Catena 170 (2018) 25–35

Contents lists available at ScienceDirect

Catena
journal homepage: www.elsevier.com/locate/catena

Spatial analysis of infiltration in agricultural lands in arid areas of Iran T

⁎
Fariba Babaei, Ali Asghar Zolfaghari , Mohammad Reza Yazdani, Ahmad Sadeghipour
Faculty of Desert science, Semnan University, Semnan, Iran

A R T I C LE I N FO A B S T R A C T

Keywords: Infiltration plays an important role in the hydrologic cycle, runoff generation, soil erosion, as well as irrigation.
Agricultural lands In the current study, we evaluated a variety of infiltration models, in order to determine the model that is best
Infiltration models suited to predict infiltration on agricultural lands in arid areas of the Semnan province in Iran. Additionally, we
Scaling factors analyzed spatial variability of the infiltration process using scaling parameters. A number of 60 points were
Scaled infiltration
determined for the measurement of water infiltration using the conditioned Latin hypercube sampling (cLHS)
method. Each infiltration measurement was carried out with a tension infiltrometer apparatus. Several in-
filtration models, including Philip, Horton, Kostiakov (KO), Modified Kostiakov (MKO) and Revised Modified
Kostiakov (RMKO) models, were fitted to the measured infiltration data. Among the mean coefficient of de-
termination (R2) values, the highest R2 values were associated with the RMKO, MKO and Horton models, re-
spectively. While base on the Akaike criterion (ACI), the MKO model was slightly better than the RMKO model
for prediction of cumulative infiltration. Cumulative infiltration was scaled using Sorptivity (αS) and trans-
missivity (αA) scaling factors. By minimizing the differences of the sum of squares between the scaled and the
average infiltration, the optimum scaling factors (αopt) were estimated. Arithmetic, geometric and harmonic
means (i.e. αm, αG, αH; respectively) of αS and αA scaling factors were calculated and the infiltration data were
scaled utilizing the mentioned scaling factors. Our findings indicated that the best result was yielded by αH.
Strong correlations were found for αG (r = 0.86) and αH (r = 0.86) with αopt. For defining the relationship
between αopt and αA, αS, αm, αG and αH data, a regression analysis was performed. According to our results,
curves reflecting the relationship between αopt and αA, αS, αm, αG and αH were sigmoid. Based on the results of
this study, infiltration in agricultural lands in the arid area displays a great spatial variability.

1. Introduction
It has been widely accepted that the hydrologic cycle, runoff gen-
eration, soil erosion and irrigation are effected by infiltration (Mirzaee
et al., 2014; Dafny and Šimunek, 2016). Precipitation, irrigation, or a
contaminated spill water, which enters the soil or creates runoff is
determined by infiltration (Radcliffe and Šimunek, 2012; Ghorbani
Dashtaki et al., 2009). It also plays a key role in controlling crop yield
for designing irrigation systems, increasing the efficiency of water and
solute transport in the soil profile and reducing water losses (Hillel,
1998). The amount of infiltrated water into the soil is one of the main
parameters for water resources management. The groundwater system
sustainability is also dependent on the amount of recharge by in-
filtrating rainfall (Chen et al., 2005). So far, a large number of modeling
approaches for infiltration have been developed by numerous soil and
water scientists (e.g., Green and Ampt, 1911; Kostiakov, 1932; Philip,
1957; Mein and Larson, 1973; Argyrokastritis and Kerkides, 2003;
Mirzaee et al., 2014). However, only a few of the proposed infiltration
models have been successfully applied using field data. Parameter es-
timation is the principle criterion for preferring one model over an-
other.
The spatial variability of the soil is considered the main problem for
modeling infiltration at the watershed scale (Machiwal et al., 2006).
Spatial variation of infiltration causes complexity to the management of
irrigation at a watershed scale, as well as drainage. Furthermore, in the
agricultural lands, a large spatial variability of infiltration is caused by
farming practice. Hence, providing an applicable general infiltration
equation for a wide range of soil conditions is pivotal. The classification
of soil mapping units and the variation coefficient are among the
available approaches in determining the spatial soil variability (Warrick
and Nielsen, 1980; Jury, 1986; Jury et al., 1987; Beven et al., 1993;
Mulla and Mcbratney, 2002). However, considerable variation in soil
hydraulic properties have been reported, even in the same soil mapping
unit (coefficients of variation greater than 100%) (Nielsen et al., 1973;
Comegna and Vitale, 1993). Several geostatistical techniques can be
utilized for the estimation of the hydraulic properties of the spatial soil

⁎
Corresponding author.
E-mail address: azolfaghari@semnan.ac.ir (A.A. Zolfaghari).

