Journal of Hydrology 233 (2000) 7285

www.elsevier.com/locate/jhydrol

Numerical simulation of inltration, evaporation and shallow

groundwater levels with the Richards equation
J.C. van Dam*, R.A. Feddes
Department of Environmental Sciences, Wageningen Agricultural University, Nieuwe Kanaal 11, 6709 PA Wageningen, The Netherlands
Received 9 August 1999; received in revised form 13 January 2000; accepted 1 March 2000

Abstract
Analysis of water and solute movement in unsaturatedsaturated soil systems would greatly benet from an accurate and
efcient numerical solution of the Richards equation. Recently the mass balance problem has been solved by proper evaluation
of the water capacity term. However, the Darcy uxes as calculated by various numerical schemes still deviate signicantly due
to differences in nodal spacing and spatial averaging of the hydraulic conductivity K. This paper discusses a versatile, implicit,
backward nite difference scheme which is relatively easy to implement. Special attention is given to the selection of a head or
ux controlled top boundary condition during the iterative solution of the Richards equation. The stability of the scheme is
shown for extreme events of inltration, evaporation and rapidly uctuating, shallow groundwater levels in case of two strongly
non-linear soils. For nodal distances of 5 cm, arithmetic means of K overestimate the soil water uxes, while geometric means
of K underestimate these uxes. At smaller nodal distances, arithmetic means of K converge faster to the theoretical solution
than geometric means. In case of nodal distances of 1 cm and arithmetic averages of K, errors due to numerical discretization
are small compared to errors due to hysteresis and horizontal spatial variability of the soil hydraulic functions. q 2000 Elsevier
Science B.V. All rights reserved.
Keywords: Inltration; Modelling; Numerical analysis; Permeability; Richards equation; Unsaturated ow

1. Introduction
Numerical simulation models of water ow and
solute transport in unsaturated soils are important
tools in environmental research and policy analysis.
Many water ow and solute transport problems near
the soil surface can only be solved numerically due to
soil heterogeneity, non-linearity of soil physical
properties, non-uniform root water uptake and rapid
changing boundary conditions. The soil water uxes
as simulated by the numerical models play a key role

* Corresponding author. Fax: 131-317-484885.

E-mail address: jos.vandam@users.whh.wau.nl (J.C. van Dam).

because of their dominant inuence in the hydrologic

cycle, solute transport, heat ow and plant growth.
Water ow in the vadose zone is predominantly
vertical, and can be simulated as one-dimensional
ow in many applications (Romano et al., 1998). By
running the one-dimensional model at various
locations, horizontal variability of meteorological
conditions, crop characteristics, soil properties and
drainage conditions is accommodated and regional
water and solute balances can be determined (Bresler
and Dagan, 1983; Hopmans and Stricker, 1989).
The Richards equation for variably saturated soil
water ow has a clear physical basis. Therefore the
Richards equation is generally applicable and can be
used for fundamental research and scenario analysis.

0022-1694/00/$ - see front matter q 2000 Elsevier Science B.V. All rights reserved.
PII: S 0022-1694(00 )00 227-4

J.C. van Dam, R.A. Feddes / Journal of Hydrology 233 (2000) 7285

Soil hydraulic data which are collected at a great

number of soil physical laboratories (Bruand et al.,
1996; Leij et al., 1996; Wosten et al., 1998) enhance
the applicability of the Richards equation. In the last
two decades, various numerical routines to solve the
Richards equation were developed. However, the
numerical solution of the Richards equation is still a
subject of research. The Richards equation is difcult
to solve because of its parabolic form in combination
with the strong non-linearity of the soil hydraulic
functions which relate water content, soil water pressure head and hydraulic conductivity. Also the abrupt
changes of moisture conditions near the soil surface,
causing steep wetting fronts in dry soils, or causing
steep drying fronts in wet soils, may pose problems.
The result is that calculated soil water uxes may
depend largely on the structure of the numerical
scheme and the applied time and space steps (van
Genuchten, 1982; Milly, 1985; Celia et al., 1990;
Warrick, 1991; Zaidel and Russo, 1992; Baker,
1995; Pan et al., 1996; Miller et al., 1998; Romano
et al., 1998).
Another reason why research continues on the
numerical solution of the Richards equation, is the
computation time needed to achieve accurate solutions for heterogeneous soils with abruptly changing
wetness conditions (Ross, 1990; Pan and Wierenga,
1995; Miller et al., 1998; Berg, 1999). Despite the
rapid advance of computation speed for personal
computers, the computation time may still be excessive for long term simulations, in regional studies or in
case of parameter optimization.
The purpose of this paper is to present a versatile
numerical scheme which is able to solve the onedimensional Richards equation with an accurate
mass balance and which converges rapidly to the
theoretically correct solution. The numerical scheme
is able to handle short duration inltration and runoff
during intensive rain showers as well as simulations
with the time scale of growing seasons. The scheme is
relatively easy to implement. Special attention is paid
to the procedure with respect to the top boundary
which is important for situations with ponded water
layers or with uctuating groundwater levels close to
the soil surface. Experiences with the numerical
scheme and top boundary procedure have been
obtained by many applications of the agrohydrological models SWATRE (Feddes et al., 1978; Belmans et

73

al., 1983), SWACROP (Kabat et al., 1992), and

SWAP (van Dam et al., 1997; Groen, 1997; Smets
et al., 1997). We will describe the numerical scheme
and the top boundary procedure and show the scheme
performance for extreme events of inltration and
evaporation and for rapidly uctuating groundwater
levels near the soil surface.
2. Model
2.1. Discretization of ow equation
Combination of Darcy's law and the principle of
mass conservation results in the Richards equation,
which in the vertical dimension can be written as:



