Geoderma 138 (2007) 261 271

www.elsevier.com/locate/geoderma

Estimating soil hydraulic properties of Fengqiu County soils in the

North China Plain using pedo-transfer functions
Y. Li a,, D. Chen a , R.E. White a , A. Zhu b,c , J. Zhang b,c
a

b
c

School of Resource Management, Faculty of Land and Food Resources, The University of Melbourne, Parkville 3010, Victoria, Australia
State Experimental Station of Agro-Ecosystem in Fengqiu, Institute of Soil Science, The Chinese Academy of Sciences, Nanjing 210008, PR China
State Key Laboratory of Soil and Sustainable Agriculture, Institute of Soil Science, The Chinese Academy of Sciences, Nanjing 210008, PR China
Received 1 December 2005; received in revised form 24 November 2006; accepted 27 November 2006
Available online 11 January 2007

Abstract
The unsaturated soil hydraulic characteristics, including soil water retention curve and hydraulic conductivity, are the crucial input data for
simulating soil water and solute transport through the unsaturated zone at regional scales using GIS, and are expensive to measure. These
properties are frequently predicted with pedo-transfer functions (PTFs) using the routinely measured soil properties (e.g. soil texture, soil bulk
density, and soil organic matter content).
In this study, 63 soil water retention curves and 36 saturated soil hydraulic conductivities of seven soil profiles collected in Fengqiu County in the
North China Plain were measured. Soil texture, bulk density and soil organic matter of these soil samples were also measured. The van Genuchten
model describing soil water retention was used to fit the measured data for quantifying the soil hydraulic parameters. The PTFs were developed by
multiple regression between soil hydraulic parameter data and basic soil properties. The double cross-validation of these PTFs is also discussed in this
paper. The locally-developed PTFs from this study were compared with several existing PTFs in predicting the soil hydraulic parameters.
The developed PTFs were used in the regional simulation of a wheat and maize cropping agroecosystem in Fengqiu County for the 19981999
rotation year, and can explain 33% of spatial variation of the observed crop yields in 409 villages.
2006 Elsevier B.V. All rights reserved.
Keywords: Soil water retention curve; Saturated soil hydraulic conductivity; Basic soil properties; Pedo-transfer functions; Multiple regression; Application

1. Introduction
A spatially distributed model of water and nutrient
management (WNMM) has been developed to study the impact
of intensive cropping systems on water resource quality in the
North China Plain (Li, 2002). Because dynamic water flow and
nitrate transport through the soil vadose zone are modelled in
WNMM, it requires that the soil hydraulic properties be known
at regional scales. The hydraulic properties include the soil
water retention curve (SWRC), which presents the relationship
between the volumetric water content () and the soil water
pressure head (h), and the hydraulic conductivity curve, which
Corresponding author. Tel.: +61 3 83447583; fax: +61 3 83444665.
E-mail addresses: yong.li@unimelb.edu.au (Y. Li),
delichen@unimelb.edu.au (D. Chen), robertw@unimelb.edu.au (R.E. White),
anzhu@mail.issas.ac.cn (A. Zhu), jbzhang@mail.issas.ac.cn (J. Zhang).
0016-7061/$ - see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.geoderma.2006.11.018

relates the conductivity (K) to the soil water pressure head (h) or
the water content.
When the temporal and spatial variability of the region is
considered, the required measurements of unsaturated soil
hydraulic properties are tremendous, time-consuming, and very
expensive. Therefore, it is necessary to develop a set of socalled pedo-transfer functions (PTFs) to estimate the unsaturated soil hydraulic properties from more easily measured or basic
soil properties in the attribute database of a digital soil survey
map, in which soil hydraulic properties are not always available.
Bouma and van Lanen (1987) first described the equations for
relating different land characteristics and soil properties as the
term PTFs even though there were many attempts in this field
before. For the recent development of PTFs and their
application, we refer to reviews by Rawls et al. (1991), van
Genuchten and Leij (1992), Pachepsky et al. (1999) and Wsten
et al. (2001).

262

Y. Li et al. / Geoderma 138 (2007) 261271

In developing PTFs, soil texture (including sand, silt and

clay contents), bulk density and organic mater content are the
most used predictors in the literatures, and additional factors
(soil particle size and distribution indices) are rarely applied
because of lack of availability in the soil databases (Wsten
et al., 2001). Furthermore, as summarized by Nemes et al.
(2003), most of PTFs are developed to estimate the soil water
retention (points at a series of matric potentials or parameters of
analytical water retention equations) and saturated hydraulic
conductivity. A small number of PTFs were proposed for the
estimation of unsaturated hydraulic conductivity, e.g. Wagner
et al. (2001). Methods for predicting soil hydraulic characteristic using PTFs are grouped by Tietje and Tapkenhinrichs
(1993) into three types: (i) estimation of the water contents at
certain matric potentials (Husz, 1967; Renger, 1971; Gupta and
Larson, 1979; Rawls et al., 1982; Puckett et al., 1985; Imam
et al., 1999; Kar et al., 2004), (ii) estimation of soil water
retention relation with a physicalconceptual model approach
(Arya and Paris, 1981; Haverkamp and Parlange, 1986; Tyler
and Wheatcraft, 1989; Baumer, 1992; van den Berg et al., 1997;
Tomasella and Hodnett, 1998; Tomasella et al., 2003), and (iii)
estimation of parameters of algebraic retention functions for
describing (h) and K() or K(h) (Pachepsky et al., 1982;
Cosby et al., 1984; Rawls and Brakensiek, 1985; Nicolaeva
et al., 1986; Wsten and van Genuchten, 1988; Rawls and
Brakensiek, 1989; Vereecken et al., 1989, 1990; Schaap et al.,
1998; Minasny et al., 1999; Wsten et al., 1999; Tomasella
et al., 2003). The third method is widely used to directly predict
hydraulic model parameters for describing soil water retention
and hydraulic conductivity properties. PTFs are usually

