You are on page 1of 28

.

••
ELSEVIER Journal of Hydrology 195 (1997) 172-199
Journal
of
Hydrology

Evaluation of multimodal hydraulic functions in characterizing


a heterogeneous field soil
Dirk Mallants a'*, Peng-H. Tseng b, Nobuo Toride e, Anthony Tirnmerman a,
Jan Feyen a
alnstitutefor Land and WaterManagement, Faculty of Agricultural and Applied Biological Sciences, KU Leuven,
Vital Decosterstraat 102, B-3000 Leuven, Belgium
blnstitate of Geophysics and Planetary Physics, University of California, Riverside, CA 92521-0412, USA
CFaculty of Agriculture, Saga University, 840 Saga, Japan
Received 19 October 1995; revised 4 June 1996; accepted 1 August 1996

Abstract

Soil water retention curves are often used to estimate the hydraulic conductivity function. Unfor-
tunately, single S-shaped functions cannot adequately describe water retention curves of structured
soil, especially near saturation. The approach of superposition of two or more unimodal retention
functions such as the van Genuchten model was used here to describe retention data of a macro-
porous soil. A total of 180 cores, 0.05 m diameter and 0.051 m long, were sampled along a 3 l-m-long
transect in three overlying soil horizons. Use of unimodal retention curves leads to an underesti-
marion of observed water contents both near saturation and in the midpore range, while an over-
estimation is found in the drier range. Superposition of two unimodal retention curves significantly
improved the estimation over the entire pressure range. However, the predictions were still not ideal
near saturation. With three unimodal curves, a perfect fit was obtained from saturation to residual
water content. Most of the multimodal parameter values were moderately heterogeneous along the
transect, with the surface horizon slightly more heterogeneous than the deeper layers. The coefficient
of variation (CV) for muitimodal parameters was generally in the range of 20 to 70%. Use of the
multimodal van Genuchten model with the conductivity estimation model of Mualem resulted in
conductivities that were generally much smaller than those estimated by the classical unimodal van
Genuchten-Mualem model. A preliminary evaluation of the estimated bimodal and trimodal
unsaturated hydraulic conductivity model was based on a comparison with independent conductivity
measurements using a combination of crust test, hot-air method, and an unsteady drainage flux
experiment on large columns. The crust and hot-air data compared best with the estimated trimodal
conductivity function. The unsteady drainage data did not match well with the crust and hot-air data
and could not be described with any of the estimated conductivity functions.

* Correspondingauthor.

0022-1694/97/$17.00 @ 1997- Elsevier Science B.V. All rights reserved


Pll S0022-1694(96)03251-9
D. Mallants et aUJournal of Hydrology 195 (1997) 172-199 173

1. Introduction

The issues of ecosystem and water quality management have received considerable
attention in the literature (e.g. Addiscott and Wagenet, 1985). Numerical simulation
models play a key role in predicting water flow and contaminant transport in variably
saturated soils to assess the impact of current or future land management on the subsurface
environment or to predict chemical migration from disposal sites.
Description of water movement in soil is most often based on Richards' equation which
requires knowledge of the soil water retention characteristic, 0(~b), and the hydraulic
conductivity function, K(~), or K(O), where 0 is the volumetric water content (cm 3 cm -3)
and ~b is the soil water pressure (cm). The former property is fairly easy to measure
whereas the latter is difficult to determine. A great number of mathematical expressions
for both water retention and hydraulic conductivity functions is currently available
(Brooks and Corey, 1964; Mualem, 1976; van Genuchten, 1980; Hutson and Cass,
1987; Rossi and Nimmo, 1994), in which several retention models ave related theoretically
to the conductivity models to estimate the unsaturated hydraulic conductivity from mea-
sured retention data (e.g. Mualem, 1986). A review of indirect estimation methods for K(O)
is given by van Genuchten and Leij (1992) and Mualem (1992). Many of the theoretical
methods are based on the pore-size distribution of a soil. This approach avoids the need for
time consuming and often unreliable measurements, but may itself lead to gross errors
when the functional description of the retention characteristic is nonoptimal. The estima-
tion methods such as the ones discussed here likely will become the only workable means
to characterize hydraulic properties of large areas of land owing to the problem of spatial
variability. Direct measurements of the unsaturated hydraulic conductivity may be
restricted to site-specific flow and transport problems.
Nonoptimal behavior of the soil water retention curve has been reported for aggregated
and structured soils near saturation (Smettem and Kirkby, 1990; Othmer et al., 1991;
Dumer, 1994). Mallants et al. (1996b) demonstrated that the use of nonoptimal retention
and conductivity functions in the simulation of free-drainage from a saturated soil profile
resulted in an underprediction of the cumulative drainage by as much as 30% for a
macroporous soil. In field observations, Othmer et al. (1991) showed that soil water
pressure was much better estimated with bimodal retention and conductivity functions
when compared with unlmodal functions. Furthermore, both field and laboratory experi-
ments (van Genuchten et al., 1990; Poletika and Jury, 1994; Mallants et al., 1994) have
revealed that in aggregated and macroporous soils solutes can bypass parts of the soil
leading to an accelerated transport from soil surface to groundwater. Simulation
models which do not properly address the physics of the subsurface processes will thus
be erroneous or misleading when used to predict contaminant transport in many field soils.
Unfortunately, the traditional theoretical conductivity estimation models have long been
used with little attention to characterize field soils. Since heterogeneity is inherent in any
field soil, there is an urgent need for a systematic investigation and comprehensive
evaluation of the validity and representativeness of using such conductivity estimation
models.
In the present study we evaluated the use of a multimodal van Genuchten retention
function, previously discussed by Durner (1994), in describing a large set of retention data
174 D. Mallants et alJJournal of Hydrology 195 (1997) 172-199

obtained from a macroporous soil. The variability of the estimated model parameters is
discussed for a bimodal and a trimodal form of the van Genuchten retention model. In
addition, estimated unsaturated hydraulic conductivity using a multimodal Mualem con-
ductivity model was evaluated using independent conductivity measurements from crust
test, hot-air method, and an unsteady drainage flux experiment.

2. Model

2.1. Unimodal water retention characteristic

The mathematical formulation of the water retention characteristic, 0(4), has been a
subject of research for the last 30 years (Brooks and Corey, 1964; Rogowski, 1972; van
Genuchten, 1980; Rossi and Nimmo, 1994). One of the most commonly used soil water
retention functions is that proposed by van Genuchten (1980), here expressed in terms of
the effective saturation, Se:

Se=
[1].
1 +(alibi)" (1)

where Se = (0 - 0r)/(0s - 0r), with 0r and 0s the residual and saturated water content
(cm 3 cm-3), respectively, 0 and ~ are as previously defined, c~(cm -l) and n are curve-fitting
parameters, and m is related to n as m = 1 - 1/n. Se is considered as a cumulative distribution
function of a capillary pore-size distribution with density function f(~b) = dSe/d~:
f(~b) = cmm(o~l~bl)n-! (1 + o~1~1")-("+ 1) (2)
Pore-size distributions are visualized using Eq. (1) of Durner (1994), i.e.
d0(~) [logc(10)]l~bld0
d logl01~bl= d~b (3)
and are smooth normal to lognormal functions. Based on a comparison with different
retention models, van Genuchten and Nielsen (1985) concluded that Eq. (1) gave an
excellent description of the retention data for most soils investigated.

2.2. Multimodal water retention characteristic

All the aforementioned retention models assume that the underlying pore-size distribu-
tion is unimodal normal to lognormal shaped. However, undisturbed soils may have very
heterogeneous pore systems, hence may display multiple-peak pore-size distributions
(Durner, 1994). Evidence of heterogeneous pore systems can be found in the retention
data presented by Ragab et al. (1981), Jacobsen (1989), Smettem and Kirkby (1990), and
Othmer et al. (1991). These heterogeneous pore systems may be the result of aggregation
(Sharma and Uehara, 1968) and/or biological processes (De Jong et al., 1992), but have
also been observed in morainic and solifluction soils (Parkes and Waters, 1980).
A better description of retention characteristics of soils exhibiting multiple-peak pore-
size distributions can be obtained by using a multimodal retention function, consisting of a
D. MaUants et alJJournal of Hydrology 195 (1997) 172-199 175
linear superposition of van Genuchten type retention models (Durner, 1994):
[ 1 ]m,
Se = Y. wi (4)
i= l 1 + (o~il~kl)n']
where S~ is as defined previously, k is the number of subcurves, wi are weighting factors
indicating the fraction of the total pore space occupied by each distributionf,(ff), with 0 <
wi < I and Ewi = 1. For each of the subcurves, parameters (¢xi, ni, mi) are defined similar to
the corresponding parameters used in the unimodal curve, i.e. o~ is the inverse of the air
entry value (cm-~), ni represents the slope of the moisture retention curve and we impose
the constraint mi = 1 - llni. The weighting factors wi have to be obtained by parameter
estimation in the same way as the other van Genuchten parameters. The approach of
superimposing two or more pore systems has been previously used by Othmer et al.
(1991), Durner (1992), and Ross and Smettem (1993) to fit retention curves assuming a
multimodal porous system. The effects of macropores on the retention characteristic can
he represented in a physically more realistic way by means of an additional function.