2h
11
2 Kh
2h
2z
2 Sh
1

Ch
2z
2t
where C is the differential water capacity du=dh
(L 21), u is the water content (L 3L 23), h is the soil
water pressure head (L), t is the time (T), K is the
unsaturated hydraulic conductivity (LT 21), z is the
vertical coordinate (positive upward) (L), and S is
the root water extraction (L 3L 23T 21). Both nite
difference and nite element methods are used to
solve Eq. (1) (Feddes et al., 1988; Celia et al., 1990;
Pan et al., 1996). Finite elements are advantageous at
an irregular geometry in 2 and 3-dimensional ow
domains. In one dimension nite difference is advantageous because it needs no mass lumping to prevent
oscillations (van Genuchten, 1982; Pan et al., 1996),
and is relatively easy to conceive and to implement in
numerical routines.
A popular method to solve Eq. (1) has been the
implicit, nite difference scheme with explicit
linearization of K, C, and S, as described by
Haverkamp et al. (1977) and Belmans et al. (1983):
hij11 2 hij

Dt j
j
Ki21=2

!
!
j11
j11
Dhi21=2
Dhi11=2
j
1 2 Ki11=2
11
Dzu
Dz`
Cij Dzi

Sij

Cij

(2)

74

J.C. van Dam, R.A. Feddes / Journal of Hydrology 233 (2000) 7285

where subscript i is the node number (increasing

downward), and superscript j is the time level. The
values of the hydraulic conductivity K and differential
water capacity were taken at the old time level
(explicit linearization). The spatial averages of K
were calculated as geometrical means. The scheme
can be solved without iteration.
Three adaptations to this scheme were made to
arrive at the scheme currently applied in SWAP
(van Dam et al., 1997). The rst adaptation is the
handling of the differential water capacity C. In Eq.
(2) C was put in the denominator of a fraction. As C
equals zero in the saturated zone, this limited the
numerical scheme to the unsaturated zone. The
saturated zone and uctuations of the groundwater
table had to be modelled separately (Belmans et al.,
1983). The numerical scheme was adapted in such a
way that only multiplication with C occurs. In this way
the ow equation can be solved for the unsaturated
and saturated zone simultaneously.
The second adaptation concerned the evaluation of
the C term. Because of the high non-linearity of C,
averaging during a time step results in serious mass
balance errors when highly transient conditions are
simulated. A simple but effective adaptation was
suggested by Milly (1985) and further analysed by
Celia et al. (1990). Instead of applying during a time
step

uij11 2 uij Cij11=2 hij11 2 hij

where Cij11=2 denotes the average water capacity

during the time step, they applied at each iteration
step:
uij11 2 uij Cij11;p21 hij11; p 2 hij11; p21 1 uij11;p21 2 uij

where superscript p is the iteration level and Cij11;p21

is the water capacity evaluated at the pressure head
value of the last iteration, hij11;p21 : At convergence
the term hij11;p 2 hij11;p21 will be small, which
eliminates effectively remaining inaccuracies in the
evaluation of C. Implementation of this mass conservation property requires an iterative solution of
the equation matrix.
The third adaptation concerns the averaging of K
between the nodes. Haverkamp and Vauclin (1979),
Belmans et al. (1983) and Hornung and Messing

(1983) proposed to use the geometric mean. In their

simulations the geometric mean increased the
accuracy of calculated uxes and caused the uxes
to be less sensitive to changes in nodal distance.
However, the geometric mean has serious disadvantages too. When simulating inltration in dry
soils or high evaporation from wet soils, the geometric
mean severely underestimates the water uxes
(Warrick, 1991) or causes convergence problems of
the iterative scheme due to steepening of the wetting
front (Zaidel and Russo, 1992). Other researchers
proposed to use the harmonic mean of K or various
kind of weighted averages (Warrick, 1991; Zaidel and
Russo, 1992; Baker, 1995; Desbarats, 1995; Romano
et al., 1998). In SWAP we selected the arithmetic
mean, which will be argued in Section 4.
The implicit, nite difference scheme of Eq. (2)
including the three adaptations results in the discretization of Eq. (1) which is currently solved in
SWAP:
Cij11;p21 hij11;p 2 hij11;p21 1 uij11;p21 2 uij
"
!
j11;p
Dt j
hi21
2 hij11;p
j
j

Ki21=2
1 Ki21=2
Dzt
Dzu
#
!
j11;p
hij11;p 2 hi11
j
j
2 Ki11=2 2 Dt j Sij
2 Ki11=2
Dz`
5
where Dt j t j11 2 t j ; Dzu zi21 2 zi ; Dz`
zi 2 zi11 ; and Dzi is the compartment thickness. All
the nodes, including the top and bottom node, are in
the centre of the soil compartments. K and S are
evaluated at the old time level j, which can be
shown to give a good approximation at ordinary
time steps of 1026 , Dt j , 0:2 d: Calculations
show that in order to simulate inltration and evaporation accurately, the distance between the nodes should
be in the order of cm's near the soil surface. This
advocates the use of a variable node spacing in the
soil prole. Application of Eq. (5) to each node,
including the prevailing boundary conditions, results
in a tri-diagonal system of equations which can be
solved efciently (Press et al., 1989). Appendix A
lists the equations for the top, intermediate and bottom
nodes, both in case of head and ux type boundary
conditions. The numerical solution of the equation

