk density; and cation exchange properties. A prediction
Gypsum (% <2 mm) 3,159 <0.1 92 0 1.7 7.9
model that is national in scope will provide consistent and improved
accuracy of LEP estimates, which will benefit all users of soil survey Organic carbon (%) 16,359 <0.1 2.9 0.37 0.58 0.57
data and their interpretations, as soil survey data are being updated
LEP (%) 16,359 <0.1 29.6 2.5 3.7 3.7
or through initial mapping.

2 www.soilsci.com © 2017 The Author(s). Published Wolters Kluwer Health, Inc.


January 2017 • Volume 182 • Number 1 Predicting Linear Extensibility Percent

in order to estimate COLE on a <2-mm basis, which would be con- clay mineralogy classes used were smectitic, kaolinitic, vermiculitic,
sistent with the properties used in predicting LEP. Percent carbonate and illitic, as well as the major mineralogy classes of mixed, sili-
clay was subtracted from the percent total clay to get noncarbonate ceous, and carbonatic (Fig. 1). Most of the remaining taxonomic
clay (nclay) because carbonate clay has less of an influence on the mineralogy classes are dominated by Fe and Al oxides, hydroxides,
shrink-swell properties than silicate clays (Nettleton et al., 1991). amorphous materials, and other less common minerals and are of
Subtracting out the carbonate clay removes the disadvantage of the minor extent and so were grouped together. Mineralogy classes in-
particle-size measurement by giving only silicate clay amounts. cluded in this group are ferritic, gibbsitic, allitic, sesquic, parasesquic,
The database was further screened for obvious inconsistencies, and ferruginous, magnesic, halloysitic, amorphic, ferrihydritic, glassy, and
samples with Db,33 > Db,o, negative values of water contents retained isotic and will be referred to as the “other” mineralogy group. In soil
at −1,500 kPa, ECEC > CEC, and the sums of sand, silt, and clay taxonomy, a mineralogy class is assigned to all mineral soils, except
less than 100% were removed. Quartzipsamments (USDA-NRCS, 2016). The LEP concept is
In the final database, only pedons that had a correlated soil classi- not applicable to organic layers, but they are assigned to mineral
fication based on laboratory data were used. Within a pedon, hori- layers of organic soils (Soil Survey Staff, 2016). All mineral ho-
zons were used only if they met the previously listed criteria. The rizons that do not fit into the mineralogy groups defined previ-
family mineralogy class was extracted from the soil classification. ously (including those mineral horizons in Histosols and
Pedon depths varied, ranging up to 200 cm. Organic C ranged from Histels) are assigned to the “mixed” mineralogy group. Com-
0% to 3%. Horizons contained no rock fragments. The −1,500 kPa plete coverage of all soils is necessary for a national model.
water-to-clay ratios were all less than 0.6 except for those horizons Similar to calcium carbonate, gypsum does not have a CEC,
in the mineralogy classes listed previously. There were no missing and thus, it is corrected for in the sample. Soils with greater than
values for total sand, silt, and nclay. There were no missing values 5% gypsum were also grouped into a separate data set, which was
for LEP and CEC or ECEC. The statistics describing the develop- split further into two data sets at a threshold of 40% gypsum. In
ment data set are presented in Table 1. most cases, NASIS does not contain clay content (particle size in-
formation) for soils with greater than 40% gypsum. Therefore, a
Data Stratification separate group is needed because clay content cannot be used as
The database was partitioned into more homogeneous soil groups to a predictive variable. Visual examination of scatter plots of LEP
improve the accuracy of the LEP predictions. The data set for model versus nclay content for the smectitic and mixed mineralogy
development was stratified by taxonomic mineralogy class. It is well groups showed a change in the relationship at approximately
established that clay mineralogy governs the shrink-swell behavior 35% and 20% clay, respectively. To improve predictions, the
of soils (Franzmeier and Ross, 1968; Nettleton and Brasher, 1983; smectitic group was split at 35% clay, and 20% clay for the mixed
Kariuki and van der Meer, 2004; Burt, 2011). The major taxonomic mineralogy group.

FIGURE 1. The taxonomic mineralogy classes of the continental United States in this study. The “other” mineralogy class includes ferritic, gibbsitic,
allitic, sesquic, parasesquic, ferruginous, magnesic, halloysitic, amorphic, ferrihydritic, glassy, and isotic.