expressed as linear or nonlinear regression equations or, more

recently, distributed as computer codes resulting from artificial
neutron network analysis (Pachepsky et al., 1996; Tamari et al.,
1996; Schaap and Leij, 1998; Minasny et al., 1999; Schaap
et al., 2001; Nemes et al., 2003).
If van Genuchten models (van Genuchten, 1980) for soil
water retention and soil hydraulic conductivity, based on the
statistical pore-size distribution model of Mualem (1976), are
applied in modelling, the parameters representing the soil
hydraulic conductivity curve can be the same or directly derived
from the soil water retention parameters, except for the saturated
soil hydraulic conductivity (Ks). This eliminates the need for the
direct measurement or indirect estimation of the hydraulic
conductivity curve if Ks is known. Hence, the van Genuchten
models of soil water retention and unsaturated soil hydraulic
conductivity are considered in this study.
Because existing PTFs for estimating soil water retention
curve and soil hydraulic conductivity in the literature are not
always applicable in other regions with acceptable accuracy
(Tietje and Tapkenhinrichs, 1993; Kern, 1995; Tietje and
Hennings, 1996; Cornelis et al., 2001; Wagner et al., 2001;
Nemes et al., 2003), we based this study on a data set covering
measured basic soil properties, soil water retention curves and
the saturated hydraulic conductivity of representative Fengqiu
County soils in the North China Plain. The objective was to
derive our own PTFs for estimating the soil water retention
parameters and saturated hydraulic conductivity. The adjusted
coefficients of determination and double cross-validation were
used to evaluate the predictive capabilities of the derived PTFs,
which will be deployed to the digital soil map of Fengqiu

Fig. 1. The sampling sites of seven soil profiles in Fengqiu County.

Y. Li et al. / Geoderma 138 (2007) 261271

County to build the spatial distribution of the soil water

retention curve and saturated hydraulic conductivity. The
performance of the derived PTFs was also compared with that
of several existing PTFs.
2. Materials and methods
Fengqiu County soils are mainly classified as two types:
Ochric Aquic Cambisol and Ustic Sandic Entisol according to
the Chinese Soil Taxonomy System (Research group of Chinese
Soil Taxonomy System, 1995). Ochric Aquic Cambisols
dominate the soil distribution in whole Fengqiu County, covering 98% more of the total soil area, and Ustic Sandic Entisol
accounts for about 2%.
Sixty three undisturbed 100-cm3 soil cores were collected from
7 representative soil subtype profiles in Fengqiu County (9 cores
in each profile from soil surface to 2 m deep) (Fig. 1). The soil
water retention data were measured on 100-cm3 soil samples
using a pressure membrane apparatus at the suctions of 0, 10, 30,
50, 100, 300, 500, and 1500 kPa (Klute, 1986). Eight-point
retention data were fitted to the equation of van Genuchten (1980):
h hr

hs hr
1 jahjn m

where denotes the soil volumetric water content (cm3 cm 3), r

and s are the soil residual and saturated volumetric water
contents (cm3 cm 3), respectively, h is the soil water pressure
head (cm), and is in cm 1, n and m are parameters defining the
SWRC's shape. The unknown parameters (r, s, , n and m)
were obtained using the nonlinear least-squares optimisation
program RETC (van Genuchten et al., 1991) from measured soil
water retention data. The dry bulk density was measured by ovendrying soil samples at 105 C for 24 h. The organic matter content
was estimated from the organic carbon content determined by the
WalkeyBlack method, using a constant 1.724 for transformation. The particle-size distribution was obtained by the pipette
method for particles with a diameter less than 0.002 mm (clay
fraction), 0.020.002 mm (silt fraction), and 0.022 mm (sand
fraction). The soil texture was determined by the international soil
texture classification. Because of limited budget for this

Table 2
Texture-grouped soil properties of 63 samples for SWRC
Texture
group

Variables

Minimum

Maximum

Mean

Std. Error

63 samples for SWRC

Soil organic matter (SOM, %)
Sand fraction (SAND, %)
Silt fraction (SILT, %)
Clay fraction (CLAY, %)
Bulk density (BD, g cm 3)

0.12
6.26
1.74
0.54
1.20

1.54
93.03
82.20
31.75
1.61

0.65
50.23
38.72
9.06
1.42

0.05
3.53
3.05
0.82
0.01

36 samples for Ks
Soil organic matter (SOM, %)
Sand fraction (SAND, %)
Silt fraction (SILT, %)
Clay fraction (CLAY, %)
Bulk density (BD, g cm 3)

0.12
8.98
1.74
0.54
1.20

1.54
93.03
79.50
27.12
1.59

0.65
53.28
35.86
8.86
1.42

0.07
4.46
3.83
1.01
0.01

Number SOM (%)

Sand
9
Loamy sand 3
Sand loam
18
Silty loam
24
Silty clay
7
loam
Silty clay
2

0.20 0.01
0.13 0.01
0.67 0.08
0.78 0.08
0.96 0.09

BD
(g cm 3)

PO a
(cm3 cm 3)

Measured s b
(cm3 cm 3)

1.47 0.02
1.41 0.03
1.45 0.02
1.38 0.02
1.42 0.03

0.466 0.008
0.487 0.009
0.474 0.007
0.498 0.006
0.483 0.010

0.460 0.011
0.477 0.003
0.483 0.008
0.519 0.005
0.480 0.019

0.67 0.05 1.43 0.04 0.481 0.013 0.479 0.061

PO denotes soil total porosity, calculated from (1 BD / 2.65).

b
Measured s is the volumetric water content at 0 kPa pressure head in
determining SWRC.
a

experiment, only two replicates for each sampling point were

taken. Therefore, the measures for each sampling point were
expressed in average, without additional standard deviation or
error information. Considering its high spatial variability in the
field, the Ks was measured on selected 36 soil layers or sampling
points from the same seven soil profiles using the method of Cook
and Broeren (1995) with 6 replicates. Table 1 summarises the soil
basic properties in this study.
Once the parameters for Eq. (1) were developed the correlation
and multiple regression analyses from the SPSS package
(Norusis, 1994) were carried out to formulate the PTFs of these
parameters as well as Ks, based on the basic soil properties. The
predictive capabilities of the PTFs were assessed using the
adjusted R2 value:



N 1
R2adj 11R2
2
N M 1
where N is the number of observations, M is the number of
independent variables in the PTF, and R2 is the coefficient of
determinations, given by
N
X