J.C. van Dam, R.A. Feddes / Journal of Hydrology 233 (2000) 7285

matrix renders the soil water pressure heads in the

unsaturated and saturated zone simultaneously. Starting in the saturated zone, the phreatic groundwater
table is simply found at h 0: Also perched water
tables, which may form above less conductive layers
in the soil prole, are determined in this way.
The convergence criterion used to be a maximum
pressure head difference uhij11;p 2 hij11;p21 u in the
iterative solution of Eq. (5). Huang et al. (1996)
proposed to use the water content difference uuij11;p 2
uij11;p21 u instead. The advantage of a criterion based
on u is that the criterion is more sensitive in pressure
head ranges with a large differential soil water
capacity, while it allows less iterations at low h-values
when soil water uxes are minor. Huang et al. (1996)
show the higher efciency of the u -criterion for a
large number of inltration problems. Moreover the
u -criterion was found to be more robust when the soil
hydraulic characteristics are extremely non-linear.
The u -criterion was implemented in SWAP and our
experiences thus far are very positive. Simulations are
performed in less time, without sacricing mass
balance accuracy. An extra criterion is needed for
saturated conditions. In that case the convergence
criterion switches to maximum pressure head differences between iterations, uhij11;p 2 hij11;p21 u:
The variable, optimal time step should minimize
the computational effort of a simulation. The number
of iterations needed to reach convergence in the
former time step, Nit, can effectively be used to derive
the optimal time step (Kool and van Genuchten,
1991). In case of a large number of iterations, many
calculations are needed to reach convergence, and the
time step is too large. In case of a small number of
iterations, a small number of calculations are needed
for convergence, and the time step can be increased.
We apply the following criteria:
Nit , 2 : multiply time step with a factor 1.25
# Nit # 4 : keep time step the same
Nit . 4 : divide time step by a factor 1.25
SWAP will determine the actual time step using
above criteria in combination with an initial time
step, times at which a signicant change of the precipitation intensity occurs and specied minimum and
maximum time steps. If after 6 iterations no convergence is reached of the numerical solution of the
Richards equation, the time step is divided by 3, and

75

the iterative solution of the Richards equation starts

again.
2.2. Top boundary condition
At moderate weather and soil wetness conditions,
the soil top boundary condition is ux-controlled. In
case of too wet weather or soil conditions, the head of
the water collecting on the soil surface starts to govern
the inltration ux. In case of prolonged dry weather
or soil conditions, the soil water pressure head at the
soil surface adapts to the air humidity and starts to
govern the evaporation ux.
An appropriate procedure for the top boundary
conditions during the iterative solution of the
Richards equation may determine the success or
failure of a numerical scheme. The soil water pressure
heads may change very rapidly near the soil surface.
For instance in case of irrigation or rainfall after a dry
period, the soil water pressure head may increase in a
few minutes from 210 6 to 0 cm. Also when saturated
soils become unsaturated, the pressure head distribution near the soil surface changes rapidly because of
the small differential water capacity near saturation of
most soils. Moreover, the top boundary condition may
switch from head-controlled to ux-controlled and
vice versa during the iterative solution of the Richards
equation.
We dene the soil surface ux qsur (LT 21) in case of
a ux-controlled condition, and the soil surface
pressure head hsur (L) in case of a head-controlled
condition. Soil water uxes are considered positive
when they are directed upward. Fig. 1 shows the
decision procedure which gradually evolved from
applications of the agrohydrological models
SWATRE and SWACROP and which is currently
used in SWAP (van Dam et al., 1997). Criterion k1l
considers if the soil column is saturated. If so,
criterion k2l determines whether at the end of the
time step, the soil column is still saturated or becomes
unsaturated. The inow Qin (L) is calculated as:
Qin qbot 2 qtop 2 qroot 2 qdrain Dt j

where qbot is the ux at the soil prole bottom (LT 21),

qtop the potential ux at the soil surface (LT 21), qroot
the total water ux extracted by roots (LT 21), and
qdrain the total lateral ux to drains or ditches (LT 21).

76

J.C. van Dam, R.A. Feddes / Journal of Hydrology 233 (2000) 7285

Fig. 1. Procedure to select head (hsur) or ux (qsur) top boundary condition. The variables are explained in the text.

where qeva is the actual soil evaporation (LT 21), qprec is

the precipitation rate at the soil surface (LT 21), qirrig is
the applied irrigation ux (LT 21), hpond is the height of
water ponding on the soil surface at t j (L), and Imax is
the maximum soil water ux at the soil surface. The
last term at the RHS of Eq. (7) follows from the
assumption that all ponded water at the soil surface
potentially enters the soil during the next time step.
Imax (LT 21) is calculated according to Darcy as:
!
hpond 2 h1j11;p21 2 z1
8
Imax K1=2
z1

rated and a head condition applies, which is equal to

Qin. If inow Qin is negative, the soil prole becomes
unsaturated and a ux condition applies, which is
equal to qtop.
When the soil column is unsaturated, criterion k3l
determines whether the soil column will remain
unsaturated or becomes saturated during the time
step. The symbol Vair (L) denotes the pore volume in
the soil prole which is lled with air. If the soil
becomes saturated, a head condition applies, which
is equal to Qin 2 Vair : If the soil remains unsaturated,
criterion k4l distinguishes between evaporation
qtop . 0 and inltration qtop # 0: In case of
evaporation, the maximum ux is limited to the maximum ux according to Darcy, Emax (LT 21):
!
hatm 2 h1j11;p21 2 z1
Emax K1=2
9
z1

where height z is positive upward and zero at the soil

surface.
If inow Qin is positive, more water enters than
leaves the soil prole. So the soil prole remains satu-

with hatm the soil water pressure head in equilibrium

with the prevailing air relative humidity (L).
In case of inltration, a head-controlled condition
applies, if the potential ux qtop exceeds the maximum

The potential ux at the soil surface qtop follows from:

qtop qeva 2 qprec 2 qirrig 2

hpond
Dt j

with qtop $ Imax

J.C. van Dam, R.A. Feddes / Journal of Hydrology 233 (2000) 7285
Table 1
Mualemvan Genuchten parameters of the soils considered
Soil

a , (m 21) n (-)

Sand 2.49
Clay 5.32

u res (-) u sat (-) Ksat, (m d 21) l (-)

1.507 0.01
1.081 0.00

0.43
0.55

0.175
0.155

20.140
28.823

inltration rate Imax as well as the saturated hydraulic

conductivity Ksat (criterion k6l). The extra condition of
qtop , 2Ksat stabilized the iterative procedure as Imax
is only a rst order approximation.
During the iterative procedure of calculating hij11;p ;
the top boundary condition is updated at each iteration
p. Appendix A lists the way ux- and head-type boundary conditions are imposed at the rst compartment.
3. Numerical experiment
The performance of the numerical scheme will be
shown for three illustrative cases of extreme conditions at bare soils of sand and clay:
1. Intensive rain at a dry soil. The rainfall rate was
1000 mm d 21 during 0.1 day. The initial conditions
were very dry, with u 0:1 for sand and h
216000 cm for clay. The inltration capacity of
both soils will be exceeded and the main part of
the rainfall becomes runoff.
2. High evaporation at a wet soil. The potential
evaporation amounted 5 mm d 21, the initial h was
2200 cm for both soils. After a while the maximum soil water ux starts to determine the actual
evaporation rate.
3. Groundwater levels uctuating near soil surface.
At 2 consecutive days intensive rain showers of