Soil Science • January 2017 • Volume 182 • Number 1 www.soilsci.com 3


Seybold and Libohova Soil Science

TABLE 2. Statistical Properties of the Validation Data Set


Validation
The validation data set consisted of 2,176 soil samples from KSSL
Property n Min Max Median Mean S.D.
corresponding to pedons sampled throughout the United States
Total clay (%) 2172 1.0 91.0 30.0 33.0 24.7 (Table 2). All properties of the validation data set were within the
range of the development data set. Organic C in this data set was es-
Carbonate clay (%) 382 <1.0 35 2.0 3.7 5.2
timated from the difference between the total and CaCO3-C (Soil
Total sand (%) 2172 <1.0 96.0 13.0 23.4 24.7 Survey Staff, 2014). Model performance was evaluated by compar-
ing measured versus predicted LEP values and by calculating the
CEC (cmol(+) kg) 1695 0.5 84.9 21.1 22.5 12.9
prediction RMSE (RMSEP) and mean error (ME) as defined in
ECEC (cmol(+) kg) 477 0.1 39.2 6.9 8.7 7.6 McBratney et al. (2011). Confidence intervals (P = 0.05) were cal-
culated for the slope and intercept of the least squares estimate line.
CaCO3 equiv. (%) 900 <0.1 87 2 7.5 12.5

Gypsum (% <2 mm) 304 <0.1 92 0 2.1 10.4


RESULTS AND DISCUSSION
Organic carbon (%) 2172 <0.1 2.9 0.60 0.80 0.69

LEP (%) 2172 <0.1 27.5 3.5 3.7 4.4


Model Development and Comparisons
The data set used to develop the prediction models had a broad range
in soil properties and represents the soils of the United States (Fig. 2).
Clay content ranged from 1.5% to 95%, and CEC ranged from 0.1
Statistical Analysis to 100.6 cmol(+) kg−1 (Table 1). The range in LEP varied from 0%
Linear extensibility percent was estimated using general linear to 29.6%. Given the wide range of soil properties, the LEP for
model procedures in SYSTAT Software (2009). For each stratum the complete unstratified data set was highly correlated with CEC
(data group), the best fit model (with the lowest root mean square er- (r = 0.834), ECEC (r = 0.708), nclay (r = 0.813), and total clay con-
ror [RMSE]) was developed. The RMSE gives the accuracy of the tent (r = 0.743). The high correlations with these properties were
estimations in terms of S.D.; the smaller the RMSE, the more accu- expected and are similar to what others have found between clay,
rate the model. Pearson correlation coefficients were computed to CEC, and some measure of the soil shrinkage and/or swelling po-
help in the selection of predictive variables. Only data elements that tential (Gill and Reaves, 1957; Seed et al., 1962; Franzmeier and
contributed significantly (P = 0.05) to predicting LEP and that con- Ross, 1968; McCormack and Wilding, 1975; Grey and Allbrook,
tributed more than 5% to the overall improvement of the coefficient 2002; Kariuki and van der Meer, 2004). The CEC and ECEC can
of determination (R2) were included in the equations. The R2 value give an indication of the clay mineralogy (Burt, 2011). In an exam-
represents the proportion of total variability in LEP data that is ex- ple, Puppala et al. (2013) used a high CEC value to indicate the
plained by a model. Scatter plots of the residuals versus the fitted presence of smectite clay minerals and a low CEC to indicate
values of each model were used to determine whether there were the presence of nonexpansive clay minerals such as kaolinite. In
nonlinearity, unequal variances, and outliers in the data. All outliers, the present study, LEP weakly correlated with calcium carbonate
as identified by the studentized residual in SYSTAT Software (r = −0.196) and organic carbon (r = 0.158) contents. Similarly,
(2009), were removed from the data groups. others have shown no relationship between COLE and organic C

FIGURE 2. Sample location of pedons (continental United States only) that make up the development data set used in the prediction of LEP.

4 www.soilsci.com © 2017 The Author(s). Published Wolters Kluwer Health, Inc.


