INDIRECT METHODS FOR ESTIMATING

THE HYDRAULIC PROPERTIES OF

UNSATURATED SOILS

Proceedings of the International Workshop on

Indirect Methods for Estimating the Hydraulic Properties
of Unsaturated Soils
Riverside, California, October 11-13, 1989

edited by

M. Th. van Genuchten and F. J. Leij

U. S. Salinity Laboratory
USDA, ARS
4500 Glenwood Drive
Riverside, CA 92501

L. J. Lund
Department of Soil and Environmental Sciences
University of California
Riverside, CA 92521

U. S. Salinity Laboratory Department of Soil and

Agricultural Research Service Environmental Sciences
U. S. Department of Agriculture University of California
Riverside, California Riverside, California
 ANALYSIS OF PREDICTED HYDRAULIC
CONDUCTMTIES USING RETC
S. R Yates, M. Th. van Genuchten, and F. J. Leij

US. Salinity Laboratory, USDA, ARS, Riverside, CA

The computer program RETC was used to analyze unsaturated hydraulic conductivity data of
36 soils obtained from the literature using the empirical relationship of v a n Genuchten [1980] to
describe the soil water retention curve, and the predictive model of Mualem [1976a] to estimate the
hydraulic conductivity. By comparing the measured and estimated conductivities for the group of
data, the accuracy of the approach could be determined for various conditions. The analysis
consisted of comparing measured and estimated conductivities using the predictive method (i.e.,
when the unsaturated hydraulic conductivity is assumed to be unknown), the simultaneous method
(i.e., when both water retention and unsaturated hydraulic conductivity data are available), a
matching point (at a point somewhat dryer than saturation) to scale the estimated value of the
conductivity, and a methods involving the adjustment of Mualem’s empirical factor, 4 to improve
the fit between the measured and estimated conductivity. The results indicate that for this group
of data, the best estimates of the unsaturated hydraulic conductivity are obtained using either the
predictive approach with a matching point, or the simultaneous approach using a 6-parameter fit
(i.e., including 0,. O,, a. n, (and K, in the fitting procedure). Surprisingly, the worst and most
biased estimates were obtained using the simultaneous approach with a S-parameter fit (i.e., when
K, was fixed).

INTRODUCTION

There is an ever increasing need for more accurate numerical models to solve
groundwater flow and contamination problems. To achieve this. better methods for
characterizing the soil hydraulic properties are needed. Currently, our ability to create
complex numerical models far exceeds our capacity to describe the physical system on
which the models are based. The basic soil hydraulic properties, and in particular the
unsaturated hydraulic conductivity, play an integral role in determining the accuracy of
any numerical solution to flow, and therefore, transport problems. This inability to
characterize the functional relationship between the unsaturated hydraulic conductivity
and the water content (or pressure head) will result in an inaccurate representation of
the simulated flow process. Incorporating an inaccurate flow representation into a
contaminant transport model, in turn. will seriously affect the overall accuracy of the
transport simulation.
The problem of determining the unsaturated hydraulic conductivity is confounded
by the expense of experimentally obtaining this relationship, and the number of
observations required due to field scale variability. This has led to the introduction of
predictive techniques for obtaining the unsaturated hydraulic conductivity from soil
water retention data which are much easier to obtain. The advantage of predictive
methods is that more measurements can be obtained for a given investment, and hence,
that a better characterization of the flow domain can be obtained given that the methods
provide adequate predictions.
The purpose of this paper is to demonstrate the use of RETC by undertaking
several analyses on a soil hydraulic data set consisting of a large number of different soil

273
 274 Yates et al.

types. water retention data, and unsaturated hydraulic conductivities. By analyzing the
group as a whole and comparing the measured and estimated unsaturated hydraulic
conductivities, it is possible to make a qualitative judgment as to the accuracy of the
predictive techniques, whether the estimated conductivities are systematically biased, and
whether measured unsaturated hydraulic conductivity values are necessary to obtain an
accurate representation of this soil property. A more detailed discussion of the results
of this study is given by Yates et al. [1992].