77

40 mm with a duration of 0.1 d occurred on both

soils with the initial groundwater level at 20 cm
below the soil surface. No runoff was allowed, so
precipitation in excess of the inltration rate accumulated on top of the soil prole. During part of the
day, the groundwater level rose above soil surface.
At the bottom of the soil prole z 240 cm a
constant downward ux of 40 mm d 21 was
adopted. The drainage amount of the total day
was thus equal to the amount of rainfall during
one shower.
The soil hydraulic functions u (h) and K(u ) were
described by the Mualemvan Genuchten model
(van Genuchten, 1980):

uh ures 1

usat 2 ures
1 1 uahun n21=n

Ku Ksat

u 2 ures
usat 2 ures

"
 12 12

10

!l

u 2 ures
usat 2 ures

!n=n21 !n21=n #2
11

where u sat is the saturated water content (cm 0), u res is

the residual water content (cm 0), Ksat is the saturated
hydraulic conductivity (cm d 21) and a (cm 21), n (-)
and l (-) are tting parameters. The parameters are
listed in Table 1 and were derived from the national
Dutch soil catalogue (Wosten et al., 1994). For both
soils, the hydraulic functions are strongly non-linear.
The sand is well-sorted and loses water quickly in the
pressure head range 2200 , h , 220 cm: For the

Table 2
Nodal distance, method of K-averaging and remarks of the simulations performed. For all simulations of the minimum time step was 10 26 d, the
maximum timestep 0.2 d and the convergence criteria uuij11;p 2 uij11;p21 u , 0:0001: The reference case is denoted R
Simulation

Nodal distance (cm)

Averaging K

Remarks

R
S1
S2
S3
S4
S5
S6
S7

0.1
1
1
5
5
1 and 5
1
1

Arithmetic 1 geometric
Arithmetic
Geometric
Arithmetic
Geometric
Arithmetic
Arithmetic
Arithmetic

upper 5 nodes Dzi 1 cm; others Dzi 5 cm

hysteresis (aw 2a )
scaling (coarse 95%)

78

J.C. van Dam, R.A. Feddes / Journal of Hydrology 233 (2000) 7285

Table 3
Cumulative inltration (mm) of the simulations listed in Table 2 for
the case of inltration under intensive rain
Simulation

Sand

Clay

R
S1
S2
S3
S4
S5
S6
S7

39
40
37
47
27
42
32
94

21
23
18
30
13
24
21
79

clay soil, K decreases sharply near saturation. At

saturation, K equals 15.50 cm d 21, while at h
21 cm; K has decreased to 0.73 cm d 21 only. A
small suction extracts water from the macro pores
and causes K to decrease sharply.
In all three cases we varied the nodal distance from
0.1 to 5 cm and applied both arithmetic and geometric
averaging of K, as summarized in Table 2 (simulations S1S4). At simulation S5, only the rst ve
compartments have a thickness of 1 cm, the remaining
compartments are 5 cm thick. The deviations due to
discretization were also compared to the effects of
hysteresis and spatial variability on soil water uxes.
For hysteresis (S6) the scaling concept as applied by
Scott et al. (1983) and Kool and Parker (1987) was
used (van Dam et al., 1997). The main wetting branch
was described by the parameter set (a w, n, u res and
u sat) with aw 2a (Table 1). For spatial variability
(S7) the similar media concept of Miller and Miller
(1956) was followed. Hopmans and Stricker (1989)
derived the distribution of scaling factors of a
650 ha catchment with sandy soils in the Netherlands
Table 4
Cumulative evaporation (mm) of the simulations listed in Table 2
for the case of high evaporation demand
Simulation

Sand

Clay

R
S1
S2
S3
S4
S5
S6
S7

11
11
4
18
1
11
11
7

12
12
11
19
12
12
12
14

(Hupselse Beek). Of this distribution a scale factor

2.25 corresponds to a 95% soil, e.g. 5% of the soils
in the area have a more coarse texture. This scale
factor was used for both the sand and clay soil.
Theoretical solutions of Richards equation are
needed to evaluate the performance of the numerical
scheme. Unfortunately, at the specied top boundary
conditions the Richards equation in combination with
the Mualemvan Genuchten model for the soil
hydraulic functions cannot be solved analytically.
However, we may derive the theoretically correct
solution by the numerical model itself. By decreasing
the nodal distance in combination with a strict convergence criterion, the hydraulic gradient and the
average K will approach their theoretical value. Whatever way of K-averaging is used, simulations
converge to the same solution. This was conrmed
in the simulations considered. The reference solution
R as listed in Table 2 is derived for a nodal distance
Dzi 0:1 cm and a convergence criterion of uuij11;p 2
uij11;p21 u , 0:0001: Smaller nodal distances and a
more strict convergence criterion hardly affect the
simulated soil water uxes. In case of the ne
temporal Dt j < 1026 d and spatial discretization
Dzi < 0:1 cm; the mass balance showed no accumulation of rounding-off errors by the numerical calculations. An independent check of the reference solution
R was achieved by using the detailed nite element
model HYSWASOR (Dirksen et al., 1993) with the
same small nodal distances. Both SWAP and
HYSWASOR showed the same reference solution R.
Note that adopting a more strict convergence
criterion without decreasing the spatial discretization,
is insufcient to derive the reference solution R.
Although a very strict convergence criterion may
result in a perfect mass balance, a large nodal distance
will still cause the calculated Darcy uxes to deviate
from the theoretical Darcy uxes (Milly, 1985;
Warrick, 1991). This discrepancy is caused by the
linear approximation of the hydraulic head gradient
2h 1 z=2z and by the approximation of the spatial
average K between the nodes.
4. Simulation results
Tables 3 and 4 show the simulated amounts of
inltration and evaporation for both the sand and the