January 2017 • Volume 182 • Number 1 Predicting Linear Extensibility Percent

TABLE 3. Model to Predict LEP for Soil Layers With <3% Organic Carbon
Models Equation† n R2 RMSE

1. Gypsum >5% 0.082 × ngclay − 0.096 × CEC + 0.008 × (CEC) + 0.263 2


218 0.88 1.444

2. Gypsum >40% and clay not available 0.157 × CEC − 0.276 52 0.75 0.254

3. Smectitic and clay 35% 0.16 × nclay + 0.18 × CEC − 4.219 2000 0.66 2.447

0.176 × nclay + 0.088 × ECEC − 1.346 274 0.42 2.863

4. Smectitic and clay <35% 0.128 × nclay + 0.088 × CEC − 1.613 956 0.52 1.288

0.14 × nclay + 0.113 × ECEC − 2.239 179 0.67 1.020

5. Kaolinitic 0.017 × nclay + 0.18 × CEC − 0.227 201 0.62 0.912

0.003 × nclay + 0.001 × (nclay) + 0.236 × ECEC − 0.067


2
385 0.51 1.768

6. Mixed and clay 20% 0.135 × nclay + 0.132 × CEC − 2.611 4619 0.67 1.434

0.06 × nclay + 0.167 × ECEC − 1.126 1049 0.59 0.976

7. Mixed and clay <20% 0.063 × nclay + 0.07 × CEC − 0.279 3408 0.43 0.640

0.05 × nclay + 0.084 × ECEC − 0.289 774 0.49 0.454

8. Siliceous‡ 0.016 × nclay + 0.17 × CEC + 0.003 × (CEC) − 0.265


2
393 0.78 0.826

0.016 × nclay + 0.199 × ECEC − 0.098 752 0.64 0.512

9. Vermiculitic 0.151 × nclay + 0.103 × CEC − 2.881 28 0.86 1.475

10. Illitic 0.026 × nclay + 0.227 × CEC − 1.398 135 0.79 0.817

0.02 × nclay + 0.223 × ECEC − 0.548 46 0.47 1.049

11. Carbonatic 0.16 × nclay + 0.092 × CEC − 1.878 332 0.73 1.271

12. Other§ 0.065 × nclay + 0.06 × CEC + 0.015 416 0.60 1.063

−0.025 × nclay + 0.002 × (nclay) + 0.049 × ECEC + 0.599


2
142 0.61 0.808

13. All other mineral horizons¶ Uses the mixed mineralogy equations
Equations were developed based on taxonomic mineralogy classes and selected soil properties. The equations are ordered to serve as a guide in the selection
of the appropriate model.
†ngclay = nclay adjusted to a gypsum-free basis (%); nclay = nclay content (%); CEC = cation-exchange capacity (cmol(+) kg−1); ECEC = effective cation-
exchange capacity (cmol(+) kg−1); nclay (%) = total clay (%) − carbonate clay (%).
‡Includes Quartzipsamments.
§Other mineralogy classes include ferritic, gibbsitic, allitic, sesquic, parasesquic, ferruginous, magnesic, halloysitic, amorphic, ferrihydritic, glassy, and isotic.
¶Includes mineral horizons of Histosols and Histels.