https://doi.org/10.1016/j.catena.2018.05.039
Received 18 July 2017; Received in revised form 26 March 2018; Accepted 29 May 2018
Available online 06 June 2018
0341-8162/ © 2018 Elsevier B.V. All rights reserved.
F. Babaei et al.
variability. However, the main limitation in utilizing geostatistical
techniques for its determination is its requirement to numerous mea-
surement points. Scaling is an alternative way for describing the general
infiltration equation. Multifractal analysis has been proposed, as the
scaling approach for determining the hydraulic properties of the spatial
soil variability (Pahlevan et al., 2016). Furthermore, scaling has been
applied as a simple method for describing the field spatial variability
(Pachepsky et al., 2003; Nielsen et al., 1998; Warrick et al., 1977),
alongside with infiltration, drainage rate and available water (Sharma
et al., 1980; Rasoulzadeh and Sepaskhah, 2003; Ahuja et al., 2007). By
measuring a representative point at different locations in a watershed,
soil hydraulic properties can be estimated by applying scaling approach
(Williams and Ahuja, 1991; Kozak and Ahuja, 2005). By using scaling
method, Philip's quasi-analytical solution was developed by Warrick
et al. (1985) for one dimensional infiltration. Rasoulzadeh and
Sepaskhah (2003) scaled infiltration using Buckingham's theorem for
the dimensional analysis. Warrick and Hussen (1993) scaled Richards'
equation in similar soils. More recently Sadeghi et al. (2011, 2012) have
developed an additional scale solution for Richards' equation. Despite
their dissimilarity, their advantage over Warrick–Hussen method was
that they produced invariant solutions for a wider range of soils.
Spatial variability of soil properties in agricultural land could be
induced by farmer's practice and/or result from edaphic factors, such as
the parent material and soil position in the catena. Therefore, spatial
variability of soil properties in agriculture lands may be greater for the
undisturbed soil. Limited number of studies have characterized the
spatial variability of soil hydraulic properties in arid agricultural lands.
However, to the best of our knowledge, no study has been carried out
for describing the spatial variability of the soil water infiltration in arid
areas of Iran. The main objective of this study is to evaluate a variety of
infiltration models and to determine the best model for agricultural
lands in arid areas of Iran. In addition, we aim to use the scaling ap-
proach to analyze the spatial variability of infiltration process.
2. Materials and methods
2.1. Study area and soil sampling
The study was carried out in the agriculture land in Semnan region
in the Semnan province located in the central region of Iran, between
latitudes of 35° 30′ to 35° 55′ N and longitudes of 53° 21′ to 53° 34′ E
(Fig. 1). The total area was approximately 12,000 ha, in which 90% is
earmarked for the production of wheat, barley and corn.
The mean annual temperature in the study area is 18.5 °C the
Fig. 1. Position of the study area and distribution of sampling points.
26
Catena 170 (2018) 25–35
precipitation is 138 mm, respectively, (Fig. 2).
Selection of sampling points were based on the conditioned Latin
hypercube sampling (cLHS) method (Minasny and Mcbratney, 2006).
Sampling of variables from their distributions can be effectively per-
formed by the stratified-random procedure in cLHS. This sampling
method has proven to be impractical in locations with accessibility
limitations (Roudier et al., 2012; Jeong et al., 2017). However, our
study area was flat and easily accessible. The prescribed distribution of
variables or auxiliary variables is utilized for determination of the
sampling points. In this study, the Landsat 8 satellite image acquired on
September 2015, freely available from the NASA server, was used as the
auxiliary variable. Auxiliary variables were consisted of four visible
bands (bands 1, 2, 3 and 4), NIR (band 5), SWIR (bands 6 and 7). A
number of 60 points were determined for the measurement of water
infiltration. Each infiltration measurement was triplicated, using a
tension infiltrometer apparatus. For determination of the physical and
chemical properties in each point, disturbed and undisturbed soil
samples were collected from at a depth of 0 to 15 cm. The disturbed soil
samples were air dried and passed through a 2 mm sieve. Soil texture
was determined by hydrometer (Gee and Bauder, 1986). Later on, soil
pH and electrical conductivity (EC at 25 °C) were determined in the
extract paste. Finally, soil organic carbon (OC) content was measured
by the Walkley–Black method (Walkley and Black, 1934).
2.2. Infiltration measurement
Infiltration data were obtained using a tension infiltrometer with a
20 cm diameter disk (Soil Measurement Systems LLC, Tucson, Arizona).
Water infiltration into the initially dry soil was measured at the inlet
(upper boundary) with a constant pressure head of 5 cm for all loca-
tions. In order to assure a proper contact between disk and soil, moist
fine sand (0.10–0.25 mm) was gently placed on the soil surface. Then,
the disk was placed on the soil surface and cumulative water infiltration
was monitored versus time. For each sample, the infiltration experi-
ments were continued until the infiltration rate reached a constant
value. For measuring bulk density (BD), an undisturbed soil sample was
collected from the surface (near to the disk place), using a core sampler
with 5 cm diameter and 5.1 cm of height (i.e. 100 cm3).
2.3. Estimation of infiltration model parameters
In this study five infiltration models including Philip, Horton,
Kostiakov, Modified Kostiakov and Revised Modified Kostiakov models
were selected. For all the studied models, a non-linear regression
F. Babaei et al. Catena 170 (2018) 25–35

Fig. 2. Average monthly precipitation and temperature (from 1993 to 2015).

procedure was used to fit the experimental infiltration data (Marquardt, similar to the α1 and β1 in Kostiakov model.
1963). The best fitting parameters were determined by minimizing the
sum of square error (SSE) between the model and the measured in- 2.3.5. Revised modified Kostiakov model (RMKO)
filtration data. The fitting process was performed using the MATLAB The Modified Kostiakov model was revised by Parhi et al. (2007)
Software Package (The MathWorks, Inc., 2007). The infiltration models and the new four parameters model was introduced as:
used in this study is briefly described below.
α4 β3+ 1 α5 1 − β4
I= t + t
β3 + 1 1 − β4 (5)
2.3.1. Philip model
The non-linear partial differential Richards equation, describing where α4, β3, α5 and β4 are parameters to be defined empirically, using
transient fluid flow in a porous medium (Richards, 1931) was solved by measured infiltration data.
Philip (1957). For cumulative infiltration, the Philip model is expressed
as follow: 2.4. Selection of the best fitting infiltration model
I= St 0.5 + At (1)
To classify the studied infiltration models, four statistic values were
where S is the sorptivity (LT-1/2) and A refers to a parameter with di- calculated using the following equations. Models with the best good-
mensions of the saturated hydraulic conductivity (LT−1) which is re- ness-of-fit were selected as the superior models.
lated to the soil properties and the initial and the saturated water Root Mean Square Error (RMSE):
contents.
N
∑i = 1 (Ipi − I 0i )2
RMSE =
2.3.2. Horton model N (6)
A semi-empirical infiltration model was proposed by Horton (1940) 2
as: Coefficient of determination (R ):
2
I = ct + m (1 − e−at ) (2) ⎧ N (∑ Io Ip) − (∑ Io )(∑ Ip ) ⎫
R2 =
where parameters c, m (LT−1), and a (T−1) are fitting parameters. ⎨ [N ∑ I 2 − (∑ Io )2] [N ∑ I 2 − (∑ Ip )2] ⎬
o p (7)
⎩ ⎭

