HYDROLOGICAL PROCESSES

Hydrol. Process. 25, 915– 925 (2011)

Published online 8 October 2010 in Wiley Online Library
(wileyonlinelibrary.com) DOI: 10.1002/hyp.7876

Development and application of a physically based landscape

water balance in the SWAT model
Eric D. White,1 Zachary M. Easton,1 * Daniel R. Fuka,1 Amy S. Collick,2 Enyew Adgo,2
Matthew McCartney,3 Seleshi B. Awulachew,3 Yihenew G. Selassie4 and Tammo S. Steenhuis1
1 Department of Biological and Environmental Engineering, Cornell University, Ithaca, NY, USA
2 Department of Water Resource Engineering, Bahir Dar University, Bahir Dar, Ethiopia
3 International Water Management Institute, Nile Basin and East Africa Office, Addis Ababa, Ethiopia
4 Amhara Regional Agricultural Institute, Bahir Dar, Ethiopia

Abstract:
Watershed scale hydrological and biogeochemical models rely on the correct spatial-temporal prediction of processes governing
water and contaminant movement. The Soil and Water Assessment Tool (SWAT) model, one of the most commonly used
watershed scale models, uses the popular curve number (CN) method to determine the respective amounts of infiltration and
surface runoff. Although appropriate for flood forecasting in temperate climates, the CN method has been shown to be less
than ideal in many situations (e.g. monsoonal climates and areas dominated by variable source area hydrology). The CN
model is based on the assumption that there is a unique relationship between the average moisture content and the CN for
all hydrologic response units (HRUs), and that the moisture content distribution is similar for each runoff event, which is not
the case in many regions. Presented here is a physically based water balance that was coded in the SWAT model to replace
the CN method of runoff generation. To compare this new water balance SWAT (SWAT-WB) to the original CN-based SWAT
(SWAT-CN), two watersheds were initialized; one in the headwaters of the Blue Nile in Ethiopia and one in the Catskill
Mountains of New York. In the Ethiopian watershed, streamflow predictions were better using SWAT-WB than SWAT-CN
[Nash–Sutcliffe efficiencies (NSE) of 0Ð79 and 0Ð67, respectively]. In the temperate Catskills, SWAT-WB and SWAT-CN
predictions were approximately equivalent (NSE >0Ð70). The spatial distribution of runoff-generating areas differed greatly
between the two models, with SWAT-WB reflecting the topographical controls imposed on the model. Results show that a
water balance provides results equal to or better than the CN, but with a more physically based approach. Copyright  2010
John Wiley & Sons, Ltd.

KEY WORDS SWAT model; water balance; Ethiopia; monsoonal climate; New York; runoff; variable source area

Received 19 April 2010; Accepted 19 August 2010

INTRODUCTION is therefore not suitable for regions where the rainfall

Non-point source runoff can contribute significant quan-
tities of nutrients, chemicals, and sediments to stream
and water bodies. To locate these ‘non-point’ sources of
pollution in a landscape, many watershed managers and
researchers frequently use distributed watershed models.
One of the most commonly used watershed scale models
is the US Department of Agriculture (USDA) Soil and
Water Assessment Tool (SWAT) (Arnold et al., 1998).
SWAT, like any water quality model, must first accu-
rately simulate the hydrologic processes before it can
realistically predict pollutant transport. Many different
approaches to modelling the hydrologic processes have
been presented in the scientific literature over the past
several decades, but SWAT currently uses two meth-
ods to model surface runoff: the curve number (CN)
(USDA–SCS, 1972) or the Green–Ampt routine (Green
and Ampt, 1911). The Green–Ampt method is a physi-
cally based infiltration excess, rainfall–runoff model, and
* Correspondence to: Zachary M. Easton, Department of Biological and
Environmental Engineering, Cornell University, Ithaca, NY, USA.
E-mail: zme2@cornell.edu
rate seldom exceeds the saturated conductivity of the
soil, such as in the northeastern part of the USA (Wal-
ter et al., 2003). As a result, the empirically based CN
method is the most commonly used runoff routine in the
SWAT model (King et al., 1999; Gassman et al., 2007).
SWAT and other CN-based models are frequently used in
watersheds around the world with little recognition that
the underlying runoff calculations were neither developed
nor validated for these regions (Gassman et al., 2007).
Indeed, CN models have commonly been applied in
the Blue Nile basin of Africa. Located in the monsoonal
climate of the Ethiopian Highlands, the temporal runoff
dynamics in the Blue Nile basin are poorly captured by
the CN method, which assumes that the moisture content
distribution of the watershed is similar for each runoff
event (Collick et al., 2009; Steenhuis et al., 2009). Pre-
vious work in the Blue Nile basin has shown that for a
given amount of rain, runoff volumes will vary through-
out the rainy season. Liu et al. (2008) demonstrated that
in the Ethiopian Highlands, less runoff was generated at
the beginning of the rainy season as compared to the
same rain volume at the end of the season. Liu et al.
(2008) showed that when a 10-day watershed discharge

Copyright  2010 John Wiley & Sons, Ltd.

