316

Assessment of Two Physically-Based

Watershed Models Based on Their
Performances of Simulating Water and
Sediment Movement

Latif Kalin, Mohamed M. Hantush

Abstract

Two physically based watershed models, GSSHA
and KINEROS-2, are evaluated and compared for
their performances on modeling flow and sediment
movement. Each model has a different watershed
conceptualization. GSSHA divides the watershed
into cells, and flow and sediments are routed through
these cells in a cascading fashion. Conversely,
KINEROS-2 divides the watershed into sub-
watersheds and channel segments having uniform
properties. GSSHA requires much longer simulation
times depending on what is simulated. KINEROS-2,
on the other hand, entails relatively less data and
effort. Simulations were performed with each model
over a small watershed for several events. Models
were calibrated using the same events and the
differences in estimated parameters were discussed.
Both models have resulted in different calibration
parameters although the underlying physics are
similar. The differences in model behaviors are
discussed.

Keywords: sediment, distributed models,
watershed, GSSHA, KINEROS-2

Introduction

Hydrologic models are useful tools in understanding
the natural processes in a watershed. For instance,
one practical application is analyzing the effect of
land use changes, such as urbanization, on runoff

Kalin is an ORISE Postdoctoral Researcher and
Hantush is a Research Hydrologist, both at the
United States Environmental Protection Agency,
National Risk Management Research Laboratory,
Cincinnati, OH 45268. E-mail: kalin.latif@epa.gov.


and sediment yield. There are numerous watershed
scale hydrologic models varying from lumped such
as the unit hydrograph concept (Sherman 1932) to
highly complex distributed models such as MIKE-
SHE (Refsgaard and Storm 1995). Each of those
models has their own advantages and disadvantages.
Depending on needs, sometime a simple lumped
model might suffice. However, to achieve TMDL
targets and implement BMPs, use of distributed
models is inevitable. The availability of high power
computers relaxed the burden of long simulation
times. Among the distributed models the physically
based ones always have edges over the empirical
ones, since the model parameters have physical
meanings and can be measured in the field. When
measurements are not available model parameters
can be still be deduced from published data in
literature based on topography, soil and land use
maps. When flow is concerned, to our knowledge
three models seem to be the most physically based
and separate themselves from others: GSSHA
(Downer and Ogden 2002), KINEROS-2 (Smith et
al. 1995) and MIKE-SHE (Refsgaard and Storm
1995). In this study we examined and compared the
former two. In what follows is a brief discussion of
each model.

GSSHA

Gridded Surface Subsurface Hydrologic Analysis
(GSSHA) is a reformulation and enhancement of
CASC2D (Downer and Ogden 2002). The CASC2D
model was initiated at Colorado State University by
Pierre Julien as a two dimensional overland flow
routing model. In its final form, it is a distributed-
parameter, physically-based watershed model. Both
single event and continuous simulations are possible.
The U.S. Army Waterways Experiment Station
considered this model as very promising and
therefore fully incorporated this model into
317
Watershed Modeling System(WMS). Watershed is
divided into cells and water and sediment is routed
from one cell to another. It uses one and two-
dimensional diffusive wave flow routing at channels
and overland planes, respectively. Although only
Hortonian flows were modeled by employing Green-
Ampt (G-A) infiltration model in the initial versions,
GSSHA considers other runoff generating
mechanisms such as lateral saturated groundwater
flow, exfiltration, stream/groundwater interaction
etc. GSSHA offers two options for long-term
simulations: G-A with redistribution (Ogden and
Saghafian 1997) and the full Richards equation. The
latter requires tremendous amount of simulation time
and is very sensitive to time step and horizontal and
vertical cell sizes (Downer and Ogden 2003).
Modified Kilinc and Richardson equation (Julien
1995) is used to compute sediment transport
capacity at plane cells. A trap efficiency measure is
used to determine how much material is transported
from the outgoing cell. Details on theory and
equations used can be found in Julien et al. 1995,
Johnson et al. 2000, and Downer and Ogden 2002.

