Geoderma 400 (2021) 115221

Contents lists available at ScienceDirect

Geoderma
journal homepage: www.elsevier.com/locate/geoderma

Evaluating models to estimate cation exchange capacity of calcareous soils

Fatemeh Razzaghi a, b, Emmanuel Arthur c, *, Ali Akbar Moosavi d
a
Department of Water Engineering, College of Agriculture, Shiraz University, Shiraz, Iran
b
Drought Research Center, Shiraz University, Iran
c
Department of Agroecology, Faculty of Technical Sciences, Aarhus University, Denmark
d
Department of Soil Science, College of Agriculture, Shiraz University, Shiraz, Iran

A R T I C L E I N F O A B S T R A C T

Handling Editor: Ingrid Kögel-Knabner Cation exchange capacity (CEC) is one of the fundamental properties of soil that indicates soil’s ability to retain
exchangeable cations. Several regression-based models exist to estimate CEC from widely available or easily
Keywords: measured basic soil properties. However, the majority of these models do not include CaCO3 as an input variable.
Calcium carbonate Consequently, the applicability of these models for calcareous soils is largely unknown. This study evaluates the
Clay
performance of selected CEC models for soils rich in calcium carbonate and assesses the influence of CaCO3 on
Hygroscopic water
CEC estimation performance. Thirty-eight soil samples that had a wide range in CEC (4 to 59 cmol(+) kg− 1) and
Pedotransfer functions
Regression-based model CaCO3 content (10 to 97%) were selected to evaluate seven published CEC models that had inputs such as clay,
silt, sand, pH, organic carbon/organic matter, CaCO3, and hygroscopic water content. The models that comprised
clay, silt, sand, and organic carbon overestimated CEC for samples with CEC > 40 cmol(+) kg− 1 and vice versa
(average RMSE = 13.7 ± 1.2 cmol(+) kg− 1 and r2 = 0.41 ± 0.03). One model that included CaCO3, performed
slightly better (RMSE = 11.7 cmol(+) kg− 1). The model based on hygroscopic water content (28% relative hu­
midity) gave the best estimates of CEC (RMSE = 6.2 cmol(+) kg− 1 and r2 = 0.88).
Apart from the water content-based model, there was a consistent trend of increasing CEC estimation errors
with increasing CaCO3 contents. A newly developed CEC model based on clay and CaCO3 contents explained the
variation in CEC better than all the published models, except the water content model.

1. Introduction advantages and drawbacks and are used for specific soils (e.g., acidic or
alkaline soils) or clays. (Sumner and Miller, 1996).
The cation exchange capacity (CEC) of soil provides an insight into To provide an alternative to the conventional methods, especially for
soil fertility and nutrient retention capacity. Besides this agronomic large-scale studies, predictive models exist to estimate CEC from widely
importance, soil CEC is important for civil engineering applications such available or more easily measured basic soil properties (e.g., Minasny
as assessment of the swelling potential of clays, and preliminary design et al., 2004). Soil properties such as clay (Cl), silt and sand contents and
estimations of materials used for construction (Yilmaz, 2006). Quanti­ organic carbon (e.g., Bell and van Keulen, 1995; McBratney et al., 2002;
tatively, CEC is a measure of negatively charged sites on the soil surface, Rehman et al., 2019), soil specific surface area, and Atterberg limits
which retain positively charged ions, mainly Al3+ and H+ in acidic soils (Yukselen and Kaya, 2006), soil water content (Arthur, 2017), and soil
and Na+, K+, Ca2+, and Mg2+ in neutral and alkaline soils. pH have been used singly or in various combinations to develop
Conventionally, CEC is determined by the ammonium acetate regression-based or more sophisticated models (e.g., Kashi et al., 2014)
method at pH 7.0 (Schollenberger and Dreibelbis, 1930), the ammonium for CEC estimation. Alternate approaches such as near-infrared spec­
chloride method (Shuman and Duncan, 1990), the sodium acetate troscopy (Rehman et al., 2019) and laser-induced breakdown spectros­
method (Bower et al., 1952), the barium chloride method at pH 8.2 copy (Pelagio et al., 2020) have also been used to estimate CEC with
(Mehlich, 1939) or the methylene blue absorption (Nevins and Wein­ varying degrees of success.
tritt, 1967). All these methods provide a direct measure of CEC but can Apart from a few instances (e.g., Asadzadeh et al., 2019), the ma­
be difficult, time-consuming, and expensive to use (Ghorbani et al., jority of the regression-based models were based on soils that had low
2015; Krogh et al., 2000). Also, these methodologies all have their amounts of calcium carbonates, did not include CaCO3 as a model

* Corresponding author.
E-mail address: [email protected] (E. Arthur).