916 E. D. WHITE ET AL.
is plotted against effective precipitation (i.e. precipitation
minus potential evapotranspiration) there was no relation-
ship until approximately 500 mm of effective precipita-
tion. Once 500 mm of effective precipitation had fallen,
there was a relatively strong, linear relationship, indicat-
ing that the proportion of the rainfall that became runoff
was constant during the remainder of the rainy season.
Of course, this relationship can be partially explained
by channel losses in the early monsoonal period, or by
the filling of landscape depressions that must become
linked prior to the outlet hydrograph responding. How-
ever, in another, similar Blue Nile watershed, Ashagre
(2009) found that the 5-min rainfall intensity very rarely
exceeded the measured soil infiltration capacity, and
thus assumed excess saturation runoff was the dominate
runoff-generating mechanism. These dynamics cannot be
predicted by the CN, and are, in fact, in direct con-
trast to the CN approach, which assumes no correlation
between antecedent precipitation and a watershed’s maxi-
mum retention beyond 5 days (USDA–Natural Resource
Conservation Service (NRCS), 2004).
Many have attempted to modify the CN model to
work better during monsoonal climates, by proposing
various temporally based values and initial abstractions.
For instance, Bryant et al. (2006) suggested that a water-
shed’s initial abstraction should vary as a function of
storm size. Although this is a valid argument, the intro-
duction of an additional variable reduces the appeal
of the one-parameter CN model. Kim and Lee (2008)
found that SWAT was more accurate when CN values
were averaged across each day of simulation, rather than
using a CN that described moisture conditions only at
the start of each day. White et al. (2009) showed that
SWAT model results improved when the CN was changed
seasonally to account for watershed storage variation
due to plant growth and dormancy. Wang et al. (2008)
improved SWAT results by using a different relation-
ship between antecedent conditions and watershed stor-
age. Although these variable CN methods improve runoff
predictions, they are not easily generalized for use out-
side the watershed they are tested mainly because the
CN method is a statistical relationship and not physically
based.
In many regions, surface runoff is produced by only a
small portion of a watershed that expands with an increas-
ing amount of rainfall. This concept is often referred to
as a variable source area (VSA), a phenomenon actually
envisioned by the original developers of the CN method
(Hawkins, 1979), but never implemented in the original
CN method as used by the Natural Resource Conser-
vation Service (NRCS). Since the methods’ inception,
numerous attempts have been made to justify its use
in modelling VSA-dominated watersheds. These adjust-
ments range from simply assigning different CNs for wet
and dry portions to correspond with VSAs (Sheridan and
Shirmohammadi, 1986; White et al., 2009) to full rein-
terpretations of the original CN method (Hawkins, 1979;
Steenhuis et al., 1995; Schneiderman et al., 2007; Easton
et al., 2008).
Copyright  2010 John Wiley & Sons, Ltd.
To determine what portion of a watershed is produc-
ing surface runoff for a given precipitation event, the
reinterpretation of the CN method presented by Steen-
huis et al. (1995) and incorporated into SWAT by Easton
et al. (2008) assumes that rainfall infiltrates when the
soil is unsaturated or runs off when the soil is satu-
rated. It has been shown that this saturated contributing
area of a watershed can be accurately modelled spatially
by linking this reinterpretation of the CN method with
a topographic index (TI), similar to those used by the
topographically driven TOPMODEL (Beven and Kirkby,
1979; Lyon et al., 2004). This linked CN–TI method
has since been used in multiple models of watersheds in
the northeastern part of the USA, including the general-
ized watershed loading function (GWLF) (Schneiderman
et al., 2007) and SWAT (Easton et al., 2008). Although
the reconceptualized CN model is applicable in temper-
ate US climates, it is limited by the fact that it imposes
a distribution of storages throughout the watershed that
needs to fill up before the runoff occurs. Although this