2.3.3. Kostiakov model (KO) Normalized RMSE (NRMSE):

Based on the curve fitting from field data, Kostiakov (1932) pre-
sented a simple empirical infiltration equation:
I = α1 t β1 (3)
where α1 and β1 are fitting parameters and can be estimated using the
observed infiltration data. The main drawback of the Kostiakov model
is the lack of long term description.
2.3.4. Modified Kostiakov model (MKO)
MKO model is a three parameters model for describing cumulative
infiltration. Smith (1972) inserted the term of ultimate infiltration ca-
pacity (α3) to the original Kostiakov model, as follow:
I = α2 t β2 + α3 t (4)
−1
where α3 refers to the final infiltration rate (L T ) and α2 and β2 are
RMSE
N RMSE = − × 100
IO (8)
where N, Io and Ip are number of observed data points, observed and
−
predicted data values, respectively. IO is average observed cumulative
infiltration.
Akaike information criterion (AIC):
In this study, five infiltration models with a different number of
parameters were used to describe infiltration. Generally, increasing the
number of parameters improves the model performance. Therefore,
there is a trade-off between goodness-of-fit and model complexity (i.e.
number of parameters). An additional criteria for model comparison
with different fitting parameters is necessary. In this study, Akaike in-
formation criterion (AIC) (Carrera and Neuman, 1986) was used to
evaluate the performance of infiltration models. The optimal model

27
F. Babaei et al. Catena 170 (2018) 25–35

− n
generally had the smallest AIC criterion value which is calculated as 1
follows:
S =
n
∑ Sj
j=1 (15)
AIC = N {ln(2π ) + ln[SSE /(N − p)] + 1} + p (9) and
N n
− 1
where p is the number of model parameters and SSE is ∑ (Ipi − I 0i )2 . A = ∑ Aj
i=1 n (16)
In this study, the Philip model with two fitting parameters was used j=1

as the reference model to be compared with the rest of the models. For Scaling of the measured cumulative infiltration I(t) can be per-
any given soil, the model with 5% smaller AIC criterion value than the formed by either αS or αA as below (Sharma et al., 1980):
reference model, was considered as a better model (Mirzaee et al.,
I∗ = αI (17)
2014).
and
2.5. A review of the scaling theory t∗ = α3t (18)
⁎ ⁎
where t and I are scaled time and cumulative infiltration, respectively.
Various studies have characterized the spatial variability of soil If two-term Philip model (Eq. 1) can sufficiently describe the in-
hydraulic properties utilizing the scaling approach (Nielsen et al., 1998; filtration with the valid similar media conditions, the scaled infiltration
Sposito, 1998; Warrick, 1990, 1998; Machiwal et al., 2006; Sadeghi should be defined as follow:
et al., 2012). The main objective of scaling is to determine the re-
− −
lationship between soil hydraulic properties at a given location with the I ∗ = S t ∗0.5 + A t ∗ (19)
reference soil by using scaling factors. The theory of microscopic geo-
Comparable sum of squares error (SSE) of the original and the
metry was applied by Miller and Miller (1956) to develop the concept of
scaled data confirms the equality between sets of scaling factors (αS,
scaling approach. The parameters of the two-term Philip's model can be
αA). Scaling factors were also calculated by statistical optimization. The
scaled as follow:
sum of squares of the differences (D) between the scaled infiltration (I*)
− − −
Sj Sr 0.5 and the average infiltration (I based on S and A ) was minimized to
λj0.5 = λr (10) calculate the optimum scaling factor αopt for each test. Subsequently, D
is given by
and
−
Aj D= ∑ (It∗∗j − It j )2
λj2 = Ar λr2 ∗
(11) tj (20)

where λ is the characteristic length of a soil at a specific site, subscript j The above summation is obtained for all values of tj for the whole
refers to the test site (i.e. 1–60) and λr is defined as the characteristic period of the infiltration test.
length of the reference soil with the average hydraulic properties.
The λj of Eqs. (11) and (12) have identical values in the case of the 3. Results and discussion
similar media conditions (Machiwal et al., 2006).
Dimensionless scaling factors can be defined as: 3.1. Statistics of soil properties

−
αj = λj λr = λj λ (12)
A statistical summary of the soil properties is presented in Table 1.
Results indicated that the data set covered a wide range of soil factors.
−
where λ is the length characteristics of a reference soil and α is scaling The values of the sand fraction varied from 45.1 to 83.2%, while clay
factor. values varied from 8. 1 to 13.8%. Meanwhile, the values of silt content
The dimensionless scaling factors for S and A, αS and αA, can be varied from 4.2 to 40.1%. According to results, the minimum and the
computed by combining Eqs. (10), (11) and (12) as follows maximum soil organic carbon were 0.82 and 2.26%, respectively. Soil
EC varied widely from 0.99 to 43 dSm−1, with an average of
2
7.85 dSm−1 in the studied region. Results indicated that EC, sand, silt
−
α sj = Sj S ( ) (13) and OC content had frequency distributions with positive skewness.
While, pH, BD, clay and saturated moisture were negatively skewed.
and
The results indicate that the highest and the lowest spatial variation
1/2 related to the EC and the pH, respectively.
−
α Aj = Aj A ( ) (14) In this study, the Kolmogorov-Smirnov test was applied for testing
− −
the distribution of the measured data. The best-fit probability dis-
where average sorptivity, S , and transmissivity, A factors can be cal- tribution for each attribute was chosen by the lowest value of the
culated as following equations: computed Kolmogorov statistic in each row (the bold values in Table 2).

Table 1
Summary statistics of the studied soil properties at the measurement points.
Parameters Mean Max Min std cv% Kurtosis Skewness

−1
EC (dSm ) 7.8516 43.85 0.996 7.9312 1.0101 7.8514 2.569
pH 7.8117 8.7 6.8 0.35324 0.04522 0.62878 −0.38379
BD (gcm−3) 1.6318 1.87 1.31 0.12535 0.07682 0.11856 −0.16146
Clay (%) 21.995 31.84 8 6.6313 0.3015 −1.0318 −0.29524
Sand (%) 59.024 83.24 45.16 9.6553 0.16358 −0.56891 0.62825
Silt (%) 18.981 40 4 7.0953 0.37381 0.46117 0.19953
OC (%) 0.8268 2.262 0.117 0.50605 0.61206 0.69036 0.93789
θs (%) 41.146 61.17 24.93 8.0175 0.19486 −0.0616 −0.01399

28
F. Babaei et al. Catena 170 (2018) 25–35

Table 2
Results of Kolmogorov-Smirnov test for the four selected probability distributions.