KINEROS-2

This is the improved version of KINEROS
(Woolhiser et al. 1990). It is event based since it
lacks a true soil moisture redistribution formulation
for long rainfall hiatus and more importantly it does
not consider evapotranspiration (ET) losses. This
model is primarily useful for predicting surface
runoff and erosion over small agricultural and urban
watersheds. Smith et al. 1995 suggest watershed size
smaller than 1,000 ha for best results. Runoff is
calculated based on the Hortonian approach using a
modified version of Smith- Parlange (Smith and
Parlange 1978) infiltration model. KINEROS-2
requires the watershed divided into homogeneous
overland flow planes and channel segments, and
routs water movement over these elements in a
cascading fashion. Mass balance and the kinematic
wave approximations to the Saint Venant equations
are solved with implicit finite difference numerical
scheme in a 1-D framework. KINEROS-2 accounts
for erosion resulting from raindrop energy and by
flowing water separately. A mass balance equation is
solved to describe sediment dynamics at any point
along a surface flow path. Erosion is based on
maximum transport capacity determined by
Engelund-Hansen equation (1967). The rate of
sediment transfer between soil and water is defined
with a first order uptake rate. A detailed description
of the model and the equations used can be found in
Smith et al. 1995 and at the official URL of the
model: http://www.tucson.ars.ag.gov/kineros.

Data

The data used in this study comes from a small
USDA experimental watershed named W-2, which
is located near Treynor, Iowa. It is approximately 83
acres. Figure 1 depicts the location and topography
of this watershed. This watershed is one of the 4
experimental watersheds established by USDA in
1964 to determine the effect of various soil
conservation practices on runoff and water-induced
erosion. Runoff and sediment load has been
measured since then. There are two rain gauges (115
and 116) around the watershed. W-2 has a rolling
topography defined by gently sloping ridges, steep
side slopes, and alluvial valleys with incised
channels that normally end at an active gully head,
typical of the deep loess soil in MLRA 107 (Kramer
et al. 1990). Slopes usually change from 2 to 4
percent on the ridges and valleys and 12 to 16
percent on the side slopes. An average slope of about
8.4 percent is estimated, using first-order soil survey
maps. The major soil types are well drained Typic
Hapludolls, Typic Udorthents, and Cumulic
Hapludolls (Marshall-Monona-Ida and Napier
series), classified as fine-silty, mixed, mesics. The
surface soils consist of silt loam (SL) and silty clay
loam (SCL) textures that are very prone to erosion,
requiring suitable conservation practices to prevent
soil loss (Chung et al. 1999). Corn has been grown
continuously on W-2 since 1964.

318

Figure 1. Study watershed.


Methodology

KINEROS-2 was already calibrated for W-2
watershed in a previous study using 3 rainfall events
(Kalin and Hantush 2003). In that study average
values were used for net capillary drive, G (35,20
cm), pore size distribution index, (0.6,0.6),
porosity, (0.47,0.50), and median particle size
diameter, D
50
(7 m). The two values given in
parentheses represent SCL and SL soil types,
respectively. Table 1 lists the parameter sets used
after calibration of KINEROS-2. In the table, n is
Mannings roughness, K
s
is saturated hydraulic
conductivity, I is interception depth, S
i
is initial
saturation, C
g
is soil cohesion coefficient and C
f
is
rainsplash coefficient. For simplicity channel and
overland roughness were assumed to be same. Since
corn has been grown on W-2, the parameters n, C
g

and C
f
were allowed to vary with season where C
g

and C
f
were assumed to decay exponentially with the
growing season. This assumption was justified over
four independent verification events (see Kalin and
Hantush 2003).

Table 1. Parameter sets used in KINEROS-2.

event n K
s
*
I
**
S
i
C
g
C
f

6/13/83 0.055 (1.8,6.5) 2.0 (0.44,0.27) 0.15 160
5/30/82 0.040 (1.5,6.0) 0.0 (0.90,0.86) 0.25 200
8/26/81 0.080 (2.0,7.0) 1.0 (0.84,0.60) 0.05 100
*
K
s
: mm/hr
**
I: mm


Flow simulations

GSSHA was run with the above events. KINEROS-
2 values were directly substituted for parameters
common to both models i.e. , , n, I, S
i
, and K
s
.
Other parameters were adjusted accordingly. The
infiltration scheme in GSSHA is the Green-Ampt
(G-A) model, whereas KINEROS-2 uses Smith-
Parlange infiltration model. G-A capillary head ()
needs to be provided in GSSHA. We approximated
as equal to G in KINEROS-2. Figure 2 shows the
comparison of the simulation results for flow with
two models. It is clear that both models behave very
differently when similar parameter sets are used as
inputs. The most striking observation is that, in all
cases GSSHA generates later responses and lower
peak flows than KINEROS-2. For instance, the
difference in time to peaks for the event 8/26/81 is
around 25 minutes which is very significant
considering the fact that the base time is around 150
minutes. Similarly, the peak flow generated by
KINEROS-2 is about 45 % larger than the peak flow
generated by GSHHA. One possible rationale to this
might be the different watershed conceptualizations
involved in each model. Flow routing in GSSHA is
only in x-y directions. In other words, flow from a
cell is allowed only in the four principal directions.
Diagonal neighboring cells can not be receivers
which well might be the reality. This results in
overestimation of the travel lengths of water
particles which might be up to 41 %. On the other
hand, the travel paths used to compute the average
travel lengths of each element in KINEROS-2 were
determined based on the D-8 methodology using the
TOPAZ algorithm (Garbrecht and Martz 1999)
which allows flow in 8 directions. Considering the
fact that flow in the study watershed is mostly
diagonal, the overestimation of travel lengths by
GSSHA resulted in longer travel time leading to
more resistance to flow, and consequently lower and
retarded peaks.