limitation does not seem to affect results in temperate
climates, it results in poor model results in monsoonal
climates.
SWAT–VSA, the CN–TI adjusted version of SWAT
(Easton et al., 2008), returned hydrological simulations
as accurate as the original CN method; however, the
spatial predictions of runoff producing areas and as a
result the predicted phosphorus export were much more
accurate. Although SWAT–VSA is an improvement upon
the original method in watersheds where the topography
drives flows, ultimately, it still relies on the CN to model
runoff processes and therefore is limited when applied to
the monsoonal Ethiopian Highlands.
Water balance models are relatively simple to imple-
ment and have been used frequently in the Blue Nile
basin (Johnson and Curtis, 1994; Conway, 1997; Ayenew
and Gebreegziabher, 2006; Kim and Kaluarachchi, 2008;
Liu et al., 2008; Collick et al., 2009; Steenhuis et al.,
2009). Despite their simplicity and improved watershed
outlet predictions, they fail to predict the spatial loca-
tion of the runoff-generating areas. Collick et al. (2009),
and to some degree Steenhuis et al. (2009), present semi-
lumped conceptualizations of runoff producing areas in
water balance models. SWAT, a semi-distributed model,
can predict these runoff source areas in greater detail,
assuming that the runoff processes are correctly mod-
elled.
In this study, we develop and test a CN-free version
of SWAT based on a simple water balance approach.
We utilize the spatial adjustments as proposed by Lyon
et al. (2004) and implemented by Easton et al. (2008),
into SWAT–VSA, and replace the CN method with a
water balance to predict runoff. This new version of
SWAT, SWAT-WB, calculates runoff volumes based on
the available soil storage capacity of a given soil, and
then partitions excess moisture to runoff and infiltrating
fractions. This can lead to a more accurate simulation of
where and when the runoff occurs in watersheds dom-
inated by saturation-excess processes. Both the original
Hydrol. Process. 25, 915–925 (2011)
 SWAT LANDSCAPE WATER BALANCE
CN method used by SWAT and the new water balance
(SWAT-WB) method are tested on two watersheds that
vary widely in climate, geology, and data availability:
one in the monsoonal Blue Nile basin in Ethiopia and
one in the Catskill Mountains of New York.
MODEL OVERVIEW
SWAT is a basin-scale model designed to simulate hydro-
logical processes, nutrient cycling, and sediment trans-
port throughout a watershed. SWAT has been applied
to catchments ranging from 0Ð015 km2 (Chanasyk et al.,
2003) to as large as 491 700 km2 (Arnold et al., 2000).
To initialize the model, SWAT requires soils data,
land use/management information, and elevation data to
drive flows and direct sub-basin routing. The hydrologic
response unit (HRU) is the smallest unit in the SWAT
model and is used to simulate processes such as runoff,
infiltration, plant dynamics (including uptake of water
and nutrients, biomass, etc), erosion, nutrient cycling, and
leaching of pesticides and nutrients. Traditionally, HRUs
are defined by the coincidence of soil type (Hydrologic
Soil Group, USDA 1972) and land use. The predictions
from each HRU are aggregated for each sub-basin, and
routed through the internal channel network. Simulations
require meteorological input data including precipitation,
temperature, wind, humidity, and solar radiation. All
these inputs are initialized using the ARCSWAT (9Ð2)
interface (Olivera et al., 2006). More details on SWAT
can be found at http://www.brc.tamus.edu/swat/doc.html.
Original CN approach
Historically, when initializing SWAT, a CN is assigned
for each specific land use/soil combination in the water-
shed, and these values are read into the model. SWAT
then calculates the upper and lower limits for each CN
following a probability function described by the NRCS
to account for varying antecedent moisture conditions
(CN–AMC) (USDA–NRCS, 2004). SWAT determines a
CN for each simulated day by using this CN–AMC dis-
tribution in conjunction with daily soil moisture values
determined by the model. This daily CN is then used to
determine a theoretical storage capacity, S, of the water-
shed for each day. This storage, S, is then indirectly used
to calculate runoff volume, Q, via