Parameters Normal Lognormal Lognormal (3P) Log-Pearson 3

Statistic Rank Statistic Rank Statistic Rank Statistic Rank Critical
value at 5%
significant
level
EC 0.215 4 0.055 1 0.074 3 0.065 2 0.172

pH 0.110 2 0.117 4 0.109 1 0.116 3 0.172

BD 0.069 1 0.083 4 0.071 2 0.078 3 0.172

clay 0.126 2 0.135 4 0.132 3 0.110 1 0.172

sand 0.143 4 0.133 3 0.085 1 0.114 2 0.172

silt 0.087 1 0.172 4 0.097 2 0.102 3 0.172

OC 0.134 4 0.095 3 0.082 2 0.071 1 0.172

θs 0.109 1 0.143 4 0.121 3 0.110 2 0.172

Note: Bold values indicate the best-fit distribution for each variable.

Results indicated that the normal probability distribution could pre-
cisely describe the bulk density (BD), silt as well as saturated water
content (θs) while the best probability distribution for describing the EC
and the content of sand followed a log normal distribution. Further-
more, log-Pearson distribution was the best-fitted distribution for clay
and OC content. This finding was in agreement with Pahlevan et al.
(2016). They proved that EC, sand, θs and BD on a transect obey log
normal distribution.
3.2. Estimation of the infiltration model parameters
without any constrains to the model parameters. Secondly, some con-
strains were applied for a more accurate parameter estimation; as an
example, after approaching the A value equal to zero, the best para-
meters were determined by minimizing the sum of square error with
respect to these constrains. It should be noted that for the initial con-
dition, at 12 test sites, the A parameter of the Philip two-term model
was negative, which indicates negative value for the predicted in-
filtration rate. Similarly, Machiwal et al. (2006) reported that the Philip
A parameter at six out of 23 test sites was negative.

We fitted the measured infiltration data to the selected infiltration
models i.e. Philip, Horton, Kostiakov, Modified Kostiakov, and Revised
Modified Kostiakov, the results are presented in Fig. 3 as an example of
the fitting of the infiltration models. The parameters of the models were
determined using the least square technique. Based on results in
Table 3, a wide range of the parameters were obtained for the in-
filtration models, indicated considerable spatial variation of soil water
infiltration and soil hydraulic properties in the agricultural land. The
parameters of the models were estimated twice. Firstly, the best para-
meters were obtained by minimizing the sum of square error (SSE),
3.3. Selection of the best-fit infiltration model
Table 4 represents the coefficient of determination (R2) of 60 sites
and for five infiltration models. The R2 values range from 0.53 to 0.99
for the cumulative infiltration over all soils and all models. While for
the infiltration rate, R2 values ranged from 0.40 to 0.99. It is obvious
from Table 4 that in the majority of the cases (i.e. 48 out of 57) the
highest value of R2 was obtained for the RMKO model, which leads to
the conclusion that it is the best-fit infiltration model for agricultural
land in arid regions in Iran. The mean R2 values showed that the highest
R2 values were associated with the RMKO, MKO and Horton models,

Table 3
Statistics of optimized parameters of the infiltration models for all soils.
Cumulative infiltration Model Range Mean Std⁎
model parameter

Philip S 0–16.01 4.9486 3.1389

A 0–10.39 1.1264 1.6587
KO α1 0.1785–1.002 0.6427 0.1840
β1 0.4027–36.66 6.3518 5.0626
MKO α2 0.000006–6.403 1.1093 1.2141
β2 0–1.003 0.3447 0.2560
α3 0.3769–36.66 6.1308 5.0230
RMKO α4 0.00001–8.749 1.5953 1.8753
α5 −1–2.913 −0.0836 0.8136
β4 0–8.602 1.7530 2.0060
β3 −1–7.127 0.0613 1.5469
Horton C 0.1388–10.65 1.8287 1.7614
a 0.003588–195.5 10.4611 27.7960
m 0.9343–42.26 8.9925 7.2292
Fig. 3. Comparison between the measured and fitted infiltration curve for a
given soil (KO-Kostiakov Model, MKO-Modified Kostiakov model, RMKO- Std - standard deviation, KO-Kostiakov Model, MKO-Modified Kostiakov model,
Revised Modified Kostiakov model). RMKO-Revised Modified Kostiakov model.

29
F. Babaei et al. Catena 170 (2018) 25–35

Table 4
Coefficient of determination values for five infiltration models at soil sample points.
Test site Coefficient of determination

Philip Horton KO MKO RMKO

Infiltration rate Cumulative Infiltration rate Cumulative Infiltration rate Cumulative Infiltration rate Cumulative Infiltration rate Cumulative
infiltration infiltration infiltration infiltration infiltration