2.3. Conductivity estimations

The hydraulic conductivity function is estimated by the model of Mualem (1976), where
the relative hydraulic conductivity is given as
2
Kr(S¢) =S~" [! dO /
J0' ~(o)]
dO]
(5)
where O is a dummy variable of integration, and ~"is a pore-interaction or tortuosity index
usually taken as 0.5 (Mualem, 1976). The value of 0.5 for r was derived from repacked
soils and should thus be considered as an approximation. From Eq. (5), the absolute
conductivity, Ka(Se) can be calculated by using a reference matching value, K~r, usually
measured at or near saturation where Se = S~f = 1, as
Kr(Se)
Ka(Se) = K~f Kr--~--) (6)
In the case of a multimodal hydraulic conductivity function, Eq. (5) has to be evaluated
using numerical approximation.

3. Materials and methods

3.1. Sampling

Three overlying soil horizons were sampled along a 31-m-long transect in an experi-
mental field located East of Leuven (Bekkevoort), Belgium. The horizons were identified
as Ap (0-0.25 m) and C1 (0.25-0.55 m) of a colluvial material, and a textural B horizon,
the C2 (0.55-1 m). Soil was classified as a well-drained sandy loam (Udifluvent or Eutric
Regosol). In each horizon, 60 undisturbed 0.05 l-m-long and 0.05-m-diameter soil cores
were collected using a Uhland core sampler with an alternating sampling distance of 0.1
176 D. MaUants et al./Journal of Hydrology 195 (1997) 172-199

//TOP VIEW\\
I @ • |

0 crn
Ap
25 cm
C1
55 cm

C2

100 c m
C3
[

Fig. 1. Layout of sampling scheme for collection of 0.051-m, (in all three horizons), 0.2-m- (only in Ap), and
1-m-long columns. For each 0.2-m-long core, there were two 0.05 l-m-long cores (top view).

and 0.9 m (Fig. 1). In the Ap horizon we also collected 0.2-m-diameter and 0.2-m-long
cores whose sampling locations were as close as possible to the sampling locations of the
0.051-m-long cores (see the top view of the sampling locations in Fig. 1). In between the
sampling locations for those small cores, thirty 1-m-long and 0.3-m-diameter columns
were collected. Detailed sampling methods of the large columns were described in
Mallants et al. (1996c).

3.2. Laboratory measurements

For each of the 180 small soil cores we determined saturated hydraulic conductivity, Ks,
using a constant-head permeameter (Klute, 1965), the drying part of the retention curve
using standard desorption techniques (Hillel, 1980), and the bulk density, Pb. Water
contents were measured in the pF 0.0-4.2 (pF = logl0i~b[) range by means of a sand-
box apparatus (for soil water pressures of pF 0.0, 0.5, 1.0, 1.5, and 2.0) and pressure cells
(for soil water pressures of pF 2.3, 2,8, 3.4, and 4.2).
Independent measurements of unsaturated hydraulic conductivities were obtained by
combining the suction crust test and the hot-air method. First, saturated hydraulic con-
ductivity was measured on 0.2-m-long and 0.2-m-diameter cylinders by putting the satu-
rated soil cylinder on a perforated disk to which a filter cloth was attached to prevent soil
from falling out. This assembly was mounted on a funnel from which the outflow could be
measured. One tensiometer was installed at 0.1 m below the soil surface and was
connected to a pressure transducer. Water was ponded on the soil surface to a constant
depth of 0.01 m. Values of Ks were obtained by recording outflow with time when pressure
Was zero.
T h e unsaturated h y d r a u l i c c o n d u c t i v i t y near saturation was subsequently m e a s u r e d on
the same cores using the suction crust test (Bouma et al., 1983). A total of three crusts each
having a thickness of 0.0| m were constructed by smearing a mixture of sand, water, and
quick-setting cement (PROMPT, Grenoble, France) on top of the soil surface. The crusts
had an increasing amount of cement content to reduce its conductivity, i.e. the volume
D. Mallants et alJdournal o f Hydrology 195 (1997) 172-199 177

percentage of cement/sand increased from 20/80 to 35/65 and 50/50. To prevent the
cement from flowing into the macropores, a filter paper was attached in between the
soil surface and crust. Immediately after the first crust was dry, water was allowed to
pond on the crust surface to a depth of 0.01 m. When the soil water pressure remained
constant, outflow was again recorded. For measurements with the second and the third
crust, each soil column was put on top of a 0.25-m-diameter and 1-m-long column filled
with uniform fine sand. Prior to the installation of the soil columns, the sand was
thoroughly wetted from the top. When a constant flux is applied to the soil core and the
underlying sand, both will drain from saturation to a condition of steady-state unsaturated
flow. Under such conditions, a unit hydraulic gradient exists in the soil column, and the
flux density is equal to the conductivity associated with the pressure head in the soil
(Klute and Dirksen, 1986). Unlike the classical crust test, crusts were not removed but
installed on top of an old crust. In this way, the initial contact between crust and soil was
not disturbed. As soon as the crust had become hard, a Perspex plate was attached on top of
the soil column. A grease layer between cylinder and plate prevented possible entry of air.
Water was applied to the soil column by means of a feeder burette with an adjustable
mariotte tube inside. The level of the mariotte tube was fixed in such a way that a positive
pressure of 0.003 m was maintained on top of the crust. After finishing the crust experi-
ments, crusts were carefully removed and soil columns were again saturated.
A dye experiment was subsequently carried out to identify and quantify the presence of
macropores in the soil. The determination of the macroporosity, expressed as blue staining
areas as percentage of total cross-sectional area, involved the addition of a methylene blue
dye solution under identical ponding conditions to 24 0.2-m-diameter and 0.2-m-long
columns (Mallants et al., 1996b) and one 0.3-m-diameter and 1-m-long column
(Timmerman and Feyen, 1996).
Following the dye experiment, the hot-air method (Arya et al., 1975) was conducted to
determine soil water diffusivity on 0.045-m-diameter and 0.1-m-long cores taken in
duplicate from the 0.2-m columns at a depth of 0.025 m below soil surface. Prior to the
sampling of the cores, the 0.2-m columns were drained to an equilibrium condition for
approximately 24 h, following the addition of a dye, with the top covered to prevent
evaporation. The equilibrium condition considering all columns corresponded to a mean
pressure of pF 1.84 ( = 1.84 kPa) with a standard deviation of pF 0.22. In total 60 cores
were exposed to a hot-air blower for 11 min. The soil was subsequently sliced into 0.001-
to 0.002-m increments near the exposed end and 0.007- to 0.01-m increments at the
nonexposed end. Weighing of all sections was done before and after oven-drying. Volu-
metric water content of each section was determined assuming a uniform bulk density
along the core.
Soil water diffusivity,D(O), as a function of water content, 0, was computed for each
core using the water content distribution following enhanced evaporation (Arya et al.,
1975):
0i
D(0)= 1 ~ [x(O)dO (7)
2tf Offl0 J
0
where tf is evaporation time, x is distance along the soil core, and 01 is initial water content.
178 D. Mallantset al./Journalof Hydrology 195 (1997) 172-199
A rectangular hyperbola model (Anderson and Cassel, 1986) was fitted to the x vs 0 data
for each core using least squares optimization:
O=x/(Ax+B)+C (8)
where A, B, and C are fitting parameters. The inverse of Ex1. (8) was determined and
subsequently integrated and differentiated analytically to solve Eq. (7).
From the soil-water diffusivity as a function of water content, D(0), we calculated the
unsaturated hydraulic conductivity, K(O), by multiplying D(0) by the soil-water capacity,
C(O)= d0/d~b, at a given water content. The slope of the soil water retention characteristic,
d0/d~, was first obtained from the unimodal van Genuchten function. Since each time two
0.051-m cores were taken adjacent to the 0.20-m columns, the mean van Genuchten
parameters were computed and used to calculate C(O) for each of the two 0.10-m-long
cores collected from within the 0.20-m columns. The calculations based on the unimodal
C(O) do not reflect the presence of a heterogeneous pore system and its effects on the
hydraulic conductivity function. To assess the effect on the calculated hydraulic con-
ductivity by using either a capacity function derived from a unimodal or a multimodal
retention curve, a second approach was adopted in which D(O) was multiplied by the slope
of a trimodal retention curve.
A third method for estimating the unsaturated hydraulic conductivity comprises simul-
taneously measuring the pressure head and water content through a detailed unsteady
drainage-flux experiment (Green et al., 1986), generally used as an in situ field method.
For this purpose, 1-m-long undisturbed soil columns sampled every meter along the
transect were used. An end cap assembly, consisting of a perforated PVC plate with a
nylon-type cloth with an air entry value of 70 mbar on top, was attached to the bottom of
each column. Columns were saturated from the bottom and allowed to drain during the
measurements until outflow became negligible. The top end of the columns was covered to
prevent evaporation, while at the bottom a constant suction of 20 mbar was applied. Time
domain reflectometry (TDR) probes were installed at six different depths (0.05, 0.15, 0.30,
0.45, 0.60, and 0.80 In). The two-rod TDR probes, each 0.25 m in length, were connected
to a 90-channel computer-controlled TDR system for automatic monitoring of soil water
content (Heimovaara and Bouten, 1990). Tensiometers were installed at precisely the
same depths but at an angle of 90* relative to the TDR probes. The soil water pressure
was measured by inserting a hypodermic needle into a rubber membrane that was mounted
on an air-filled reservoir. The needle was connected to a pressure transducer that converted
the pressure to a digital value on a LCD screen (Thies-Clima, Germany). During the first
7 h of the experiment, the water content and pressure head were measured every half hour.
The time interval between measurements was subsequently increased as the drainage
process progressed.
The drainage flux method assumes one-dimensional flow which can be described by
Richards' equation:

#0 #-~[K(~b)(~x
if- 1)] (9,
#t
where t is time and x, K(~b), and ~b are as previously defined. The working equation
of the detailed drainage flux method may be found by time integration of Richards'
D. Mallants et al./Journal of Hydrology 195 (1997) 172-199 179

equation:
Zl

~ fOdz=K(O) ~-~-[%b(Z,t)+Z]Jz, (10)


J OZ
0

where z l is an arbitrary depth in the soil, and a zero flux condition has been imposed at the
soil surface (z = 0). When measurements of 0 and ~b are available at different depths and
times during the drainage experiment, Eq. (10) can be solved for K(O) at various depths in
the soil profile.
Parameters of the unimodal and multimodal retention functions are obtained by mini-
mizing the objective function
N
Z(P) = j~ c0j(0y-0(~j, p))2 (11)
in which P = {0s, 0 , c~, ni, mi, wi} r is the parameter vector, N is the number of 0(~) data
values, c0t are the weighting factors for the least squares optimization, and 0r and O(~bi,P)
are the measured and estimated water contents at pressure ~b~,respectively. Because we
assumed all data values were of equal quality, the weighting factors o:i for the optimization
were put equal to one. In addition, the parameter 0s was put equal to the largest measured
water content. In the current study the nonlinear parameter estimation code Soil Hydraulic
Properties FITting (SHYPFIT) (Durner, 1993) was used to obtain parameter values for the
uni-, bi-, and trimodal van Genuchten model (Eq. (4)) for each of the 180 measured
moisture retention characteristic curves. Statistical parameters were calculated for the three
sets (for three soil layers) of hydraufic properties (0, 0s, oq, hi, wi) where i is 1, 2, or 3
depending on the number of subcurves. In addition to the traditional van Genuchten para-
meters o~ and n~, the parameters w~ are also site-specific soil hydraulic parameters and may
thus be subjected to spatial variability. Spatial correlation along the transect was examined for
pore space parameters wt, w2, and w3 and for cq and ni for all three horizons by constructing
the experimental semivatiogram and subsequent fitting of a theoretical model.

4..Results and discussion

4.1. Multimodal hydraulic functions

A comparison between fitted uni-, bi-, and trimodal retention curves is shown in Fig. 2
for six measured retention curves which best represent the different types of retention
behavior in the soil profile. Optimized parameter values for Eq. (4) are given in Table 1.
Most of the measured retention data show a rapid change in water content as a function of
pF near saturation (except samples b and e). The presence of a considerable number of
macropores in our soil as demonstrated previously by Mallants et al. (1994) in a methylene
blue dye experiment has caused this rapid change phenomenon. In the pore range of pF 1-
2, a second 'jump' in the retention curve is often observed (samples c, d, e, and f). This
behavior may be attributed to aggregation processes typical for loamy soils (Smettem and
Kirkby, 1990). Since the data does not show a simple sigmoidal curve, several subcurves
are usually required to obtain a better fit. Although in a few cases (samples b, c, and f) the
180 D. MaUants et alJJournal of Hydrology 195 (1997) 172-199

0°5 0.5
.J unlmodll
• - ........ blmodd "
0.4-- .-. 04

0.3 0.3

0.2 ~ ~ . ~ _ a) 0.2 \ d)

0.1 .......... 0.1

0.0
' I ' I ' ' i ' I ' I o.o ' I ' I ' I ' I ' I ' I
1 2 3 4 5 6 0 1 2 3 4 5 6

0.5 - - 0.5
°

CD 0.4 0.4

0.3 0.3

.- 0.2 0.2

0.1 0.1

0.0
' I , I , I , I , I , I 0"0 '1'1' I' I'1'1
0 0 1 2 3 4 5 6
1 2 3 4 5 60.5_
0.5--

0.4

0.3--
c)
0.2 0.2--
I
0.1

0.0 , i , i , i , i , i , i °.° ' I ' I ' I ' I ' I ' I


0 1 2 3 4 5 6 0 1 2 3 4 5 6
Soil w a t e r pressure, pF Soil w a t e r pressure, pF

Fig. 2. Representative measured and estimated retention curves for Ap (a through c) and CI (d through f) horizon
using uni-, bi-, and trimodai van Oenuchten models.
D. Mallants et alJJournal of Hydrology 195 (1997) 172-199 181

Table 1
Estimated parmnete~for typicalretention curves
a b c d e f

Unimodal
0~ (cm3cm-3) 0.4106 0.3809 0.4476 0.3714 0.3309 0.3426
Or (cm3 cm-3) 0.0204 0.0577 0.0442 0.0000 0.0673 0.0260
(cm-I) 0.0033 0.0055 0.0125 0.0124 0.0016 0.0054
n 1.6323 1.9279 1.5232 1.3048 2.1235 1.3718

0s" (crn3 cm-3) 0.460 0.383 0.461 0.398 0.356 0.365


0r (cm3 cm-3) 0.061 0.063 0.076 0.038 0.061 0.072
wi 0.747 0.450 0.350 0.290 0.530 0.5 ! 0
c~i (cm-I) 0.0015 0.0026 0.0006 0.0006 0.0008 0.0014
n, 2.663 2.321 8.223 5.283 4.545 10.00
w2 0.253 0.550 0.650 0.710 0.470 0.490
c~2 (cm-I) 0.500 0.0085 0.103 0.050 0.050 0.223
n2 1.633 2.060 3.654 1.313 1.287 1.204

Trimodal
0,' (cm3cm-3) 0.460 0.383 0.461 0.398 0.356 0.365
Or (cm3 cm-3) 0.065 0.055 0.078 0.058 0.036 0.081
w, 0.543 0.667 0.365 0.350 0.480 0.490
o~, (cra-I) 0.001 0.005 0.0008 0.0006 0.0006 0.0012
nl 3.367 2.175 5.102 5.314 7.105 6.010
w2 0.293 0.322 0.550 0.540 0.450 0.450
o~2(crn-I) 0.0078 0.0076 0.009 0.0223 0.0375 0.0424
n2 2.020 1.553 4.789 1.461 1.200 1.308
w3 0.163 0.0!0 0.085 0.110 0.070 0.060
o~3(era-I) 0.459 0.648 0.411 0.568 0.463 0.423
n3 5.296 10.00 4.908 3.668 1.358 15.00
• 0~ was fixed during optimization.

bimodal van Genuchten model gives a fairly good fit from saturation to pF 4.2, it fails to
describe the j u m p both near saturation and in the midpore range for other cases (i.e.
samples a, c, and d). For these cases, a trimodal curve would better describe the multiple
j u m p phenomena observed in laboratory measurements.
Pore'size distributions based on Eq. (3) pertaining to the fitted retention curves of Fig. 2
are illustrated in Fig. 3. It is evident that completely different pore systems are represented
when using different models. U n i m o d a l distributions are fairly similar in the location of
the m a x i m u m pore-size density ranging from pF 1.9 (sample c) to pF 2.8 (sample e). Pore-
size distributions based on the bimodal retention curve display a secondary (structural)
pore system at the left side of the primary (textural) pore system. The secondary pore
system is due to the effect of mesopores and aggregation. Although the samples come from
the same transect, the different shapes of the curves suggest the existence of a variety of
pore-size distributions. This is also true in the case when a trimodal retention curve is used.
In addition to a secondary pore system and a fine-textured pore system around pF 3, the
trimodal system has almost always a p o r e class near saturation caused by macropores.
182 D. Mallants et alJJournal of Hydrology 195 (1997) 172-199