METHODS

A data set consisting of 23 different soil types and 36 distributions for the moisture
retention relationship and the unsaturated hydraulic conductivity were obtained from
the literature and used in the following analysis. These data are reported in Table 1 and
include the Mualem [ 1976b] soil index, the soil name, and values of four parameters in
-the water retention function of van Genuchten [ 1980]. The connectivity parameter I
[van Genuchten and Nielsen, 1985; Luckner et al., 1989] was also used as a potential
unknown. The program RETC, described in more detail by van Genuchtenet al. [ 1991]
and Leij et al. [1992]. was used to determine the vaiues for the soil hydraulic properties
using a least squares parameter optimization method. The same set of initial
parameters were used for each soil type considered, and were modified only if the
minimization process failed to converge.
The analysis consisted of using RETC in several ways. The first method, termed the
predicted method, determines the optimal soil hydraulic parameters (6, 19, I, and n)
using only the measured water retention data. These soil hydraulic parameters are
subsequently used to predict the relative unsaturated hydraulic conductivity (assuming
a saturated conductivity, K,. of unity). Since the measured hydraulic conductivity data
for each soil listed in Table 1 are known, and assuming that the data are free from
error, a direct comparison can be made between the predicted and measured
unsaturated hydraulic conductivity giving some indication of the quality of the predictor.
If the unsaturated hydraulic conductivity is known at only one moisture content or
pressure head, but not at saturation, it is possible to scale the estimated hydraulic
conductivity using the method proposed by Luckner er al. [ 1989]. When combined with
the predictive method described above, this approach is termed the scaled-predictive
method.
Using RETC it is also possible to include known values of the unsaturated hydraulic
conductivity in the parameter optimization process to simultaneously fit the water
retention and unsaturated hydraulic conductivity functions. In this case, two additional
parameters (i.e., I and K,) may be adjusted during the fitting procedure. For our
analysis, a 5-parameter fit indicates that (was adjusted, while a 6-parameter fit indicates
that both ( and K, were adjusted during the optimization. The scaling technique
proposed by Luckner et al. [1989] was also used in conjunction with the simultaneous
fitting procedure.
The final analysis, termed the predictive mode with adjustable I. indicates the use
of fitted soil hydraulic parameters resulting from the predictive method (i.e., when the
measured unsaturated hydraulic conductivity is not used), fixing these values. and
subsequently using RETC to find the best value for the parameter (. In this manner,
only the parameter 1 is adjusted and the sensitivity of the method to the parameter I
can be obtained.
Analyzing Conductivity Predictions with RETC 275

Table 1. Basic Soil Hydraulic Data Set and Parameter Estimation Results
using the Predictive Method

soil
Index Soil Name 8, 0 (1/cm) n
3310 Silt Loam GE 3 0.139 0.394 0 . 0 0 4 1 4 2.15
---- Yolo Light Clay K(WC) 0.205 0.499 0.02793 1.71
---- Yolo Light Clay K(H) 0.205 0.499 0.02793 1.71
4130 Hygiene Sandstone 0.000 0.256 0.00562 3.27
1003 Lambcrg Clay 0.000 0.502 0.140 1.93
1006 Beit Netofa Clay Soil 0.000 0.447 0.00156 1.17
1101 Shiohot Silty Clay 0.000 0.456 183 1.17
2001 silt Columbia 0.146 0397 0.0145 1.85
2002 Silt Mont Cenis 0.000 0.425 0.0103 134
2004 Slate Dust 0.000 0.498 0.00981 6.75
3001 Weld Silty Clay Loam 0.159 0.496 0.0136 5.45
3101a Rideau Clay Loam, Wetting 0.279 0.419 0.0661 1.89
3101b Rideau Clay Loam, Drying 0.290 0.419 0.0177 3.18
3301a Caribou Silt Loam, Drying 0.000 0.451 0.00845 1 . 2 9
3301b Caribou Silt Loam, Wetting 0.000 0.450 0.140 1.09
3302a Grenville Silt Loam, Wetting 0.013 0523 0.0630 1.24
3302c Grenville Silt Loam, Drying 0.000 0.488 0.0112 1.23
3304 Touchet Silt Loam 0.183 0.498 0.0104 5.78
3402a Gilat Loam 0.000 0.454 0.0291 1.47
3403 Pachapa Loam 0.000 0.472 0.00829 1.62
3404 Adelanto Loam 0.000 0.444 0.00710 1.26
3405a Indio Loam 0.000 0.507 0.00847 1 . 6 0
3407a Guclph Loam 0.000 0.563 0.0275 1.27
3407b Guclph Loam 0.236 0.435 0.0271 262
3501a Rubicon Sandy Loam 0.000 0393 0.00972 2.18
350lb Rubicon Sandy Loam 0.000 0.433 0.147 1.28
3503a Pachapa Fme Sandy Clay 0.000 0.340 0.0194 1.45
3504 Gilat Sandy Loam 0.000 0.432 0.0103 1.48
4101a Plainfield Sand (210-250 pm) 0.000 0351 0 . 0 2 3 6 12.30
4101b Plainfield Sand (210-250 gm) 0.000 0312 0.0387 4.48
4102a Plainfield Sand (177-210 pm) 0.000 0361 0.0207 10.m
4102b Plainfield Sand (177-210 pm) 0.022 0309 0.0328 6.23
4103a Plainfield Sand (149-177 pm) 0.000 0387 0.0173 7.80
4103b Plainfield Sand (149-177 pm) 0.025 0321 0.0272 6 . 6 9
4104a Plainfield Sand (l25-149 pm) 0.000 0377 0.0145 10.60
4104b Plainfield Sand (125-149 urn\ 0.000 0.342 0.0230 5.18