W-
IA
N
319
6/13/83
0
1
2
3
4
60 90 120 150
time (min)
flow
(m
3
/s)
kineros-2
gssha
observed
5/30/82
0.0
0.1
0.2
0.3
0.4
0.5
40 80 120 160 200
time (min)

8/26/81
0.00
0.05
0.10
0.15
0.20
0.25
30 80 130 180
time (min)
flow
(m
3
/s)

Figure 2. Comparison of hydrographs generated with
GSSHA and KINEROS-2 based on KINEROS-2
calibrated parameters.


The differences in flow volumes do not seem to be
significant. With this set of parameters KINEROS-2
seems to simulate events having multi-modal shapes,
such as the one in 5/30/82, better than GSSHA. In
fact GSSHA completely misses the first and second
humps in 5/30/82 as opposed to KINEROS-2.
KINEROS-2, to some extent, performs better than
GSSHA in simulating the small hump seen on the
observed data of 8/26/81.

It is important to keep in mind that all these
observations are based on simulations with the
parameters calibrated for KINEROS-2. Therefore,
we recalibrated the GSSHA parameters for the same
events. This time each event was calibrated
individually and parameters were compared to
KINEROS-2 calibrated parameters. We accept that
we did not follow the traditional model calibration/
verification methodology. However, we need to
mention that the aim of this study is basically a
comparison of the two models rather than a model
calibration effort. Keeping this in mind, we kept I, S
i

and the overland plane roughness (n
p
) same and
recalibrated channel roughness (n
c
) and K
s
. Figure 3
shows the hydrographs after calibration. For the
event 6/13/83 both model performs equally. For
5/30/82 GSSHA is still underestimating the first and
second humps, but interestingly it does a better job
than KINEROS-2 in representing the shape.
Although KINEROS-2 could not simulate the first
(the smallest hump in the figure) happening
approximately at 45 minutes, GSSHA does a fairly
good job in catching the both humps. Finally, when
we look at the last event we see that GSSHA almost
perfectly reproduces the observed hydrograph shape
while KINEROS-2 suffers to simulate the first peak.

5/30/82
0.0
0.1
0.2
0.3
0.4
0.5
40 80 120 160 200
time (min)
6/13/83
0
1
2
3
4
60 90 120 150
time (min)
flow
(m
3
/s)
kineros-2
gssha
observed

8/26/81
0.00
0.05
0.10
0.15
0.20
0.25
30 80 130 180
time (min)
flow
(m
3
/s)

Figure 3. Comparison of hydrographs generated with
GSSHA and KINEROS-2. GSSHA is recalibrated.


The recalibrated parameters for GSSHA are
summarized in Table 2. In the table C is the USLE
crop factor which will be discussed later. The value
of n
c
had to be decreased dramatically for each event
that is clearly expected from Figure 2 as GSSHA
generated later responses in each case. One
remarkable observation is that n
c
values are very
close to each other which confirms the comments of
Larry Kramer (personal communication) who has
extensive experience on Treynor watersheds. He
stated that channels are covered with bromegrass
and they are cultivated such a way that channel
roughness can be assumed invariable year around.
K
S
values are very close to KINEROS-2 values.