R2 1

SSE SSR

1 i1
N
P
SSQ SSQ

N
X

yi y i 2

yi
yi

i1

i1
N
P

y i
y i 2

3
yi y i

i1

with
y

Table 1
Value range and sample distribution of basic soil properties

263

N
1X
yi
N i1

yi and i are the measured and predicted hydraulic model

parameters under investigation, SSQ is the total sum of squares,
Table 3
Texture-grouped soil properties of 36 samples for Ks
Texture group

Number

SOM (%)

BD (g cm 3)

Ks

Sand
Loamy sand
Sand loam
Silty loam
Silty clay loam
Silty clay

4
3
12
13
3
1

0.21 0.02
0.13 0.01
0.67 0.10
0.81 0.12
1.02 0.14
0.62

1.45 0.01
1.41 0.02
1.45 0.02
1.37 0.02
1.43 0.05
1.39

45.21 8.04
21.50 0.68
9.85 3.24
19.96 4.35
20.89 6.10
7.85

264

Y. Li et al. / Geoderma 138 (2007) 261271

Table 4
The mean and standard error of the fitted retention parameters
Texture
groups

Number s

Sand

Loamy sand

Sandy loam

18

Silty loam

24

Silty clay
loam
Silty clay

All

2
63

0.458 0.011 0.1403

0.0269
0.468 0.001 0.0047
0.0006
0.484 0.008 0.0320
0.0065
0.478 0.008 0.0120
0.0020
0.430 0.019 0.0054
0.0034
0.430 0.056 0.0025
0.0007
0.477 0.006 0.0346
0.0070

R2

1.322 0.023 0.986 0.007

1.358 0.010 0.987 0.001

distribution pattern of 36 soil samples is similar to the 63 soil

samples, even the relationships of SOM and BD vs. soil texture
(Table 3). Within the six texture classes, the sand class has the
highest Ks of 45 cm d 1; loamy sand, silty loam and silty clay
loam classes have the similar Ks of around 20 cm d 1, in the
middle range; sand loam and silty clay classes have the lowest
Ks of 810 cm d 1.

1.349 0.046 0.976 0.005

1.305 0.026 0.954 0.006
1.290 0.037 0.915 0.007
1.228 0.028 0.900 0.020
1.318 0.017 0.960 0.004

SSR is the regression sum of squares (explained by the PTFs), and

SSE is the residual sum of squares of error. Adjusted R2 measures
the proportion of the variation that can be accounted for by the
regression models.
In order to validate the developed PTFs, we applied the
double cross-validation method (Green and Caroll, 1978) to
evaluate predictions and stability of the PTFs. To use this
method, the complete set of observations was randomly split in
two equal subsets. Regression analysis was carried out on each
of these subsets using the independent variables retained from
the entire data. Subsequently, the regression equation derived
from one set was applied to another, and vice versa, while the
coefficient of determination of two regression analyses as well
as the coefficients of Pearson correlation between predicted and
observed hydraulic parameters for the two subsets were
calculated.
3. Results and discussion
3.1. Variations of the soil properties
The 63 soil samples for SWRC are grouped into six textural
classes according to the international soil texture classification,
in which the loam classes account for 78% of total samples
(Table 2). There are twelve samples in sand classes, but only
two in the clay class. Therefore, soil samples are mainly coarsetextured. As seen in Table 2, as soil texture increases, soil
organic matter (SOM) increases except the silty clay class, but
there is no obvious trend for bulk density (BD) and soil total
porosity (PO). For the measured s, it increases as soil texture
reaches silty loam class, then it starts to decrease with soil
texture increasing. Basically, PO is slightly greater than
measured s, except for the silty loam class. If a soil particle
density greater than the average value of 2.65 is used, which
may be true for these low SOM and river sedimentary soils in
the North China Plain, PO will be apparently greater than
measured s.
Because 36 soil samples for Ks were taken within the sites of
63 soil samples for SWRC, it is not surprising that the texture

Fig. 2. The soil water retention curves described by Eq. (5): (A) sand with
SAND = 91.21%, SILT = 2.35%, CLAY = 4.44%, SOM = 0.24%, and BD = 1.45 g
cm 3; (B) sandy loam with SAND = 76.61%, SILT = 20.19%, CLAY = 1.20%,
SOM = 0.17%, and BD = 1.51 g cm 3; and (C) Silty clay loam with
SAND = 25.23%, SILT = 56.51%, CLAY = 16.26%, SOM = 0.65%, and
BD = 1.32 g cm 3. Note that open circles denote the measured data, and the
solid lines present the fitted curves.

Y. Li et al. / Geoderma 138 (2007) 261271

265

Table 5
Correlation matrix of soil hydraulic parameters via basic soil properties
s

ln e(s)

ln()

ln(n)

Ks

ln(Ks)

a
b
c
d
e

Pearson Correlation
Sig. b (2-tailed)
Nc
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N
Pearson Correlation
Sig. (2-tailed)
N

SAND

SILT

CLAY

SOM

BD

ln(SAND)

ln(SILT)

ln(CLAY)

ln(SOM)

ln(BD)

0.020
0.874
63
0.573 a
0
63
0.128
0.319
63
0.044
0.731
63
0.682 a
0
63
0.135
0.29
63
0.050
0.535
36
0.010
0.971
36

0.086
0.504
63
0.575 a
0
63
0.057
0.659
63
0.064
0.619
63
0.667 a
0
63
0.064
0.617
63
0.110
0.641
36
0.040
0.824
36

0.407 a
0.001
63
0.331 a
0.008
63
0.338 a
0.007
63
0.428 a
0
63
0.456 a
0
63
0.344 a
0.006
63
0.080
0.335
36
0.120
0.497
36

0.071
0.581
63
0.316 d
0.012
63
0.349 a
0.005
63
0.075
0.557
63
0.260 d
0.04
63
0.362 a
0.004
63
0.170
0.784
36
0.070
0.668
36

0.465 a
0
63
0.060
0.641
63
0.302 d
0.016
63
0.454 a
0
63
0.098
0.444
63
0.309 d
0.014
63
0.290
0.091
36
0.370 d
0.028
36

0.203
0.111
63
0.484 a
0
63
0.069
0.589
63
0.229
0.071
63
0.684 a
0
63
0.074
0.565
63
0.170
0.974
36
0.110
0.511
36

0.180
0.158
63
0.699 a
0
63
0.090
0.482
63
0.16
0.209
63
0.651 a
0
63
0.106
0.407
63
0.010
0.061
36
0.200
0.246
36

0.244
0.054
63
0.287 d
0.023
63
0.570 a
0
63
0.262 d
0.038
63
0.345 a
0.006
63
0.565 a
0
63
0.32
0.421
36
0.160
0.360
36

0.048
0.708
63
0.395 a
0.001
63
0.366 a
0.003
63
0.058
0.649
63
0.312 d
0.013
63
0.378 a
0.002
63
0.140
0.324
36
0.220
0.198
36

0.471 a
0
63
0.061
0.633
63
0.306 d
0.015
63
0.46 a
0
63
0.104
0.417
63
0.313 d
0.013
63
0.29
0.086
36
0.37 d
0.027
36

Correlation is significant at the 0.01 level (2-tailed).