J.C. van Dam, R.A. Feddes / Journal of Hydrology 233 (2000) 7285

79

Fig. 2. Inltration rate of sand for simulations R and S1S4 (Table 2) in case of intensive rain at a dry soil, showing the effect of nodal distance
and K-averaging.

clay. We will discuss the results of the sandy soil, as

the differences between the simulations are more
pronounced at this soil. However the simulations
S1S7 show the same trend for the sand and clay soil.
4.1. Intensive rain at a dry soil
At the reference case R, until t 0:008 d; the
hydraulic head gradient at the soil surface is large
enough to absorb the high rain ux of 1000 mm d 21,
as shown in Fig. 2. At t 0:008 d; h at the soil surface
becomes zero and the ux condition is replaced by a
head condition hsur 0:0 cm: Gradually the inltration rate declines, ultimately to the value of Ks
175 mm d21 : The total amount of inltration is
39 mm out of 100 mm of rainfall (Table 3), the
remaining amount is runoff. In general, use of
arithmetic averages results in larger hydraulic
conductivities and thus larger soil water uxes than
use of geometric averages. In case of Dzi 5 cm;
arithmetic averages of K seriously overestimate the
inltration rate (S3:47 mm) while geometric averages
seriously underestimate the inltration rate
(S4:27 mm). The very steep wetting front due to low
geometric K-averages causes inltration rate oscillations at S2 and S4. These oscillations gradually
decrease when smaller nodal distances are used, but
convergence to the nal solution is relatively slow

(Fig. 2). Harmonic means (not shown here) underestimate the mean K at the wetting front and the
inltration rate even more than the geometric mean.
However, in case of arithmetic averages with Dzi
1 cm (S1), the calculated inltration rate is close to
that of the reference (R). To obtain proper results, the
nodal distance needs only to be smaller near the soil
surface, as is shown for S5. Although only the rst 5
compartments have Dzi 1 cm; the cumulative
inltration is 42 mm, compared to 39 mm for reference R (Table 3). The inltration curve of S5 is very
close to that of S1 in Fig. 2. In case of hysteresis (S6),
the inltration decreases from 40 to 32 mm. If we
change the soil texture to a 95% coarse sand (S7),
94 mm inltrates instead of 40 mm! Thus the deviations due to the numerical discretization at Dzi
1 cm and with arithmetic averages of K, are considerably less than the deviations caused by hysteresis
and horizontal spatial variability of soil hydraulic
functions.
4.2. High evaporation at a wet soil
Fig. 3 shows the simulated actual evaporation rate
of sand for R and S1S4. At the reference case R,
initially the potential soil water ux is large enough
to meet the potential soil evaporation rate qsur
5 mm d21 : At t 1:1 d the upper boundary condition

80

J.C. van Dam, R.A. Feddes / Journal of Hydrology 233 (2000) 7285

Fig. 3. Evaporation rate of sand for simulations R and S1S4 (Table 2) in case of high evaporation at a wet soil, showing the effect of nodal
distance and K-averaging.

changes from ux- to head-controlled hsur hatm

and the evaporation rate gradually decreases. After
5 days, the cumulative actual evaporation amounts
11 mm (Table 4), while the cumulative potential
evaporation equals 25 mm. The nodal distance and
the type of K-averaging has a large effect on the
evaporation rate. Similar to inltration, geometric
averaging underestimates the soil water ux, while
arithmetic averaging overestimates the soil water
ux. Choosing a nodal distance of 1 cm, results in
(see Table 4) a cumulative evaporation of 11 mm
for arithmetic averaging (S1) and of 4 mm for
geometric averaging (S2), compared to 11 mm for
the reference (R). The low evaporation in case of
geometric averaging is caused by the low conductivity
for the sand at hatm 21377 m: Also when Dzi is
decreased to 0.1 cm, cumulative evaporation for the
geometric mean still equals 4 mm, compared to
11 mm for arithmetic mean. In this case no convergence was achieved between geometric and
arithmetic averaging, which we attribute to the very
low conductivity of sand at pressure heads near hatm.
Harmonic means of K underestimate the evaporation
rate even more severely than geometric means.
Increase of the nodal distance at larger soil depth is
allowed, as simulation S5 gives the same results as S1.
Hysteresis (S6) hardly affects the evaporation process.
Under conditions of spatial variability (S7), the 95%

coarse sand evaporates 7 mm in stead of 11 mm.

Hysteresis and spatial variability thus affect evaporation uxes relatively less than inltration uxes.
However the effects of hysteresis and spatial variability on simulated evaporation uxes are larger
than the effects of numerical discretization using
Dzi 1 cm and arithmetic averages of K.
4.3. Groundwater levels uctuating near soil surface
Heavy rain (40 mm in 0.1 d) is simulated at a sandy
soil with initial groundwater level at 20 cm depth. Fig.
4 shows the initial inltration rate for 0 , t ,
0:04 day: The way of averaging K and the nodal
distance near the surface (S1S4) had a minor effect
on the inltration rate. At t 0:003 d; Imax becomes
less than 400 mm/d. Therefore at the upper boundary
the ux-type condition qsur 400 mm=d is replaced
by a head-type condition hsur 0: At t 0:014 d;
the soil prole becomes saturated. At that moment a
ponding layer starts and the inltration rate declines
sharply to the soil prole bottom ux (40 mm d 21).
Also hysteresis (S6) had hardly effect on the inltration pattern, in contrast to the 95% coarse soil (S7). At
S7 the inltration rate is maximal until t 0:024 d;
when the soil is almost saturated. More water is
needed to saturate this more coarse soil, starting

