You are on page 1of 21

Advances in Water Resources 24 (2001) 11731193

www.elsevier.com/locate/advwatres

A new surface runo parameterization with


subgrid-scale soil heterogeneity for land surface models
Xu Liang *, Zhenghui Xie

Department of Civil and Environmental Engineering, University of California, Berkeley, CA 94720-1710, USA
Received 13 June 2000; received in revised form 23 December 2000; accepted 30 March 2001

Abstract
Soil heterogeneity plays an important role in determining surface runo generation mechanisms. At the spatial scales represented
by land surface models used in regional climate model and/or global general circulation models (GCMs) for numerical weather
prediction and climate studies, both inltration excess (Horton) and saturation excess (Dunne) runo may be present within a
studied area or a model grid cell. Proper modeling of surface runo is essential to a reasonable representation of feedbacks in the
landatmosphere system. In this paper, a new surface runo parameterization that dynamically represents both Horton and Dunne
runo generation mechanisms within a model grid cell is presented. The new parameterization takes into account of eects of soil
heterogeneity on Horton and Dunne runo. A series of numerical experiments are conducted to study the eects of soil heterogeneity on Horton and Dunne runo and on soil moisture storage under dierent soil and precipitation conditions. The new parameterization is implemented into the current version of the hydrologically based variable inltration capacity (VIC) land surface
model and tested over three watersheds in Pennsylvania. Results show that the new parameterization plays a very important role in
partitioning the water budget between surface runo and soil moisture in the atmosphereland coupling system. Signicant underestimation of the surface runo and overestimation of subsurface runo and soil moisture could be resulted if the Horton runo
mechanism were not taken into account. Also, the results show that the Horton runo mechanism should be considered within the
context of subgrid-scale spatial variability of soil properties and precipitation. An assumption of time-invariant spatial distribution
of potential inltration rate may result in large errors in surface runo and soil moisture. In addition, the total surface runo from
the new parameterization is less sensitive to the choice of the soil moisture shape parameter of the distribution. 2001 Elsevier
Science Ltd. All rights reserved.
Keywords: Rainfall; Surface runo; Subgrid-scale spatial variability; Soil heterogeneity; Land surface models; Hydrologic models;
Soil moisture

1. Introduction
Numerous eld studies have shown that surface
runo is mainly generated by two mechanisms, inltration excess runo (Hortonian ow) and saturation excess runo (Dunne ow); and that the spatial variability
of soil properties, antecedent soil moisture, topography,
and rainfall will result in dierent surface runo generation mechanisms that produce storm hydrograph. For
a large area (e.g., a model grid size of a regional climate
model or a general circulation model), these runo
*

Corresponding author. Tel.: +1-510-642-2648; fax: +1-510-6427483.


E-mail address: liang@ce.berkeley.edu (X. Liang).
1
Current address: ICCES and LASG, Institute of Atmospheric
Physics, Chinese Academy of Science, Beijing, China.

generation mechanisms are commonly present at different portions of a grid cell simultaneously. Missing
one of the two major runo generation mechanisms and
lacking of the consideration of spatial soil variability
can result in signicant under/over estimation of surface
runo which can introduce directly large errors in soil
moisture states over each model grid cell. Therefore,
proper modeling of surface runo is essential to a
reasonable representation of feedbacks in a land
atmosphere system. This is because the generated
amount of surface runo directly aects soil moisture
which can signicantly aect the regional atmospheric
budgets of water and energy at dierent time scales. The
partitions of rainfall into runo, and soil moisture storage, however, vary widely among the current generation
of land surface models used for numerical weather prediction and climate studies. This is clearly illustrated by

0309-1708/01/$ - see front matter 2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 3 0 9 - 1 7 0 8 ( 0 1 ) 0 0 0 3 2 - X

1174

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

the dierent study phases of the project for intercomparison of land-surface parameterization schemes
(PILPS) [7,25,27,39,46]. Parameterizations of surface
runo in the current generation of land surface models
and in some of the widely used hydrological models can
be classied into the following major categories:
(1) Inltration excess runo without consideration of
subgrid-scale spatial variability. Models in this category
allow surface runo only when rainfall rate exceeds a
single, spatially averaged inltration capacity that is
usually calculated based on diusion equations. The
saturation excess runo is an ``all or nothing'' process
[43]. Atmosphereland-surface interaction scheme
(ALSIS), best approximation of surface exchanges
(BEST), parameterization for landatmospherecloud
exchange (PLACE), Oregon State University (OSU),
among others, belong to this category [11,21,28,42,47].
(2) Saturation excess runo without consideration of
subgrid-scale spatial variability. Models in this category
allow surface runo only when the surface soil layer
becomes saturated. BUCK model (the simplest surface
runo parameterization in land surface models) described in the work of Manabe, Canadian land surface
scheme (CLASS), European Centre for Medium-Range
Weather Forecasts (ECMWF), surface energy and water
balance (SEWAB), interaction soil biosphere atmosphere (ISBA), and most of other force-restore
models, among others, belong to this category
[29,31,34,40,41]. The Sacramento model (lumped hydrological model) [5] which does not generate inltration excess runo over pervious area also falls into this
category. Sacramento model is an operational rainfall
runo model used for river forecast by the National
Weather Service. The biosphereatmosphere transfer
scheme (BATS) parameterizes surface runo as a proportion of net water applied to the surface empirically
[12]. In the surface runo parameterization of BATS,
the two surface runo generation mechanisms (Horton
and Dunne) are not physically represented. Also, the
subgrid-scale spatial variability of soil properties affecting surface runo generations are not considered.
(3) Inltration excess runo considering subgrid-scale
spatial variability. The importance of subgrid-scale
spatial variability has been addressed by many researchers (e.g. [1,19,44], among others). Models in this
category generate inltration excess runo by considering the subgrid-scale spatial variability to some extent.
For example, simplied simple biosphere model (SSiB)
and the land surface model of National Centers for
Environmental Prediction (NCEP) belong to this category. NCEP uses the parameterization of the simple
water balance (SWB) model [38] to estimate its surface
runo [6], where the spatial heterogeneity of rainfall and
inltration capacity is considered. Within each grid cell,
the surface runo of SWB model is derived based on
probabilistic averaging of the point inltration-excess

equation [32] by assuming exponential distribution


functions of precipitation and inltration capacity
[19,38]. The Stanford watershed model (lumped hydrological model) also falls into this category [10].
(4) Saturation excess runo considering subgrid-scale
spatial variability. Models in this category generate saturation excess runo by considering the subgrid-scale
spatial variability to some extent. For example, MOSAIC [20] assumes each model grid cell consists of wet
(saturated) and dry soil. Precipitation falling on the wet
portion of the soil generates surface runo immediately,
while precipitation falling on the dry portion must ll in
the top soil layer before it can generate surface runo.
Within the portion of dry soil, the spatial variability of
soil properties is not considered. Also, precipitation is
assumed to fall uniformly over a fraction of a model grid
cell. A constant value for the fraction of receiving the
precipitation is assumed for convective and frontal
storms, respectively. The hydrologically based two-layer
(or three-layer) variable inltration capacity (VIC-2L/
VIC-3L) land surface model generates saturation excess
surface runo by considering both the subgrid-scale
spatial variability of soil and precipitation [2224]. To
account for the eects of subgrid-scale soil variability,
ISBA and SEWAB later incorporated the VIC approach
in generating their saturation excess surface runo
[31,46]. The hydrological model, TOPMODEL, also
falls into this category [3].
Other models that consider the eects of subgrid-scale
spatial variability of soil properties and precipitation are
briey reviewed by Liang et al. [23]. Among them, the
models by Entekhabi and Eagleson [15] and Famiglietti
and Wood [16] are the two that consider both inltration and saturation excess runo within a grid cell.
However, the surface runo computed by both models is
a point average over the fraction of a grid cell on which
precipitation falls. In other words, both models assume
that each point is independent from each other and that
each point has the same statistical distribution in precipitation, soil moisture, and/or topography soil index,
respectively. SWB model applies similar statistical approach in this regard [38].
From the above brief discussion, it can be seen that
the current generation of land surface models do not
take into account of the inltration and saturation excess runo within a model grid cell eectively under the
context of subgrid-scale spatial variability of soil properties and rainfall. However, for a large area (e.g., a
GCM or regional climate model grid cell), the eects of
soil heterogeneity play an important role in surface
runo generations which would directly aect soil
moisture states, and water and energy partitions. In the
pilot phase study of the ongoing global soil wetness
project (GSWP), a large spread, similar to the results
from dierent phases of PILPS studies [7,25,27,39,46], in
the estimations of soil moisture as well as the par-