0.5 -- 0.5 N

0.4 ...... 22
O.4

0.3

0.2
a) 0.3
0.2 '
L d)
0.1 0.1 / "~

0.0
'1'1 ' I ' I' I ' I 0.0 I ' I ' I ' I ' I ' I ' I
0 1 2 3 4 5 6 0 1 2 3 4 5 6

0.5-- 0.7 --

0.6
0.4
0.5
"0 0.3 b) 0.4

"~ 0.2 0.3

0.2
nO 0.1
0.1

0.0 A o.o
' 1 ' 1 ' I'1 ' I ' I
0 1 2 3 4 5 6 0 1 2 3 4 5 6
0.7-- 1.0--
; 0.9
0.0
-

1 0.8
-_
0.5 ,, 0.7
. $1
" 0.6 .
0.4

0.3--

0.2--
,
i \
IL 1
i
0.5
0,
0.3 .
"
~ 0.2 -
0.1
o.1 "2
0.0
, I , I , I , I , I--I °'° ' 1 ' 1 ' 1 ' 1 ' '1
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Soil w a t e r pressure, pF Soil water pressure, pF
Fig. 3. Typical pore-size distributions for uni, bi-, and trimodal retention curves, calculated from d#(~)/d log ,o1@1.
D. MaUants et al./Journal of Hydrology 195 (1997) 172-199 183

0o6
C2-horlzon /

y
0.5 trimedal . j

0.4
0.3

0.2

0.1
f 54o aaTA ~ : ~ a
0.0 '
t I, I, I, I, I, I
0.0 0.1 0.2 0.3 0.4 0.5 0.6

0°6 m

y
i 0.5
•- 0.4

i 0.3
--! 0.2

f 540DATAPOINT8
0.0
'1'1'1'1'1'1
0.0 0.1 0.2 0.3 0.4 0.5 0.6

0.6

0.5--
unimodal
0.4--
r 2 -0.848
0.3

0.2

0.1
540DATAPOINT8
0.0
~' I' I' I' I' I' I
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Measured water content
Fig. 4. Esfinlated versus measured soil water content using uni-, bi-, and trimodal retention curves for C2 horizon.
184 D. Mallants et alJJournal of Hydrology 195 (1997) 172-199

Water desorption methods for measuring pore-size distributions are a commonly used
technique. To what degree such retention-based pore-size distributions reflect the true
underlying pore-size structure can be evaluated by comparison with pore-size distribu-
tions obtained from mercury intrusion porosimetry measurements (Ragab et al., 1982;
Danielson and Sutherland, 1986). More promising measurement techniques include com-
puterized tomography (CT) (Crestana et al., 1986) and nuclear magnetic resonance (NMR)
(Liu et al., 1994). Although the comparison of the pore-size distributions derived from the
retention curve with more direct techniques is beyond the scope of this paper, further
investigation is needed to validate multimodal retention functions in terms of pore size
distributions.
The overall performance of the multimodal retention curves was evaluated by compar-
ing the measured water content with the estimated water content for all soil samples
pertaining to a given soil horizon. Fig. 4 shows, for the C2 horizon, scatter plots for
uni-, bi-, and trimodal retention curves, together with the line of perfect agreement.
Similar results were obtained for the other two horizons (results not shown). The predic-
tions improve when more subcurves are included, with the coefficient of determination r 2
increasing from 0.848 for the unimodal system to 0.991 for the trimodal system. The lower
panel of Fig. 4 clearly shows that the unimodal curves overestimate the water contents
between 0.1 and 0.2 cm 3 cm -3, whereas an underestimation can be found in the ranges of
0.2 to 0.4 cm3 cm-3. Close to saturation, there is a tendency of either overprediction or
underprediction, although the later is more pronounced. Introducing a secondary pore
system (Fig. 4 (middle)) results in an excellent agreement in the dry range, and an
improved description in the wet range. Using a trimodal retention curve gives a perfect
fit over the entire range of water contents.

Table 2
Statistical parameters for unimodal van Genuchten function

Mean Variance CV (%)

Ap
0r (cm 3 cm -3) 0.040 5 × 10-4 57.8
0~ (cm 3 cm -3) 0A20 9 × 10-4 7.2
a (cm -I) 0.007 9 × 10 -e 45
nI 1.754 0.153 22

CI
0r (cm 3 cm -3) 0.012 4 x 10-4 156.4
0~ (cm 3 cm -3) 0.360 4 x 10-4 5.1
c~ (cm -I) 0.013 4 x 10 -s 47
n° 1.386 0.024 11

C2
8r (era 3 a-n -3) 0.044 6 × 10 -4 54.9
e~ (cm 3 crn -3) 0.430 0.001 7.6
a (cm -I) 0.0038 0.002 53
na 1.788 0.094 17

• Lognormally distributed.
D. Mailants et alJJournal of Hydrology 195 (1997) 172-199 185

Statistical parameters for the unimodal as well as the bimodal and trimodal retention
parameters are presented in Tables 2 and 3, respectively. As was already observed from
Fig. 2, there exist a variety of shapes for the bimodai as well as for the trimodal retention
curves. As a result, optimized parameter values change considerably along the transect.
Based on CV values of optimizedparameters for the unimodal case Gable 2), highest

Table 3
Statistical parameters for muitimodal van Gennchten function (0, was fixed during optimization)

Mean Bimodal CV (%) Mean Trimodal CV (%)


Variance Variance

0, (cm 3 cm -3) 0.423* 0.0009 7.1 0.423* 0.0009 7.1


0r (cm 3 cm -3) 0.057 b 0.00034 32.2 0.053 b 0.0006 47.6
w, 0.498 • 0.028 33.9 0.353* 0.007 24.7
6, (cm -I) 0.0023' 4 x I0 "6 88 0.004" 2 x I0 -5 99.2
nI 5.315' 9.47 57.9 5.981 * 9.66 51.9
w2 0.502 b 0.0285 33.6 0.465 b 0.03 37.7
a2 (cm-') 0.0687' 0.0019 20.2 0.0074* 2.8 x 10 -s 99.1
n2 2.163 • 2.515 73.3 2.03' 2.14 72.1
w3 0.181 * 0.583 112.4
c~3 (cm-') 0.303 b 0.109 109.2
r/3 7.008 • 10.69 46.7

CI
0, (cm 3 cm -3) 0.382' 0.00043 5.5 0.382 • 0.00043 5.5
Or (era 3 cm -3) 0.046 b 0.00037 41.5 0.0567 • 0.00057 42.2
wi 0.379 b 0.0137 30.8 0.357 b 0.007 23.3
aa (cm -t) 0.0016" 8.8 x 10 -7 58.9 0.0014" 1.1 x 10 -e 73.9
nj 4.89" 7.78 57 4.877' 7.44 55.9
w2 0.62b 0.0137 18.9 0.557b 0.0056 13.5
62 (cm -I) 0.113" 0.0187 121.7 0.027 b 0.000132 42.2
n2 1.328 ' 0.0177 10 1.489' 0.084 19.4
w3 0.086' 0.0025 57.7
a3 (cm-') 0.386 b 0.04 52.7
n3 7.397 b 14.77 51.9

C2
0, (cm 3 cm -3) 0.46' 0.0012 7.4 0.46' 0.0012 7.4
Or (cm 3 cm -3) 0.0183* 0.00029 93.1 0.047 b 0.0005 46
w, 0.536 b 0.0015 7.3 0.428 b 0.006 17.5
O, (cm -I) 0.0015* !.1 × 10 -7 22.7 0.0012 b 6.6 x 10 4 21.2
n, 3.44 • ! .44 34.9 5.17 • 4.32 40.2
w2 0.464 b 0.0015 8.4 0.492 • 0.0056 15.2
a2 (cm-') 0.263 • 0.02 54.2 0.0154" 8 x I0 -5 60.2
n2 1.163* 0.0026 4.4 1.417 • 0.04 13.9
w3 0.08 • 0.0013 44.6
a3 (cm-') 0.394' 0.003 13.6
n3 7.97 b 8.79 37.2

• Parameter is lognomudly distributed.