J.C. van Dam, R.A. Feddes / Journal of Hydrology 233 (2000) 7285

81

Fig. 4. Inltration rate of sand for simulations S1S4 and S7 (Table 2) in case of groundwater levels uctuating near soil surface, showing the
effect of nodal distance, K-averaging and soil texture.

from hydrostatic equilibrium with groundwater level

at 20 cm depth.
Fig. 5 shows the groundwater level and inltration
rate of S1 and S7 for the total period of 2 days. The 2
rain showers of 40 mm occur at 0:0 , t , 0:1 d and
1:0 , t , 1:1 d: At t 0:91 (S1) and 0.77 d (S7) the
ponded water of the rst shower has inltrated into the
sand soil. The numerical scheme solved the rapid

pressure head change at the transition from saturated

to unsaturated conditions without problems. Gradually the groundwater levels return to their original
level. In case of the more coarse sand (S7), the
groundwater level rises more slowly and the ponding
height is less as more water can be stored in the soil
prole (Fig. 5). The second day shows the same
pattern.

Fig. 5. Inltration rate and groundwater level of sand for simulations S1 and S7 (Table 2) in case of groundwater levels uctuating near soil
surface, showing the effect of soil texture.

82

J.C. van Dam, R.A. Feddes / Journal of Hydrology 233 (2000) 7285

At the clay soil a similar pattern can be shown. The

water balance was closed for all simulations. The
simulations illustrate that in case of shallow uctuating groundwater levels, the transition from ux to
head controlled and from saturated to unsaturated
soil prole is well simulated under extreme eld
conditions.
5. Concluding remarks
Reliable numerical schemes of the Richards equations are needed in order to use numerical soil water
ow models in research and policy analysis routinely.
This paper discusses an implicit, nite difference
scheme, which applies simultaneously to the unsaturated and saturated zone and which is relatively easy
to comprehend and implement in numerical models.
The convergence criteria, in combination with the
iterative solution of the capacity term as suggested
by Milly (1985) and Celia et al. (1990), result in a
closed water balance.
The simulation results show that nodal distances of
5 cm or larger may seriously over- or underestimate
inltration and evaporation uxes at the soil surface in
case of arithmetic and geometric averages of the
hydraulic conductivity. In two-dimensional problems
we tend to work with relatively large grid sizes. In
such cases the accuracy of calculated upper boundary
uxes requires more attention.
At ne nodal grid, spatial averaging of K with arithmetic means performs better than geometric means.
This supports the use of arithmetic averages in
commonly applied nite element numerical schemes
(Kool and van Genuchten, 1991; Simunek et al., 1992,
1994). However, the soil water uxes near the soil
surface will only be accurate when a nodal distance
of about 1 cm is used.
When using nodal distances # 1 cm and arithmetic
averages of K, the effects of the numerical discretization in extreme inltration and evaporation events are
less than typical effects of hysteresis and spatial
variability. This suggests that in one-dimension
further improvement of the numerical discretization
schemes with various kind of weighting functions
for K is less important than the incorporation of
hysteresis and spatial variability of the soil hydraulic
functions.

Romano et al. (1998) show that inaccuracies due to

K-averaging may also occur at sharp texture transitions within the soil prole. Renement of the nodal
grid near these texture transitions with the described
scheme, or application of the algorithm as presented
by Romano et al. (1998), may improve the simulation
results near these transitions.
A proper procedure for the top boundary condition
during the iterative solution of the Richards equation
may determine the success or failure of a numerical
scheme when simulating eld conditions. The
discussed procedure, allows to calculate accurately
inltration and runoff, reduced soil evaporation and
transitions from saturated to unsaturated soil and vice
versa. The paper shows three illustrative cases with
extreme eld conditions. In addition, Miller et al.
(1998) describe a test case with sudden ponding
water on a relatively dry soil, at which many current
numerical solvers of the Richards equation fail. We
simulated the test case with above procedure and
experienced no problems.
The calculation time should be small if a large
number of simulations are performed, e.g. at long
term simulations, in regional studies or in case of
parameter optimization. The run of an ordinary year
with SWAP using an ordinary PC (Pentium 200 MMX
PC) takes about 15 s. Further reduction of the calculation time might be achieved by transformation of the
soil water pressure head (Pan and Wierenga, 1995) or
by Hermite spline interpolation of tabulated soil
hydraulic functions (Miller et al., 1998).
Appendix A
Set of equations to solve Richards equation including boundary conditions. The implicit, backward,
nite difference solution as discussed in Section 2,
reads (Eq. (5)):
Cij11;p21 hij11;p 2 hij11;p21 1 uij11;p21 2 uij
"
!
j11;p
Dt j
hi21
2 hij11;p
j
j

Ki21=2
1 Ki21=2
Dzi
Dzu
#
!
j11;p
hij11;p 2 hi11
j
j
2 Ki11=2
2 Ki11=2 2 Dt j Sij
Dz`
A1

J.C. van Dam, R.A. Feddes / Journal of Hydrology 233 (2000) 7285

Application of (A1) to each node results in a tridiagonal matrix, for which we may dene the
following coefcients:
32 j11;p 3
2
h1
b1 g1
7
76
6
6
76 h j11;p 7
6a b g
7
76 2
6 2
2
2
7
76
6
76 j11;p 7
6
7
7
6
a
b
g
3
3
3
7
h3
76
6
7
76
6
7
76
6
7
6
76
6
7
76
6
7
76
6
7
76
6
7
76
6
76 j11;p 7
6
7
7
6
an21 bn21 gn21 56 h
7
4
4 n21 5
an
bn
hnj11;p
3
2
f1
7
6
6 f 7
6 2 7
7
6
7
6
6 f3 7
7
6
7
6
7
A2
6
7
6
7
6
7
6
7
6
7
6
7
6
6 fn21 7
5
4
fn
In this Appendix the expressions for the coefcients
a i, b i, g i, and fi are listed for each node and for both
ux and head controlled boundary conditions.
A.1. Intermediate nodes
Rearrangement of (A1) to (A2) results in the coefcients:

ai 2

Dt j
Kj
Dzi Dzu i21=2

bi Cij11;p21 1

gi 2
fi

Dt j
Dt j
j
Ki21=2
1
Kj
Dzi Dzu
Dzi Dz` i11=2
A4

Dt j
Kj
Dzi Dz` i11=2

Cij11;p21 hij11;p21
1

A3

A5

uij11;p21

uij

Dt j j
j
K
2 Ki11=2
2 Dt j Sij :
Dzi i21=2

A6

83

A.2. Top node

A.2.1. Flux boundary condition qsur
The right hand side of (A1) transforms to:
"
#
!
Dt j
h1j11;p 2 h2j11;p
j
j
2qsur 2 K11=2
2 K11=2
Dzi
Dz`
2 Dt j S1j