(Schafer and Singer, 1976; de Jong et al., 1992) and have found soils, a second set of eight prediction equations was developed that
COLE to have no correlation or to be negatively correlated with use ECEC as a predictor variable in place of CEC (Table 3). Because
the carbonate content of the soil (de Jong et al., 1992; Dinka gypsum generally does not occur in low pH soils, no additional
et al., 2013). The high correlations with CEC and nclay suggest equations were needed. In low pH soils, the ECEC and nclay content
they would be the most useful in predicting LEP. were the most useful variables in predicting LEP. For six of the equa-
The development data set was stratified into 12 data sets based on tions, the ECEC and nclay content explained between 42% and 67%
their gypsum content, taxonomic mineralogy class, and clay content of the variability in LEP (Table 3). Noncarbonate clay, (nclay)2, and
(Table 3). For each data set, a general linear model was developed ECEC explained 51% of the variation in LEP for the kaolinitic min-
for a total of 12 prediction equations (Table 3). The CEC and nclay eralogy group (Table 3). Similarly, for the “other” mineralogy data
content were found to be the most useful in predicting LEP. The set, nclay, (nclay)2, and ECEC explained 61% of the variation in
CEC and nclay content alone explained between 43% and 86% of LEP. No other combination of predictor variables provided equa-
the variability in LEP for nine of the equations (Table 3). For soils tions with a lower RMSE.
with greater than 5% gypsum, nclay (on a gypsum-free basis) and The RMSEs ranged from 0.25 to 2.86 among all the equations
CEC explained 88% of the variability in LEP. For soils with greater (Table 3). In general, the RMSEs were slightly higher for the smec-
than 40% gypsum, CEC alone explained 75% of the variability in titic equations compared with the other groups. They also have the
LEP. Noncarbonate clay, CEC, and (CEC)2 explained 78% of the greatest range in LEP values among the mineralogy groups, because
variation in LEP for the siliceous mineralogy group (Table 3). No RMSE depends on the scale or range of the data.
other combination of predictor variables provided equations with The numbered models serve as a guide to help select the correct
a higher R2 and lower RMSE for either of the data sets. equation in predicting LEP (Table 3) and provide flexibility for the
In soil survey, CEC is available only for soils that have a pH user. Based on the properties of the soil and taxonomic classifica-
greater than 5.5, and ECEC is available for low pH layers (USDA- tion, the user selects the prediction equation from Table 3 starting
NRCS, 2016). To make the prediction of LEP available for low pH with Model 1. If gypsum is present in the soil and whether total clay

Soil Science • January 2017 • Volume 182 • Number 1 www.soilsci.com 5


Seybold and Libohova Soil Science

Model Validation

Each soil sample from individual pedons in the validation data set
was run through the appropriate predictive model to estimate LEP,
based on the models in Table 3. All developed prediction models
were used as indicated by the breakdown of the soil taxonomic min-
eralogy classes in the validation data set (1% has >5% gypsum, 24%
smectitic, 5% kaolinitic, 61% mixed, 3% siliceous, 0.5% vermiculitic,
1% illitic, 2% carbonatic, and 2.5% other). This is very similar to the
mineralogy class breakdown of the development data set (1.5% has
> 5% gypsum, 21% smectitic, 3.5% kaolinitic, 60% mixed, 7% sili-
ceous, 0.5% vermiculitic, 1% illitic, 2% carbonatic, and 3.5% other).
A plot of measured versus predicted LEP values for all other equa-
tions is presented in Fig. 3. The collective model (all equations
using CEC) explained 87% of the variation in the LEP data. Accu-
racy of the predictions produced an RMSEp of 1.44% and ME of
−0.16% (Fig. 3).
Equations using ECEC as a predictor variable were validated as a
separate group. Validation of the equations using ECEC as a predic-
tor variable for low pH soils is shown in Fig. 4. The collective model
explained 77% of the variation in the LEP data. Accuracy of the pre-
dictions produced an RMSEp of 1.29% and ME of −0.034% (Fig. 4).
Prediction accuracy for equations using ECEC was slightly lower
FIGURE 3. A scatter plot of the measured versus predicted LEP than that for equations using CEC. It should be noted that the range
representing prediction equations that use CEC as one of the predictor in LEP is smaller for equations using ECEC than that for equations
variables, as listed in Table 3. using CEC. Validation results from Schafer and Singer (1976) on
a model developed from 16 California soils could predict LEP
is available will determine which gypsum equations to use in predicting within 1% to 2%. Also, simple linear regressions using total clay
LEP (Table 3). Next, if gypsum is not present in the soil, then the for predicting LEP were developed from laboratory characterization
taxonomic mineralogy class and total clay content are used to deter- data for each of the taxonomic mineralogy groups (montmorillon-
mine which equation to use (Table 3). For example, if the soil has itic, mixed, and kaolinitic) that resulted in R2 ranging from 0.44 to
mixed mineralogy and the clay content is less than 20%, then the ap- 0.77 (Soil Survey Laboratory Staff, unpublished data, 1981). There
propriate equation is used to predict LEP (depending whether CEC are no other published studies that attempted to predict LEP or ad-
or ECEC is available). The models provide an LEP estimate for ev- dressed such a wide range of soils as in the present study.
ery soil sample that has an organic carbon content of 3% or less,
given that the CEC or ECEC, nclay content, and the soil classifica-
tion are known. If rock fragments are present in the soil, then the pre-
TABLE 4. Validation Results for the Prediction of LEP for Each of the
dicted LEP is adjusted to a whole-soil basis: (1 − volume fraction of
Mineralogy Groups
fragments)  LEP.
Mineralogy n R2 RMSEp ME