https://doi.org/10.1016/j.geoderma.2021.115221
Received 2 December 2020; Received in revised form 26 April 2021; Accepted 2 May 2021
Available online 24 May 2021
0016-7061/© 2021 Elsevier B.V. All rights reserved.
F. Razzaghi et al. Geoderma 400 (2021) 115221

variable, and hence the applicability of these models for calcareous soils
is largely unknown. Calcareous soils are those soils with high CaCO3
whose physical problems of land and water use for crop production are
primarily dominated by the high CaCO3, especially the active fraction
with high specific surface area (Tadry, 1973). Calcareous soils generally
have more than 15% of CaCO3 in various forms (Taalab et al., 2019).
These soils cover at least 1000 million ha worldwide (WRB, 2015), with
a substantial amount of the soils occurring naturally in arid and semi-
arid regions, such as Iran, because of limited leaching. Moreover, the
measurement of soil CEC in calcareous soil can be fraught with signifi­
cant measurement errors (Silva et al., 2010). The errors occur due to
partial dissolution of calcite, as a result of the interaction with the ex­
change solution, rich in electrolytes, with a resulting underestimation of
total CEC and overestimation of exchangeable calcium (Ca2+) and
magnesium (Mg2+) (Dohrmann, 2006).
Due to the widespread nature of calcareous soils and the fact that the
majority of existing models do not consider CaCO3 as an input variable,
this study set out to (i) examine the performance of selected CEC models
on soils rich in calcium carbonate, and (ii) assess the influence of CaCO3
on CEC estimation performance.

2. Materials and methods

2.1. Investigated soil samples Fig. 1. Distribution of the investigated samples in the USDA textural triangle.

The 38 soil samples considered for the study originated from the Walkley–Black chromic acid wet oxidation method (Nelson and
different locations in Iran (29 samples), Texas, USA (four samples), Cuba Sommers, 1996) and then converted to organic carbon by assuming that
(five samples), and Kenya (one sample). The samples were chosen to approximately 50% of organic matter (OM) is organic carbon (OC)
encompass a wide range in both CEC and calcium carbonate content (Pribyl, 2010).
(Table 1 and Table S1). The samples had a wide range for the relevant
soil properties: all the USDA textural classes, except sand, silt loam, and 2.2.2. Cation exchange capacity and calcium carbonate equivalent
silt, were present (Table 1; Fig. 1). The CaCO3 content ranged from 10 to Cation exchange capacity was measured with the ammonium acetate
97% (mean = 38%), and CEC ranged from 4 to 59 cmol(+) kg− 1 (mean = extraction method (Sumner and Miller, 1996) buffered at pH 8.2. Cal­
22 cmol(+) kg− 1). The samples were generally low in OC contents (mean cium carbonate equivalent was measured with the neutralization
= 0.75%) and average pH of 8.0 (range from 7.4 to 8.9). The clay method by 1.0 N hydrochloric acid and titration with sodium hydroxide
mineralogy of the samples was dominated by smectite, illite, and trace (Loeppert and Suarez, 1996).
amounts of kaolinite and vermiculite (Arthur, 2017; Arthur et al., 2020).
2.2.3. Soil water content determination
The soil water content for the samples was measured in triplicate
2.2. Laboratory measurements with one of two methods. For the samples from Iran, three replications
of ~ 10 g were weighed into a pre-weighed can. The weighted soil
2.2.1. Soil texture and organic matter contents samples were kept in a humidity and temperature-controlled (20 ◦ C)
The particle size distribution (texture) of the samples was deter­ laboratory for two weeks to allow for equilibration of the samples to
mined in triplicate with a combination of wet sieving and pipette ambient relative humidity (RH) of 28%. Thereafter, the soil samples and
methods (Gee and Or, 2002). During the measurement procedure, soil the cans were re-weighed and placed in an oven at 105 ◦ C for 48 h.
organic matter was removed by H2O2, and despite the high contents of Samples were subsequently cooled down in desiccators and weighed
carbonates, soil particle dispersion was attained by adding 5% sodium again to determine the oven-dry weight. The gravimetric water content
hexametaphosphate (addition of larger amounts of sodium hexameta­ at 28% RH (w28) was estimated using Equation (1):
phosphate risks flocculation of the soil particles, which will interfere [ ]
with the particle size estimates). Soil organic matter was determined by M28 − MOD
w28 % = × 100 (1)
MOD
Table 1
where w28 is the soil gravimetric water content at 28% RH, M28 is the soil
Descriptive statistics of soil properties for investigated samples (n = 38).
mass after equilibration at 28%RH, and MOD denotes the soil mass after
Property Min Max Mean Median CV Skewness oven drying (Arthur et al., 2017).
%
For the remaining nine samples, the adsorption water content at 28%
Clay %w/w 9 76 38 37 40 0.28 RH was determined with an automated vapor sorption analyzer as
Silt 8 61 30 28 41 0.74
described in Arthur et al. (2014).
Sand 4 83 32 26 64 0.65
OC 0.07 2.34 0.75 0.66 71 1.49
CaCO3 10 97 51 53 42 − 0.03
w28 0.66 8.14 3.05 2.64 57 1.06
2.3. Selected models to estimate cation exchange capacity
pH – 7.4 8.9 8.0 7.9 4.3 0.91
CEC cmol(+) 4 59 22 18 66 1.33 Seven published models were evaluated in this study (Table 2). The
kg− 1 first set of models [M1 and M2] utilize a combination of clay, silt, sand,
CV, coefficient of variation; OC, soil organic carbon; CaCO3, calcium carbonate pH, and OC or OM; one of the models [M3] considers the CaCO3 content,
equivalent; w28, soil water content measured at 28% relative humidity; CEC, the third set of models [M4 to M6] employ only the clay and OC con­
cation exchange capacity. tents, whiles the last model (M7) utilize only w28. The number of