A7

Rearrangements of (A1) to the rst line of (A2) gives

the coefcients:

b1 C1j11;p21 1
g1 2

Dt j
Kj
Dz1 Dz` 11=2

Dt j
Kj
Dz1 Dz` 11=2

A8
A9

f1 C1j11;p21 h1j11;p21 2 u1j11;p21 1 u1j

1

Dt j
j
2qsur 2 K11=2
2 Dt j S1j :
Dz1

A10

A.2.2. Head boundary condition hsur

The right hand side of (A1) transforms to:
"
!
Dt j
hsur 2 h1j11;p
j
j
K1=2
1 K1=2
Dz1
Dzu
#
!
h1j11;p 2 h2j11;p
j
j
2 K11=2
2 K11=2 2 Dt j S1j
Dz`
A11
Rearrangements of (A1) to the rst line of (A2) gives
the coefcients:

b1 C1j11;p21 1

g1 2

Dt j
Dt j
j
K1=2
1
Kj
Dz1 Dzu
Dz1 Dz` 11=2
A12

Dt j
Kj
Dz1 Dz` 11=2

A13

f1 C1j11;p21 h1j11;p21 2 u1j11;p21 1 u1j

1

Dt j
j
K j 2 K11=2

Dz1 1=2

Dt j
K j h 2 Dt j S1j :
Dz1 Dzu 1=2 sur

A14

84

J.C. van Dam, R.A. Feddes / Journal of Hydrology 233 (2000) 7285

A.3. Bottom node

fn Cnj11;p21 hnj11;p21 2 unj11;p21 1 unj

A.3.1. Flux boundary condition qbot

The right hand side of (A1) transforms to:
"
#
!
j11;p
Dt j
hn21
2 hnj11;p
j
j
Kn21=2
1 Kn21=2 1 qbot
Dzn
Dzu
2 Dt j Snj

Dt j
an 2
Kj
Dzn Dzu n21=2
Dt j
Kj
Dzn Dzu n21=2

A16

A17

fn Cnj11;p21 hnj11;p21 2 unj11;p21 1 unj

1

Dt j
K j
1 qbot 2 Dt j Snj :
Dzn n21=2

(A18)

A.3.2. Head boundary condition hbot

The right hand side of (A1) transforms to:
"
!
j11;p
Dt j
hn21
2 hnj11;p
j
j
1 Kn21=2
Kn21=2
Dzn
Dzu
#
!
hnj11;p 2 hbot
j
j
2 Kn11=2
2 Kn11=2 2 Dt j Snj
Dz`
A19
Rearrangements of (A1) to the last line of (A2) gives
the coefcients:

an 2

Dt j
Kj
Dzn Dzu n21=2

bn Cnj11;p21 1

Dt j
j
K j
2 Kn11=2

Dzn n21=2

Dt j
Kj
h 2 Dt j Snj :
Dzn Dz` n11=2 bot

A22

(A15)

Rearrangements of (A1) to the last line of (A2) gives

the coefcients:

bn Cnj11;p21 1

A20

Dt j
Dt j
j
Kn21=2
1
Kj
Dzn Dzu
Dzn Dz` n11=2
A21

References
Baker, D.L., 1995. Darcian weighted interblock conductivity means
for vertical unsaturated ow. Ground Water 33, 385390.
Belmans, C., Wesseling, J.G., Feddes, R.A., 1983. Simulation of the
water balance of a cropped soil: SWATRE. J. Hydrol. 63, 271
286.
Berg, P., 1999. Long-term simulation of water movement in soils
using mass-conserving procedures. Adv. Water Resour. 22,
419430.
Bresler, E., Dagan, G., 1983. Unsaturated ow in spatially variable
elds. 2. Application of water ow models to various elds.
Water Resour. Res. 19, 421428.
Bruand, A., Duval, O., Wosten, H., Lilly, A. 1996. The use of
pedotransfer in soil hydrology research in Europe. Workshop
Proc., 1012 October 1996, Orleans, France, pp. 211.
Celia, M.A., Bouloutas, E.T., Zarba, R.L., 1990. A general massconservative numerical solution for the unsaturated ow
equation. Water Resour. Res. 26, 14831496.
van Dam, J.C., Huygen, J., Wesseling, J.G., Feddes, R.A., Kabat, P.,
van Walsum, P.E.V., Groenendijk, P., van Diepen, C.A., 1997.
Theory of SWAP version 2.0. Simulation of water ow, solute
transport and plant growth in the SoilWaterAirPlant
environment. Report 71, Department of Water Resources,
Wageningen Agricultural University, Tech. Docu. 45, DLO
Winand Staring Centre, Wageningen, the Netherlands.
Desbarats, A.J., 1995. An interblock conductivity scheme for nite
difference models of steady unsaturated ow in heterogeneous
media. Water Resour. Res. 31, 28832889.
Dirksen, C., Kool, J.B., Koorevaar, P., van Genuchten, M.Th., 1993.
Hyswasorsimulation model of hysteretic water and solute
transport in the root zone. In: Russo, D., Dagan, G. (Eds.),
Water Flow and Solute Transport in soils: Modeling and Application, Advanced Series in Agricultural Sciences, vol. 20.
Springer, Berlin, pp. 99122.
Feddes, R.A., Kowalik, P.J., Zaradny, H., 1978. Simulation of eld
water use and crop yield. Simulation Monographs, Pudoc,
Wageningen, the Netherlands, pp. 189.
Feddes, R.A., Kabat, P., van Bakel, P.J.T., Bronswijk, J.J.B.,
Halbertsma, J., 1988. Modelling soil water dynamics in the
unsaturated zonestate of the art. J. Hydrol. 100, 69111.
van Genuchten, M.Th., 1980. A closed form equation for predicting
the hydraulic conductivity of unsaturated soils. Soil. Sci. Soc.
Am. J. 44, 892898.
van Genuchten, M.T., 1982. A comparison of numerical solutions of