Sig. indicates significance.
N is the number of observed samples.
Correlation is significant at the 0.05 level (2-tailed).
ln means natural logarithm.

3.2. Parameterisation of the soil water retention curve

The 63 measured soil water retention data were fitted to Eq.
(1) with the restriction of m = 1 1 / n and n N 1. However, the
value of soil residual water content (r) derived from nonlinear
regression analysis is not reasonable since some r values of 63
soil retention curves exceed 0.20 or even more than 0.30 cm3
cm 3. The high soil residual water content was concerned. After
carefully checking the raw retention data, it could be explained
by the non-equilibrium measurement of soil water content in the
1500 kPa-pressure chamber, which was too close to the water
content at pressure of 500 kPa. The last point of soil water
retention curve was removed and replaced by air-dry water
content corresponding to the soil water pressure head of
approximate 22,000 kPa (White, 1997) instead. The second
parameterisation results showed that only nine of 63 samples
had the term of r in their soil water retention equation formula
in which r ranged from 0.002 cm3 cm 3 to 0.100 cm3 cm 3.
r is set to be zero if the fitted r is less than 0.001 cm3 cm 3 as
estimated by RETC. Based on the above pre-analysis on the
best equation to fit the SWRC, the original van Genuchten
model (1980) without r of SWRC:
h

hs
1 jahjn 11=n

was chosen to better describe the soil water retention characteristic of Fengqiu County soils. The mean and standard error

of the retention parameters in Eq. (5) of the third parameterisation are listed in Table 4. Apparently, the adjusted R 2 values for
nonlinear regression given in Table 4 indicate that Eq. (5) is
able to describe the coarse textured soils better than the fine
textured soils in Fengqiu County. There is no other consistent
pattern in the variation of s, , and n according to soil texture.
Fig. 2 shows the measured and fitted retention curves of a sand,
a sandy loam and a silty clay loam, respectively. It also gives
visual information about the goodness of fitting the measured

Table 6
PTFs for estimating soil hydraulic parameters of the van Genuchten models as a
function of basic soil properties
Model
Parameters

Regression equations

ln(s)

1.531 + 0.212*ln(SAND) +
0.355 0.206
0.006*SILT 0.051*SOM
0.566*ln(BD)
67.408 0.040*SILT
120.0
76.12
0.670*ln(SILT) 2.189*SOM +
1.410*ln(SOM) + 78.400*BD
121.331*ln(BD)
1.488 + 0.002*ln(SILT) + 0.013*CLAY
0.642 0.524
0.248*ln(CLAY) + 0.048*ln(SOM) +
0.451*ln(BD)
13.262 1.914*ln(SAND)
25.94 10.13
0.974*ln(SILT) 0.058*CLAY
1.709*ln(SOM) +
2.885*SOM 8.026*ln(BD)

ln()

ln(Ks)

SSR

SSE

R2adj
0.61

0.57

0.51

0.66

266

Y. Li et al. / Geoderma 138 (2007) 261271

retention data to Eq. (5) for the soil samples with different
textures listed in Table 4.
The poor prediction for the SWRC parameters of the claytextured soil using Eq. (5) could be explained by several
reasons. For example, the laboratory measurements are not
done very well because the clay soil always releases water
very slowly when drying during the measurement. This results
in overestimating soil water contents at high pressure heads.
Other reasons mentioned by van Genuchten (1980) regarding
this same problem do not apply here. Is this because of r
being set to be zero? If r is flexible, some of the predicted
value of r will be overestimated as mentioned earlier. In
addition, s of clay-textured soils is always underestimated by
RETC in this study. Conceptually, r is very low at huge
pressure head condition even though clay r is considered
greater than sand r. It seems that the poor prediction for the
SWRC parameters of clay-textured soils results from the
inability of van Genuchten SWRC model to match the experimental soil water retention data because it assumes unimodal pore-size distributions underlying all soils. Durner
(1994) introduced the concept of multimodal pore-size distributions in estimating soil hydraulic properties, e.g. (h) and
K(h), but this approach increases the complexity of the expressions of soil hydraulic properties and does not comply
with the ultimate purpose of developing robust PTFs in this
study, which is based on the universal national soil survey

database. Furthermore, according to van Genuchten (1980),

limited data at low water contents leave some doubt about the
accuracy of the fitness, and r needs to be estimated by other
independent procedure.
3.3. Correlation analysis of soil hydraulic parameters vs. basic
soil properties
In order to develop the fundamental relationships between
soil hydraulic parameters and the basic soil properties, the
correlation analysis of SPSS package was applied, and the
statistical results are shown in Table 5.
For the soil hydraulic parameters, it is found that a high
correlation exists between s and CLAY and BD, as well as
between and SAND, SILT, CLAY, and SOM. The n parameter
has significant correlation with CLAY, SOM, and BD. It is seen
that there is no significant correlation between Ks and basic soil
properties. When the logarithmical transformation of Ks is
considered during the correlation analysis, BD becomes
significantly related to Ks. From Table 5, it can be seen that in
most cases the natural logarithmical transformation of variables
increases the correlation with other variables considered.
Therefore, s, and K s would be better transformed
logarithmically. The distributions of ln(s), ln() and ln(Ks)
are closer to the normal distribution compared with the nontransformed values, and give higher correlations with different

Fig. 3. Measured vs. predicted A) ln(s), B) ln(), C) n and D) ln(Ks) by PTFs of this study (), Vereecken et al. (1989, 1990) (), ROSETTA () and HYPRES ().
The straight line in each plot is the 1:1 line.