b Parameter is normally distributed.
186 D. MaUants et olJJournal of Hydrology 195 (1997) 172-199

variability was found for the residual water content and lowest variability for the saturated
water content. The relatively high variability of 0r is the result of obtaining several zero
values through optimization. The tx parameter is moderately variable whereas the n para-
meter shows a low variability for all horizons. A comparison between the unimodal and
trimodal (Table 3) CV values for curve shape parameters oq and ni pertaining to the
surface horizon (Ap) reveals a much higher variability for the trimodal function. This
indicates that the parameter variabilities obtained by using the unimodal function may be
underestimated because the actual shape of the retention curve is not well represented
leading to fitted curves similar in shape and hence a reduced parameter variability. For the
remaining two horizons (C1 and C2) the differences in variability between unimodal and
trimodal a, i parameters are smaller. In the case of the ni parameter, again lower variability
is observed for the unimodal function in both C1 and C2 horizons.
For the bimodal case (Table 3), mean values for the weighting factors w i and w2 for the
Ap are almost identical, indicating that on average the secondary pore system occupies
50% of the total pore space. Note the larger value for the mean slope parameter for the
primary midrange pore system compared with the secondary system, i.e. n i = 5.315 for the
former versus n2 = 2.163 for the latter. This reflects the steeper slope in the range pF 2 - 3 in

1.0 I "~ 10

c
o.8I!
0.6
0.4
B
8 ~"

0
i
0.2 i •
2
" 0"00 10 20 300
C1

"8 0.2 2
E
u. 0.00 10 20 300
Distance (m)
Fig. 5. Variabilityof estimatedweightingfactorsfor the trimodalpore systemalongthe transectfor the Ap (top),
C 1 (middle),and C2 (bot~m) horizon. Each factorcan be regardedas a fractionof the totalpore spaceoccupied:
w3 (top area), w2 (middle area), and w~ (bottom area). White lines representmeasuredKs (cm day-I) using the
same soil cores.
D. Mallants et alJJournal o f Hydrology 195 (1997) 172-199 187

comparison with the slope for the secondary pore system which lies in the pore range of pF
1-2. Based on the coefficient of variation, CV, largest variability is observed for the ot~
(88% and 20%) and nt (58% and 73%) parameters. For the C1 horizon, the secondary pore
system occupies 60% of the total pore space (w2 = 0.62). Variability is largest for or2 (CV =
122%), oil, (CV = 59%), and nl (CV = 57%). In case of the C2 horizon, pore space is again
almost equally dislributed between main and secondary pore systems. Coefficients of varia-
tion for the remaining parameters are slightly smaller compared with the other two horizons.
Based on the Wilk's W test for normality the majority of the parameters' probability density
function (p.d.f.) is slightly better described by a lognormal than a normal distribution.
The statistical analysis results for the trimodal retention curve parameters are also
summarized in Table 3. For the Ap, an average of 18% (w3 = 0.18) in the total pore
space is taken by the tertiary (macro)pore system (closest to saturation), 47% (w2 = 0.47)
by the secondary pore system (for slightly drier conditions than the former), and 35% (w i =
0.35) by the primary dry-range pore system. Spatial variability of these three pore systems
along the transect is depicted in Fig. 5 for all three horizons. Because the tertiary pore
system can be directly related to macroporosity, the variability of the "3 parameter may be
interpreted as the variability in macropores (top area in each graph in Fig. 5). The varia-
bility in the tertiary pore system (w3) is much larger in the Ap horizon thanin the C1 and
C2 horizon, presumably as the result of the higher macroporosity in the Ap (Fig. 6).
0 ,

lO

2o

3o

4o C1 ~ -

7O
-

9O

11111
' I ' I ' [ ' I ' I '
0 1 2 3 4 5
Macroporosity(%)
Fig. 6. Vertical variability of nmcropor~ity (blue staining areas as % of total area) based on (i) one 0.3-m-
diameter column ( - - O - - ) and (ii) the mean of 24 0.2-m-diameter columns ( - - • - - ) , with error bars indicating
the 95% confidence limits for the mean. Arrows indicate sampling depths of 0.051-m-long cores.
188 D. Mallant$ et alJJ ournal of Hydrology 195 (1997) 172-199

According to the position of the arrows in Fig. 6, sampling of the 0.051-m-long cores
pertaining to the C1 and C2 horizons occurred in a layer with a reduced macroporosity.
The data from both the 0.2-m-long and the 1-m-long columns indicate a decreasing trend
of the macroporosity with depth. Using the concept of the equivalent pore sizes derived
from a capillary model (Bouma, 1991), the macropores visualized by means of the color-
ing dye represent pores that drain in the pressure range 0 to -4.3 cm water ( = pF 0.63
0.43 kPa). Interestingly, the tertiary (macro)pore system depicted in Fig. 3 corresponds to

0.25

e) 0.20
J= a al• a

E
m
0.15 f
.o~ 0.10
1,= 0.135 ÷ 0.05 [I.5(hl4) - 0.S(hl4) $] 0 < h < 4 in

"E
¢/)
0.05 - ~,= 0.14 hal4m

0.00
' I ' I ' I
0.05 --

0.04 --

m
0.03

i 0.02 °

,- 0.01e + 0.015 [1.5(h/3.5) - 0.5(h/3.5) 3] 0 < h < 3.5 in


"~ 0.01 -
U)
,= 0.033 h , 3.5 in

0.00
I ' I ' I
0.03 - -

S 0.02

0
m
.> 0.01
E
y : 0.015 ÷ 0.00711.5 (h/4) - 0.0(h/4) $] 0 < h < 4 m
O0
"(: 0.022 h t'4 m
0.00
' I ' I ' I
5 10 15
Lag distance (m)

Fig. 7. Experimental and fitted spherical semivariogram for fraction of pore space w n (top), w2 (middle), and w3
(bottom) for C2 horizon.
D. Mallants et al./Journal of Hydrology 195 (1997) 172-199 189

a similar pressure range. These observations suggest the validity of w3 as being represen-
tative for the macropore system when derived from a multimodal retention curve (to be
discussed later).
Fig. 5 further reveals that for the Ap horizon, regions of large variability in the tertiary
pore system around 10 and 20 m alternate with regions of low variability, around 5, 15, and
25 m. This apparent periodicity in the w3 parameter was previously also found for Ks
measurements on 0.051-m-long cores (Mallants et al., 1997). The latter data are included
in Fig. 5 as white lines, in an attempt to relate Ks to the pore system parameters w l, w2, and
w3. However, none of these three parameters was found to be correlated with Ks.
In a number of cases, the pore system near saturation (parameter w3) represents as much
as 60% of the total pore space in the Ap horizon. For the pore fraction parameters, largest
variability existed for w3 (CV = 112%) and smallest for wl (CV = 25%). The overall
largest parameter variability was observed for ol3 (CV = 109%). Also Otl and or2 had large
values of CV (99%).
For the C1 horizon, only 9% of the total pore space is occupied by the macropore system
(w3 = 0.086), whereas the other two pore size classes account for 55.7 (w2) and 35.7%
(w t). The spatial variability of the parameters is generally smaller than the corresponding
parameters in the Ap horizon. For example, the overall maximum CV is 74% for the oq
parameter.
Results of the C2 parameters wl, w2, and w3 are similar to those of CI. The hetero-
geneity in the C2 horizon is further reduced resulting in a maximum overall CV of 60%
(for the or2 parameter). For all three horizons an equal number of parameters behaves like a
normal or a lognormal p.d.f.
A geostatistical analysis revealed spatial correlation structure for parameters w ~, w2, and
w3 can be found only in the C2 horizon (textural B) and can be described with a theoretical
spherical semivariogram model as depicted in Fig. 7. The range of spatial correlation was
3.5 to 4 m, which is almost identical to the spatial range of 4 m previously found for
unimodal retention parameters 0s, '~, and n (Mallants et al., 1996a). This indicates that at
least for the C2 horizon macro-, meso-, and micro-pores are all spatially correlated in more
or less the same way. The observed spatial variability of the pore systems may be seen as
the underlying controlling factors which govern the spatial correlation of hydraulic para-
meters ot and n described previously (Mallants et al., 1996a). Spatial variability of intrinsic
factors such as pore-size distributions may be due to specific patterns of extrinsic factors
such as root growth and decay, and earthworm burrows. The spatial variability of macro-
porosity in the Ap horizon was investigated by Mallants et al. (1997) using dye stains to
visualize macropores. They found that the CV for macroporosity at three observation
depths ranged from 65 to 109%, whereas this study showed CV for weighting coefficient
w3 (tertiary pore system) in Ap was 112%. Percentage of macropore area measured at
0.025 m soil depth was positively correlated (r = 0.54) with w3, whereas at the depths of
0.125 and 0.165 m no correlation with w3 was found. These independent data suggest that
at least for the Ap horizon the variability in weighting factor w3 may reflect the spatial
variability in maeropores. However, until more experimental evidence becomes available
on the true nature of the weighting factors, they can be interpreted as shape factors.
Most of the other trimodal parameters did not display spatial correlation. Exceptions
were the n~ parameter for Ap horizon with a spatial range of 8 m, and ,vl and 0~3 for C1
190 D. Mallants et al./Journal of Hydrology 195 (1997) 172-199

1[+0
1Eol \ 1E-1 "~ \ " , \

1E-3 "1~ ~ \ 1E-3 ~1~ "~\ \

1E-6~ ~j \ 1E-6 -i~ ~


1E-7 \ 1E-7 - i ~'t \

0 1 2 3 4 ,5 6 0 1 2 3 4 5 6

i lE+O-m--------~ 1E+O-~.. n ....