J.C. van Dam, R.A. Feddes / Journal of Hydrology 233 (2000) 7285
the one-dimensional unsaturatedsaturated ow and transport
equations. Adv. Water Resour. 5, 4755.
Groen, K.P., 1997. Pesticide leaching in polders. Field and model
studies on cracked clays and loamy sand. PhD thesis, Agricultural University, Wageningen, the Netherlands, pp. 295.
Haverkamp, R., Vauclin, M., 1979. A note on estimating nite
difference interblock hydraulic conductivity values for transient
unsaturated ow problems. Water Resour. Res. 15, 181187.
Haverkamp, R., Vauclin, M., Touma, J., Wierenga, P.J., Vachaud,
G., 1977. A comparison of numerical simulation models for
one-dimensional inltration. Soil Sci. Soc. Am. J. 41, 285294.
Hopmans, J.W., Stricker, J.N.M., 1989. Stochastic analysis of soil
water regime in a watershed. J. Hydrol. 105, 5784.
Hornung, U., Messing, W., 1983. Truncation errors in the numerical
solution of horizontal diffusion in saturated/unsaturated media.
Adv. Water Resour. 6, 165168.
Huang, K., Mohanty, B.P., van Genuchten, M.Th., 1996. A new
convergence criterion for the modied Picard iteration method
to solve the variably saturated ow equation. J. Hydrol. 178,
6991.
Kabat, P., van den Broek, B.J., Feddes, R.A., 1992. SWACROP: A
water management and crop production simulation model. ICID
Bulletin 92 (41), 6184.
Kool, J.B., Parker, J.C., 1987. Development and evaluation of
closed form expressions for hysteretic soil hydraulic properties.
Water Resour. Res. 23, 105114.
Kool, J.B., van Genuchten, M.Th., 1991. HYDRUS, Onedimensional variably saturated ow and transport model
including hysteresis and root water uptake. Research Report
124, US Salinity Laboratory, USDA, ARS, Riverside, CA.
Leij, F.J., Alves, W.J., van Genuchten, M.Th., Williams, J.R., 1996.
The UNSODA unsaturated soil hydraulic database. User's
manual version 1.0 Soil Salinity Laboratory, Riverside, CA.
Miller, E.E., Miller, R.D., 1956. Physical theory for capillary ow
phenomena. J. Appl. Phys. 27, 324332.
Miller, C.T., Williams, G.W., Kelly, C.T., Tocci, M.D., 1998.
Robust solution of Richards equation for nonuniform porous
media. Water Resour. Res. 34, 25992610.
Milly, P.C.D., 1985. A mass conservative procedure for timestepping in models of unsaturated ow. Adv. Water Resour. 8,
3236.

85

Pan, L., Wierenga, P.J., 1995. A transformed head-based approach

to solve Richards equation for variably saturated soils. Water
Resour. Res. 31, 925931.
Pan, L., Warrick, A.W., Wierenga, P.J., 1996. Finite elements
methods for modelling water ow in variably saturated porous
media: numerical oscillation and mass distributed schemes.
Water Resour. Res. 32, 18831889.
Press, W.H., Flannery, B.P., Teukolsky, S.A., Vetterling, W.T.,
1989. Numerical Recipes. The Art of Scientic Computing,
Cambridge University Press, Cambridge, p. 702.
Romano, N., Brunone, B., Santini, A., 1998. Numerical analysis of
one-dimensional unsaturated ow in layered soils. Adv. Water
Resour. 21, 315324.
Ross, P.J., 1990. Efcient numerical methods for inltration using
Richards equation. Water Resour. Res. 26, 279290.
Scott, P.S., Farquhar, G.J., Kouwen, N., 1983. Hysteretic effects on
net inltration. Advances in Inltration, ASAE, St. Joseph, MI,
pp. 163170.
Simunek, J., Suarez, D.L., 1994. Two-dimensional transport model
for variably saturated porous media with major ion chemistry.
Water Resour. Res. 30, 11151133.
Simunek, J., Vogel, T., van Genuchten, M.Th., 1992. The SWMS2D code for simulating water ow and solute transport in twodimensional variably saturated media. Version 1.1, Res. Rep.
126, US Salinity Lab., ARS, USDA, Riverside, CA.
Smets, S.M.P., Kuper, M., van Dam, J.C., Feddes, R.A., 1997.
Salinization and crop transpiration of irrigated elds in
Pakistan's Punjab. Agric. Water Mgmt. 35, 4360.
Warrick, A.W., 1991. Numerical approximations of Darcian ow
through unsaturated soil. Water Resour. Res. 27, 12151222.
Wosten, J.H.M., Veerman, G.J., Stolte, J., 1994. Water retentionand conductivity characteristics of topen subsoils in the
Netherlands: the Staring Series. Tech. Docu. 18, DLO Winand
Staring Centre, Wageningen, the Netherlands, in Dutch, pp. 66.
Wosten, J.H.M., Lilly, A., Nemes, A., le Bas, C., 1998. Using existing soil data to derive hydraulic parameters for simulation
models in environmental studies in land use planning. Report
156, Winand Staring Centre, Wageningen, the Netherlands.
Zaidel, J., Russo, D., 1992. Estimation of nite difference interblock
conductivities for simulation of inltration into initially dry
soils. Water Resour. Res. 28, 22852295.