Gypsum content >5% 19 0.95 0.813 0.073

Smectitic 534 0.80 1.80 −0.390

35% clay 434 0.72 1.787 −0.450

<35% clay 100 0.58 0.732 −0.130

Kaolinitic 108 0.60 1.750 −0.001

Mixed† 1313 0.77 1.086 −0.063

20% clay 913 0.71 1.144 −0.005

<20% clay 400 0.34 0.498 −0.196

Siliceous‡ 71 0.74 0.418 −0.331

Vermiculitic 10 0.90 1.297 −0.590

Illitic 25 0.62 1.079 0.056

Carbonatic 37 0.61 1.218 −0.689

Other¶ 53 0.45 1.580 0.609


†Includes mineral horizons of Histosols and Histels and other classifica-
tions where the mineralogy class is not used.
‡Includes Quartzipsamments.
FIGURE 4. A scatter plot of the measured versus predicted LEP ¶Other mineralogy classes include ferritic, gibbsitic, allitic, sesquic,
representing prediction equations that use ECEC as one of the predictor parasesquic, ferruginous, magnesic, halloysitic, amorphic, ferrihydritic, glassy,
variables, as listed in Table 3. and isotic.

6 www.soilsci.com © 2017 The Author(s). Published Wolters Kluwer Health, Inc.