1E-1 ~ - ' ~ ' ~ 1E-1 -1~",

•~ 1E-6-~ ~' 1E-6 I ~ \

1E-8 ] , I' [' I ''1'' I' I 1E'e I' I' I' I ;'1 ' I ' I
0 1 2 3 4 5 6 0 1 2 3 4 5 6

IE-I -~ \ \', IE-I-~,L \


~.~-~ ~-~\ 1~.~-~-,,,\ \
1E'3 -I~ \'~'l 1E-3 -~ ~
~1E-5~-~ ~\~
~
c) ~1E-5~-~ "-~\~ ~ I')
1E-6 ~ /! \ 1E-6 -~ ~\
~-~ ,\ ~-~ !\
1E-8 I'1' I' I' I' I' I 1E'e I ' I' I'1" I "1 ' I
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Soil water pressure, pF Soil water pressure, pF
Fig. 8. Estimated unsaturated hydraulic conductivity using Mualem's model with uni-, bi-, and trimodal retention
curvcs.
D. MaUants et alJJournal of Hydrology 195 (1997) 172-199 191

horizon with a range of, respectively, 4 and 3.5 m, when a spherical variogram model was
fitted to the data.

4.2. Estimated hydraulic conductivities

Relative hydraulic conductivities estimated using Eq. (5) are shown in Fig. 8. In
general, the estimated change in conductivity as a function of pressure head for the
unimedal system is considerably smaller than that of the multimodal system. In the wet
range, the differences in estimated conductivity value can be several orders of magnitude.
This is because the unimodal retention curve underestimates the retention data near satura-
tion (as depicted in Fig. 2), and the shape of K(~) is directly related to that of 0(~b) through
Eq. (5). This fast decrease in conductivity as a function of pressure head is a typical pattern
for macroporous soils (Othmer et al., 1991; Chen et al., 1993) and requires a bimodal or
even a trimodal retention curve to describe this behavior. From a strict mathematical point
of view the prediction of K(~b) by means of a statistical pore-size distribution model
(Mualem, 1992) near saturation poses no immediate difficulty, as long as sufficient sub-
curves are included. It also seems appropriate to measure more 0 - ~b data points near
saturation for a better description of the true retention characteristic in the wet range.
However, as was previously pointed out by Bouma (1991), measurements of the moisture
retention curve at pressure heads higher than approximately - 2 0 cm (pF 1.3) for soils with
macropores are usually unreliable. As a result, Bouma concluded that equivalent pore
sizes cannot be accurately estimated for sizes larger than 150 ttm. The dramatic effect of
the shape of the retention curve near saturation on the estimated hydraulic conductivity
indicates (i) the uncertainty involved in using prediction models for K(O) based on the
available water retention data, and (ii) the need for more accurate measurement tech-
niques. A potentially very useful method to circumvent the above problems would be to
use inverse optimization methods applied to outflow data as suggested by Durner (1992).
Finally, whereas in the range close to saturation any estimation of the conductivity char-
acteristic from a retention curve is unreliable, regardless of the chosen modality of the
retention model, the intermediate shape of the conductivity function is probably better
estimated from a multimodal function, which closely fits the measured retention data (to
be discussed later).

4.3. Measured unsaturated conductivity: combined crust and hot-air methods

Mean values and coefficient of variation (between parentheses) for estimated para-
meters of the rectangular hyperbola model, i.e. A, B, and C in Eq. (8) are, respectively,
2.53 (36%), 0.40 (70%), and -0.21 (369%). Unsaturated hydraulic conductivity calculated
from the hot-air method was combined with data from the crust method as depicted in
Fig. 9 for three samples (a, b, and c). The estimated curves for sample b are the same as
those shown in Fig. 8(c). Soil water pressure values used in the crust method ranged from
~b = 0 cm to ~b = - 27 cm. Values of Ks determined previously were used to scale all
conductivities. Note that for each soil column, two measured conductivity curves were
obtained for the hot-air method.
Shown in Fig. 10 is a scatter diagram with hydraulic conductivities measured with the
192 D. Mallants et aL/.lournal of Hydrology 195 (1997) 172-199

1E+O '~$,..;- o.... "~ unimode


-,1-" "%~-"" . o ~o" ="==
1E-2 "~ .,~\ • =am
:I - ~ o ~,,i.~
1E'3 "1-1 = .~k\ " Int./r-3

1E'
IIE-51E.6
4 ~ 0 - -

IE-7 ~
1E..8
0 1 2 3 4 5
1E+O .........

.- 1E-2 -1 ~l:~t~j~

1E-4
i 1E-5 -~
1E-6 -~ \
1E-7 -~ \ b)
1E-8 , , I" , I

0 1 2 3 4 5
1E+O "l~'~.~t..~,~
1E-I"IJ ~ ****o~lk

1E-3 "~
t

1E-4 °

1E~
1E-7 c)
1E-8 , I I I I

0 1 2 3 4 5
Soil water pressure, pF
Fig. 9. Measured and estimated relative hydraulic conductivity of three samplesfor Ap horizon. Hot-air-I uses
unimodal and hot-air-3 uses trimodal capacity function C to calculate K.
D. MaUants et aUJournal of Hydrology 195 (1997) 172-199 193

I['+I _ ,,,,~ "'"I '""~ :"

IE-1

1E-3 1
=

1E-6 -

1E-S - ;' • • ** ' *"

1E-g -

1s-10 Is-0 lse l s 7 ~ s 4 ~ s ~ ~ s 4 ~s-3 ~ - = ~ s l ~,o ~s~l

Unimod~ memumd conductivity


Fig. 10. Unimodal versus trimodal measured conductivities using hot-air method. In the calculation of the K the
soil water capacity C was either unimodal or tfimodal.

hot-air method. The first approach (x-axis) used the water capacity function derived from
an unimodal retention curve, whereas the second approach (y-axis) considered a trimodal
0(~). The solid fine is a 1:1 line. This figure indicates that the hydraulic conductivity falls
within one order of magnitude from the 1:1 line in the wet range and within two to five
orders of magnitude in the dry range. In the dry range, conductivities based on the trimodal
0(~), and hence trimodal C(0), are much lower than the values calculated from an
unimodal 0(~) curve. These findings illustrate the sensitivity of the conductivities to the
functional description of C(O) when hot-air measurements are used. Because the
unimodal 0(~) function gave a poor description of the d Ct~ near saturation, K values
obtained from considering a trimodal 0(~) may he preferred. In Fig. 9, conductivities
from both approaches (indicated as open circles for the former and open triangles for
the latter) are included and illustrate to what differences the two calculation procedures
may lead.
Results from the crust method clearly indicate the quick drop of hydraulic conductivity
near saturation. For instance, in Fig, 9(b), the conductivity drops quickly from 6.2 cm h -~
at saturation to 0.8 cm h -I when the first crust is present. Although saturated conditions
may still exist, with soil water pressures approximately zero, the presence of a crest
significantly reduces fluxes because macropores are partially empty. This behavior was
previously demonstrated by Bouma (1982), who introduced the K(m) notation for this
condition. With higher negative pressures for the second and third crust, the conductivity
further decreases. Comparison between measured and estimated conductivities using the
uni-, bi-, and trimodal Mualem model reveals that generally the best agreement is obtained
for the trimodal case. Especially the intermediate shape of the conductivity function
corresponds relatively well with the data. In addition, the trimodal case can properly
194 D. MaUants et aiJJournal o f Hydrology 195 (1997) 172-199

a)

0.40

1
0"35 I ~
~ 80
6O

40
0.30 ...... A - 1 20

0.25 , I l j 0 i i

0 10 20 30 0 10 20 30

b)
0.45- CO0 1

~ 0.40
_o
80.35
Q
~40
~0.35
| 20
0.25 l I w l O~ 0
0 10 20 30 0 10 20 30
Time ( d a y s ) Time (days)
c)
0.45 100

8O
0.40
60
0.35
40
0.30
~ 20

0.25 l I , i 0 I

0 10 20 30 0 10 20 30
Time (days) Time (days)
Fig. 11. Water content and soil water pressure distributions at selected times in unsteady drainage experiments
used in the calculation of K given in Fig. 9.
D. MaUantset al./Journal of Hydrology 195 (1997) 172-199 195

link the crust and hot-air data by implementing a sudden change of the slope of the curve
between pF 0.5 and 1. Note again that for each conductivity model four hot-air curves are
presented since (i) cores were taken in duplicate, and (ii) conductivities were calculated
considering unimodal and trimodal C(O) functions. The unimodnl model consistently
overpredicts the measured conductivity, with largest deviations occurring between satura-
tion and pF 1.5. The same is true for the bimodal model, although in Fig. 9(a) the agree-
ment is fairly good. The results shown here illustrate the possible errors caused by a
unimodai description and the potential appropriateness of using a multimodal approach
in characterizing hydraulic properties of a heterogeneous soft. It is also shown that large
errors in estimated conductivity may result from using nonoptimal retention curves in
combination with conductivity estimation models.