RESULTS

Shown in Figure 1 is a scatter diagram of the measured versus estimated values for
the unsaturated hydraulic conductivity using the predictive method. The solid line is a
1:l line and denotes the location on the graph where the measured and calculated
conductivities are equal. It is apparent from this figure that the hydraulic conductivity
shows considerable variation around the 1:l line. Except for a few outliers, most of the
values fall within one to three orders of magnitude from the 1:l line in the wet region,
and within three to five orders of magnitude in the dry range.
276 Yates et al.

lo+ k c"d "nd n 'llld

10-~10~10-710-610-s10--410-'10--210--)100 10'

Measured Conductivity

Fig. 1. Scatter diagram of measured versus estimated values for the unsaturated
hydraulic conductivity using the predictive method.

Shown in Figure 2 is a scatter diagram when the water retention and unsaturated
hydraulic conductivity curves are fitted simultaneously. The K, parameter for this
example was fixed. It is apparent from this figure and the mean sum-of-deviations given
in Tables 2 and 3 that the calculated unsaturated hydraulic conductivity consistently
underestimates the measured values, and hence, that the estimated values are biased.
Shown in Figure 3 are measured versus estimated hydraulic conductivities using a
simultaneous tit assuming 6 parameters. This figure indicates that there is considerable
improvement in the correspondence between the measured and estimated values for
most of the soils.
Figure 4 shows the results when the predictive method is used but with the
estimated values scaled according to the method described by Luckner er al. [ 1989]. This
method uses one value of the unsaturated hydraulic conductivity measured at a point
near (but not at) saturation as a matching point. This technique improves the
correspondence between the measured and estimated values. Comparing Figures 1 and
4 demonstrate a considerable improvement in the estimated value, especially in the
near-saturated region. and illustrates the difficulty of obtaining accurate values for the
soil hydraulic properties at water contents very close to saturation. In the dryer regions,
the scaling technique Seems to lead to conductivity values which slightly overestimate
the measured values.
Figure 5 demonstrates how the scaling technique can be used to remove the bias
that occurs from using a simultaneous fitting procedure with 5 parameters (see Figure
2). For the data set used in this study, use of a matching point to scale the unsaturated
conductivity removes the bias and improves the estimates near saturation; however, the
matching factor also seems to increase the Scatter in the dry region.
Analyzing Conductivity Predictions with RETC 281

Measured Conductivity
Fig. 6. Measured and estimated conductivitics using a matching point
near saturation (K, variable).