Y. Li et al. / Geoderma 138 (2007) 261271

Table 7
Summary of the double cross-validation test for the developed PTFs

ln(s)

A
B

ln()

ln(Ks)

variables in the model. The final multiple regression equations

and statistical information, under the assumption of no further
statistical improvement, are given in Table 6. The inclusion of
interaction terms of basic soil properties were considered as
increasing the PTFs complexity rather than improving statistical
significance. The performance of all PTFs was assessed by the
2
values of Radj
.
The logarithmic form of saturated water content (s) was
positively related to SAND and SILT, and negatively related to
BD and SOM. The regression explained about 61% of the
variance by these arguments. As discussed in the correlation
analysis, the parameter for SWRC is presented in the
logarithmic form in its regression equation, which explains 57%
of the variance for ln(). The value of ln() is mainly estimated
by SILT, SOM and BD. The equation for n consists of SILT,
CLAY, SOM, and BD as predictors, and explains 51% of the
variance in n. It means that SAND is not an important variable
2
to predict the value of n. Hence, as seen from the values of Radj
,
among these three parameters, n is the poorest estimated by
PTFs. The goodness of measured vs. predicted ln(s), ln() and
n is shown in Fig. 3, respectively. The information of regression
standardised residuals against regression standardised predicted
values of ln(s), ln() and n from multiple regression analysis
also indicates that most of the regression standardised residuals
lie randomly between 2 and + 2, and thus there is no
significant correlation between standardised residual and
standardised predicted value. Therefore, the PTFs for predicting
ln(s), ln() and n satisfy the assumption of linearity and
statistical non-bias, and are regarded as reliable.
Compared to other studies, our results show that the PTFs
for predicting soil water retention parameters of Fengqiu
County soils are worse than those reported by Vereecken et al.
(1989) for Belgian soils, Wsten and van Genuchten (1988) for
Dutch soils, Rajkai et al. (1996) for Swedish soils, and
Goncalves et al. (1997) for Portuguese soils even though we
only have a total of 63 soil samples, but slightly better than
Tomasella et al. (2003) for Brazilian soils. It is possible that the
particle-size distribution is insufficient and not detailed enough
to identify the individual contribution of soil particles to the soil
water retention characteristics. Another reason could be that the
studied soil textures are quite similar, very close to sandy loam.
This suggests to us that the PTFs have to be used with caution
when extrapolation of PTFs using basic soil properties is
performed.

R2adj r2

Model
Subsets Regression equations
Parameters
1.641 + 0.228*ln(SAND) + 0.007*SILT
0.028*SOM 0.560*ln(BD)
1.339 + 0.190*ln(SAND) + 0.005*SILT
0.057*SOM 0.719*ln(BD)
1.028 0.066*SILT 0.291*ln(SILT)
7.251*SOM + 4.044*ln(SOM) +
10.830*BD 29.816*ln(BD)
108.438 0.021*SILT 1.088*ln(SILT)
0.192*SOM + 0.324*ln(SOM) +
121.195*BD 181.381*ln(BD)
1.184 + 0.037*ln(SILT) + 0.008*CLAY
0.172*ln(CLAY) + 0.022*ln(SOM) +
0.892*ln(BD)
1.647 + 0.013*ln(SILT) + 0.018*CLAY
0.269*ln(CLAY) + 0.052*ln(SOM) +
0.017*ln(BD)
16.753 2.333*ln(SAND)
1.303*ln(SILT) 0.074*CLAY
1.688*ln(SOM) + 3.605*SOM
11.106*ln(BD)
10.039 1.884*ln(SAND)
0.802*ln(SILT) 0.065*CLAY
2.210*ln(SOM) + 3.653*SOM
3.270*ln(BD)

267

0.55 0.59
0.63 0.54
0.74 0.56

0.50 0.59

0.45 0.47

0.49 0.45

0.74 0.62

0.58 0.61

For ln(s), ln() and n, subsets A and B have 31 samples and 32 samples,
respectively. For ln(Ks), both subsets A and B have 18 samples.

soil properties (Vereecken et al., 1989; Goncalves et al., 1997).

The above correlation analysis is performed to detect the linear
relationships among the variables, and is used to advise the
PTFs structure.
3.4. PTFs for soil water retention parameters
The derivation of PTFs for soil water retention parameters
was performed for the 63 soil samples through multiple
regression using the basic soil properties, including their
logarithmically-transformed values and their interaction terms.
Because the sample size was not large enough, the multiple
regression analysis was not carried out in textural groups. A
backward method regression (Norusis, 1994) selected variables
at 0.10 significance level for entry in the regression model,
while a 0.05 significance level was applied to retain the

Table 8
Comparison of the performance of different PTFs on the complete SWRC data set for Fengqiu County soils
PTFs

ln(s)

ln()

ln(Ks)

SSE

SSE

SSE

SSE

This study
Vereecken et al. (1989, 1990)
ROSETTA (Schaap et al., 2001)
HYPRES (Wsten et al., 1999)

0.21
1.64
2.84
1.14

0.80
0.30
0.16
0.55

b0.001
b0.05
n.s. a
b0.001

76
161
150
457

0.78
0.71
0.50
0.67

b0.001
b0.001
b0.001
b0.001

0.52
10.32
26.97
1.26

0.74
0.24
0.05
0.47

b0.001
n.s.
n.s.
b0.001

10
280
140
90

0.85
0.54
0.37
0.08

b0.001
b0.001
b0.05
n.s.

r denotes the coefficient of Pearson correlation, and p is the probability. Note that the mass fraction has been transformed to the US system when PTFs of Vereecken
et al. (1989, 1990), ROSETTA and HYPRES are applied.
a
n.s. denotes not significant.