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

titionings of surface energy (latent heat and sensible heat


ux) and water (runo and evapotranspiration) among
the participating schemes is present [14]. Dirmeyer et al.
[14] suggested that the most signicant reason for the
wide spread was that the subgrid-scale variability in
inltration, whether due to the heterogeneity in soil
properties or the distribution of rainfall within a grid
box, had played an important role. Lohmann et al. [27]
showed that a number of land surface schemes in the
PILPS-2C have too much runo in the summer due to
their wet soil moisture, while others have too little runo
most of the time. That study suggested that inltrationexcess runo process under dry conditions should be
improved.
As described above, VIC simulates saturation excess
runo by considering the eects of subgrid spatial
variability of soil heterogeneity and precipitation, but it
does not simulate the inltration excess runo. Nijssen
et al. [33] coupled the VIC with a simple grid based
network routing scheme to study streamow simulations
for continental-scale river basins. They found that VIC
performs quite well over the moist areas where saturation excess runo component dominates. For the arid
areas where inltration excess runo component dominates, however, they indicated some diculties in reproducing monthly observed streamows.
Inltration and saturation excess runo can be
present at dierent locations of a studied area (or model
grid cell) due to the spatial variability of soil inltration
rate and soil moisture capacity. However, the representation of Horton and Dunne surface runo generation mechanisms is missing in the current generation of
land surface models for model grid sizes where subgridscale soil heterogeneity is signicant. Lacking of the
inclusion of both surface runo generation mechanisms
and of the consideration of spatial soil heterogeneity
could under/over estimate surface runo signicantly,
unless the model parameters are well calibrated. However, for applications in large areas with no observations
(e.g., ungaged regions), especially for applications of
simulations of coupling systems of regional climate
model or GCM where only limited observations are
available, large errors in soil moisture could be resulted
which could signicantly aect the water and energy
budgets of the coupling systems.
This paper presents a new model for surface runo
parameterization used for regional climate and GCM
studies. The new parameterization dynamically represents both Horton and Dunne runo generation
mechanisms within a model grid cell. Also, the new
model takes into account of eects of subgrid-scale
spatial variability of soil heterogeneity and precipitation
on Horton and Dunne runo. The model structure and
its development are described in Section 2. A series of
numerical experiments under dierent conditions are
analyzed in Section 3. Comparisons of the new surface

1175

runo parameterization with observations from three


watersheds in Pennsylvania are presented in Section 4.
Primary conclusions are provided in Section 5.
2. Model development
2.1. General description
The new surface runo parameterization is described
in this section. The new parameterization dynamically
represents both Horton and Dunne runo generation
mechanisms within a model grid cell together with a
consideration of the subgrid-scale soil heterogeneity.
Fig. 1 shows a schematic representation of a typical
hydrologic concept where saturation and inltration
excess runo occur within a studied basin or a model
grid cell. The saturation excess runo (represented by
the dark shaded areas in Fig. 1) generally occurs adjacent to streams, while inltration excess runo (represented by the shaded areas with broken lines in Fig. 1)
occurs at other places where the local inltration capacity is exceeded by the local precipitation (or
throughfall) rate.
The saturation excess runo in the new parameterization is computed following the concepts used in the
VIC-3L/VIC-2L model (three-layer/two-layer variable
inltration capacity model). The VIC-3L model has
three soil moisture layers. They are a thin layer (usually
taken as 100 mm deep from the surface), an upper layer
which includes the top thin layer, and a lower layer [24].
The thin layer allows a quick bare soil evaporation
following small summer rainfall events. The upper layer
is designed to represent the dynamic response of soil to
rainfall events, and the lower layer is used to characterize seasonal soil moisture behavior. In this paper, we
will focus on the upper soil layer because this layer is

Fig. 1. Schematic representation of the saturation excess runo (dark


shaded area) and inltration excess runo (shaded area with broken
lines) over a studied area.

1176

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

closely related to the surface runo generation. For


more details of the VIC model, the reader is referred to
[8,2224].
The soil moisture capacity in the upper layer is assumed to vary within a studied area or a model grid cell.
This spatial variability of soil moisture capacity is
mainly due to the heterogeneity in the depth of the soil
column and in the soil properties. The spatial variation
of soil moisture capacity is expressed by a spatial
probability distribution as follows [45,48]:
h
i
1=b
i im 1 1 A ;
1
where i and im are the point soil moisture capacity
(called inltration capacity in VIC-3L/VIC-2L) and
maximum soil moisture capacity (called maximum inltration capacity in VIC-3L/VIC-2L), respectively, A is
the fraction of an area for which the soil moisture capacity is less than or equal to i, and b is the soil moisture
capacity shape parameter, which is a measure of the
spatial variability of the soil moisture capacity, dened
as the maximum amount of water that can be stored in
the upper layer of the soil column. It is a surrogate for
spatial heterogeneity of soil properties. In the new
parameterization, saturation excess runo (represented
by R1 ) is parameterized to be generated from the fraction of the area As that is initially saturated and the
fraction of the area A0s As that becomes saturated
during the time step (see Fig. 2(a)). The inltration excess runo (represented by R2 ) is parameterized to be
generated from the fraction of 1 As and is ``redistributed'' over the entire area for numerical representation
as indicated by the shaded-area with back-slashed-lines
in Fig. 2(a). The actual amount of R2 in Fig. 2(a) is
determined by the R2 illustrated in Fig. 2(b) which is to
be described. In Fig. 2(a), P represents the amount of
precipitation (or throughfall) over a time step Dt. It
has a dimension of length [L]. This amount of water
(i.e., P) is partitioned into saturation excess runo R1 ,
inltration excess runo R2 , and the amount of water

that inltrates into the soil DW . All of these terms have


a dimension of length [L]. The symbol Wt in Fig. 2(a)
represents the soil moisture content at time t before the
arrival of precipitation P, and has a dimension of
length [L].
The inltration excess runo generated over the
fraction of 1 As is due to the subgrid-scale variability
of soil inltration rate. This is the surface runo component that was missing in the previous VIC parameterizations. Similar to the concept of spatial
variation of soil moisture capacity over a studied area,
the point potential inltration rate (i.e., soil inltration
capacity with a dimension of L=T also varies spatially.
It is worth mentioning that the terminology of inltration capacity used here is dierent from the one used in
[2224] (see the explanation above). Studies in the literature have shown dierent distributions for approximating the spatial variation of potential inltration rate.
For example, Burgin and Luthin [4] suggested to use a
triangular distribution based on some experimental inltration measurements. Burgin and Luthin's suggested
distribution is a special case of the following beta distribution when the parameter B below is set to be 1. The
beta distribution can be expressed as
h
i
1=B
f fm 1 1 C
;
2
where f and fm are the point potential inltration rate
(i.e., inltration capacity) and the maximum potential
inltration rate, respectively, C is the fraction of an area
for which the potential inltration rate is less than or
equal to f, and B is the potential inltration rate shape
parameter, which is a measure of the spatial variability
of the potential inltration rate, dened as the maximum inltration rate of each point when the surface is
ponded. The spatial distribution of Eq. (2) with B 1
was the distribution adopted by Stanford model [10].
Loague and Gander [26] reported inltration results
for a rangeland experimental watershed which is
0:1 km2 in the Experimental Watershed of Little

Fig. 2. Schematic representation of upper soil layer (a) spatial distribution of soil moisture capacity ([L]) for saturation excess runo, and (b) spatial
distribution of potential inltration rate L=T (i.e., inltration capacity) for inltration excess runo.

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

Washita (USA) where a total of 157 inltrometer


measurements were made. They concluded that a lognormal distribution gave a better t to the data than a
normal distribution.
Schaake et al. [38] assumed an exponential distribution for the potential inltration rate over a studied area
based on the experimental ndings by Berndtsson [2].
Berndtsson reported that the parameters of Horton and
Philip's equations varied spatially, but the spatial coefcient of variation of all of the parameters and of the
cumulative amounts of inltration was close to 1.0.
Berndtsson's above nding was based on 52 inltrometers placed over a 19:2 km2 catchment in northern
Tunisia.
Since there are not enough eld measurements to
support one type of distribution over the other, we
assume that the potential inltration rate over a studied area follows the suggestion of Burgin and Luthin
[4]. That is, we will apply Eq. (2) in our study. The
fraction of an area C in Eq. (2) varies between 0 and 1
after being scaled over 1 As . To keep the generality,
we use B instead B 1 in the derivation below.
However, in the numerical experiments and applications presented in this paper, B is taken to be 1. In
addition, it should be mentioned that our methodology
is not limited to the usage of Eq. (2). It can be applied
to other spatial distributions assumed for the potential
inltration rate.
2.2. Model derivation
From the general description above, it is clear that we
need to nd expressions for the saturation excess runo
R1 , inltration excess runo R2 , soil moisture change
DW at each time step according to Eqs. (1) and (2),
given the amount of precipitation P for the same time
period. From the water balance equation, we have (see
Fig. 2(a)):
P R1 y R2 y DW y;

where
y  1 R1 y DW y:

That is, y represents the vertical depth shown in Fig.