Figure 6 shows that there is little improvement from using a matching point to scale
the unsaturated conductivity if the saturated hydraulic conductivity is one of the
parameters used in the fitting procedure. For these circumstances, the parameter K ,
acts in a similar manner as the matching point and provides a means to linearly
translate the estimated values to more closely correspond to the measured values.
Figure 7 demonstrates the effect of using the parameter, ( (i.e., the exponent in the
tortuosity factor), as the only fitting parameter and with the parameters O,, 19, , a, and
n fiied at values which resulted from the predictive method (see Figure 1). For this
data set, it appears that the estimated values of the conductivity are somewhat
insensitive to the value of the connectivity parameter (. Hence, adjusting this parameter
offers only slight improvement in the correspondence between the measured and
estimated values of the conductivity as compared to the predictive method.
Tables 2 and 3 give several statistics comparing the degree of correspondence
between the measured and estimated conductivity for each of the methods. Tables 2
and 3 report, respectively, the results for the hydraulic conductivity and the log,,
transformed conductivity. The two tables also give an indication of the accuracy of the
estimator in the near saturated region and the dryer region for each of the methods
described in this paper. The mean sum-of-deviation and mean sum-of-squares provides
information on how close the estimates fall to the 1:l line. The other statistics in Tables
2 and 3 give information about the best possible linear regression line through the
estimates.
 Yates et al.

10°
10-l
1o-2

lo-'

lo-'

1o-5
.
1o-6
.
. . .
lo-'

10-a .
10-g l&LudJA ’ a*ld ’ “od
10-g10"10-710-610-510-410--310-210-1100 10'

Measured Conductivity
Fig. 7. Measured and calculated conductivities using the predictive method,
but with adjustment of the connectivity parameter, t.

CONCLUSIONS

RETC has been successfully used to analyze 36 water retention curves from 23
different soils. The nonlinear least-squares optimization technique employed by RETC
offers an attractive and efficient method for predicting the unsaturated hydraulic
conductivity from water retention data when measured conductivities are not available.
For the data set of this study, the use of a matching point improved the correspondence
between measured and calculated conductivity. When used in combination with RETC's
predictive capabilities offers the most accuracy with the minimum cost since only one
value for the unsaturated hydraulic conductivity need be known. Trying to tit measured
and calculated values using only the parameter ( as a fitting parameter did not improve
the results significantly. For the data set used here, the results seem to be insensitive
to this parameter.

REFERENCES

Leij, F. J.. M. Th. van Genuchten, S. R. Yates, W. B. Russell. and F. Kaveh. 1992.
RETC: A computer program for analyzing soil water retention and hydraulic
conductivity data. p. 263-272. In M. Th. van Genuchten. F. J. Leij, and L J. Lund
(ed.) Proc. Int. Workshop on Indirect Methods for Estimating the hydraulic Properties
of Unsaturated Soils. University of California, Riverside, CA.
Luckner, L. M. Th. van Genuchten, and DR. Nielsen. 1989. A consistent set of
parametric models for the flow of immiscible fluids in the subsurface. Water Resour.
Res. 25:2187-2193.
Analyzing Conductivity Predictions with RETC 283

Mualem, Y. 1976a. A new model for predicting the hydraulic conductivity of

unsaturated porous media. Wafer Resour. Res. 12:5 13-522.
Mualem, Y. 19766. A catalogue of the hydraulic properties of unsaturated soils.
Research Project 442 100 pp. Technion Israel Inst. Tech., Haifa, Israel.
van Genuchten, M. Th. 1980. A closed-form equation for predicting the hydraulic
conductivity of unsaturated soils. Soil Sci Soc. Am J. 44:892-898.
van Genuchten, M. Th., and D. R. Nielsen. 1985. On describing and predicting the
hydraulic properties of unsaturated soils. Ann. Geophys. 3:615-628.
van Genuchten, M. Th., F. J. Leij and S. R. Yates. 1991. The RETC code for
quantifying the hydraulic functions of unsaturated soils. EPA/600/2-91/065. 93 pp.
R. S. Kerr Environ. Res. Lab., U.S. Environmental Protection Agency, Ada, OK
Yates, S. R.. M. Th. van Genuchten, A. W. Warrick and F. J. Leij. 1992. Analysis of
measured, predicted, and estimated hydraulic conductivity using the RETC computer
program. Soil Sci Soc. Am J.56:347-354.