268

Y. Li et al. / Geoderma 138 (2007) 261271

3.5. PTFs for saturated soil hydraulic conductivity

Total 36 soil samples were used to develop the PTFs for
saturated hydraulic conductivity using a backward multiple
regression method, as described for the soil water retention
parameters. Table 6 lists the regression results based on the
basic soil properties.
It is seen from Table 6 that Ks was logarithmical transformed
for better normal distribution and better regression output. The
PTF equation for estimating Ks is composed of all the basic soil
properties, and explains about 66% of variance. The fact that
more than 97% of the regression standardised residuals lie
randomly between 2 and + 2 from this multiple regression
analysis indicates that there is no significant correlation between
standardised residual and standardised predicted value. Hence,
it implies that this PTF for predicting ln(Ks) satisfies the
assumption of linearity statistically, and is reliable.
3.6. Validation of PTFs
The double cross-validation test for the above developed
PTFs in terms of estimating ln(s), ln(), n and ln(Ks) was
carried out, and the results were summarized in Table 7. As seen
in Table 7, the signs of the subset regression equations were
stable, but the degree of significance changed, which was
probably caused by the relatively small sample size. The
coefficients of determination for the regression equations

2
) were similar to those obtained
developed for the subsets (Radj
for PTFs for the complete set. In addition, the square of the
Pearson correlation coefficients between predicted and measured parameters, defined as cross-validity coefficients (r2),
relative to each of the two subsets used for the cross-validation
were not significantly different at the 0.05 level.

3.7. Comparison with existing PTFs

Three sets of existing PTFs: Vereecken et al. (1989, 1990),
ROSETTA (Schaap et al., 2001) and HYPRES (Wsten et al.,
1999) were applied to the complete SWRC data set of Fengqiu
County soils, and their performance in estimating the soil
hydraulic parameters were compared to that of the PTFs derived
in this study (Table 8 and Fig. 3). As seen in Table 8, none of the
three existing PTFs did a better job than PTFs of this study in
estimating ln(s), ln(), n and ln(Ks) at all, but briefly based on r
and p, HYPRES performed slightly better than Vereecken et al.
(1989, 1990) and ROSETTA was the worst. In particular, all of
the three existing PTFs provided better prediction for ln() than
that for ln(s), n and ln(Ks), but with varying accuracy. Among
them, only HYPRES had the capability to predict n, and
ROSETTA and HYPRES failed to estimate ln(s) and ln(Ks),
respectively.
The limitation of applying PTFs developed from one region
to other regions is obvious through the comparison analysis in
this paper.

Fig. 4. Spatial distribution of saturated soil water conductivity (Ks, cm d 1) of the fourth soil layer (6080 cm) estimated by using locally-developed pedo-transfer
functions in Fengqiu County of the North China Plain.

Y. Li et al. / Geoderma 138 (2007) 261271

269

Fig. 5. Observed and predicted crop yields (wheat and maize) in Fengqiu County for the 19981999 rotation year.

3.8. Application of PTFs

The parameters for soil water retention and saturated
hydraulic conductivity are the intermediate characteristics
needed to compute other information with more practical
meaning, e.g. water balance and crop yield. A number of studies
used soil water simulation models to evaluate the performance
of estimated soil hydraulic properties through the simulation of
aircropsoil agroecosystems (Wsten et al., 1995; Espino
et al., 1996; van Alphen et al., 2001; Nemes et al., 2003).
The PTFs derived from this study were used in Fengqiu County
in the North China Plain to estimate parameters of soil hydraulic
properties for van Genuchten (1980) models in order to simulate
water balance and interaction with C and N cycling in a wheatmaize cropping agroecosystem using the Water and Nitrogen
Management Model WNMM (Li, 2002) at the county scale from
October 1998 to September 1999. The spatial soil information,
including SAND, SILT, CLAY, SOM and BD, was derived from
the soil map and soil survey report produced in the second national
soil survey of China in the early 1980s. Fig. 4 demonstrates an
example, the estimated spatial distribution of the saturated soil
hydraulic conductivity at 6080 cm soil depth of Fengqiu County
soils using locally-developed PTFs. The spatial distribution of
crops was based on the latest landuse map. A comprehensive field
survey was carried out in the fall of 1999, covering all the
information of agricultural practices for the 19981999 rotation

year in 605 individual villages. For detailed information of the

WNMM simulation regarding mechanism of WNMM, settings of
initial conditions in the agroecosystem and simulation outputs, we
refer to the paper of Li et al. (in press). Because of lack of spatially
measured soil water content at the county scale, the surveyed crop

Fig. 6. Linear regression between the observed and predicted crop yields (wheat
and maize) by WNMM in Fengqiu County for the 19981999 rotation year.

270

Y. Li et al. / Geoderma 138 (2007) 261271

yields in 605 villages were alternatively used to assess the

performance of the derived PTFs. As seen in Fig. 5, the villageaveraged pattern of spatial variation of grain yields of wheat and
maize predicted by WNMM is similar to that of the surveyed crop
grain yields, with a determining coefficient R2 of 0.33 (P b 0.001)
for 409 valid study villages (Fig. 6). The performance of PTFs in
using WNMM predicting crop yield in Fengqiu County is
acceptable when considering other variations caused by damages
of diseases and insects, soil salinity and deficiency of other
nutrients, e.g. phosphorus.
4. Conclusions
The van Genuchten model (1980) without a residual water
content term was selected as the optimal equation to describe
the soil water retention characteristic of Fengqiu County soils.
PTFs for estimating soil hydraulic characteristics were derived from basic soil properties (particle-size distribution, soil
organic matter, and bulk density). Among the three parameters
of Eq. (5), the saturated water content (s) was best predicted
through the entire soil data set, while prediction of the value of
2
n was the poorest, according to the assessment of Radj
of the
developed regression equations. The developed regression
models for estimating ln(s), ln(), n and ln(Ks) were tested
for their stability and predictability by the double crossvalidation method. It was found that the signs of the regression
coefficients and the determination coefficients were stable.
The PTFs obtained from this study appear superior in
predicting the soil hydraulic parameters, compared to three
existing PTFs: Vereecken et al. (1989, 1990), ROSETTA and
HYPRES. This confirms the limitation of applying PTFs
developed from one region to other regions.
The PTFs derived in this study were used to estimate soil
hydraulic properties for the simulation of a wheat and maize
cropping agroecosystem in Fengqiu County for the 19981999
rotation year. The simulation result of crop yield is comparable
to the field observations (R2 = 0.33, n = 409, p b 0.01).
To improve the performance of PTFs, more information on soil
properties such as soil structure may be required to reconstruct
PTFs.
Acknowledgments
This study was funded by the Australian Centre for
International Agricultural Research (Project LWR/96/164) and
the Knowledge Innovation Program of the Chinese Academy of
Sciences (Grant No.kzcx2-yw-406).
References
Arya, L.M., Paris, J.F., 1981. A physicoempirical model to predict the soil
moisture characteristic from particle-size distribution and bulk density data.
Soil Science Society of America Journal 45, 10231030.
Baumer, O.M., 1992. Predicting unsaturated hydraulic parameters. In: van
Genuchten, M.Th., et al. (Ed.), Proceedings of the International Workshop
on Indirect Methods for Estimating the Hydraulic Properties of Unsaturated
Soils. Riverside, CA, 1113 Oct. University of California, Riverside, CA,
pp. 341354.