2(a). Based on Eq. (1) and VIC parameterization [22,24],
the saturation excess runo R1 y and soil moisture
change DW y in Eq. (4) can be represented as follows,
respectively,
8

b1 
b1 
>
i0 y
i0
im
>
1
;
< y b1 1 im
im
R1 y
0 6 y 6 im i0 ;
>
>
:
R1 yjyim i0 y im i0 ; im i0 < y 6 P ;
5
and

DW y

8 
>
im
>
1
< b1
>
>
:

im

i0
im

1177

b1


1

0 6 y 6 im i0
i0 R1 yjyim

i0 ;

i0 y
im

im

b1 

i0 < y 6 P ;
6

where i0 is the point soil moisture capacity corresponding to the soil moisture Wt in Fig. 2(a). It should be
mentioned that the saturation excess runo R1 is expressed as a function of y rather than P as before. This is
because when the inltration excess runo is considered,
less precipitation would inltrate into the soil. Thus, a
smaller fraction of the studied area would become saturated.
The expression for the inltration excess runo R2
in Eq. (3) is obtained by applying Eq. (2) (see Fig. 2(b)).
It is obvious that the amount of R2 , the shaded area in
Fig. 2(b) multiplied by Dt, should be equal to the R2
depicted in Fig. 2(a). Also, the total amount of water
inltrated into the upper layer DW , as illustrated in
Fig. 2(a) and (b), should be equal to each other. Thus R2
can be expressed as follows by applying Eq. (2) with an
available water input rate wp
wp

R2 y

R1 y
;
Dt
8
>
>
>
< P R1 y
>
>
>
:P

P R1 y
fm Dt

7

fmm Dt 1


1

P R1 y
fm Dt

B1


;

61

R1 y

fmm Dt;

P R1 y
fm Dt

P 1;
8

where fmm is the average potential inltration rate (i.e.,


average inltration capacity) over the area of 1 As ,
which can be expressed as:
Z 1
fm
1=B
fmm
:
9
fm 1 1 C dC
1B
0
It can be seen from Eqs. (5), (6), and (8) that each term,
except for the precipitation P, in Eq. (3) can be expressed as a function of y. Therefore, if Eq. (3) has a
solution for y, then we can obtain the R1 , R2 , and DW
accordingly. This implies that the precipitation (or
throughfall) P can be partitioned into R1 , R2 , and DW
through their relationships with y as illustrated in Fig.
2(a). If Eq. (3) did not have a solution for y, it would
imply that we could not nd such a partition among R1 ,
R2 , and DW for P. In other words, we could not write
the water balance equation (i.e., Eq. (3)) and Eq. (4) in
the way described above. It can be proved mathematically that not only does such y exist, but also there is an
unique y to partition the P into R1 , R2 , and DW . That is,
Eq. (3) has an unique solution. Detailed proof is illustrated in Appendix A of this paper.
As described above, we use two spatial distributions
(Eqs. (1) and (2)) to represent the spatial heterogeneity

1178

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

of soil properties. Among the two spatial representations, Eq. (1) does not vary with time, while Eq. (2) does.
This is because Eq. (1) describes the spatial distribution
of the point maximum amount of water that can be
stored in the upper layer of the soil column over the
studied area (or grid cell). Such maximum amount of
water that the soil column can hold at each point can be
assumed to be the same over the time scale of our interest. The spatial distribution of the potential inltration rate over the area of 1 As (i.e., Eq. (2)), on the
other hand, varies with time because potential inltration rates are a function of soil moisture. The magnitude
of the potential inltration rates are a function of soil
moisture. The magnitude of the potential inltration
rate at each point generally decreases when the soil
moisture increases at that point, as is evidenced by eld
measurements and the widely used inltration equations
(e.g., Horton and Philip). For some special cases where
inltration capacity decreases with the decrease of soil
moisture, such as the situation with a ``dry thin crust
layer'' at the soil surface, a dierent inltration relationship should be used. We assume a general situation where inltration capacity decreases with the
increase of soil moisture. We also assume that the shape
of the spatial distribution of the potential inltration
rate is the same at each time t, that is, it follows the same
distribution described by Eq. (2) at each time. Therefore,
the decrease of the potential inltration rate with time
should be represented by the distribution parameter fm
in Eq. (2), while the shape parameter B is kept the same
(i.e., B 1). As we know, fm is related to fmm by Eq. (9),
thus we only need to estimate fmm for each time t. Since
fmm is the average potential inltration rate (i.e., average
inltration capacity) over the area of 1 As , it can be
estimated by applying an inltration equation (for example, Horton equation [18], Philip equation [35],
Green and Ampt [17] (referenced by Dingman [13])).
The value of fmm at each time is determined by applying
the time compression analysis (TCA) similar to that of
Salvucci and Entekhabi [37]. TCA is an approximation
commonly used to relate the maximum inltration rate
at any given time during an event to the initial soil
moisture and the cumulative moisture exchange up to
that time. The approach used in this paper is briey
described below.
Let us assume that the spatially averaged point inltration rate in the studied area can be represented by
a point inltration function fp t (whether empirical or
analytical). Therefore, the inltrated water over time
0; tf that is obtained by applying fp t should be equal
to the total amount of water that is actually inltrated
during the time period 0; t Dt over the studied area.
The time tf is determined by the following equation:
Z tf
fp t dt Wt DW ;
10
0

where Wt is the soil moisture at time t, and DW is the


change of soil moisture between t and t Dt (see Fig.
2(a)). Therefore, fmm can be obtained by
fmm fp tf :
11
In Eq. (10), the value of DW is very small compared to
the Wt in general. We have compared the two approaches of including DW versus excluding it. The results showed that the inclusion of DW has little eects on
the solution of tf . Therefore, we exclude DW term in
Eq. (10) in our study.
It is worth mentioning that the well-known Stanford
hydrologic model also assumes that the parameter fm in
Eq. (2) varies with time. However, in the Stanford model,
the variation of fm with time is determined empirically
based on the results from a number of dierent watersheds [10]. The SWB model [38] assumes that its spatial
distribution of the inltration capacity is not a function
of time t, but a function of the time step Dt through its
spatially averaged inltration capacity parameter.
3. Numerical experiments and results
Numerical experiments of applying the new surface
runo parameterization are presented and discussed in
this section. In the applications, we assume that the
parameter B 1 in Eq. (2), and that the spatially averaged point inltration function fp t is represented by
the Horton inltration equation. We use the Horton
equation due to its simplicity. However, the methodology is not limited to the Horton equation, one can use
any other inltration equation that is most appropriate
to his/her question at hand. The Horton equation takes
the following form:
fH t fc f0

fc e

kt

12

where fH t is the inltration capacity L=T at time t, f0


is the initial inltration capacity L=T , fc is the nal
capacity L=T , and k is an empirical constant T 1 .
Magnitudes of the parameters (f0 , fc and k) for dierent
soil types can be found in the literature. Three soil types
with the characteristics of high, moderate, and low inltration rates are used in our study. The values of the
three soil types are listed in Table 1 [9,13,18,36]. With
fp t represented by the Horton equation, the fmm in Eq.
(11) can be expressed as
fmm fH tf fc f0

fc e

ktf

13

Three types of numerical experiments are conducted.


The rst type is to apply the new surface runo
parameterization to dierent soils combined with different precipitation intensities, and dierent degrees of
soil heterogeneity. Twenty-seven combinations in this
category are studied, which include three dierent types
of soils listed in Table 1, three dierent precipitation
intensities P 6 mm=h; P 2 mm=h; P 0:6 mm=h,

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

1179

Table 1
Soil characteristics used in the experiments
Soil type

hs

f0 (mm/h)

fc (mm/h)

Soil depth (mm)

High inltration rate


Moderate inltration rate
Low inltration rate

0.435
0.451
0.492

375.2
111.5
16.51

45.0
6.10
4.14

19.6
8.15
0.63

1500
1500
1500

0:6 mm=h are taken to represent high, normal, and low


precipitation conditions, respectively. For simplicity, the
three Horton inltration parameters (i.e., f0 , fc , and k)
are kept unchanged for the three dierent precipitation
rates over the same soil type. Also, the three Horton
parameters are assumed to be valid for the assumed
initial soil moisture condition (i.e., W0 0:1Wm ). Each of
the 27 experiments were run continuously with constant
rainfall (or throughfall) intensities for 24 h with a time
step of one hour (i.e., Dt 1 h). Evaporation is assumed
to be negligible at each time step. This assumption is
reasonable during the courses of storms when the atmosphere is humid with weak wind. The 24 h cumulative inltration and saturation excess runo of the 27
cases are shown in Tables 24 for the soils with low,
moderate, and high inltration capacities, respectively.
Also, the total amount of water inltrated (Inf. W), the
runo ratio, and the ratio Rf of inltration excess
runo to the total surface runo,
Pt24
t1 R2 t
Rf Pt24
15
Pt24
R
t
t1
R2 t
1
t1

three dierent values for parameter b b 0:1; 0:3; 0:5,


and one initial soil moisture conditions W0 0:1Wm .
The W0 and Wm are the initial soil moisture and the areal
mean soil moisture capacity for the upper soil layer.
From Eq. (1), we can obtain the relationship between
the areal mean soil moisture capacity Wm and the
maximum point soil moisture capacity im as follows:
Wm

im
:
1b

14

For the three types of numerical experiments presented


below, the spatial subgrid-scale variability of precipitation is considered in the same way as described in Liang
et al. [23]. In other words, the dierences in results are
not due to the dierence in spatial variability of precipitation, but other factors.
3.1. Type one numerical experiments
The soil moisture capacity over the studied area
would be more heterogeneously distributed if the value
of parameter b is larger. The three dierent rainfall (or
throughfall) intensities of P 6, P 2, and P

are shown in Tables 24.