January 2017 • Volume 182 • Number 1 Predicting Linear Extensibility Percent

In the plots of measured versus predicted LEP in Fig. 3, the 95% Bohn H. I., B. L. McNeal, and G. A. O’Connor. 1979. Soil Chemistry. John
confidence intervals for the slope (0.825, 0.856) do not include 1, Wiley & Sons, Inc., New York.
suggesting a significant difference from unit 1 slope. The 95% con- Brasher B. R., D. P. Franzmeier, V. Valassis, and S. E. Davidson. 1966. Use of
fidence intervals for the intercept (0.574, 0.787) do not include 0, saran resin to coat natural soil clods for bulk-density and water-retention
measurements. Soil Sci. 101:108.
which indicates the intercept is significantly different from 0. Simi- Bronswijk J. J. B. 1989. Prediction of actual cracking and subsidence in clay
larly, in plots of measured versus predicted LEP in Fig. 4, the 95% soils. Soil Sci. 148:87–93.
confidence intervals for the slope (0.760, 0.836) do not include 1, Burt R. 2011. Soil Survey Laboratory Information Manual. Soil Survey In-
and the 95% confidence intervals for the intercept (0.410, 0.739) vestigations Report No. 45. Version 2.0. US Government Printing Office,
do not include 0. For both plots, this means that more than 95% of Washington, DC.
the time similarly constructed intervals will not contain unit 1 slope de Jong E., L. M. Kozak, and H. B. Stonehouse. 1992. Comparison of shrink-
and 0 intercept. In other words, the model slightly overestimates swell indices of some Saskatchewan soils and their relationships to stan-
LEP starting at the intercept (0 LEP), then crosses over the 1:1 line dard soil characteristics. Can. J. Soil Sci. 72:429–439.
Dinka T. M., C. L. S. Morgan, K. J. Mcinnes, A. S. Kishné, and R. D. Harmel.
at some point, and beyond that point, the model underestimates LEP 2013. Shrink-swell behavior of soil across a Vertisol catena. J. Hydrology.
as LEP increases. The crossover point is at LEP of 4% in Fig. 3 and 476:352–359.
at 2.75% in Fig. 4. However, the overall ME is small, −0.16% for Favre F., P. Boivin, and M. C. S. Wopereis. 1997. Water movement and soil
equations using CEC and −0.034% for equations using ECEC. swelling in a dry, cracked Vertisol. Geoderma. 78:113–123.
The small negative ME indicates an overall underestimation. Franzmeier D. P., and S. J. Ross Jr. 1968. Soil swelling: Laboratory measure-
Validation results for the different mineralogy groups are pre- ment and relation to other soil properties. Soil Sci. Soc. Am. Proc.
sented in Table 4. The RMSEp ranged from 0.42% to 1.80% 32:573–577.
among the mineralogy groups, with the smectitic group having Gill W. R., and C. A. Reaves. 1957. Relationship of Atterberg limits and
cation-exchange capacity to some physical properties of soil. Soil Sci.
the largest RMSEp. However, the smectitic group also has the Soc. Am. Proc. 21:491–494.
largest range in LEP. The siliceous group had the lowest RMSEp. Grey C. W., and R. Allbrook. 2002. Relationships between shrinkage indices
In general, the lower clay content groups tended to have lower and soil properties in some New Zealand soils. Geoderma. 108:287–299.
RMSEp. The range in the RMSEp was lower than that for the Grossman R. B., B. R. Brasher, D. P. Franzmeier, and J. L. Walker. 1968. Lin-
RMSE of the estimate and tended to be lower or similar for most ear extensibility as calculated from natural-clod bulk density measure-
of the mineralogy groups. The MEs or bias ranged from −0.689 to ments. Soil Sci. Soc. Am. Proc. 32:570–573.
0.609 among the different mineralogy groups (Table 4). The “other” Hallberg G. R. 1977. The use of COLE values for soil engineering evaluation.
mineralogy group had the largest positive ME of 0.609, which indi- Soil Sci. Soc. Am. J. 41:775–777.
Holmgren G. S. 1968. Nomographic calculations of linear extensibility in
cates an overestimation, whereas the carbonatic group had the largest
soils containing coarse fragments. Soil Sci. Soc. Am. Proc. 32:568–570.
negative ME of −0.689, which indicates an underestimation of LEP. Six Ikizler A. B., M. Aytekin, M. Mekli, and F. Kocabas. 2010. Prediction of
of the nine mineralogy groups had a small negative ME (Table 4). In swelling pressures of expansive soils using artificial neural networks.
general, the MEs were small, indicating that these models can pre- Adv. Eng. Softw. 41:647–655.
dict LEP reasonable well for the wide variety of soils of the United States. Kariuki P. C., and F. van der Meer. 2004. A unified swelling potential index
for expansive soils. Eng. Geol. 72:1–8.
Mandal D. K., C. Mandal, and M. V. Venugopalan. 2005. Suitability of
CONCLUSIONS cotton cultivation in shrink-swell soils in central India. Agric. Sys.
Models were developed for predicting LEP that can be applied 84:55–75.
to soils of the United States. The data were stratified based on Mbagwu J. S. C. 1992. Evaluation of shrink-swell potential of soils by two
the taxonomic mineralogy class and pH, which resulted in 20 procedures. Pedologie. 42:69–82.
general linear models for predicting LEP. The models apply to McBratney A. B., B. Minasny, and G. Tranter. 2011. Necessary meta-data for
pedotransfer functions. Geoderma. 160:627–629.
mineral soils with less than 3% organic carbon. Noncarbonate clay
McCormack D. E., and L. P. Wilding. 1975. Soil properties influencing swell-
and CEC were the variables most highly correlated with LEP. ing in Canfield and Geeburg soils. Soil Sci. Soc. Am. J. 39:496–502.
Noncarbonate clay and CEC were able to explain between 43% Nettleton W. D., and B. R. Brasher. 1983. Correlation of clay minerals and
and 86% of the variation in LEP for nine of the models, whereas properties of soil in the western United States. Soil Sci. Soc. Am. J.
nclay and ECEC (low pH soils) explained between 42% to 67% 47:1032–1036.
of the variability in LEP for six of the models. Validation results Nettleton W. D., B. R. Brasher, and S. L. Baird. 1991. Carbonate clay charac-
showed a prediction accuracy (RMSEp) of 1.44% with an ME of terization by statistical methods. In: Nettleton W. D. (ed.). Occurrence,
−0.16% for equations using CEC as a predictor variable and an Characteristics, and Genesis of Carbonates, Gypsum, and Silica Accumu-
RMSEp of 1.29% with an ME of −0.034% for equations using lations in Soils. SSSA Spec. Publ. no 26. 75–87.
ECEC as a predictor variable. The RMSEp ranged from 0.42% to Pruška J., and M. Šedivý. 2015. Prediction of soil swelling parameters. Proc.
1.80% among the different mineralogy groups, which were gener- Earth Planet. Sci. 15:219–224.
ally lower than the RMSE of the estimate. The models are consid- Puppala A. J., T. Mhammanoon, and B. C. S. Chittoori. 2013. Swell and
shrinkage characterizations of unsaturated expansive clays from Texas.
ered adequate for use in soil surveys. Eng. Geol. 164:187–194.
The LEP prediction models developed in this study are an im- Ross G. J. 1978. Relationships of specific surface area and clay content to
provement over previous estimation procedures based on visual esti- shrink-swell potential of soils having different clay mineralogical compo-
mations of shrink-swell potential and on a discrete class assignment. sitions. Can. J. Soil Sci. 58:159–166.
The models will be made available as a calculation within the NASIS Schafer W. M., and M. J. Singer. 1976. Influence of physical and mineralog-
database for field soil scientists to use when measured data are not ical properties on swelling of soil in Yolo County, California. Soil Sci.
available. Initial soil mapping is still being conducted in the United Soc. Am. J. 40:557–562.
Seed H. B., R. J. Woodward, and R. Lundgren. 1962. Prediction of swell-
States, and many areas are undergoing updates. Improvements in
ing potential for compacted clays. J. Soil. Mech. Foundations Div.
LEP estimates will improve interpretations generated from soil 88:53–87.
survey data, which benefits all users of soil survey information. Soil Survey Division Staff. 1993. Soil Survey Manual. Soil Conservation Ser-
vice. US Department of Agriculture, Handbook no. 18. US Gov. Print.
Office, Washington, DC.
REFERENCES Soil Survey Staff. 1999. Soil Taxonomy: A Basic System of Soil Classification
Al-Rawas A. A. 1999. The factors controlling the expansive nature of the for Making and Interpreting Soil Surveys. US Department of Agriculture,
soils and rocks of northern Oman. Eng. Geol. 53:327–350. Handbook no. 436. 2nd ed. US Gov. Print. Office, Washington, DC.