1 0.603 0.995 0.993 1.000 0.672 0.990 0.994 1.000 0.994 1.000
2 0.882 0.999 0.961 1.000 0.854 0.998 0.981 1.000 0.982 1.000
3 * 0.999 * 0.999 * 0.999 * 0.999 * 0.999
4 0.420 0.999 0.764 0.999 0.284 0.998 0.759 0.999 0.801 1.000
5 0.684 0.999 0.970 1.000 0.604 0.997 0.974 1.000 0.974 1.000
6 * 0.999 * 0.999 * 0.999 * 0.999 * 0.999
7 0.420 0.999 0.764 0.999 0.290 0.998 0.759 0.999 0.801 1.000
8 0.602 0.994 0.982 0.999 0.738 0.989 0.983 0.999 0.985 0.999
9 0.486 0.975 0.993 0.998 0.892 0.975 0.993 0.999 0.993 0.999
10 0.816 0.999 0.954 1.000 0.807 0.996 0.949 0.999 0.952 1.000
11 0.612 0.978 0.979 0.991 0.924 0.978 0.965 0.991 0.965 0.991
12 0.703 0.998 0.883 0.995 0.872 0.999 0.910 0.999 0.926 0.999
13 0.576 0.995 0.862 0.998 0.613 0.991 0.863 0.998 0.867 0.998
14 0.830 0.999 0.926 0.999 0.813 0.998 0.937 0.999 0.944 0.999
15 0.621 0.997 0.950 0.998 0.654 0.993 0.950 0.999 0.961 0.999
16 0.800 0.999 0.912 0.999 0.796 0.997 0.920 0.999 0.920 1.000
17 0.817 0.999 0.937 0.996 0.940 0.999 0.975 0.999 0.993 0.999
18 0.770 0.999 0.881 0.999 0.764 0.997 0.905 0.999 0.924 0.999
19 0.845 0.998 0.919 0.999 0.877 0.998 0.903 0.999 0.906 0.999
21 0.761 0.999 0.872 0.997 0.762 0.998 0.880 0.999 0.927 0.999
22 0.701 0.995 0.946 0.997 0.846 0.993 0.950 0.998 0.951 0.998
23 0.692 0.999 0.941 0.998 0.707 0.997 0.944 0.999 0.959 1.000
24 0.465 0.957 0.993 0.999 0.914 0.963 0.991 0.998 0.992 0.998
25 0.706 0.999 0.942 0.999 0.698 0.997 0.950 1.000 0.961 1.000
26 0.835 0.999 0.988 0.996 0.868 0.998 0.992 1.000 0.992 1.000
27 0.833 1.000 0.956 0.999 0.826 0.998 0.974 1.000 0.978 1.000
28 0.871 0.999 0.952 0.998 0.907 0.999 0.969 1.000 0.971 1.000
29 0.623 0.999 0.741 0.999 0.567 0.998 0.673 0.999 0.708 0.999
30 0.562 0.971 0.984 0.998 0.893 0.985 0.984 0.998 0.984 0.998
31 0.444 0.817 0.986 0.981 0.958 0.977 0.988 0.997 0.991 0.997
32 0.467 0.859 0.987 0.993 0.958 0.975 0.981 0.992 0.969 0.992
33 0.362 0.982 0.970 0.990 0.587 0.984 0.970 0.984 0.971 0.984
34 0.621 0.987 0.983 0.999 0.780 0.982 0.983 0.998 0.983 0.999
35 0.667 0.994 0.624 0.997 0.665 0.996 0.668 0.996 0.693 0.996
36 0.681 0.994 0.650 0.994 0.678 0.995 0.689 0.995 0.479 0.995
38 0.769 0.997 0.955 0.997 0.891 0.997 0.952 0.998 0.932 0.998
39 0.744 0.998 0.949 0.998 0.794 0.996 0.956 0.999 0.964 0.999
40 0.806 0.982 0.826 0.987 0.806 0.983 0.806 0.983 0.812 0.983
41 0.326 0.999 0.603 0.999 0.257 0.999 0.603 0.999 0.605 0.999
42 0.271 0.928 0.656 0.947 0.274 0.918 0.656 0.947 0.905 0.997
44 0.719 0.999 0.804 0.999 0.663 0.999 0.810 0.999 0.810 0.999
45 0.902 0.999 0.943 1.000 0.892 0.999 0.939 0.999 0.941 1.000
46 0.312 0.996 0.629 0.997 0.277 0.995 0.629 0.997 0.636 0.997
47 0.463 0.683 0.994 0.974 0.982 0.961 0.994 0.989 0.994 0.989
48 0.605 0.992 0.964 0.997 0.708 0.989 0.964 0.997 0.964 0.997
49 0.389 0.986 0.404 0.987 0.386 0.987 0.408 0.987 0.408 0.987
50 0.847 0.999 0.931 0.999 0.846 0.999 0.948 0.999 0.948 0.999
51 0.799 0.999 0.978 0.998 0.922 0.999 0.984 0.999 0.984 1.000
52 0.457 0.990 0.876 0.996 0.489 0.990 0.876 0.997 0.926 0.997
53 0.350 0.917 0.889 0.957 0.658 0.906 0.889 0.957 0.929 0.972
54 0.411 0.631 0.632 0.930 0.532 0.819 0.532 0.819 0.532 0.819
55 0.631 0.987 0.981 0.998 0.856 0.985 0.976 0.997 0.977 0.997
56 0.835 0.999 0.926 0.997 0.902 0.999 0.947 0.999 0.947 0.999
57 0.617 0.927 0.980 0.960 0.953 0.991 0.976 0.991 0.982 0.994
58 0.766 0.968 0.835 0.991 0.889 0.970 0.889 0.970 0.889 0.970
59 0.717 0.994 0.756 0.994 0.703 0.995 0.770 0.995 0.770 0.995
60 0.581 0.977 0.980 0.996 0.746 0.974 0.980 0.996 0.982 0.997
mean 0.632 0.966 0.881 0.991 0.734 0.986 0.883 0.992 0.893 0.994

Note: Bold values indicate the highest value of coefficient of determination at a site.
KO-Kostiakov Model, MKO-Modified Kostiakov model, RMKO-Revised Modified Kostiakov model.

respectively. Philip two parameters model yielded the lowest mean R2
value which is in agreement with Mirzaee et al. (2014). In contrast,
Machiwal et al. (2006), reported that Philip two parameters model was
the best model for describing cumulative infiltration.
Fig. 4 illustrates the box plot for R2 percentiles of the selected
models for all the studied soils. The range of R2 values was greater for
the models fitted on infiltration rate compared with models fitted on
cumulative infiltration data (Fig. 4). Over all the soils and models, R2
values were smaller for the models fitted on infiltration rate compared
to cumulative infiltration data (Table 4).
Fig. 5 illustrates the boxplot of variation of ACI criteria for selected
models. Mean AIC statistics of cumulative infiltration for MKO, RMKO,

30
F. Babaei et al. Catena 170 (2018) 25–35

Fig. 4. Boxplots of variations of the coefficient of determination (R2) for infiltration rate (a) and cumulative infiltration (b) (KO-Kostiakov Model, MKO-Modified
Kostiakov model, RMKO- Revised Modified Kostiakov model).