Table 2
Summary results for the soil with low inltration capacity property
W0 =Wm

P (mm/day)

R1 (mm/day)

R2 (mm/day)

Rf (%)

Runo ratio (%)

Inf. W (mm)

0.1
0.1
0.1
0.1
0.1
0.1
0.1
0.1
0.1

0.1
0.1
0.1
0.3
0.3
0.3
0.5
0.5
0.5

244
48
14.4
144
48
14.4
144
48
14.4

1.57
0.54
0.15
3.98
1.35
0.37
5.76
1.94
0.53

48.49
5.17
0.44
46.83
5.00
0.43
45.63
4.86
0.42

96.87
90.59
75.20
92.16
78.68
53.85
88.79
71.48
44.26

34.76
11.89
4.10
35.29
13.20
5.55
35.69
14.17
6.60

93.94
42.29
13.81
93.18
41.66
13.60
92.61
41.20
13.45

Table 3
Summary results for the soil with moderate inltration capacity property
W0 =Wm

P (mm/day)

R1 (mm/day)

R2 (mm/day)

Rf %

Runo ratio (%)

Inf. W (mm)

0.1
0.1
0.1
0.1
0.1
0.1
0.1
0.1
0.1

0.1
0.1
0.1
0.3
0.3
0.3
0.5
0.5
0.5

144
48
14.4
144
48
14.4
144
48
14.4

1.96
0.56
0.15
4.94
1.42
0.37
7.09
2.03
0.54

33.94
3.72
0.32
32.52
3.58
0.31
31.51
3.49
0.31

94.55
86.81
68.62
86.82
71.64
45.70
81.63
63.18
36.42

24.93
8.92
3.28
26.01
10.42
4.78
26.81
11.50
5.85

108.11
43.72
13.93
106.54
43.00
13.71
105.40
42.48
13.56

1180

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

Table 4
Summary results for the soil with high inltration capacity property
W0 =Wm

P (mm/day)

R1 (mm/day)

R2 (mm/day)

Rf %

Runo ratio (%)

Inf. W (mm)

0.1
0.1
0.1
0.1
0.1
0.1
0.1
0.1
0.1

0.1
0.1
0.1
0.3
0.3
0.3
0.5
0.5
0.5

144
48
14.4
144
48
14.4
144
48
14.4

2.85
0.62
0.15
7.02
1.55
0.38
9.94
2.22
0.55

4.48
0.48
0.04
4.21
0.46
0.04
4.03
0.45
0.04

61.12
43.68
21.21
37.49
22.94
9.39
28.86
16.82
6.58

5.08
2.30
1.34
7.80
4.20
2.93
9.70
5.56
4.07

136.68
46.90
14.21
132.77
45.98
13.98
130.03
45.33
13.81

Table 2 shows the results of soil with low inltration


capacity for three dierent values of the shape parameter b. From the table, it can be seen that the inltration excess runo R2 plays a signicant role in the
surface runo generation mechanisms for the investigated conditions. The inltration excess runo accounts
for about 90% of the total surface runo generated when
the storm has high intensity (e.g., P 6 mm=h. This is
true for all three dierent b parameter values, although
there is a small decrease trend in Rf (i.e., from 97% to
89%) with the increase of b values from 0.1 to 0.5. The
runo ratio, however, is close to a constant at about
35% in each case, which is due to the increase of saturation excess runo with the increase of b values. In
other words, with the increase of b values, a larger
portion of the studied area will contribute to saturation
excess runo R1 , and a smaller portion will contribute
to the inltration excess runo R2 . The increase in R1
and the decrease in R2 compensate each other to a certain extent. Such compensation decreases the variation
of total surface runo with the choices of b values. The
total amount of water inltrated into the soil and the
runo ratio have small variations among the cases with
dierent b values, as expected, when the precipitation is
the same. This property is also shown for soils with
moderate and high inltration rates (see Tables 3 and 4).
In comparison, however, the sensitivity of the runo
ratios and the amount of water inltrated are more
sensitive to the choice of b values in the original VIC
surface runo parameterization (see Table 5). In the
original VIC surface runo parameterization, it only has
the saturation excess runo which is aected by the
values of b parameter. In the new surface runo
parameterization, dierent b values redistribute the
available water between saturation excess runo R1
and inltration excess runo R2 , which results in less

eects on the total amount of surface runo generated.


This implies that the total surface runo from the new
runo parameterization is less sensitive to the choice of
the values of the shape parameter b which are not easy
to be accurately estimated with the knowledge at
present. More studies in this aspect need to be explored
in the future.
From this type of experiments, we can see that the
inltration excess runo plays an important role in the
generation of total surface runo at the spatial scales
where subgrid-scale variability of soil properties is signicant. This is true especially for high rainfall intensities. The ratios of Rf range from 97% to 29% for soils
with low inltration capacity to soils with high inltration capacity under the high rainfall intensity (Tables
24). For precipitation with the moderate intensity, the
ratios of Rf vary from 91% to 17% for soils with low to
high inltration capacities. Even for precipitation with
the low intensity, the values of Rf range from 75% to
36% for the soils with low and moderate inltration
capacities (see Tables 2 and 3). For soil with high inltration capacity under the low rainfall intensity, saturation excess runo will dominate the surface runo.
The inltration excess runo only accounts for small
percentages of the total surface runo. Table 4 shows
that the ratios of Rf vary between 21% to 7% in this case.
3.2. Type two numerical experiments
In the second type of the numerical analysis, results
of the new runo parameterization (called dynamic
results) are compared with two other congurations
under otherwise the same conditions. One conguration is to assume that the parameter fm in Eq. (2)
does not vary with time, but a constant which is
represented by the maximum saturated hydraulic

Table 5
Eects of the b parameter with VIC parameterization for the soil with moderate inltration capacity property
Model

W0 =Wm

P (mm/day)

R1 (mm/day)

R2 (mm/day)

Rf %

Runo ratio (%)

Inf. W (mm)

VIC
VIC
VIC

0.1
0.1
0.1

0.1
0.3
0.5

144
144
144

2.95
7.26
10.26

0
0
0

0
0
0

2.05
5.05
7.12

141.05
136.74
133.74

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

conductivity of the studied area for the selected soil.


Our choice of taking the maximum saturated hydraulic conductivity is because it is used by others
(e.g. [13]). Taking a constant for parameter fm is
equivalent to the concept of keeping the spatial distribution of the potential inltration rate unchanged
as is implied in the work of Schaake et al. [38] and
Manley's [30] work described by Dingman [13]. The
results from this conguration are called static results.
The other conguration is the original VIC parameterization where the surface runo is constituted
by the saturation excess runo only. The subsurface
runo from the lower layer of the original VIC
model is not considered in this comparisons. Also, the
interactions of water movement between the upper
and lower soil layers that are described in the original
VIC model [2224] are not considered here either.
This is because the focus here is to compare the eects
of dierent surface runo parameterizations on the
total surface runo and the soil moisture in the upper
layer.

Figs. 35 show the comparisons among the three


congurations for soils with low, moderate, and high
inltration capacities, respectively. Fig. 3 shows the
comparisons of time series of R1 , R2 , R1 R2 , Rf , the
amount of water inltrated, and the soil moisture for the
precipitation rate of P 2 mm=h. Figs. 4 and 5 show
the same comparisons for the precipitation rate of
P 6 mm=h. The original VIC conguration produces
the least total surface runo and the static conguration
produces the most in all cases of the three dierent soils.
Consequently, more water inltrated into the soils with
the original VIC parameterization than the new surface
runo parameterization with the dynamic conguration,
and less amount of water inltrated in the static conguration than in the dynamic one. Therefore, the soil
moisture in the original VIC parameterization is significantly wetter than the dynamic conguration, while the
soil moisture in the static conguration is much drier
than that in the dynamic conguration. Such over- and
under-estimation of soil moisture by the original VIC
and static congurations can result in signicant errors

0.2

0.6

R1(mm/h)

R2(mm/h)

Dynamic
VIC
Static

0.15
0.1

0.05
0

(a)

10
15
Time (hour)

20

Total runoff (mm/h)

R2/(R1+R2)

0.6
0.4

(c)

10
15
Time (hour)

20

10
15
Time (hour)

20

25

10
15
Time (hour)

20

25

10
15
Time (hour)

20

25

0.6
0.4
0.2
0

25
(d)

60

120
Soil moisture (mm)

Water infiltrated (mm)

0.8

0.2

(e)

0.2

(b)

0.8

40
20
0

0.4

25

1181

10
15
Time (hour)

20

100
80
60

25
(f)

Fig. 3. Comparison of surface runo parameterizations with dynamic, VIC, and static congurations for the soil of low inltration capacity with
constant precipitation of P 2 mm=h for 24 h at Dt 1 h: (a) saturation excess runo, (b) inltration excess runo, (c) fraction of inltration
excess runo over the total surface runo (i.e., Rf ), (d) total surface runo (i.e., R1 R2 ), (e) amount of water inltrated into the upper soil layer, and
(f) soil moisture in the upper layer.