2
F. Razzaghi et al. Geoderma 400 (2021) 115221

Table 2 of CEC, particularly for models that do not have CaCO3 content as an
Selected models for estimating cation exchange capacity from other soil input variable. For this, we used three approaches:
properties.
Model Equation No. Reference (i) examined the relationship between the percent estimation error
M11 9.47 + 0.48 × Cl − 0.24 × Si − 0.35 × 65 Khaledian et al.
([measured CEC– estimated CEC]/measured CEC × 100) and the
Sa + 3.13 × pH − 0.233 × OC (2017) CaCO3 content,
M22 42.8 − 5.36 × pH + 0.30 × OM − 2.04 × 148 Bell and van Keulen (ii) through partial correlation analyses, quantify the relationship
Cl + 0.36 × Cl × pH (1995) between CaCO3 content and CEC while excluding the effect of
M3 0.027 + 0.811 × Cl − 0.045 × Sa + 334 Asadzadeh et al. clay content.
0.168 × OM − 0.091 × CaCO3 (2019) (iii) through multiple regression analyses, identify the contribution of
M43 − 29.25 + 8.14 × Cl + 0.25 × OC 1930 McBratney et al. CaCO3 content to the measured CEC considering the soil particle
(2002)
size fractions and organic carbon contents, and,
M51 0.60 + 0.61 × Cl + 2.00 × OC 188 Rehman et al. (2019)
M6 7.71 + 0.26 × Cl + 4.06 × OC 1141 Khodaverdiloo et al.
(2018)
The regression and partial correlation analyses were conducted in R
M7 2.46 + 7.55 × w28 203 Arthur (2017) v4.0.2 (R Core Team, 2020) using the lm and the pcor.test functions,
1
respectively.
Clay (Cl), Silt (Si), Sand (Sa), Calcium carbonate equivalent (CaCO3), water
content at 28% relative humidity (w28) and organic carbon (OC) are in weight %;
2 3. Results and discussion
OM is in g kg− 1; 3estimated CEC is in mmol(+) kg− 1; No. denotes the number of
samples used for developing the models.
3.1. Correlation of CEC to selected soil properties