Fig. 5. Boxplots of variations of the Akaike criteria (AIC) for cumulative infiltration (a) and infiltration rate (b) (KO-Kostiakov Model, MKO-Modified Kostiakov
model, RMKO- Revised Modified Kostiakov model).

Fig. 6. Statistical analysis of infiltration models for all soils using Akaike cri-
Fig. 7. Relation between S and A parameter in study area.
terion (ACI) (KO-Kostiakov Model, MKO-Modified Kostiakov model, RMKO-
Revised Modified Kostiakov model).
their superiority over the Philip model (Fig. 6). Hence, the MKO model
Horton, Philip and KO model were 13.35, 13.46, 18.66, 25.48 and was slightly better than the RMKO model in prediction of cumulative
25.64, respectively. This result indicates that the cumulative infiltration infiltration.
could be estimated by the MKO, RMKO, Horton, Philip and KO models,
respectively (Fig. 5). Results showed that the AIC value for the RMKO, 3.4. Properties of the Philip's model parameters
MKO, Horton and KO models were significantly smaller than that of the
Philip model in 55.6, 56.4, 54 and 25% of soils, respectively, indicating The spatial variability was determined by Philip's infiltration model

31
F. Babaei et al. Catena 170 (2018) 25–35

Table 5
Results of Kolmogorov-Smirnov test for the four selected probability distributions.
Goodness of Kolmogorov fit

Parameter Normal Lognormal Lognormal (3P) Log-Pearson 3 Gamma

Statistic Rank Statistic Rank Statistic Rank Statistic Rank Statistic Rank Critical value at 5% significant level

S 0.109 3 0.357 4 0.065 1 0.453 5 0.088 2 0.172

K 0.249 2 0.375 5 0.374 4 0.273 3 0.199 1 0.172
αS 0.217 2 0.357 3 0.360 4 0.453 5 0.112 1 0.172
αK 0.102 1 0.375 5 0.346 4 0.273 3 0.200 2 0.172
αOPT 0.193 4 0.171 1 0.299 5 0.171 2 0.174 3 0.172
αM 0.158 5 0.065 3 0.056 1 0.061 2 0.099 4 0.172
αG 0.147 2 0.277 5 0.264 4 0.100 1 0.218 3 0.172
αH 0.153 2 0.314 5 0.310 4 0.086 1 0.238 3 0.172

Note: Bold values indicate the best- fit distribution for each parameter.

Fig. 8. Infiltration curves of 60 sites scaled by different scaling factors.

32
F. Babaei et al. Catena 170 (2018) 25–35

Table 6 parameters (i.e. S and A). The variation of the parameters S and A
The NRMSE results of fitting Eq. (18) on scaling infiltration. showed no definite pattern or significant relation (Fig. 7) which is in
Scaling factors NRMSE (%) R2 S⁎ A⁎ agreement with Talsma (1969) who reported a weak relationship be-
tween the saturated hydraulic conductivity and the sorptivity. By con-
αs 47 0.923 6.197 0 trast, Sharma et al. (1980), Machiwal et al. (2006) and Hunt et al.
αA 40 0.917 5.889 0.731
(2017) reported a high correlation between S and A.
αm 46 0.923 7.708 0
αG 29 0.964 8.274 0.217
The results of the Kolmogorov-Smirnov test are presented in
αH 24 0.974 7.306 0.5103 Table 5. It is obvious that sorptivity, S, could be perfectly described by
αopt 11 0.987 7.991 0.698 the Lognormal (3P) probability distribution; while, the best probability
distribution describing A parameter was the Gamma distribution.

Fig. 9. Sigmoid relationships between αopt and αA, αS, αm, αG and αH data.