1182

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

in evapotranspiration calculation during the period after


the storms.
From Figs. 35, we can also see clearly that the dynamic conguration generates signicantly less inltration excess surface runo than that of the static
conguration. This is true because at earlier hours of
each storm, the potential inltration rate is generally
higher, if the soil is not saturated at the beginning of the
storms, than the saturated hydraulic conductivity used
in the static conguration, thus more water can inltrate
into the soil with less water generated as inltration
excess runo in the dynamic conguration than that in
the static conguration. As expected, the original VIC
conguration does not generate any inltration excess
runo, and all of the water (except for the R1 ) can inltrate into the soil. Thus, the original VIC parameterization results in higher soil moisture in the soil layer
than the other two, with the static one the driest. The
saturation excess runo in all three congurations result
in similar amount. Therefore, the total surface runo is
the largest for the static one, and the smallest for the

original VIC. The fraction of the inltration excess


runo (i.e., Rf ) accounts for about 80% or more with the
dynamic conguration for the cases considered in Figs.
3 and 4. For the soils with high inltration capacity, the
Rf varies between 50% and 25% with the dynamic conguration (see Fig. 5). Tables 68 list the same items as
those listed in Tables 24. From the tables, we see again
that the dynamic conguration produces more surface
runo and less amount of water inltrated into the soil
than that of the original VIC conguration. Also from
Table 8 and Fig. 5, it can be seen clearly that when the
soil has high inltration capacity, the dynamic conguration results in similar results as that of the original
VIC conguration, which suggests that VIC would work
well at places where saturation excess runo dominates.
It is worth mentioning that the results reported in this
section are related to the case where the parameter fm
takes a value of the maximum saturated hydraulic
conductivity. The dierences between the dynamic and
static results may be smaller if a dierent static value for
parameter fm is taken. However, there may be dierent
4

0.4

R2(mm/h)

Dynamic
VIC
Static

R (mm/h)

0.6

0.2
0

(a)

10
15
Time (hour)

20

Total runoff (mm/h)

R /(R +R )

1
2

10
15
Time (hour)

20

10
15
Time (hour)

20

25

10
15
Time (hour)

20

25

10
15
Time (hour)

20

25

2
1

(d)

150

200

100

(e)

25

Soil moisture (mm)

Water infiltrated (mm)

(c)

50
0

0.2
0

(b)

0.8

0.4

25

0.6

10
15
Time (hour)

20

150
100
50
0

25
(f)

Fig. 4. Comparison of surface runo parameterizations with dynamic, VIC, and static congurations for the soil of moderate inltration capacity
with constant precipitation of P 6 mm=h for 24 h at Dt 1 h: (a) saturation excess runo, (b) inltration excess runo, (c) fraction of inltration
excess runo over the total surface runo (i.e., Rf ), (d) total surface runo (i.e., R1 R2 ), (e) amount of water inltrated into the upper soil layer, and
(f) soil moisture in the upper layer.

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

0.6
Dynamic
VIC
Static

0.2
0

(a)

10
15
Time (hour)

20

Total runoff (mm/h)

R2/(R1+R2)

10
15
Time (hour)

20

25

10
15
Time (hour)

20

25

10
15
Time (hour)

20

25

0.8

0.6
0.4
0.2
0

(c)

10
15
Time (hour)

20

0.6
0.4
0.2
0

25
(d)

200
Soil moisture (mm)

150
100
50
0

(b)

0.8

0.2
0

25

Water infiltrated (mm)

0.4

0.4

R (mm/h)

R1(mm/h)

0.6

1183

(e)

10
15
Time (hour)

20

150
100
50

25
(f)

Fig. 5. Comparison of surface runo parameterizations with dynamic, VIC, and static congurations for the soil of high inltration capacity with
constant precipitation of P 6 mm=h for 24 h at Dt 1 h: (a) saturation excess runo, (b) inltration excess runo, (c) fraction of inltration
excess runo over the total surface runo (i.e., Rf ), (d) total surface runo (i.e., R1 R2 ), (e) amount of water inltrated into the upper soil layer, and
(f) soil moisture in the upper layer.

static values for parameter fm for dierent applications.


Therefore, the dynamic approach is preferred.
3.3. Type three numerical experiments
In the third type of the numerical analysis, one storm
with high sampling rates of 5 s (i.e., Dt 5 s) that is
measured by the weather station of the Hydrometeorology Laboratory (HML) at the University of
Iowa is used with the dynamic and VIC congurations.
The 5 s rainfall time series is aggregated into 10 min,
30-min, and 1-h time intervals. The storm started at
18:02:19 on 3 May 1990, and lasted over 9 h. The time
series of the storm are shown in Figs. 68 for Dt 10,

30, and 60 min, respectively. The rainfall intensities


decrease with the increase of the time intervals.
Comparisons are conducted between the new surface
runo parameterization and the original VIC applied to
the upper layer. As discussed previously, the evaporation, subsurface runo, and interactions of water
movement between soil layers are not considered in both
the new and the original VIC parameterizations in the
comparison studies. The static conguration is not
considered in this comparison because of its clear
weaknesses compared with the dynamic conguration.
Figs. 68 show the comparison studies for the storm
with three dierent time intervals (i.e., Dt 10, 30, and
60 min). The information on soil type is not available at

Table 6
Comparisons among the dynamic, static and VIC congurations for the soil with low inltration capacity property
Model

W0 =Wm

P (mm/day)

R1 (mm/day)

R2 (mm/day)

Rf %

Runo ratio (%)

Inf. W (mm)

VIC
Dynamic
Static

0.1
0.1
0.1

0.3
0.3
0.3

48
48
48

1.55
1.35
1.12

0
5.00
11.17

0
78.68
90.90

3.23
13.20
25.6

46.45
41.66
35.71

1184

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

Table 7
Comparisons among the dynamic, static and VIC congurations for the soil with moderate inltration capacity property
Model

W0 =Wm

P (mm/day)

R1 (mm/day)

R2 (mm/day)

Rf %

Runo ratio (%)

Inf. W (mm)

VIC
Dynamic
Static

0.1
0.1
0.1

0.3
0.3
0.3

144
144
144

7.26
4.94
2.86

0
32.52
68.03

0
86.82
95.96

5.05
26.01
49.23

136.74
106.54
73.10

Table 8
Comparisons among the dynamic, static and VIC congurations for the soil with high inltration capacity property
Model

W0 =Wm

P (mm/day)

R1 (mm/day)

R2 (mm/day)

Rf %

Runo ratio (%)

Inf. W (mm)

VIC
Dynamic
Static

0.1
0.1
0.1

0.3
0.3
0.3

144
144
144

7.35
7.02
6.67

0
4.21
8.74

0
37.49
56.7

5.11
7.80
10.70

136.65
132.77
128.59

the measured site. Thus, it is assumed that the soil


considered in Figs. 68 has a moderate inltration capacity dened by the soil parameters in Table 1. The
initial soil moisture is assumed to be at W0 0:1Wm as
before.
Figs. 68 show the comparisons of the time series of
saturation excess runo R1 , total surface runo (i.e.,

R1 R2 ), the fraction of inltration excess runo over


the total surface runo (i.e., Rf ), and the water inltrated
into the soil between the new parameterization and the
original VIC with Dt 10, 30, and 60 min, respectively.
In the same gures, the dierences of total surface runo
between the two parameterizations are also shown in
panel f. It can be seen clearly that the original VIC

Fig. 6. Comparison of surface runo with dynamic and VIC parameterizations for the soil of moderate inltration capacity with a storm from Iowa
at Dt 10 min: (a) precipitation time series of the storm, (b) saturation excess runo, (c) fraction of inltration excess runo over the total surface
runo (i.e., Rf ), (d) total surface runo (i.e., R1 R2 ), (e) amount of water inltrated into the upper soil layer, and (f) dierence of total surface runo
between dynamic and VIC parameterizations.

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

1185

Fig. 7. Comparison of surface runo with dynamic and VIC parameterizations for the soil of moderate inltration capacity with a storm from Iowa
at Dt 30 min: (a) precipitation time series of the storm, (b) saturation excess runo, (c) fraction of inltration excess runo over the total surface
runo (i.e., Rf ), (d) total surface runo (i.e., R1 R2 ), (e) amount of water inltrated into the upper soil layer, and (f) dierence of total surface runo
between dynamic and VIC parameterizations.

surface runo parameterization signicantly under-estimates the total surface runo in all of the three dierent
time intervals due to the lack of the inltration excess
runo mechanism. Although the fraction Rf varies over
a wide range from about 0.1 to somewhere close to 1.0,
for most of the time during the course of storm, the
fraction Rf is greater than 0.6 for all three dierent time
intervals (Figs. 6(c)8(c)). The rate of the inltration
excess runo decreases as the time interval Dt increases. For Dt 10 min, the peak of the rate of inltration excess runo can be as high as 3 mm/h (see Fig.
6(d)), while the peak reduces to less than 1 mm/h for
Dt 60 min (see Fig. 8(d)).
The large dierence of surface runo between the
two parameterizations results in large dierence in the
amount of water inltrated into the soil (see Figs.
6(e)8(e)). Table 9 shows the summaries of the accumulated amount of precipitation, accumulated total
surface runo, and accumulated amount of water inltrated into the soil over the entire storm period for
the two parameterizations. For the VIC parameter-

ization, the accumulated total surface runo is actually the accumulated saturation excess runo since the
inltration excess runo is zero. From Table 9, it can
be seen that the runo ratios with VIC conguration
are about ve times lower than those with the dynamic conguration in all three cases of dierent time
intervals. Also, 18% more water inltrates into the soil
in the VIC parameterization than that in the new
parameterization over the storm period of about 9 h.
Such large dierence in inltrated water can directly
result in large dierence in evapotranspiration during
the period following the storm. For the other two
types of soil (i.e., soil with high or low inltration
capacity property), similar results as those shown in
Figs. 68 are obtained (gures are not shown here).
For the soil with high inltration capacity, the saturation excess runo accounts more in the total surface
runo than those shown in Figs. 68. On the other
hand, the inltration excess runo accounts more than
those shown in Figs. 68 for the soil with low inltration capacity.