samples, their sources, and inputs for the models are summarized below:
The CEC of the samples were correlated significantly to the clay (r =
Model M1 (Khaledian et al., 2017) was developed based on 65 Ira­
0.64***), sand (r = − 0.62***), and CaCO3 (r = − 0.75***) contents, and
nian soil samples and uses clay, silt, sand, pH, and OC as input param­
w28 (r = 0.94***) and weakly correlated to silt (r = 0.23) content
eters. Model M2 (Bell and van Keulen, 1995) relates CEC to soil pH, clay,
(Table 3). There was no correlation between the CEC and OC content of
and OM contents based on 148 soil samples from four sites in Mexico.
the samples. Although the impact of OC on soil CEC, relative to clay
Model M3 (Asadzadeh et al., 2019) used 417 calcareous samples from
content, is small, CEC estimation models indicate a positive contribution
East-Azerbaijan province in northwest Iran, and the model considers the
of OC to CEC (e.g., Asadzadeh et al., 2019; McBratney et al., 2002). This
clay, silt, sand, OM, and CaCO3 contents. Models M4 to M6 all use clay
is, however, not always the case. In soils dominated by montmorillonitic
and OC to estimate CEC; M4 (McBratney et al., 2002) was based on 1930
clay, OC tends to decrease CEC, whereas, in kaolinite-rich or high
soil samples, M5 (Rehman et al., 2019) was developed based on 188
organic soils, OC tends to increase the CEC (Manrique et al., 1991;
samples from several countries, and M6 (Khodaverdiloo et al., 2018)
Ramos et al., 2018; Tan and Dowling, 1984). For the present work, the
was based on 1141 samples from arid and semi-arid regions in Iran. The
low average OC content (0.75%), large clay and silt contents, and the
M7 model (Arthur, 2017) utilizes w28 and was developed using 203
confounding effects of CaCO3 content could have contributed to the
samples from several countries across different continents. The general
apparent lack of an effect of OC on the CEC. Considering the sources of
form of the M7 model is CEC = a + b × wRH, where a and b are the
cations in soil (primarily clay, silt, and OC contents), sand content has
intercept and the water content coefficient, respectively, is the gravi­
little or no real impact on soil CEC. Despite this, several studies report a
metric water content at a specified RH. For RH = 28%, a = 2.46 cmol(+)
strong negative correlation between CEC and sand content (e.g., Asad­
kg− 1 and b was computed from Equation (2):
zadeh et al., 2019; Khaledian et al., 2017). The apparent link between
b = 15.42 − 0.45 × RH + 0.007 × RH 2 − 3.44 × 10− 5 × RH 3 (2) sand and CEC is thus due to the strong correlation between clay and sand
content in the cited studies. This is also seen in the present study, where
Thus b = 7.55 and M7 becomes CEC = 2.46 + 7.55 × w28. clay content was significantly correlated to sand content (r = − 0.81***).
The strong dependence of CEC on clay content is well established; for a
2.4. Evaluation of model performance specific clay mineralogy, increasing clay contents results in higher CEC
values. This explains the presence of clay content in the majority of CEC
The performance of the models was evaluated with three indices; the estimation models, such as the M1 to M6 models (Table 2). The strong
mean error/bias (ME), the root mean squared error (RMSE), and the positive correlation between CEC and w28 is because, during the water
Nash-Sutcliffe efficiency coefficient (NSE; Nash and Sutcliffe, 1970). sorption process, water molecules first cluster around cations or ex­
change sites before the formation of layers on the particle surface (Laird,
1∑ n
1999; Woodruff and Revil, 2011). Thus, the mass of water sorbed is
ME = (CECe − CECm ) (3)
n i=1 directly proportional to the cations available, and by extension, the CEC.
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ The CaCO3 content was negatively correlated to the CEC, clay, sand, and
1∑ n w28 contents (Table 3).
RMSE = (CECe − CECm )2 (4)
n i=1
3.2. Performance of CEC models
∑n 2
(CECe − CECm )
NSE = 1 − ∑i=1
n
(5) The relationships between measured CEC and estimated CEC using
i=1 (CECm − CEC)2
the different models are illustrated in Fig. 2 and the model performance
indicators are presented in Table 4. In Fig. 2a, M1 and M2 models, based
where CECm is the measured CEC, CECe is the estimated CEC, CEC is the
on clay, silt, sand, OC/OM, and pH, tended to overestimate CEC (RMSE
mean of the measured data, and n is the sample size.
of 16.9 and 16.3 cmol(+) kg− 1, respectively). Except for five samples that
were underestimated, the rest of the samples had significantly larger
2.5. Analyses of the effect of calcium carbonate equivalent on model estimated CEC values compared to the measured CEC. This is reflected in
performance the positive ME values of 0.34 and 0.31, respectively, coupled with the
negative NSE values (Table 4). The poor performance of M1 and M2
We hypothesized that the amount of CaCO3 influences the estimation could be due to (i) the small number of samples (65 for M1) that were

3
F. Razzaghi et al. Geoderma 400 (2021) 115221

Table 3
Correlation coefficients for the considered soil properties.
CEC Clay Silt Sand OC CaCO3 w28

CEC 1
Clay 0.64*** 1
Silt 0.23 0.08 1
Sand − 0.62*** − 0.81*** − 0.66*** 1
OC 0.01 − 0.02 0.32 − 0.17 1
CaCO3 − 0.75*** − 0.55*** − 0.17 0.52*** − 0.12 1
w28 0.94*** 0.72*** 0.16 − 0.64*** − 0.12 − 0.75*** 1
***
p < 0.001; ** p < 0.01; * p < 0.05. CEC, cation exchange capacity; OC, soil organic carbon; CaCO3, calcium carbonate equivalent; w28, soil water content measured
at 28% relative humidity.

Fig. 2. Scatterplots of the measured versus estimated CEC for the evaluated models (M1 to M7). Acronyms in plots indicate the variables in the models (Cl, Si, Sa, OM
and OC indicate clay, silt, sand, organic matter and organic carbon contents, respectively, w28 indicates the water content at 28% relative humidity).