33
F. Babaei et al.
Prediction of the parameters of Philip two-term model and subse-
quently the infiltration rates at certain probability levels would be
possible by using the best-fit probability distributions. Acquaintance
with the probability distribution of soil hydraulic properties is im-
portant in precision agriculture, planning and irrigation management.
Hence, a stochastic approach could be used for the planning and de-
velopment of agriculture land in arid land areas.
3.5. Scaling of model parameters
From 60 sets of the values for S and A, αS and αA were obtained by
Eqs. (13) and (14); while, the cumulative infiltration data were scaled
according to Eqs. (17) and (18). Infiltration curves scaled by different
scaling factors are presented in Fig. 8. The following equation was fitted
on scaled infiltration and scaled time data.
I ∗ = S ∗. t ∗0.5 + Ap∗ . t ∗ (21)
The NRMSE results of fitting Eq. (21) on scaling infiltration curve
for different scaling factors are presented in Table 6. Results showed
that the best scaling was achieved by αopt. Statistical analysis indicated
a strong difference between αS and αA scaling factors. Like Machiwal
et al. (2006) study, a poor correlation coefficient was obtained for αS
and αA. Based on the similar porous media concept, the pore-size dis-
tribution and the porosity are expected to be identical at different lo-
cations; while, they should be constant over time or at various degrees
of saturation (Miller and Miller, 1956). However, empirically, some
swelling or shrinking was displayed at various degree of saturation.
Accordingly, the similar media theory could not be considered true in
any field soil. On the other hand, if all conditions of the similar media
theory were satisfied, scaling factors could be estimated by either of the
infiltration parameters (i.e. S or A). Therefore, significant difference
among the scaling factors (αS and αA) was related to the intrinsic var-
iation in initial soil water content that had a considerable effect on
sorptivity (S) and also the changes in soil porosity and pore-size dis-
tribution. Apparently, αA (NRMSE = 41%) was preferred for scaling of
the infiltration process compared to the αS (NRMSE = 47%).
αopt was calculated with minimized objective function (Eq. 21). The
infiltration data scaled by αopt, showed a reasonable reduction in the
scatter points around the best-fitted curve; consequently, reduced the
NRMSE (11%). The values of αopt fall between αS and αA. Therefore,
arithmetic, geometric and harmonic scaling factors (i.e. αm, αG, αH;
respectively) were calculated by using these scaling factors. Compared
to αS, utilization of αm did not result in a considerable reduction in
NRMSE. However, significant reductions in the NRMSE were obtained
by αG, SA and αH scaling factors. The best result was yielded by αH.
Correlation coefficients between αopt and various scaling factors was
determined, which indicated the strong correlations between αG
(0.862) and αH (0.866) with αopt.
For describing the distribution of scaling factors, Kolmogorov-
Smirnov test was applied. Based on the frequency analysis depicted in
Table 5, the best-fitted probability distributions for αS and αA were
normal and Gamma distributions, respectively. In addition, Log-
Pearson 3 was the best-fitted probability distribution for both αG, and
αH scaling factors. The optimum scaling factors, αopt, and the arithmetic
mean scaling factor, αm, obeyed log normal and Lognormal (3P) dis-
tributions, respectively. Our results could be utilized for describing the
spatial variability of the soil infiltration properties with application of
stochastic approaches.
For defining the relationship between αopt and αA, αS, αm, αG and
αH, a regression analysis was performed. The curves reflecting the re-
lationship between αopt and αA, αS, αm, αG and αH were sigmoid
(Fig. 9). Machiwal et al. (2006) reported that the αopt of 17 infiltration
sites were correlated with αA, αS, αm, αG and αH, following a linear
trend. Yet, our results showed that the sigmoid regression could also
provide a better description than the simple linear regression. As
34
Catena 170 (2018) 25–35
illustrated in Fig. 9 a significant sigmoidal relation exists between αopt
and αm, αG and αH while the sigmoidal relations of αopt with αA and αS
are weak.
4. Conclusions
In this study the spatial variability of the infiltration process was
demonstrated in agricultural lands of Iranian arid areas. The RMKO
model could be considered as the optimum prediction model according
to R2 values for most agricultural soils in the studied area. While base
on the ACI criterion MKO model was the best model for prediction of
cumulative infiltration.
Parameters of the Philip two-term model (i.e. A (αA) and S (αS))
were used to scale cumulative infiltration. A more precise scaling was
obtained by αA based on the normalized RMSE. There was a significant
difference between αS and αA scaling factors. Therefore, it can be in-
ferred that no similar porous media conditions exists in the studied
area. The optimum scaling factor αopt was used to scale cumulative
infiltration by reducing the sum of squares error. Parameters of the
scaled model (i.e. S and A) were 7.99 and 0.699, respectively.
Arithmetic, geometric, and harmonic means of αS and αA scaling factors
were also used for scaling of the infiltration data. The average scaling
factors had significant correlation with αopt, and the strongest corre-
lation was obtained by the average harmonic scaling factor αH.
Frequency analysis elucidated that αopt and αm obeyed log normal
distribution; while, αH and αG could be best described by Gamma dis-
tribution.
References
Ahuja, L.R., Kozak, J.A., Andales, A.A., Ma, L., 2007. Scaling parameters of the Lewis-
Kostiakov water infiltration equation across soil textural and extension to rain in-
filtration. Trans. ASABE 50 (5), 1525–1541.
Argyrokastritis, I., Kerkides, P., 2003. A note to the variable sorptivity infiltration
equation. Water Resour. Manag. 17, 133–145.
Beven, K.J., Henderson, D.E., Reeves, A.D., 1993. Dispersion parameters for undisturbed
partially saturated soil. J. Hydrol. 143, 19–43.
Carrera, J., Neuman, S.P., 1986. Estimation of aquifer parameters under transient and
steady state conditions: 1. Maximum likelihood incorporating prior information.
Water Resour. Res. 22, 199–210.
Chen, J.F., Lee, C.H., Yeh, T.C., Yu, J.L., 2005. A water budget model for the Yun-Lin
Plain, Taiwan. Water Resour. Manag. 19, 483–504.
Comegna, V., Vitale, C., 1993. Space–time analysis of water status in a volcanic Vesuvian
soil. Geoderma 60, 135–158.
Dafny, E., Šimunek, J., 2016. Infiltration in layered loessial deposits: revised numerical
simulations and recharge assessment. J. Hydrol. 538, 339–354.
Gee, G.W., Bauder, J.W., 1986. Particle size analysis. In: Klute, A. (Ed.), Methods of soil