1186

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

Fig. 8. Comparison of surface runo with dynamic and VIC parameterizations for the soil of moderate inltration capacity with a storm from Iowa
at Dt 60 min: (a) precipitation time series of the storm, (b) saturation excess runo, (c) fraction of inltration excess runo over the total surface
runo (i.e., Rf ), (d) total surface runo (i.e., R1 R2 ), (e) amount of water inltrated into the upper soil layer, and (f) dierence of total surface runo
between dynamic and VIC parameterizations.
Table 9
Comparisons of the new parameterization and the original VIC for Iowa storm
Name

W0 =Wm

Total P (mm)

Total R1 (mm)

Total R1 R2 (mm)

Runo ratio (%)

Inf. W (mm)

Dt (min)

Dynamic
VIC
Dynamic
VIC
Dynamic
VIC

0.1
0.1
0.1
0.1
0.1
0.1

0.3
0.3
0.3
0.3
0.3
0.3

25.3
25.3
25.1
25.1
25.1
25.1

0.60
0.72
0.61
0.71
0.62
0.71

4.41
0.72
3.73
0.71
3.43
0.71

17.4
2.85
14.9
2.83
13.7
2.83

20.9
24.6
21.4
24.4
21.7
24.4

10
10
30
30
60
60

From these comparisons, it can be seen clearly that


the new parameterization plays a very important role
in partitioning the water budget between surface
runo and soil moisture in an atmosphereland
coupling system. Also, the comparisons show that the
dierences in surface runo rate and soil moisture are
more signicant with shorter time intervals or with
higher rainfall intensities. This implies that appropriate surface runo parameterization could have significant eects on an atmosphereland coupling system
because sub-hour time intervals are widely used in the
coupling systems.

4. Validation
Due to lack of available separated measurements on
Hortonian and Dune runo, the new surface runo
parameterization is tested in this section only through
comparisons between the model simulated total runo
and observed streamows. The surface runo generation
mechanism in the current version of VIC-3L is replaced
by the new parameterization (called new VIC-3L hereafter). Also, the total runo from the new VIC-3L is
compared with the one from the current version of VIC3L (called old VIC-3L hereafter). The dierence between

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

1187

Table 10
Information on streamow stations and watersheds
Station

Latitude

Longitude

Basin name

Watershed index

Drainage km2

01573810
03049800
01572190

39490 0600
40310 1300
40280 4500

77060 2600
79560 1800
76310 5200

Brush Run near Mcsherrystown


Little Pine Creek near Etna
Swatara Creek near Inwood

1
2
3

1
15
428

the new and old VIC-3L is the lack of consideration of


the inltration excess runo generation mechanism in the
old VIC-3L, while everything else is kept the same. Three
watersheds from Pennsylvania are used in the comparison study. Among the three watersheds, two of them are
in the Lower Susquehanna basin with drainage areas of 1
and 428 km2 , respectively, and one in the Lower Allegheny with a drainage area of 15 km2 (see Table 10).
The forcing data needed to run VIC-3L are obtained
from nearby surface meteorologic stations (see Table
11). Information on vegetation, soil properties, and VIC
parameters is obtained from the corresponding grid cells
(at 1/8 resolution) compiled by University of Washington. The three parameters related to Horton inltration equation are determined based on the
corresponding soil properties. No calibrations on VIC3L model parameters (both new and old) were conducted. It should be mentioned that in this application,
Horton inltration equation is used. However, the
methodology described in this paper is not limited to
Horton equation, other inltration equations can be
used if appropriate for the problems at hand. In fact, the
new surface runo parameterization with Philip's inltration equation provides similar results, if not better, as
the Horton equation. Due to page limitation, results
with the Philip equation will not be reported here. The
two versions of VIC-3L model were running for a few
water years for each watershed. The model simulated
total runo is routed through the channel and then
compared with the observed streamow. The routing
scheme used here is a modied version used in the study
by Nijssen et al. [33]. Results from the last year simulations were taken for comparison analysis to limit the
eects of model initializations.
Comparisons between model simulated total runo
and observed streamow, and between the two versions
of VIC-3L on surface runo, subsurface runo, and soil
moisture in the upper and lower layers at the three
watersheds are presented in Figs. 911. Fig. 9 shows the
comparison for watershed 1 from July to September
Table 11
Information on surface meteorologic stations
Station

Latitude

Longitude

360656
366993
364896

39560
40300
40200

77150
80140
76280

1991 during which period its soil moisture becomes wet


after one dry month. There was not much rainfall and
streamow in July and early August of 1991 (Figs. 9(a)
and (b)). The soil moisture stayed dry. The dierences
between the new and old VIC-3L on the surface and
subsurface runo, and on the upper and lower soil
moisture were very small during this period (Figs. 9(b)
(d)). In the mid-August, a relative large storm did not
result in large oods but to increase the soil moisture
level in the upper layer. During this period, the dierences between the new and old VIC-3L on surface runo
and soil moisture started to increase, but still small. In
late September, a big storm came in with a peak of
about 50 mm/day. This storm produced a high streamow peak of about 15 mm/day (Fig. 9(b)). In comparison, the new VIC-3L produced a peak ow of about 12
mm/day for this storm, while the old VIC-3L produced a
peak ow of about 6 mm/day (Fig. 9(b)). Although both
new and old versions of VIC-3L underestimated the
observed peak ow, the new VIC-3L gave much better
peak ow simulation than the old VIC-3L. The large
dierence of the total runo in Fig. 9(b) was mainly
caused by the large dierence in the surface runo
component rather than the subsurface runo component
which is almost the same for the new and old cases (Fig.
9(c)). As expected, the new VIC-3L produced much
higher surface runo for the big storm than that of the
old VIC-3L. For the rest of the small storms during the
period, both new and old VIC-3L gave similar results.
Fig. 9(d) shows that corresponding to this September
storm, the old VIC-3L produced higher soil moisture
due to the fact that less water became the surface runo
and thus more inltrated into the soil. However, the
subsurface runo dierences between the two versions
are negligible since the moisture level in the lower layer
is still low. Results shown in Fig. 9 are consistent
with the results discussed in the sensitivity analysis of
Section 3.
Fig. 10 shows the comparison for watershed 2 from
October to December of 1997. The soil moisture also
has a similar transition from dry to wet as shown in Fig.
9, but at a higher level. For the storms prior to the big
one around November 10, both the new and old VIC-3L
overestimated the small streamow peaks (Fig. 10(b)),
and the overestimation of the new parameterization is
more signicant. However, for the large storm with a
peak over 30 mm/day around November 10, the total
runo simulated from the new VIC-3L compares quite

1188

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

well with the observation while the old VIC-3L underestimated the observation by about half (Fig. 10(b)).
Fig. 10(c) shows again that the main reason for such a
signicant underestimation in old VIC-3L is due to its
signicant underestimation of surface runo rather than
the subsurface runo. Due to such a signicant underestimation in the old VIC-3L parameterization, the soil
moisture in both layers with the old VIC-3L is higher
than that with the new VIC-3L (Fig. 10(d)). For the
second largest storm right after the big one, the total
runo dierence between the new and old VIC-3L reduced signicantly and both underestimated the observation (Fig. 10(b)). The reason for the reduced dierence

between the new and old VIC-3L is due to the larger


contribution of the subsurface runo from the old VIC3L, which is resulted from its higher soil moisture (Fig.
10(b)(d)). After the two big storms, the old VIC-3L
overestimated low ows due to higher subsurface runo
caused by less surface runo but more inltration from
previous large storms. However, the low ows estimated
by the new VIC-3L compared well with the observations
(Fig. 10(b)).
Fig. 11 shows the comparison for watershed 3 from
December 1996 to April 1997. The soil moisture in lower
layer (Fig. 11(d)) is generally wetter than that shown in
Figs. 9 and 10. From Fig. 11, it can be seen clearly that

Fig. 9. Comparison of daily observed streamow with total runo from the new and old VIC-3L model at watershed 1 between July and September
1991: (a) daily precipitation time series; (b) comparison of daily total runo among the observed streamow (dotted line), new VIC-3L (solid line),
and old VIC-3L (dashdot line); (c) comparison of daily surface and subsurface runo between the new VIC-3L (solid line for surface runo and
dashed line for subsurface runo) and old VIC-3L (dashdot line for surface runo and dotted line for subsurface runo); and (d) comparison of
daily upper and lower layer soil moisture between the new VIC-3L (solid line for upper layer and dashed line for lower layer) and old VIC-3L (dash
dot line for upper layer and dotted line for lower layer).