Soil Science • January 2017 • Volume 182 • Number 1 www.soilsci.com 7


Seybold and Libohova Soil Science

Soil Survey Staff. 2014. Kellogg soil survey laboratory methods manual. Soil USDA-NRCS. 2016. National Soil Survey Handbook, title 430-VI. [Online]
Survey Investigations Report No. 42. Version 5.0. US Department of Ag- Available at http://www.nrcs.usda.gov/wps/portal/nrcs/detail/soils/ref/?
riculture, Natural Resources Conservation Service, p. 998. Available at cid=nrcs142p2_054242. Accessed July 7, 2016.
https://www.nrcs.usda.gov/wps/portal/nrcs/detail/soils/ref/?cid= Vaught R., K. R. Brye, and D. M. Miller. 2006. Relationship among coeffi-
nrcs142p2_054247 (accessed January 5, 2017). cient of linear extensibility and clay fractions in expansive, stony soils.
Stolpe N. B., and D. T. Lewis. 1990. Mineralogical implications of high ratios Soil Sci. Soc. Am. J. 70:1983–1990.
of 1500 kPa water content to percent measured clay for two soils of west- Yule D. F., and J. T. Ritchie. 1980. Soil shrinkage relationships of Texas
ern Nebraska. Geoderma. 46:1–16. Vertisols: I small cores. Soil Sci. Soc. Am. J. 44:1285–1291.
SYSTAT Software. 2009. SYSTAT 13 for Windows. Version No. 13.1. Zhang X., Y. Hao, Y. Zhao, F. Min, and S. Song. 2016. Study on the dif-
SYSTAT Software Inc., San Jose, CA. ferences of Na- and Ca-montmorillonites in crystalline swelling re-
Thomas P. J., J. C. Baker, and L. W. Zelazny. 2000. An expansive index for gime through molecular dynamics simulation. Adv. Powder Technol.
predicting shrink-swell potential. Soil Sci. Soc. Am. J. 64:268–274. 27:779–785.

8 www.soilsci.com © 2017 The Author(s). Published Wolters Kluwer Health, Inc.

You might also like