4.4. Measured unsaturated conductivity: unsteady drainage flux method

The water content and pressure head measured at selected times and two observation
depths in the 1-m-long column are given in Fig. 11. For both depths, the soil water content
decreased rapidly after the initiation of the drainage process, and reached a constant value
after 30 days of drainage. At early times in the drainage process, the pressure head
increases quickly, while the rate of increase slows down as time elapses. The three
cases shown (a through c) in Fig. 11 were used in the calculation of the hydraulic con-
dnctivity and were compared with the crust and hot-air data as given in Fig. 9.
Unsaturated hydraulic conductivity was calculated for the two shallowest observation
depths in the 1-m large columns, i.e. 0.05 and 0.15 In, which are both within the Ap
horizon. Time series of water content, 0, and soil water pressure, ~b, obtained from three
columns were used for this purpose. The columns were chosen such that they were only
0.5 m away from the 0.20-m columns previously used for the crust and hot-air method.
Fig. 9 shows that near saturation the differences in hydraulic conductivity between drain-
age flux and crust method are relatively large. The agreement between the drainage flux
and the hot-air method is relatively poor, with the slope of the former curve turning to the
abscissa at lower pressure values (wetter soil) in comparison with the latter curve. The
typical fast decrease of the hydraulic conductivity near saturation as detected by the crust
method as well as estimated based on the trimodal form of the Muaiem model is not
present when using the drainage flux method. Hydraulic conductivity values obtained with
the unsteady drainage flux method show a fairly smooth decrease, which is in contrast with
the fast decrease revealed by the crust data. This may be due in part to the different
observation scale of both measurement methods relative to the length scale of the macro-
pores. This behavior is most prominent in Fig. 9(b), with the difference between crust and
drainage data being almost two orders of magnitude around pF 1. These findings suggest
that macropores present in the 0.2-m crust columns presumably were continuous through
the core and had drained completely. In contrast, release of water through macropores in
the 1-m-long column was limited because pores were not continuous across the entire
column length (see also Fig. 6). Another possible reason is the relatively large sampling
volume of the TDR probes in monitoring soil water content. The two-rod TDR probes used
in this study were 0.25 m long with an effective sampling diameter of approximately
0.035 cm, which is 1.4 times the spacing between the rods (Topp and Davis, 1985).
196 D. Mallants et alJJournal of Hydrology 195 (1997) 172-199

The soil water content measured by TDR is thus an average over a cylindrical soil volume
of 0.240 x 10 -3 m 3. The presence of any macropore and its effect on the local water content
may thus be masked. Sources of error which may have caused the disparities between the
drainage flux method and the crust method include, among others, the limited numbers of
observations in space and time resulting in inaccurate estimates of the hydraulic gradient
(Tseng and Jury, 1993). This is especially true when pressure increases rapidly at early
times in the drainage process. Furthermore, because 0 and ~b measurements were made at
different locations in the soil column, different volumes of soil were sampled giving rise to
additional errors. Spatial variability may also contribute to errors in K(~b) in the order of
20-30% in the wet range and more than 100% in the dry range (Fltthler et al., 1976). The
above findings clearly illustrate the need for combining several measurement methods in
an attempt to test the validity of conductivity estimation methods. Because the data used to
test the models also contain errors, more accurate measurement techniques are needed
which are able to characterize the detailed behavior of the unsaturated conductivity for a
broad range of soil water pressures.

5. Conclusions

Soil water retention data of a macroporous sandy loam soil were investigated to test the
suitability of applying a multimodal form of the van Genuchten 0(4) model and the
subsequent estimation of hydraulic conductivities with Mualem's K(O) model. A majority
of the measured water retention characteristics exhibited a multimodal shape due to both a
secondary (structural) pore system and a macropore system, in addition to the primary
(textural) pore system. Using a bimodal retention curve improves the description of the
retention data for most of the pressure range, but some underprediction of water content
near saturation is still present. With a trimodal retention curve, the data can be perfectly
described over the entire range of water contents. When the Mualem conductivity estima-
tion model was used in combination with a uni-, hi-, or trimodal van Genuchten retention
model, the differences in conductivity estimations can be up to several orders of magni-
tude especially,in the pressure range near saturation. These results indicate that the use of
nonoptimal retention curves in combination with a conductivity estimation model may
lead to unacceptably large errors in the estimated hydraulic conductivity, hence in any
water flow and solute transport simulations. Attempts to draw conclusions from the simu-
lation results can thus be erroneous or misleading.
All mult/modal retention curve parameters were moderately heterogeneous along the
transect with coefficients of variation generally in the range of 20 to 70%. Higher CV
values (with a maximum of 355% for ~2) were found for trimodal retention parameters
pertaining to the Ap horizon. Furthermore, variability was generally higher for trimodal
than for unimodal parameters, especially in the Ap horizon. These results indicate that the
true field variability may be somewhat underestimated if unimodal retention functions are
used that cannot accurately describe the retention behavior of soils with a heterogeneous
pore system. The heterogeneity of the weighting c~fficient w3 in the trimodal retention
curve seems to reflect the spatial variability of macropores, as was confirmed using
observations of macropores obtained from dye stain patterns.
D. Mallants et al./Journal of Hydrology 19.5 (1997) 172-199 197

Measured unsaturated hydraulic conductivity obtained by combining crust and hot-air


data was best described by a trimodal Mualem model, whereas the bi- and unimodal
models generally overestimated the measurements, especially at near-saturation. The
results suggest that fitting of multimodal van Genuchten retention curves to measured
retention data and subsequent estimation of unsaturated conductivity using a multimodal
form of the conductivity model of Mualem may be a viable approach in current practices
to characterize heterogeneous soils when there are no direct measurements of K(~b) avail-
able. The results also suggest that more accurate and detailed conductivity measurements
are needed to further evaluate the validity of the estimated hydraulic conductivities using
either the unimodal or multimodal approach.
Hydraulic conductivity measurements obtained from a drainage experiment on large
columns did not show the fast decrease of conductivity near saturation which is the typical
pattern for field soils with inherent heterogeneous pore systems. This behavior resulted in
considerable differences with the more accurate crust data. The possible causes to these
disparities may be (i) the volume-averaged moisture content readings obtained by TDR
which may mask the effect of macropores and (ii) the use of 1-m-long columns that are
longer than the length scale of macropores such that their effect on the drainage process is
reduced in comparison with short columns. Since drainage experiments have been widely
used as a data-acquisition technique in field practices, the data obtained from such experi-
ments should be evaluated carefully.

References

Addiscot~ T.M. and Wagenet, RJ., 1985. Concepts of solute leaching in soils: a review of modelling approaches.
J. Soil Sci., 36: 176-184.
Anderson, S.H. and Cassel, D.K., 1986. Statistical and autoregressive analysis of soil physical properties of
Portbsmouth sandy loam. Soil Sci. Soc. Am. J., 50: 1096-1104.
Arya, L.M., Farell, D.A. and Blake, G.R., 1975. A field study of soil water depletion patterns in presence of
growing soybean roots: I. Determination of hydraulic propeRies of soil. Soil Sci. Soc. Am. Proc., 39: 424-
430.
Bouma, J., 1982. Measuring the hydraulic conductivity of soil horizons with continuous macropores. Soil $ci.
Soc. Am. J., 46: 438---441.
Bouma, J., 1991. Influence of soil macroporosity on environmental quslity. Adv. Agron., 46: 1-37.
Bouma, J., Boimans, C., Dekker, L.W. and Jeurissen, WJ.M., 1983. Assessing the suitability of soils with
macropores for subsurface liquid waste disposal. J. Environ. Qual., 12: 305-311.
Brooks, R.H. and Corey, A.T., 1964. Hydraulic properties of porous media. Hydrol. Pap. 3, Colo. State Univ.,
Fort Collins, pp. 1-27.
Chert, C., Thomas, D.M., Green, R.E. and Wagenet, RJ., 1993. Two-domain estimation of hydraulic properties in
macropore soils. Soil Sci. Soc. Am. J., 57: 680-686.
Cresmna, S., Cesareo, R. and Mascarenhas, S., 1986. Using a computed tomography miniscanner in soil science.
Soil Sci., 142: 56-61.
Danieison, R.E. and Sutherland, P.L., 1986. Porosity. In: A. Klute (Editor), Methods of Soil Analysis, Part 1,
Physical and Mineralogical Methods. Agron. Monogr. 9, American Society of Agronomy, Madison, WI.
De Jong, R., Topp, G.C. and Reynolds, W.D., 1992. The use of measured and estimated hydraulic properties in
the simulation of soil water profiles: a case study. In: M.T. van Gennchten, F.J. Leij and LJ. Land (Editors),
Proceedings of the International Workshop, Indirect Methods for Estimating the Hydraulic Properties of
Unsaturated Soils. University of California, Riverside, pp. 569-584.
198 D. Mallants et al,/Journal of Hydrology 195 (1997) 172-199