used for developing the models, (ii) the limited number of sites (4) used (M4, M5, and M6), the M5 model performed slightly better (RMSE =
for the M2 model, (iii) inclusion of sand content, which is generally 11.6 cmol(+) kg− 1 and NSE = 0.33) than the M4 and M6 models (NSE ≤
correlated to clay, and (iv) the potential influence of CaCO3 on the CEC. 0.25). Additionally, all three models tended to overestimate CEC for
In Fig. 2b, the M3 model performed creditably well (RMSE = 11.7 samples with low CEC and underestimated CEC for samples that had
cmol(+) kg− 1 and NSE = 0.31) compared to the M1 and M2 models. The CEC > 40 cmol(+) kg− 1 (Fig. 2c).
M3 model included CaCO3 as an input variable together with clay, sand, As previously explained, the clay content is a major contributor to
and OM contents. Since all the variables, except CaCO3, were included in the CEC of mineral soils and this is why almost all the considered models
the M1 and M2 models, it is reasonable to conclude that the better include it. However, in addition to clay content, clay mineralogy is
performance of the M3 model is primarily due to the presence of CaCO3 equally important in determining soil CEC. For a given clay content, soils
in the model. Despite the improved performance, there was a significant dominated by non-expandable 1:1 clay minerals like kaolinite have
underestimation of CEC for low-CEC samples (Fig. 2b). smaller CEC compared to soils with smectite or illite clay minerals
Among the models that had clay and OC contents as input variables (Parfitt et al., 1995). This could partly explain the observed estimation

4
F. Razzaghi et al. Geoderma 400 (2021) 115221

Table 4 limits (Cerato, 2001). Furthermore, Holford and Mattingly (1975)

Performance of models used to estimate the cation exchange capacity of the showed an inverse relationship between soil specific surface area and
investigated samples. CaCO3 for samples with greater than 12% CaCO3. This can be due to
Model ME* RMSE (cmol(+) kg− 1) NSE r2 obscuration of the available sorption sites on clay, silt, and OM fractions
M1 0.34 16.9 − 0.44 0.45
by carbonate crystals that have a correspondingly lower specific surface
M2 0.31 16.3 − 0.34 0.46 area (Arthur et al., 2017). This effect is likely already incorporated in the
M3 0.10 11.7 0.31 0.48 CEC, which is reflected in the amount of water sorbed. During the
M4 0.18 13.2 0.13 0.41 sorption process, water molecules cluster around exchangeable cations
M5 0.10 11.6 0.33 0.40
and form multilayers, and the cation charge, as well as size, plays a
M6 − 0.02 12.2 0.25 0.32
M7 0.10 6.2 0.81 0.88 significant role in the amount of water sorbed (Khorshidi and Lu, 2017;
*
Orchiston, 1959).
ME; mean error; RMSE: root mean square error; NSE, Nash-Sutcliffe effi­
ciency coefficient; r2: square of Pearson correlation coefficient.
3.3. Effect of calcium carbonate content on CEC estimation errors

errors for the models that utilize clay content and OC content.
The investigated samples had a wide range in both CEC and CaCO3
The model with the best performance was the M7 model that had
content (Table 1). This suggests that to identify any potential sources of
only w28 as the input variable (Fig. 2d). The model estimates of CEC
errors; it can be beneficial to scale the CEC estimation error with the CEC
closely matched the measured values (RMSE = 6.2 cmol(+) kg− 1 and
value, rather than the use of absolute errors. In Fig. 3, the relationship
NSE = 0.81). The amount of water sorbed by soil depends on the
between the scaled percentages of the CEC estimation errors versus the
available sorption sites that in turn depend on the total specific surface
CaCO3 contents for the evaluated models is depicted.
area. In the majority of cases, this total surface correlates well with the
The M1 and M2 models showed a consistent percentage increase in
CEC (Yukselen and Kaya, 2006). The significantly better CEC estimation
estimation errors as the CaCO3 content increased (Fig. 3a). The average
of the calcareous samples in this study is possibly because of the po­
of the absolute errors was similar (96 and 91%) between the two models
tential effect of CaCO3 on the amount and availability of sorption sites.
(M1 and M2). A similar but weaker trend of increasing errors with
Calcium carbonate in soils is reported to decrease the water content at
CaCO3 content was observed for the M3 model, the average absolute
–1.5 MPa (Khodaverdiloo et al., 2011), and decrease plastic and liquid
error (66%) was much smaller for M3 than for the two previous models

Fig. 3. Scatterplots of the cation exchange capacity (CEC) estimation error percentage versus calcium carbonate content for the evaluated CEC models (M1 to M7).
Acronyms in plots indicate the model variables (Cl, Si, Sa, OM and OC indicate clay, silt, sand, organic matter, and organic carbon contents, respectively, w28 in­
dicates the water content at 28% relative humidity). Numbers in the legend are the average of the absolute values of the estimation error for the models.