Bouma, J., van Lanen, H.A.J., 1987. Transfer functions and threshold values:
from soil characteristics to land qualities. In: Beek, K.J., et al. (Ed.),
Quantified Land Evaluation. International Institute for Aerospace Survey
and Earth Sciences. . ITC Publication, vol. 6. Enschede, the Netherlands,
pp. 106110.
Cook, F.J., Broeren, A., 1995. Six methods for determining sorptivity and
hydraulic conductivity with disk permeameters. Soil Science 157, 211.
Cornelis, V.M., Ronsyn, J., van Meirvenne, M., Hartmann, R., 2001. Evaluation
of pedotransfer functions for prediction the soil moisture retention curve.
Soil Science Society of America Journal 65 (3), 638648.
Cosby, B.J., Hornberger, G.M., Clapp, R.B., Ginn, T.R., 1984. A statistical
exploration of the relationship of soil moisture characteristic to the physical
properties of soils. Water Resources Research 20, 682690.
Durner, W., 1994. Hydraulic conductivity estimation for soils with heterogeneous pore structure. Water Resources Research 30, 211223.
Espino, A., Mallants, D., Vanclooster, M., Feyen, J., 1996. Cautionary notes on
the use of pedotransfer functions for estimating soil hydraulic properties.
Agricultural Water Management 29, 235253.
Goncalves, M.C., Pereira, L.S., Leij, F.J., 1997. Pedo-transfer functions for
estimating unsaturated hydraulic properties of Portuguese soils. European
Journal of Soil Science 48, 387400.
Green, P.E., Caroll, J.D., 1978. Analysing Multivariate Data. John Wiley &
Sons, New York.
Gupta, S.C., Larson, W.F., 1979. Estimating soil water characteristic from
particle-size distribution, organic matter percent, and bulk density. Water
Resources Research 15, 16331635.
Haverkamp, R., Parlange, J.Y., 1986. Predicting the water retention curve from
particle-size distribution: 1. Sandy soils without organic matter. Soil Science
142, 325339.
Husz, G., 1967. The determination of pF-curves from texture using multiple
regression. Zeitschrift fr Pflanzenernhrung und Bodenkunde 116 (2),
2329.
Imam, B., Sorooshian, S., Mayr, T., Schaap, M.G., Wsten, J.H.M., Scholes, R.J.,
1999. Comparison of pedotransfer functions to compute water holding capacity
using the van Genuchten model in inorganic soils report to IGBP-DIS soil
data Tasks. IGBP-DIS Working Paper No. 22, IGBP-DIS, Tolouse, Cedex,
France.
Kar, G., Singh, R., Verma, H.N., 2004. Spatial variability studies of soil hydrophysical properties using GIS for sustainable crop planning of a water shed
of eastern India and its testing in a rainfed rice area. Australian Journal of
Soil Research 42, 369379.
Kern, J.S., 1995. Evaluation of soil water retention models based on basic soil
physical properties. Soil Science Society of America Journal 59, 11341141.
Klute, A. 1986. Water Retention: Laboratory methods. In: A. Klute et al.
(Editors), Methods of soil analysis: Part 1, Physical and mineralogical
methods, 2nd edition. Agronomy Monograph No. 9, American Society of
Agronomy and Soil Science Society of America, Madison, Wisconsin.
pp. 635662.
Li, Y., 2002. A spatially referenced mode for identifying optimal strategies for
managing water and fertilizer nitrogen under intensive cropping in the
North China Plain, Ph.D thesis, 257 p., The University of Melbourne,
Australia.
Li, Y., White, R.E., Chen, D., Zhang, J.B., Li, B.G., Zhang, Y.M., Huang, Y.F.,
Edis, R., in press. A spatially referenced Water and Nitrogen Management
Model (WNMM) for (irrigated) intensive cropping systems in the North
China Plain. Ecological Modelling.
Minasny, B., Mcbratney, A.B., Bristow, K.L., 1999. Comparison of different
approaches to the development of pedotransfer functions for water retention
curves. Geoderma 93, 225253.
Mualem, Y., 1976. A new model for predicting the hydraulic conductivity of
unsaturated porous media. Water Resources Research 12 (3), 513521.
Nemes, A., Schaap, M.G., Wsten, J.H.M., 2003. Functional evaluation of
pedotransfer functions derived from different scales of data collection. Soil
Science Society of America Journal 67, 10931102.
Nicolaeva, S.A., Pachepsky, Ya.A., Shcherbakov, R.A., Shcheglov, A.I., 1986.
Modelling of moisture regime for ordinary chernozems. Pochvovedenie 6,
5259.
Norusis, J.M., 1994. SPSS professional statistics 6.1. SPSS Inc., Chicago, Ill.