Dumer, W., 1992. Predicting the unsaturated hydraulic conductivity using multi-porosity water retention curves.
In: M.T. van Genuchten, FT. Leij and L.J. Lurid (Editors), ~ g s of the International Workshop,
Indirect Methods for Kcfimafing the Hydraulic Properties of Unsaturated Soils. University of California,
Riverside, pp. 185-202.
Dumer, W., 1993. SHYPHT User's Manual, Version 0.2. Fakultat fttr Biologie, Chemie und Geowissenschaften
der Universitat Bayreuth, Germany.
Durner, W., 1994. Hydraulic conductivity estimation for soils with heterogeneous pore structure. Water Resoor.
Res., 30:211-223.
HOMer, H., Ardakani, M.S. and Smlzy, L.H., 1976. Error propagation in determining hydraulic conductivities
from successive water content and pressure head profiles. Soil Sci. Soc. Am. J., 40: 830-836.
Green, R.E., Ahuja, L.R. and Chong, S.K., 1986. Hydraulic conductivity, diffusivity, and sorptivity of unsaturated
soils: field methods. In: A. Klute (Editor), Methods of Soil Analysis, Part 1, Physical and Mineralogical
Methods. Agron. Monogr. 9, American Society of Agronomy, Madison, WI.
Heimovaar& TJ. and Bonten, W., 1990. A computer-cont~ulled 36-channel time domain reflectometry system for
monitoring soil water contents. Water Resour. Res., 26:2311-2316.
Hillel, D., 1980. Applications of Soil Physics. Academic Press, New York.
Hutson, J.L. and Cass, A., 1987. A retentivity function for use in soil-water simulation models. J. Soil Sci., 38:
105-113,
JacobSen, O.H., 1989. Unsaturated hydraulic conductivity for some Danish soils, methods and characterization of
soils. Tidsskr. Plunteavi., 93($2030): 1-60 (in Danish).
Klute, A., 1965. Laboratory measurement of hydraulic conductivity of saturated soil. In: C.A. Black et al.
(Editors), Methods of Snil Analysis, Part I. Agronomy, 9: 210-220.
Klute, A. and Dirksen, C., 1986. Hydraulic conductivity and diffusivity: laboratory methods. In: Methods of Snil
Analysis, Part 1, physical and Mineralogical Methods. Agron. Monogr. 9, American Society of Agronomy,
Madison, WI.
Liu, J.W.Y., Wong, S.T.S. and Waldron, L.J., 1994. Application of NMR imaging to study preferential water flow
through root channels. In: S.A. Anderson and J.W. Hopmans (Editors), Tomography of Soil-Water-Root
Processes. SSSA Special Publication no. 36, Madison, WI.
Mallants, D., Vanclooster, M., Meddahi, M. and Feyen, L., 1994. Estimating transport parameters from undis-
turbed soil columns using TDR. J. Contain. Hydroi., 17: 91-109.
Mailants, D., Mohanty, B.P., Jacques, D. and Feyen, L., 1996a. Spatial variability of hydraulic properties in a
multi-layered soil pmlile, Soil Sci., 161: 167-181.
Mallants, D., Jacques, D., Vanclooster, M., Diels, J. and Feyen, J., 1996b. A stochastic approach to simulate water
flow in a macropomus soil. Geoderma" 70: 299-314.
Mallants, D., Vanclooster, M., Toride, N., Vanderborght, J., van Genuchten, M.Th. and Feyen, J., 1996c.
Comparison of three methods to calibrate TDR for solute movement in soil columns. Soil Sci. Soc. Am.
J., 60: 747-754.
Mallants, D., Mohanty, B.P., Vervoort, A. and Feyen, J., 1997. Spatial analysis of saturated hydraulic conduc-
tivity for a soil with macropores. Soil Technoi., 10: 115-131.
Mualem, Y., 1976. A new model for predicting the hydraulic conductivity of unsaturated porous media. Water
Re,sour. Res., 12: 513-522.
Mualem, Y., 1986. Hydraulic conductivity of unsaturated soils: prediction and formulas. In: A. Klute (Editor),
Methods of Soil Analysis, Pan 1, physical and Mineralogical Methods. Agron. Monogr. 9, American Society
of Agronomy, Madison, WI.
Mualem. Y., 1992. Modeling the hydraulic conductivity of unsaturated porous media. In: M.T. van Genuchten,
F.J. Leij and LJ. Lund (Editors), Proceedings of the International Workshop, Indirect Methods for Estimating
the Hydraulic Properties of Uusamrated Soils. University of California, Riverside, pp. 15-36.
Othmer, H., Diekkrttger, B. and Kutilek, M., 1991. Bimodal porosity and unsaturated hydraulic conductivity. Soil
Sci., 152: 139-149.
Parkes, M.E. and Waters, P.A., 1980. Comparison of meusured and estimated unsaturated hydraulic conductivity.
Water Resour. Res., 16: 749-751.
Poletika" N.N. and Jury, W.A., 1994. Effects of soil surface management on water flow distribution and solute
dispersion. Soil Sci, Soc. Am. J., 58: 999-1006.
D. MaUants et alJJournal of Hydrology 195 (1997) 172-199 199

Ragab, R., Feyon, J. and l-fillel, D., 1981. Comparative study of numerical and laboratory methods for
determining the hydraulic conductivity function of a sand. Soil Sci., 131: 375-388.
i t c h , R., Feyen, J. and Hiilel, D., 1982. Effect of the method for determining pore size distribution on prediction
of the hydraulic conductivity function and of infiltration. Soil Sci., 134: 141-145.
Rogowski, A.S, 1972. Estimation of the soil moisture characteristic and hydraulic conductivity: comparison of
models. Soil Sci., 144: 423-429.
Ross, Pj. and Smettem, K.RJ., 1993. Describing soil hydraulic properties with the sum of simple functions. Soil
Sci. Soc. Am. J., 57: 26-29.
Rnssi, C. and Nimmo, J.R., 1994. Modeling of soil water retention from saturation to oven dryness. Water Resonr.
Res., 30: 701-708.
Sharma, M.L. and Uehara, G., 1968. Influence of soil structure on water relations in low humic latosols, I, water
retention. Soil Sci. Soc. Am. Proc., 32: 765-771.
Smettem, K.RJ. and Kirkby, C., 1990. Measuring the hydraulic properties of a stable aggregated soil. J. Hydrol.,
177: 1-13.
Timmerman, A. and Feyen, J., 1996. Measurement of near-saturated conductivity in a nutcroporous soil. In: W.
Durner, J. Halbertsma and M. Ciserova (Editors) Extended Abstracts of the E ~ Workshop on Advanced
Methods to Determine Hydraulic Properties of Soils. Thurnan, Germany, pp. 7-10.
Topp, G.C. and Davis, J.L., 1985. Time-domain reflectometry (TDR) and its applications to irrigation scheduling.
In: D. Hillel (Editor), Advances in Irrigation. Vol. 3, Academic Press, New York, pp. 107-127.
Tseng, P.-H. and Jury, W.A,, 1993. Simulation of field measurement of hydraulic conductivity in unsaturated
heterogeneous soil. Water Resour. Res., 29: 2087-2099.
van Genuchton, M.Th., 1980. A closed-form equation for predicting the hydraulic conductivity of unsaturated
mils. Soil Sci. SUC.Am. J., 44: 892-898.
van Gonuchten, M.Th. and Leij, FJ., 1992. On estimating the hydraulic properties of unsaturated soils. In: M.T.
van Gonuchton, FJ. Leij and L.J. Lurid (Editors), Proceedings of the International Workshop, Indirect
Methods for Estimating the Hydraulic Properties of Unsaturated Soils. University of California, Riverside,
pp. 1-14.
van Gonuchten, M.Th. and Nielsen, D.R., 1985. On describing and predicting the hydraulic properties of unsa-
turated soils. Ann. Geophys., 3: 615-628.
van Genuchten, M.Th, Rolston, D.E. and Gormann, P.F. (Editors), 1990. Transport of Water and Solutes in
Macropores. Genderma (Special Issue), 46.

You might also like