5
F. Razzaghi et al. Geoderma 400 (2021) 115221

(Fig. 3b). It can also be seen that the average error of M3 was strongly
affected by one sample with a percentage error greater than 200%. That
sample had CEC of 4 cmol(+) kg− 1 and CaCO3 content of 80%. When that
sample was excluded, the average error dropped to 56%. Similar to
models M1 and M2, the M4, M5, and M6 models also showed a strong
increase in percentage errors as CaCO3 increased (Fig. 3c). These
observed estimation errors can be due to complications arising from the
removal of CaCO3 before the determination of soil particle size distri­
bution. For calcareous soils, carbonates are often removed before soil
texture measurements (Francis and Aguilar, 1995), but for samples with
very large contents of CaCO3, this can be problematic as the carbonates
may exist as precipitated carbonate grains that should be expressed in
the textural analyses. For the samples used here, the CaCO3 content was
determined independently of the soil texture. Contrary, to the previous
six models, the M7 model did not show a trend of increasing errors
associated with the CaCO3 content (Fig. 3d). Apart from the one sample
that had the estimated CEC larger than twice the measured one (>100%
scaled error), the rest of the samples were distributed evenly around zero
in Fig. 3d. The one sample is the same one reported earlier in Fig. 3b for
the M3 model.
Fig. 4. Scatterplot of the measured cation exchange capacity (CEC) and the
In Table 3, there was a significant correlation between CEC and
estimated CEC based on clay (Cl) and calcium carbonate (CaCO3) contents of
CaCO3 (r = − 0.75***). However, this correlation is influenced by the
the investigated samples. RMSE, root mean square error; NSE, Nash-Sutcliff
impact of clay, silt, or OC on the magnitude of CEC. Since there was no efficiency coefficient.
significant effect of OC and silt on CEC (Table 3) for the investigated
samples, that effect of both variables was discounted. Therefore, to
by a linear relationship between clay content, CaCO3 and CEC.
clarify further the effect of CaCO3 content on the magnitude of CEC, it is
necessary to take out the effect of clay content on CEC. Subsequently, a
partial correlation analysis was performed between CEC and CaCO3 Declaration of Competing Interest
content, where the effect of clay and silt content on CEC was removed.
The analyses showed a significant negative relationship between CEC The authors declare that they have no known competing financial
and CaCO3 content (partial correlation coefficient = − 0.62***). This interests or personal relationships that could have appeared to influence
confirms that the amount of CaCO3 reduces the expected CEC for soil the work reported in this paper.
with a given amount of clay. This explains the large errors observed for
the models that did not incorporate CaCO3 as an input variable (Fig. 3b). Acknowledgements
To confirm further the effect of CaCO3, the data for the investigated
samples were used in a stepwise regression involving Cl, Si¸ Sa, and This research was funded partially by a research grant (13162) from
CaCO3 to predict CEC. The two variables that could estimate CEC were VILLUM FONDEN. Authors also thank Shiraz University for financial
the Cl and CaCO3 contents (Equation (6)): support and providing required laboratory facilities. Special thanks are
due to Dr. Siroos Shakeri and Engr. Hasan Mozaffari for providing soil
CEC = 29.9 + 0.298Cl − 0.389CaCO3
(6) samples and assisting with laboratory work, respectively. Furthermore,
p < 0.001; adj. R2 = 0.62; SEE = 8.84 we are grateful to Dr. Cristine Morgan, formerly of Texas A&M Uni­
versity, and The International Soil Reference and Information Centre
where SEE is the standard error of the estimate (cmol(+) kg− 1).
(ISRIC) for providing soil samples for the research.
Equation (6) was cross-validated with the same samples and shown
in Fig. 4. There was no consistent under or overestimation of CEC after
Appendix A. Supplementary data
cross-validation, and the estimation performance was better than
models M1 to M6 but lower than model M7. This suggests that there are
Supplementary data to this article can be found online at https://doi.
other interacting factors in play that are not fully captured by a linear
org/10.1016/j.geoderma.2021.115221.
relationship between clay content, CaCO3, and CEC.