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

1189

Fig. 10. Comparison of daily observed streamow with total runo from the new and old VIC-3L model at watershed 2 between October and
December 1997: (a) daily precipitation time series; (b) comparison of daily total runo among the observed streamow (dotted line), new VIC-3L
(solid line), and old VIC-3L (dashdot line); (c) comparison of daily surface and subsurface runo between the new VIC-3L (solid line for surface
runo and dashed line for subsurface runo) and old VIC-3L (dashdot line for surface runo and dotted line for subsurface runo); and (d)
comparison of daily upper and lower layer soil moisture between the new VIC-3L (solid line for upper layer and dashed line for lower layer) and old
VIC-3L (dashdot line for upper layer and dotted line for lower layer).

both the new and old VIC-3L compare well with observed
streamow for a series of small rainfalls associated with
wet soil moisture conditions. For the observed peak ow
in early December, the new VIC-3L simulates higher peak
with more narrow width, while the old VIC-3L simulates
lower peak with wider width which is contributed by its
larger subsurface runo (Fig. 11(b)(d)).
The comparisons between observed streamow and
the total runo from the new and old VIC-3L clearly indicated again that it is very important to consider the inltration excess surface runo to simulate more
reasonably both large peak ows (Figs. 911), and low
ows (Fig. 10). The underestimation of peak ows with

the old VIC-3L can result in signicant higher low ows


(Fig. 10). The comparison also shows that the dierences
between the new and old VIC-3L could be small
with smaller rainfalls and moist soil moisture conditions
(Fig. 11).
5. Conclusions
A new surface runo parameterization for land surface models is presented in this paper. The new
parameterization dynamically considers the two major
surface runo generation mechanisms, inltration excess

1190

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

Fig. 11. Comparison of daily observed streamow with total runo from the new and old VIC-3L model at watershed 3 between December 1996 and
April 1997: (a) daily precipitation time series; (b) comparison of daily total runo among the observed streamow (dotted line), new VIC-3L (solid
line), and old VIC-3L (dashdot line); (c) comparison of daily surface and subsurface runo between the new VIC-3L (solid line for surface runo
and dashed line for subsurface runo) and old VIC-3L (dashdot line for surface runo and dotted line for subsurface runo); and (d) comparison of
daily upper and lower layer soil moisture between the new VIC-3L (solid line for upper layer and dashed line for lower layer) and old VIC-3L (dash
dot line for upper layer and dotted line for lower layer).

runo and saturation excess runo, within a studied


area or a model grid cell. Also, the eects of spatial
subgrid-scale variability of soil heterogeneity on surface
runo is considered eectively in the new parameterization. Although the new parameterization is developed
under the context of VIC model frame work, the concept and methodology of the new parameterization can
be applied to other land surface models. In this study,
three types of numerical experiments were conducted.
Also, the new surface runo parameterization was implemented into the current version of VIC-3L and tested
over three watersheds in Pennsylvania. The primary
conclusions are summarized below.

1. Inltration and saturation excess runo are two important surface runo generation mechanisms. Lacking one of them could result in signicant errors in
producing the total runo and soil moisture for applications at the spatial scales where subgrid-scale
spatial variability of soil heterogeneity is signicant.
2. The new parameterization plays a very important role
in partitioning the water budget between surface runo and soil moisture in the atmosphereland coupling
system. The original VIC approach can result in signicant underestimation of the surface runo and
overestimation of soil moisture during the period of
large storms over the studied cases. The overestima-

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

3.

4.

5.

6.

7.

tion of soil moisture can result in signicant overestimation of low ows following the large storms.
It is very important to consider the inltration excess
runo mechanism within the context of subgrid-scale
spatial variability of soil heterogeneity and precipitation for land surface models, because the Horton runo is closely related to the intensity of precipitation.
The numerical experiments show that larger surface
runo dierences are resulted with shorter time steps
between the new and original VIC surface runo
parameterizations. This implies that for studies with
similar soil characteristics over similar spatial scales
(e.g., similar size of model grid box), the inltration
excess runo mechanism may be more important
for applications in the atmosphereland coupling system than for the general hydrological applications.
This is because the computational time step Dt in
the atmosphereland coupling system is generally
much shorter (e.g., less than or equal to 1 h) than that
used in general hydrological applications (e.g., 6 h or
daily time step).
Due to the close relationship between the potential inltration rate (i.e., inltration capacity) and soil
moisture state, the spatial distribution of potential inltration rate should vary with time. An assumption
of time-invariant spatial distribution of potential inltration rate may result in large errors in surface runo
and soil moisture as shown in the numerical experiments between the dynamic and static congurations.
Comparison studies with the dynamic, static, and
original VIC congurations show that VIC generates
the least surface runo and results in the most moist
soil, the static conguration produces the largest surface runo and the driest soil moisture state. The results of dynamic conguration are bounded between
the original VIC and static congurations.
The total surface runo from the new parameterization is less sensitive to the choice of the values of
the shape parameter b than that of the original VIC
parameterization.
Comparisons using real storm data from Iowa show
that the runo ratio with VIC is about ve times lower
than that with the dynamic conguration, and that
about 18% more water inltrates into the soil with
the VIC than that with the new parameterization over
about 9 h.

Acknowledgements
We thank Dr. A. Allen Bradley for the Iowa rainfall
data, and Dr. Dennis P. Lettenmaier for the Pennsylvania vegetation and soil information. The research reported herein was supported by the National Oceanic
and Atmospheric Administration under grant
NA86GP0595 to University of California at Berkeley.

1191

Appendix A. Description for unique solution of Eq. (3)


As discussed in Section 2, there exists an unique
solution y to partition the precipitation into saturation
excess runo, inltration excess runo, and soil moisture
change. A detailed proof for the unique solution y of Eq.
(3) is provided below.
Theorem 1. For precipitation P > 0, Eq. (3) has an
unique solution y 2 0; P , where R1 y, DW y, and R2 y
are expressed by (5), (6), and (8), respectively.
Proof. From Eq. (4), Eq. (3) is equivalent to
F y 0;

A:1

where F y R2 y y P . Therefore, we only need to


prove that Eq. (A.1) has an unique solution in 0; P .
Let
gy R1 y fm Dt;

y 2 0; P :

A:2

It follows from Eqs. (A.2) and (5) that


dgy dR1 y

dy
dy

1 1

1;

i0 y b
;
im

0 6 y 6 im i0 ;
im i0 6 y 6 P :

A:3

Therefore, for every y 2 0; P ; dgy=dy > 0. Hence,


gy is a strictly increasing function in 0; P . In order to
prove that Eq. (A.1) has an unique solution, we need
to further express R2 y represented by Eq. (8) according
to the following three dierent cases: (a) g0 P P ; (b)
gP 6 P ; and (c) g0 < P < gP .
(a) g0 P P . In this case, for every y 2 0; P ,
gy P P , and P R1 y=Dt 6 fm . Thus, Eq. (8) becomes
"

B1 #
P R1 y
R2 y P R1 y fmm Dt 1
1
:
fm Dt
A:4
(b) g0 6 P . In this case, for every y 2 0; P , gy 6 P ,
and P R1 y=Dt P fm . Thus, Eq. (8) becomes
R2 y P

R1 y

fmm Dt:

A:5

(c) g0 < P < gP . In this case, there is an unique


y0 2 0; P such that gy0 P , since gy is strictly increasing in 0; P . Hence P R1 y=Dt P fm when
0 6 y 6 y0 , and P R1 y=Dt 6 fm when y0 6 y 6 P .
Thus, Eq. (8) becomes
8


B1 
>
P R1 y
>
1
;
>
> P R1 y fmm Dt 1
fm Dt
>
<
y0 6 y 6 P ;
R2 y
>
>
>
P
R1 y fmm Dt;
>
>
:
0 6 y 6 y0 :
A:6

1192

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193

With R2 y now expressed for the above three dierent


cases, we can prove that Eq. (A.1) has an unique solution y 2 0; P .
For case (a),
h
i
F y y R1 y fmm Dt 1 1 T yB1 ;
A:7
where T y P R1 y=fm Dt. It is obvious that
0 < T 0 6 1, 0 < T P < 1. Hence
F 0

fmm Dt1

T 0B1 < 0

A:8

and
F P fmm DtB 1T P

1 1

B1

T P

> 0:

A:9

d
y
dy
"
 1

R1 y

1

d
y
dy
1


1

fmm Dt

R1 y
fm Dt


R1 y 1
1

T yB

d
dy
B1 #

R1 y
fm Dt

B

d
R1 y
dy

 d
R1 y;
dy

A:10

where Eq. (9) is used. We now estimate the second term


in the right-hand side of (A.10). From (A.3) we have
0<

dR1 y
6 1;
dy

8y 2 0; P :

A:11

It is obvious that for y 2 0; P


0 < T y < 1 in the case of a:

A:14
In case (c), T y0 1 and T y < 1 when y0 < y < P .
Therefore, from Eqs. (A.14) and (A.11), we have

> 0; y0 < y < P ;
F 0 y
A:15
P 0; 0 < y < y0 :
Hence F y is increasing in 0; P , and is strictly increasing in y0 ; P . We can obtain that

From Eqs. (A.8) and (A.9), and the continuity of


function F y in 0; P , we can see that F y 0 has at
least one solution in 0; P . We now prove that F y 0
is strictly increasing in 0; P , and hence F y 0 has at
most one solution in 0; P .
From Eq. (A.7) we have
F 0 y

which is continuous in y 2 0; P . Therefore,


(
1 1 1 T yB dRdy1 y ; y0 < y < P ;
0
F y
1 d1dyy ;
0 < y < y0 :

A:12
0

It follows from Eqs. (A.10)(A.12) that F y > 0 for


y 2 0; P . Hence F y is a strictly monotone increasing
function in 0; P . Therefore, F y 0 (i.e., Eq. (A.1))
has an unique solution in 0; P for case (a). Similarly,
we can prove that F y 0 has an unique solution in
0; P for case (b).
For case (c), we have
8
B1
>
y R1 y fmm Dt1 1 T y Dt;
>
>
>
< y 6 y 6 P;
0
F y
>
y
R1 y fmm Dt;
>
>
>
:
0 6 y 6 y0 ;
A:13

F y < F y0 y0

P < 0;

0 6 y < y0

A:16

and
F P fmm DtT P B 1
> 0:

1 1

T P B1
A:17

Eqs. (A.16) and (A.17), together with the continuity


of function F y, imply that F y 0 has at least one
solution in y0 ; P .
Since F y is a strictly increasing function in y0 ; P
and F y < 0 in 0; y0 , F y 0 has at most one solution
y 2 0; P . Therefore, F y 0 has an unique solution in
case (c). This completes the proof of the theorem. 
References
[1] Avissar R, Pielke RA. A parameterization of heterogeneous land
surface for atmospheric numerical models and its impact on
regional meteorology. Mon Weather Rev 1989;117:211336.
[2] Berndtsson R. Application of inltration equations to a catchment with large spatial variability in inltration. Hydrol Sci J
1987;32(3):399413.
[3] Beven KJ, Kirkby MJ. A physically based variable contributing
area model of basin hydrology. Hydrol Sci Bull 1979;24(1):4369.
[4] Burgin RH, Luthin JN. A test of the single and double ring type
inltrometers. Trans Am Geophys Union 1956;37:18992.
[5] Burnash RJC. The NWS river forecast system catchment
modeling. In: Singh VP, editor. Computer models of watershed
hydrology. Highlands, CO: Water Resources Publ; 1995. p. 311.
[6] Chen F, Mitchell K, Schaake J, Xue Y, Pan H-L, Koren V, Duan
QY, Ek M, Betts A. Modeling of land surface evaporation by four
schemes and comparison with FIFE observations. J Geophys Res
1996;101(D3):725168.
[7] Chen T et al. The project for intercomparison of land-surface
parameterization schemes (PILPS) phase 2(c) Cabauw experimental results from the project for intercomparison of landsurface parameterization schemes (PILPS). J Clim 1997;10:1194
215.
[8] Cherkauer KA, Lettenmaier DP. Hydrologic eects of frozen soils
in the upper Mississippi river basin. J Geophy Res 1999;104(D16):
19599610.
[9] Chow VT, Maidment DR, Mays LW. In: Applied hydrology.
New York: McGraw-Hill; 1988. p. 572.
[10] Crawford NH, Linsley RK. Digital simulation in hydrology:
Stanford watershed model IV. Technical Report, Department of
Civil Engineering, Stanford University, 1966, p. 210.

X. Liang, Z. Xie / Advances in Water Resources 24 (2001) 11731193


[11] Desborough CE, Pitman AJ. The BASE land surface model.
Global Planet Change 1998;19:318.
[12] Dickinson RE, Henderson-Sellers A, Kennedy PJ. Biosphere
atmosphere transfer scheme (BATS) version 1e as coupled to the
NCAR community climate model. NCAR/TN-387, 1993, p. 72.
[13] Dingman SL. Physical hydrology. Englewood Clis, NJ: PrenticeHall; 1994. p. 575.
[14] Dirmeyer PA, Dolman AJ, Sato N. The pilot phase of the global
soil wetness project: a pilot project for global land surface
modeling and validation. Bull Amer Meteor Soc 1999;80:85178.
[15] Entekhabi D, Eagleson PS. Land surface hydrology parameterization for atmospheric general circulation models including
subgrid scale spatial variability. J Clim 1989;2:81631.
[16] Famiglietti JS, Wood EF. Evapotranspiration and runo from large
land areas; land surface hydrology for atmospheric general circulation models. In: Wood EF, editor. Land surfaceatmospheric
interactions for climate modeling: observations, models and analysis. Norwell, MA: Kluwer Academic Publishers; 1991. p. 179204.
[17] Green WH, Ampt A. Studies on soils physics, I. Flow of air and
water through soils. J Agric Sci 1911;4:124.
[18] Horton RE. Analysis of runo plat experiments with varying
inltration capacity. Trans Am Geophys Union 1939;20:693711.
[19] Koren VI, Finnerty BD, Schaake JC, Smith MB, Seo DJ, Duan
QY. Scale dependencies of hydrologic models to spatial variability
of precipitation. J Hydrol 1999;217:285302.
[20] Koster RD, Suarez MJ. Energy and water balance calculations in
the Mosaic LSM. NASA Tech Memorandum 104606 1996;9:58.
[21] Irannejad P, Shao Y. Description and validation of the atmosphereland-surface interaction scheme (ALSIS) with HAPEX
and Cabauw data. Global Planet Change 1998;19:87114.
[22] Liang X, Lettenmaier DP, Wood EF, Burges SJ. A simple
hydrologically based model of land surface water and energy
uxes for general circulation models. J Geophy Res 1994;99(D7):
1441528.
[23] Liang X, Lettenmaier DP, Wood EF. One-dimensional statistical
dynamic representation of subgrid spatial variability of precipitation in the two-layer variable inltration capacity model. J
Geophy Res 1996;101(D16):2140322.
[24] Liang X, Lettenmaier DP, Wood EF. Surface soil moisture
parameterization of the VIC-2L model: evaluation and modication. Global Planet Change 1996;13:195206.
[25] Liang X, Wood EF, Lettenmaier DP, Lohmann D, Boone A,
Chang S, et al. The project for intercomparison of land-surface
parameterization schemes (PILPS) phase 2(c) Red-Arkansas river
basin experiment: 2. Spatial and temporal analysis of energy
uxes. Global and Planet Change 1998;19:13759.
[26] Loague K, Gander GA. Spatial variability of inltration on a
small rangeland catchment. Water Resour Res 1990;26:95771.
[27] Lohmann D, Lettenmaier DP, Liang X, Wood EF, Boone A,
Chang S, et al. The project for intercomparison of land-surface
parameterization schemes (PILPS) phase 2(c) Red-Arkansas river
basin experiment: 3. Spatial and temporal analysis of water uxes.
Global Planet Change 1998;19:16179.
[28] Mahrt L, Pan H. A two-layer model of soil hydrology. Boundary
Layer Meteorol 1984;29:120.
[29] Manabe S. Climate and the ocean circulation, 1, The atmospheric
circulation and the hydrology of the earth's surface. Mon Weather
Rev 1969;97:73974.

1193

[30] Manley RE. The soil moisture component of mathematical


catchment simulation models. J Hydrol 1977;35:34156.
[31] Mengelkamp HT, Warrach K, Raschke E. SEWAB a parameterization of the surface energy and water balance for atmospheric and hydrologic models. Adv Water Resour 1999;23:
16575.
[32] Moore RJ. The probability-distributed principle and runo
production at point and basin scales. Hydrol Sci J 1985;30(2):
27397.
[33] Nijssen B, Lettenmaier DP, Liang X, Wetzel S, Wood EF.
streamow simulation for continental-scale river basins. Water
Resour Res 1997;33:71124.
[34] Noilhan J, Planton S. A simple parameterization of land surface
processes for meteorological models. Mon Weather Rev
1989;117:53649.
[35] Philip JR. The theory of inltration, 4, Sorptivity and algebraic
inltration equations. Soil Sci 1957;84:25764.
[36] Rawls W, Yates P, Asmussen L. Calibration of selected inltration equations for the Georgia coastal plain. US Department of
Agriculture, Agric. Res. Serv., ARS-S-113, Washington, DC,
1976, p. 110.
[37] Salvucci GD, Entekhabi D. Equivalent steady soil moisture
prole and the time compression approximation in water balance
modeling. Water Resour Res 1994;30:273749.
[38] Schaake JC, Koren VI, Duan QY, Mitchell K, Chen F. Simple
water balance model for estimating runo at dierent spatial and
temporal scales. J Geophys Res 1996;101(D3):746175.
[39] Shao Y, Henderson-Sellers A. Validation of soil moisture simulation in landsurface parameterization schemes with HAPEX
data. Global Planet Change 1995;13:1146.
[40] Verseghy DL. CLASS-A Canadian land surface scheme for
GCMS. I. Soil model. Int J Climato 1991;11:11133.
[41] Viterbo P, Beljaars CM. An improved land surface parameterization scheme in the ECMWF model and its validation. J Clim
1995;8:271648.
[42] Wetzel PJ, Boone A. A parameterization for landatmosphere
cloud exchange (PLACE): documentation and testing of a detailed
process model of the partly cloudy boundary layer over heterogeneous land. J Clim 1995;8(7):181037.
[43] Wetzel PJ, Liang X, Irannejad P, Boone A, Noilhan J, Shao Y,
Skelly C, Xue Y, Yang Z-L, et al. Modeling vadose zone liquid
water uxes: inltration, runo, drainage, interow. Global
Planet Change 1996;13:5771.
[44] Wood EF. Global scale hydrology: advances in land surface
modeling. US Natl Rep Int Union Geod Geophys 19871990 Rev
Geophys 1991;29(Suppl.):193201.
[45] Wood EF, Lettenmaier DP, Zartarian VG. A land-surface
hydrology parameterization with subgrid variability for general
circulation models. J Geophys Res 1992;97(D3):271728.
[46] Wood EF, Lettenmaier DP, Liang X, Lohmann D, Boone A,
Chang S, et al. The project for intercomparison of land-surface
parameterization schemes (PILPS) phase 2(c) Red-Arkansas
River basin experiment: 1. Experiment description and summary
intercomparisons. Global Planet Change 1998;19:11535.
[47] Xue Y, Sellers PJ, Kinter JL, Shukla J. A simplied biosphere
model for global climate studies. J Clim 1991;4(3):34564.
[48] Zhao RJ. The Xianjiang model applied in China. J Hydrol
1992;135:37181.

You might also like