Table 2. Calibrated parameters with GSSHA.

event n
c
K
s
(mm/hr) C
6/13/83 0.025 (2.0, 7.7) 0.042
5/30/82 0.020 (1.5, 6.0) 0.150
8/26/81 0.025 (1.8, 6.5) 0.050

320
Erosion simulations

GSSHA requires silt and sand percentages for
sediment computations. Assuming that D
50
is 0.25
mm for sand, 0.016 mm for silt and 0.003 for clay,
compositions of each soil class were determined as
sand % (10,25) and silt % (56,61) so that the overall
average D
50
is 7 mm, which is the value used in
KINEROS-2. The sediment routine in GSSHA is
based on the USLE concept that requires three
parameters K, C and P. It is not practical to infer
estimates of these parameters from the KINEROS-2
soil parameters; i.e., C
g
and C
f
. Therefore, by
keeping KP product constant C was calibrated for
each event, since it is only the product of K, C, and
P that matters. The values of K and P are (0.37,0.48)
and (0.01,0.01), correspondingly. The estimated C
values are listed in Table 2. The pattern observed in
KINEROS-2 that is erodibility decreases with the
growing season, is not observed between the C
values here. The C values obtained for the event
8/26/1981 is unexpectedly high, even higher than the
value of 6/13/83. Figure 4 compares the
sedimentographs obtained by KINEROS-2 and
GSSHA. The general observation is that GSSHA
generates narrower sedimentographs than
KINEROS-2 generates. We do not have a clear
reasoning for this. Further, this can not be attributed
to flow, since such a behavior is not monitored in
Figure 3.

It is interesting to note that the erosion parameters, c
f

and c
g
, found after calibration for KINEROS-2 are
well above the recommended values given in
Woolhiser et al. (1990) and the calibrated C
parameters for GSSHA are well below the literature
values. What this means is that when literature
values are used, GSSHA overestimates erosion
compared to KINEROS-2. Slope is an important
factor in both models erosion formulation. The
smaller the computational element, which is the grid
size for GSSHA and the average length of overland
flow planes in KINEROS-2, the greater the erosion.
Because, as the element size increases the tendency
of smoothing the topography increases, and this
results in loss of areas with steep slopes meaning
reduction in erosion. KINEROS-2 uses far less
elements than GSSHA, thus leading to loss of local
slope information in the former. This probably
elucidates the difference in estimates of soil erosion.
A very good discussion on this topic can be found in
Rosalia 2002.

6/13/1983
0
150
300
450
60 90 120 150
time (min)
Qs
(kg/s) kineros-2
gssha
obs
5/30/1982
0
10
20
30
40 80 120 160
time (min)
Qs
(kg/s)

8/26/1981
0
2
4
6
40 80 120 160
time (min)
Qs
(kg/s)

Figure 4. Comparison of sedimentographs generated
with GSSHA and KINEROS-2.

Discussions and Conclusions

It is known that in numerical solutions involving
finite difference schemes, as the grid size decreases
the required time interval should also decrease. In
fact, this is reflected in the Courant Condition as a
stability criteria which can be stated as U<x/t
where U is velocity, and t and x are time and
space increments, respectively (Chapra 1997). The
grid size used for W-2 in GSSHA simulations was
10 m. This is an unusually small grid size for such
simulations. In fact, 5 m horizontal resolution DEM
data is also available for this area, but because of the
interaction between t and x we decided to use 10
m. Using coarser grid size than 10 m would lead to
inaccurate representation of the watershed since it is
only 83 acres. In a review of several watershed scale
hydrologic and non-point source pollution models,
Borah (2002) refers to a study on CASC2D, the
older version of GSSHA, where Molnar and Julien
(2000) found that for a 150 m grid size the required
time step was about 5 seconds. This number
decreased to 1 second when the grid size was
reduced to 30 m. The smallest time interval allowed
by GSSHA is 1 second which is the value used in
our simulations. This might have introduced
additional uncertainty.

One of the deficiencies of the GSSHA is that erosion
in channels is not transport limited. GSSHA can
generate sediment which has a volume larger than
flow. This is physically impossible; however there is
321
nothing in the GSSHA formulation to prevent this
from happening once sediment reaches the channels
(Downer, personal communication). When we
initially used the literature values for C, K, and P
parameters we observed this effect. Eventually we
had to decrease these parameters dramatically to get
more realistic results. This itself is enough to claim
that the sediment routine in KINEROS-2 is more
robust than the routine used in GSSHA. In fact, there
is a contract between US Army Corps of Engineers
and Fred Ogden, University of Connecticut, one of
model developers, to completely reformulate the
sediment routine of GSSHA (Downer and Ogden,
personal communication). It would be interesting to
redo this whole exercise once that project is
completed.

Acknowledgments

The U.S. Environmental Protection Agency through
its Office of Research and Development funded and
managed the research described here through in-
house efforts. This research was supported in part by
an appointment to the Postgraduate Research
Program at the National Risk Management Research
Laboratory administered by the Oak Ridge Institute
for Science and Education through an interagency
agreement between the U.S. Department of Energy
and the U.S. Environmental Protection Agency. The
authors acknowledge Charles Downer and Carl
Unkrich for their helpful discussions on the GSSHA
and KINEROS-2 models, respectively. The
reviewers, Dennis Lai and Mingyu Wang are also
appreciated.