References
4. Conclusion
Arthur, E., 2017. Rapid estimation of cation exchange capacity from soil water content.
The study evaluated the ability of seven published models to esti­ Eur. J. Soil Sci. 68 (3), 365–373.
mate the cation exchange capacity (CEC) of 38 calcium carbonate-rich Arthur, E., Razzaghi, F., Moosavi, A.A., 2017. Validation of water sorption-based clay
prediction models for calcareous soils. J. Plant Nutr. Soil Sci. 180 (3), 347–354.
soil samples gathered from different locations in the world. Five Arthur, E., Tuller, M., Moldrup, P., de Jonge, L.W., 2014. Evaluation of a fully automated
models that used various combinations of the particle size distribution analyzer for rapid measurement of water vapor sorption isotherms for applications in
(clay, silt, sand), soil organic matter, and pH, but did not include CaCO3 soil science. Soil Sci. Soc. Am. J. 78 (3), 754–760.
Arthur, E., Tuller, M., Moldrup, P., de Jonge, L.W., 2020. Clay content and mineralogy,
content, revealed systematic estimation errors for samples with CEC
organic carbon and cation exchange capacity affect water vapour sorption hysteresis
above or below 40 cmol(+) kg− 1. These errors correlated to the calcium of soil. Eur. J. Soil Sci. 71 (2), 204–214.
carbonate percentage. The presence of CaCO3 in one of the models Asadzadeh, F., Maleki-Kakelar, M., Shabani, F., 2019. Predicting cationic exchange
capacity in calcareous soils of East Azerbaijan province, northwest Iran. Commun.
tended to improve the model accuracy, but the model that yielded the
Soil Sci. Plan 50 (9), 1106–1116.
best estimates of CEC was the one that had only hygroscopic water Bell, M.A., van Keulen, H., 1995. Soil pedotransfer functions for 4 Mexican soils. Soil Sci.
content as an input. Further, a multiple linear regression based on clay Soc. Am. J. 59 (3), 865–871.
content and CaCO3 for the investigated samples showed that the two Bower, C.A., Reitemeier, R.F., Fireman, M., 1952. Exchangeable cation analysis of saline
and alkali soils. Soil Sci. 73 (4), 251–261.
variables could explain only 62% of the variation in CEC. This suggests Cerato, A.B., 2001. Influence of specific surface area on geotechnical characteristics of
that there are other interacting factors in play that are not captured fully fine-grained soils, MSc Dissertation. University of Massachusetts, Department of
Civil and Environmental Engineering, Amherst, MA, 315 p.