analysis: part 1 agronomy handbook no vol. 9. America Society of Agronomy and Soil
Science Society of America, Madison, WI, pp. 383–411.
Ghorbani Dashtaki, S.H., Homaee, M., Mahdian, M., Kouchakzadeh, M., 2009. Site-de-
pendence performance of infiltration models. Water Resour. Manag. 23, 2777–2790.
Green, W.H., Ampt, C.A., 1911. Studies on soil physics, I. Flow of air and water through
soils. J. Agric. Sci. 4, 1–24.
Hillel, D., 1998. Environmental Soil Physics. Academic Press, Sand Diego, CA.
Horton, R.E., 1940. Approach toward a physical interpretation of infiltration capacity.
Soil Sci. Soc. Am. J. 5, 339–417.
Hunt, A.G., Holtzman, R., Ghanbarian, B., 2017. A percolation-based approach to scaling
infiltration and evapotranspiration. Water 9, 1–14.
Jeong, G., Oeverdieck, H., Jin Park, S., Huwe, B., Lie, ß., 2017. Spatial soil nutrients
prediction using three supervised learning methods for assessment of land potentials
in complex terrain. Catena 154, 73–84.
Jury, W.A., 1986. Spatial variability of soil properties. In: Hern, S.C., Melancon (Eds.),
Vadose Zone Modeling of Organic Pollutants. Lewis Publishers, Chelsea, MI.
Jury, W.A., Russo, D., Sposito, G., Elabd, H., 1987. The spatial variability of water and
solute transport properties in unsaturated soil: I. Analysis of property variation and
spatial structure with statistical models. Hilgardia 55, 1–32.
Kostiakov, A.N., 1932. On the dynamics of the coefficient of water percolation in soils and
on the necessity of studying it from a dynamic point of view for the purposes of
amelioration. In: Transactions of the Sixth Congress of International Society of Soil
Science, Moscow. Russian Part A, pp. 17–21.
Kozak, J.A., Ahuja, L.R., 2005. Scaling of infiltration and redistribution of water across
soil textural classes. Soil Sci. Soc. Am. J. 69 (3), 816–827.
Machiwal, D., Jha, M.K., Mal, B.C., 2006. Modelling infiltration and quantifying spatial
soil variability in a wasteland of Kharagpur, India. Biosyst. Eng. 95 (4), 569–582.
Marquardt, D., 1963. An algorithm for least-squares estimation of nonlinear parameters.
SIAM J. Appl. Math. 11, 431–441.
F. Babaei et al.
Mein, R.G., Larson, C.L., 1973. Modeling infiltration during a steady rain. Water Resour.
Res. 9, 384–395.
Miller, E.E., Miller, R.D., 1956. Physical theory for capillary flow phenomena. J. Appl.
Phys. 27, 324–332.
Minasny, B., Mcbratney, A.B., 2006. A conditioned Latin hypercube method for sampling
in the presence of ancillary information. Comput. Geosci. 32, 1378–1388.
Mirzaee, S., Zolfaghari, A.A., Gorji, M., Dyck, M., Ghorbani Dashtaki, S., 2014. Evaluation
of infiltration models with different numbers of fitting parameters in different soil
texture classes. Arch. Agron. Soil Sci. 60 (5), 681–693.
Mulla, D.J., Mcbratney, A.B., 2002. Soil spatial variability. In: Warrick, A.W. (Ed.), Soil
Physics Companion. CRC Press, Florida, pp. 343–373.
Nielsen, D.R., Biggar, J.W., Erh, K.T., 1973. Spatial variability of field-measured soil-
water properties. Hilgardia 42, 215–259.
Nielsen, D.R., Hopmans, J.W., Reichardt, K., 1998. An emerging technology for scaling
field soil water behavior. In: Sposito, G. (Ed.), Scale Dependence and Scale Invariance
in Hydrology. Cambridge University Press, Cambridge, pp. 136–166.
Pachepsky, Y., Radcliffe, D.E., Selim, H.M. (Eds.), 2003. Scaling Methods in Soil Physics.
CRC Press, Boca Raton, Fla..
Pahlevan, N., Yazdani, M.R., Zolfaghari, A.A., Ghodrati, M., 2016. Multifractal analysis of
soil hydraulic properties in arid areas. Soil Res. 54 (8), 914–925.
Parhi, P.K., Mishra, S.K., Singh, R., 2007. A modification to Kostiakov and modified
Kostiakov infiltration models. Water Resour. Manag. 21, 1973–1989.
Philip, J.R., 1957. The theory of infiltration: moisture profiles and relation to experiment.
Soil Sci. 84, 163–178.
Radcliffe, D.E., Šimunek, J., 2012. Water flow in soils. In: Handbook of Soil Science
Properties and Processes: Part 1 Soil Physics. CRC Press Taylor & Francis Group,
London, New York.
Rasoulzadeh, A., Sepaskhah, A.R., 2003. Scaled infiltration equations for furrow irriga-
tion. Biosyst. Eng. 86 (3), 375–383.
Richards, L.A., 1931. Capillary conduction of liquids in porous mediums. Physics 1,
318–333.
Roudier, P., Beaudette, D.E., Hewitt, A.E., 2012. A conditioned Latin hypercube sampling
algorithm incorporating operational constraints. In: Minasny, B., Malone, B.,
McBratney, A. (Eds.), Digital Soil Assessments and Beyond. CRC Press, Boca Raton,
pp. 227–231.
35
Catena 170 (2018) 25–35
Sadeghi, M., Ghahraman, B., Davary, K., Hasheminia, S.M., Reichardt, K., 2011. Scaling
to generalize a single solution of Richards' equation for soil water redistribution. Sci.
Agric. (Piracicaba, Braz). 68, 582–591.
Sadeghi, M., Ghahraman, B., Ziaei, A.N., Davary, K., Reichardt, K., 2012. Invariant so-
lutions of Richards' equation for water movement in dissimilar soils. Soil Sci. Soc.
Am. J. 76 (1), 1–9.
Sharma, M.L., Gander, G.A., Hunt, C.G., 1980. Spatial variability of infiltration in a wa-
tershed. J. Hydrol. 45, 101–122.
Smith, R.E., 1972. The infiltration envelope: results from a theoretical infiltrometer. J.
Hydrol. 17, 1–21.
Sposito, G., 1998. Scale Dependence and Scale Invariance in Hydrology. Cambridge
University Press, Cambridge, UK.
Talsma, T., 1969. In situ measurement of sorptivity. AJSR 7, 269–276.
The MathWorks, Inc, 2007. MATLAB: the Language of Technical Computing. Version 7.5.
The MathWorks Inc, Ismaning.
Walkley, A., Black, I.A., 1934. An examination of the digestion method for determining
soil organic matter, and a proposed modification of the chromic acid titration
method. Soil Sci. 37, 29–38.
Warrick, A.W., 1990. An analytical solution to Richards' equation for a draining soil
profile. Water Resour. Res. 26 (2), 253–258.
Warrick, A.W., 1998. Spatial variability. In: Hillel, D. (Ed.), Environmental Soil Physics.
Academic Press, San Diego, CA, pp. 655–676.
Warrick, A.W., Hussen, A.A., 1993. Scaling of Richards equation for infiltration and
drainage. Soil Sci. Soc. Am. 57, 15–18.
Warrick, A.W., Nielsen, D.R., 1980. Spatial variability of soil physical properties in the
field. In: Hillel, D. (Ed.), Applications of Soil Physics. Academic Press, New York, pp.
319–344.
Warrick, A.W., Mullen, G.J., Nielsen, D.R., 1977. Scaling of field measured hydraulic
properties using a similar media concept. Water Resour. Res. 13, 355–362.
Warrick, A.W., Lomen, D.O., Yates, S.R., 1985. A generalized solution to infiltration. Soil
Sci. Soc. Am. J. 49, 34–38.
Williams, R.D., Ahuja, L.R., 1991. Estimating soil water characteristics using physical
properties and limited data. p. 405. In: van Genuchten, M.Th. (Ed.), Indirect Methods
for Estimating the Hydraulic Properties of Unsaturated Soils. Proc. Int. Works.
Riverside, CA. 11–14 Oct. 1989. Univ. of Calif. Press, Riverside.