References

Borah, D.K. 2002. Watershed scale non-point source
pollution models: Mathematical bases. 2002
American Society of Agricultural Engineers Annual
International Meeting/CIGR World Congress,
Chicago, IL. Paper number 022091.

Chapra, S.C. 1997. Surface-Water Quality
Modeling. McGraw Hill, New York.

Chung, S.W., P.W. Gassman, L.A. Kramer, J.R.
Williams, and R. Gu. 1999. Validation of EPIC for
two watersheds in southwest Iowa. Journal of
Environmental Quality 28:971-979.

Downer, C.W., and F.L. Ogden. 2002. GSSHA
users manual, gridded surface subsurface
hydrologic analysis version 1.43 for WMS 6.1.
ERDC Technical Report, Engineering Research and
Development Center, Vicksburg, MS.

Downer, C.W., and F.L. Ogden. 2003. Appropriate
vertical discretization of Richards equation for two-
dimensional watershed-scale modeling.
Hydrological Processes (in press).

Engelund, F., and E. Hansen. 1967. A Monograph
on Sediment Transport in Alluvial Streams. Teknisk
Vorlag, Copenhagen.

Garbrecht, J., and L.W. Martz. 1999. An automated
digital landscape analysis tool for topographic
evaluation, drainage identification, watershed
segmentation, and sub-catchment parameterization.
Report GRL 99-1, Grazinglands Research
Laboratory, USDA, Agricultural Research Service,
El Reno, OK.

Johnson, B.E., P.Y. Julien, D.K. Molnar, and C.C.
Watson. 2000. The two-dimensional upland erosion
model CASC2D-SED. Journal of American Water
Resources Association 36:31-41.

Julien, P.Y. 1995. Erosion and Sedimentation. Press
Syndicate of the University of Cambridge, New
York.

Julien, P.Y., B. Saghafian, and F.L. Ogden. 1995.
Raster-based hydrologic modeling of spatially-
varied surface runoff. Water Resources Bulletin
31:523-536.

Kalin, L., and M. M. Hantush. 2003. Modeling of
sediment yield in a small agricultural watershed with
KINEROS-2. In J. D. Williams and D. W. Koplin,
eds., American Water Resources Association 2003
Spring Specialty Conference on Agricultural
Hydrology and Water Quality, Kansas City, MO.
CD-ROM.

Kramer, L.A., E.E. Alberts, A.T. Hjelmfelt, and
M.R. Gebhardt 1990. Effect of soil conservation
systems on groundwater nitrate levels from three
corn-cropped watersheds in southwest Iowa. In J.
Lehr, ed, Proceedings of the 1990 Cluster of
Conference, Kansas City, MO.

322
Molnar, D.K., and P.Y. Julien. 2000. Grid size
effects on surface water modeling. Journal of
Hydrologic Engineering 5:8-16.

Ogden, F.L., and B. Saghafian. 1997. Green and
Ampt infiltration with redistribution. Journal of
Irrigation and Drainage Engineering 123:386-393.

Refsgaard, J.C., and B. Storm. 1995. MIKE-SHE. In
V.J. Singh, ed, Computer Models of Watershed
Hydrology, pp. 809-846. Water Resources
Publication, Highlands Ranch, CO.

Rosalia, R.S. 2002. GIS-based Upland Erosion
Modeling, Geovisualization and Grid Size Effects on
Erosion Simulations with CASC2D-SED. Ph.D.
Thesis, Colorado State University, Fort Collins, CO.

Sherman, L.K. 1932, Streamflow from rainfall by
the unit-graph method. Engineering News Record
108:501-505.

Smith, R.E., D.C. Goodrich, D.A. Woolhiser, and
C.L. Unkrich. 1995. A Kinematic Runoff and
Erosion Model. In V.J. Singh, ed., Computer Models
of Watershed Hydrology, pp. 697-732, Water
Resources Publication, Highlands Ranch, CO.

Smith, R.E., and J.Y. Parlange. 1978. A parameter-
efficient hydrologic infiltration model. Water
Resources Research 14:553-538.

Woolhiser, D.A., R.E. Smith, and D.C. Goodrich.
1990. KINEROS-A kinematic runoff and erosion
model: Documentation and user manual. USDA-
ARS, ARS-77.