6
F. Razzaghi et al. Geoderma 400 (2021) 115221

Dohrmann, R., 2006. Cation exchange capacity methodology I: An efficient model for the Nevins, M.J., Weintritt, D.J., 1967. Determination of cation exchange capacity by
detection of incorrect cation exchange capacity and exchangeable cation results. methylene blue adsorption. Am. Ceram. Soc. Bull. 46 (6), 587-+.
Appl. Clay Sci. 34 (1–4), 31–37. Orchiston, H.D., 1959. Adsorption of water vapor: V. Homoionic illites at 25[degrees]C.
Francis, R.E., Aguilar, R., 1995. Calcium-carbonate effects on soil textural class in Soil Sci. 87 (5), 276–282.
semiarid wildland soils. Arid Soil Res. Rehab. 9 (2), 155–165. Parfitt, R.L., Giltrap, D.J., Whitton, J.S., 1995. Contribution of organic matter and clay
Gee, G.W., Or, D., 2002. Particle-size analysis. In: J.H. Dane, G.C. Topp (Eds.), Methods minerals to the cation exchange capacity of Soils. Commun. Soil Sci. Plan. 26 (9–10),
of soil analysis. Part 4. SSSA Book Series No. 5. SSSA, Madison, WI, pp. 255-293. 1343–1355.
Ghorbani, H., Kashi, H., Moghadas, N.H., Emamgholizadeh, S., 2015. Estimation of soil Pelagio, M.C., Navarro, D.A., Janik, L.J., Lamorena, R.B., 2020. Potential application of
cation exchange capacity using multiple regression, artificial neural networks, and laser-induced breakdown spectroscopy (LIBS) data for the determination of cation
adaptive neuro-fuzzy inference system models in Golestan province. Iran. Commun. exchange capacity (CEC) of Agricultural Soils. Chemistryselect 5 (13), 3798–3804.
Soil Sci. Plan. 46 (6), 763–780. Pribyl, D.W., 2010. A critical review of the conventional SOC to SOM conversion factor.
Holford, I.C.R., Mattingly, G.E.G., 1975. Surface areas of calcium carbonate in soils. Geoderma 156 (3), 75–83.
Geoderma 13 (3), 247–255. R Core Team, 2020. R: A language and environment for statistical computing. R
Kashi, H., Emamgholizadeh, S., Ghorbani, H., 2014. Estimation of soil infiltration and Foundation for Statistical Computing, Vienna, Austria http://www.R-project.org/.
cation exchange capacity based on multiple regression, ANN (RBF, MLP), and ANFIS Ramos, F.T., Dores, E.F.G.D., Weber, O.L.D., Beber, D.C., Campelo, J.H., Maia, J.C.D.,
models. Commun. Soil Sci. Plan. 45 (9), 1195–1213. 2018. Soil organic matter doubles the cation exchange capacity of tropical soil under
Khaledian, Y., Brevik, E.C., Pereira, P., Cerdà, A., Fattah, M.A., Tazikeh, H., 2017. no-till farming in Brazil. J. Sci. Food Agr. 98 (9), 3595–3602.
Modeling soil cation exchange capacity in multiple countries. Catena 158, 194–200. Rehman, H.U., Knadel, M., de Jonge, L.W., Moldrup, P., Greve, M.H., Arthur, E., 2019.
Khodaverdiloo, H., Homaee, M., van Genuchten, M.T., Dashtaki, S.G., 2011. Deriving Comparison of cation exchange capacity estimated from Vis-NIR spectral reflectance
and validating pedotransfer functions for some calcareous soils. J. Hydrol. 399 (1–2), data and a pedotransfer function. Vadose Zone J. 18 (1).
93–99. Schollenberger, C.J., Dreibelbis, F.R., 1930. Analytical methods in base exchange
Khodaverdiloo, H., Momtaz, H., Liao, K., 2018. Performance of soil cation exchange investigations on soils. Soil Sci. 30 (3), 161–173.
capacity pedotransfer function as affected by the inputs and database size. Clean-Soil Shuman, L.M., Duncan, R.R., 1990. Soil exchangeable cations and aluminum measured
Air Water 46 (3). by ammonium chloride, potassium chloride, and ammonium acetate. Commun. Soil
Khorshidi, M., Lu, N., 2017. Determination of cation exchange capacity from soil water Sci. Plan. 21 (13–16), 1217–1228.
retention curve. J. Eng. Mech. 143 (6). Silva, S., Baffi, C., Spalla, S., Cassinari, C., Lodigiani, P., 2010. Method for the
Krogh, L., Breuning-madsen, H., Greve, M.H., 2000. Cation exchange capacity determination of CEC and exchangeable bases in calcareous soils. Agrochimica 54
pedotransfer functions for Danish soils. Acta Agriculturae Scandinavica, Section B — (2), 103–114.
Soil & Plant. Science 50 (1), 1–12. Sumner, M.E., Miller, W.P., 1996. Cation exchange capacity and exchange coefficients.
Laird, D.A., 1999. Layer charge influences on the hydration of expandable 2: 1 In: Sparks, D.L., Page, A.L., Helmke, P.A., Loeppert, R.H. (Eds.), Methods of Soil
phyllosilicates. Clay Clay Miner. 47 (5), 630–636. Analysis Part 3—Chemical Methods. Soil Science Society of America, American
Loeppert, R.H., Suarez, D.L., 1996. Carbonate and Gypsum. In: Sparks, D.L., Page, A.L., Society of Agronomy. SSSA Book Series, Madison, WI, pp. 1201–1229.
Helmke, P.A., Loeppert, R.H. (Eds.), Methods of Soil Analysis Part 3—Chemical Taalab, A., Ageeb, G.W., Siam, H., Mahmoud, S., 2019. Some characteristics of
Methods. SSSA Book Series. Soil Science Society of America, American Society of calcareous soils: A review. Middle East J. Agric. 8, 96–105.
Agronomy, Madison, WI, pp. 437–474. Tadry L.K., 1973. Distribution of calcareous soils in the Near East region, their
Manrique, L.A., Jones, C.A., Dyke, P.T., 1991. Predicting cation exchange capacity from reclamation and land use measures and achievements. In Calcareous Soils. FAO Soils
soil physical and chemical properties. Soil Sci. Soc. Am. J. 55 (3), 787–794. Bulletin #21. Rome.
McBratney, A.B., Minasny, B., Cattle, S.R., Vervoort, R.W., 2002. From pedotransfer Tan, K.H., Dowling, P.S., 1984. Effect of organic matter on CEC due to permanent and
functions to soil inference systems. Geoderma 109 (1–2), 41–73. variable charges in selected temperate region soils. Geoderma 32 (2), 89–101.
Mehlich, A., 1939. Use of triethanolamine acetate barium hydroxide buffer for the Woodruff, W.F., Revil, A., 2011. CEC-normalized clay-water sorption isotherm. Water
determination of some base exchange properties and lime requirement of soil. Soil Resour. Res. 47.
Sci. Soc. Am. J. 3 (C), 162–166. WRB, I.W.G., 2015. World Reference Base for Soil Resources 2014, update 2015.
Minasny, B., Hopmans, J.W., Harter, T., Eching, S.O., Tuli, A., Denton, M.A., 2004. International soil classification system for naming soils and creating legends for soil
Neural networks prediction of soil hydraulic functions for alluvial soils using maps. World Soil Resources Reports No. 106. FAO, Rome.
multistep outflow data. Soil Sci. Soc. Am. J. 68 (2), 417–429. Yilmaz, I., 2006. Indirect estimation of the swelling percent and a new classification of
Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models part I soils depending on liquid limit and cation exchange capacity. Eng. Geol. 85 (3–4),
— A discussion of principles. J. Hydrol. 10 (3), 282–290. 295–301.
Nelson, D.W., Sommers, L.E., 1996. Total carbon, organic carbon, and organic matter. In: Yukselen, Y., Kaya, A., 2006. Prediction of cation exchange capacity from soil index
Sparks, D.L., Page, A.L., Helmke, P.A., Loeppert, R.H. (Eds.), Methods of Soil properties. Clay Miner. 41 (4), 827–837.
Analysis Part 3—Chemical Methods. SSSA Book Series. Soil Science Society of
America, American Society of Agronomy, Madison, WI, pp. 961–1010.