Y. Li et al. / Geoderma 138 (2007) 261271

Pachepsky, Ya.A., Shcherakov, R.A., Varallyay, G., Rajkai, K., 1982. Statistical
analysis of water retentions with other physical properties of soils.
Pochvovedenie 2, 4252.
Pachepsky, Ya.A., Timlin, A.D., Varallyay, G.V., 1996. Artificial neural
networks to estimate soil water retention from easily measurable data. Soil
Science Society of America Journal 60, 727733.
Pachepsky, Ya.A., Rawls, W.J., Timlin, D.J., 1999. The current status of pedotransfer functions: Their accuracy, reliability, and utility in field-and regionalscale modelling. In: Corwin, D.L., et al. (Ed.), Assessment of Non-point Source
Pollution in the Vadose Zone. . Geophysical Monograph, vol. 108. American
Geophysical Union, Washington, DC, pp. 223234.
Puckett, W.E., Dane, J.H., Hajek, B.F., 1985. Physical and mineralogical data to
determine soil hydraulic properties. Soil Science Society of America Journal
49, 831836.
Rajkai, K., Kabos, S., van Genuchten, M.Th., Jansson, P.E., 1996. Estimation of
water retention characteristics from the bulk density and particle-size
distribution of Swedish soils. Soil Science 161 (12), 832845.
Rawls, W.J., Brakensiek, D.L., 1985. Predictions of soil water properties from
hydrologic modelling. In: Jones, E., Ward, T.J. (Eds.), Watershed
Management. Eighties Proceedings of Symposium of ASCE, Denver, CO.
ASCE, New York, pp. 293299.
Rawls, W.J., Brakensiek, D.L., 1989. Estimation of soil water retention and
hydraulic properties. In: Morelseytoux, H.J. (Ed.), Unsaturated Flow in
Hydrologic Modelling. Theory and Practice. Kluwer Academic Publishers,
Dordrecht, pp. 275300.
Rawls, W.J., Brakensiek, D.L., Saxton, K.E., 1982. Estimation of soil water
properties. Transactions of the ASAE 108, 13161320.
Rawls, W.J., Gish, T.J., Brakensiek, D.L., 1991. Estimating soil water retention
from soil physical properties and characteristics. Advances of Soil Science
16, 213234.
Renger, M., 1971. The estimation of pore size distribution from texture, organic
matter and bulk density. Zeitschrift fr Kulturtechnik und Flurbereinigung
130, 5367.
Research Group of Chinese Soil Taxonomy System, 1995. Chinese Soil
Taxonomy System. China Agricultural Science and Technology Press,
Beijing.
Schaap, M.G., Leij, F.J., 1998. Database-related accuracy and uncertainty of
pedotransfer functions. Soil Science 163 (10), 765779.
Schaap, M.G., Leij, F.J., van Genuchten, M.Th., 1998. Neural network analysis
for hierarchical prediction of soil hydraulic properties. Soil Science Society
of America Journal 62, 847855.
Schaap, M.G., Leij, F.J., van Genuchten, M.Th., 2001. ROSETTA: A computer
program for estimating soil hydraulic parameters with hierarchical
pedotransfer functions. Journal of Hydrology 251 (34), 163176.
Tamari, S., Wsten, J.H.M., Ruiz-Suarea, J.C., 1996. Testing an artificial neural
network for predicting soil hydraulic conductivity. Soil Science Society of
America Journal 60, 771774.
Tietje, O., Hennings, V., 1996. Accuracy of the saturated hydraulic conductivity
prediction by pedo-transfer functions compared to the variability within
FAO textural classes. Geoderma 69, 7184.

271

Tietje, O., Tapkenhinrichs, M., 1993. Evaluation of pedo-transfer functions. Soil

Science Society of America Journal 57, 10881095.
Tomasella, J., Hodnett, M.G., 1998. Estimating soil water retention characteristics from limited data in Brazilian Amazonia. Soil Science 163, 190202.
Tomasella, J., Pachepsky, Ya.A., Crestana, S., Rawls, W.J., 2003. Comparison of
two techniques to develop pedotransfer functions for water retention. Soil
Science Society of America Journal 67, 10851092.
Tyler, S.W., Wheatcraft, S.W., 1989. Application of fractal mathematics to soil
water retention estimation. Soil Science Society of America Journal 53,
987996.
van Alphen, B.J., Booltink, H.W.G., Bouma, J., 2001. Combining pedotransfer
functions with physical measurement to improve the estimation of soil
hydraulic properties. Geoderma 103, 133147.
van den Berg, M., Klant, E., van Reeuwijk, L.P., Sombroek, G., 1997.
Pedotransfer functions for the estimation of moisture retention characteristics of Ferrasols and related soils. Geoderma 78, 161180.
van Genuchten, M.Th., 1980. A close-form equation for predicting the hydraulic
conductivity of unsaturated soils. Soil Science Society of America Journal
44, 892898.
van Genuchten, M.Th., Leij, F.J., 1992. On estimating the hydraulic properties
of unsaturated soils. In: van Genutchen, M.Th. (Ed.), Proceedings of
International Workshop on Indirect Methods for Estimating the Hydraulic
Properties of Unsaturated Soils, Riverside, CA, 1113 Oct 1989. University
of California, Riverside, CA, pp. 114.
van Genuchten, M.Th., Leij, F.J., Yates, S.R., 1991. The RETC code for
quantifying the hydraulic functions of unsaturated soils. EPA/600/2-91/065.
U.S. Environmental Protection Agency, Ada, OK.
Vereecken, H., Maes, J., Feyen, J., Darins, P., 1989. Estimating the soil moisture
retention characteristic from texture, bulk density, and carbon content. Soil
Science 148, 389403.
Vereecken, H., Maes, J., Feyen, J., 1990. Estimating unsaturated hydraulic
conductivity from easily measured soil properties. Soil Science 149, 112.
Wagner, B., Tarnawski, V.R., Hennings, V., Mller, U., Wessolek, G., Plagge,
R., 2001. Evaluation of pedotransfer functions for unsaturated soil hydraulic
conductivity using an independent data set. Geoderma 102, 275297.
White, R.E., 1997. Principles and Practice of Soil Science: The Soil as a Natural
Resource, 3rd edition. Blackwell Science Ltd, Melbourne, Australia, p. 101.
Wsten, J.H.M., van Genuchten, M.Th., 1988. Using texture and other soil
properties to predict the unsaturated soil hydraulic functions. Soil Science
Society of America Journal 52, 17621770.
Wsten, J.H.M., Finke, P.A., Jansen, M.J.W., 1995. Comparison of class and
continuous pedotransfer functions to generate soil hydraulic characterises.
Geoderma 66, 227237.
Wsten, J.H.M., Lilly, A., Nemes, A., Le Bas, C., 1999. Development and use of
a database of hydraulic properties of European soils. Geoderma 90,
169185.
Wsten, J.H.M., Pachepsky, Ya.A., Rawls, W.J., 2001. Pedotransfer functions:
bridging the gap between available basic soil data and missing soil hydraulic
characteristics. Journal of Hydrology (Amsterdam) 251, 123150.