P
Ia 2
QD
1

P
Ia  C S
where S is the watershed storage, P the precipitation, and
Ia the initial abstraction. All terms are in millimetre of
water, and by convention Ia is assumed to be equal to
0Ð2 ð S. The problem with SWAT, and other CN-based
models, is that runoff is calculated via Equation (1) prior
to any infiltration occurring in the watershed; thus the
model directly assumes an excess infiltration approach to
the runoff generation.
Copyright  2010 John Wiley & Sons, Ltd.
917
Water balance approach
A daily soil–water balance was used to determine the
saturation deficit (and by extension the runoff volume)
of each HRU by SWAT instead of the CN method.
Although SWAT’s soil moisture routine greatly simplifies
processes that govern water movement through porous
media (in particular, partly saturated regions), for a daily
basin scale model the approach is generally acceptable
(Guswa et al., 2002). In SWAT, the soil moisture levels
are estimated by considering several processes: plant
uptake of water, macropore and micropore drainage,
evaporation, redistribution between soil layers, lateral
drainage, and drainage of groundwater. A full description
of the equations used and their general applicability is
provided in the official SWAT documentation (Neitsch
et al., 2005) as well as in numerous articles in the
scientific literature. Thus, the model already provides
a convenient platform that can be expanded upon in
order to determine the surface runoff via a water balance.
SWAT’s existing soil moisture routines are then used by
SWAT-WB to determine the degree of saturation deficit
for each soil profile for each day of simulation. This
saturation deficit (in mm H2 O) is termed the available
soil storage, i and is a function of the soil properties
and watershed moisture state
i D
EDCi εi
it di
2
where EDCi is the effective depth of a given soil profile
i (unitless), εi the soil porosity (mm) of a given soil
i, it the volumetric soil moisture of a given soil i, for
each day, t (mm), and di the soil profile depth of soil
i (mm). The porosity, εi , is a constant value for each
soil type, whereas it varies in time and is determined
by SWAT soil moisture routines. The effective depth
coefficient, EDCi , a parameter ranging from 0 to 1,
is used to partition soil moisture in excess of εi into
infiltrating (groundwater) and runoff fractions (including
rapid shallow interflow). By including this adjustment
to the available storage, the amount of water able to
infiltrate each day is controlled by the EDCi . EDCi is
spatially varied based on a saturation probability defined
by a soil wetness index (Easton et al., 2008). EDCi
values approaching 1 are assigned to regions expected
to produce little saturation-excess runoff, whereas values
approaching 0 indicate an area likely to produce large
saturation-excess runoff volumes. This spatially adjusted
available storage is then used to determine what portion
of rainfall events will infiltrate and what portion will
runoff, qi (mm)


0 if P < i
qi D
3
P
i if P > i
The available storage, i , is calculated each day prior to
the start of any rain event. Once precipitation starts, a
portion of the rain, equal in volume to i , will infiltrate
the soil. If the rain event is larger in volume than i , the
soil profile will saturate and surface runoff will occur.
If the rainfall is less than i , the soil is unsaturated and
Hydrol. Process. 25, 915– 925 (2011)
918 E. D. WHITE ET AL.

there will be no surface runoff and SWATs’ internal soil

moisture routing will calculate the flux.
The model is available online at http://soilandwater.
bee.cornell.edu/research/swatwb/swatwb.html.

HRU definition
Traditionally, HRUs are defined in SWAT as being
unique occurrences of soil type, land cover, and slope
class. Any parcels of land within one sub-basin that share
the same combination of these three features will be con-
sidered one HRU. SWAT models all landscape processes
for each unique HRU in the watershed independently of
position within each sub-basin. In basins dominated by
VSA hydrology, this HRU definition has been shown to
be less than ideal for describing the spatial and tempo-
ral evolution of hydrological processes (Schneiderman
et al., 2007; Easton et al., 2008). In VSA watersheds, Figure 1. Wetness classes for Gumera, Ethiopia
runoff-generating areas are likely to occur in portions of
the landscape with shallow, low conductive soils, large
contributing areas, mild slopes, or any combination of
the three. Although SWATs’ inclusion of slope classes
in HRU delineation begins to address these issues, there
is currently no way to include upslope contributing area
while defining HRUs. To correct for this, a TI was inte-
grated with existing soils data to create a soil TI (STI),
which is then used in the SWAT-WB HRU definition
process (Easton et al., 2008).
TIs and their various derivatives have been used to
model runoff-contributing areas for quite some time (e.g.
TOPMODEL; Beven and Kirkby, 1979). Recently, soil
TIs have been incorporated into CN-based watershed
models for use in VSA-dominated regions (Lyon et al.,
2004; Schneiderman et al., 2007; Easton et al., 2008).
SWAT–VSA integrated a STI into SWAT in order to
Figure 2. Wetness classes for Town Brook, located in the Catskill
improve delineation of runoff-generating areas and the Mountains of New York State
subsequent nutrient loads in the Catskills Mountains of
New York (Easton et al., 2008). SWAT–VSA provided are the result of either a large contributing area or small
more accurate predictions of runoff source areas (as values for slope, soil depth, or saturated conductivity, and
validated by distributed measures in the watershed) than therefore are indicative of areas with a higher probability
the original SWAT; thus, we included an HRU definition for saturation.
process similar to SWAT–VSA in SWAT-WB. Following the process outlined for SWAT–VSA (Eas-
To initialize SWAT-WB, the first step was to create ton et al., 2008), an areally weighted STI (e.g. wetness
a soil TI for the watershed being modelled. The STI is classes) is used to represent a location’s likelihood to sat-
defined as (Beven and Kirkby, 1979; Beven, 1986) urate. Although using the individual STI values directly
  is, in theory, possible, in practice this would introduce far
˛
 D ln
4 too much complexity and computational time to model
T0 tan ˇ
initialization, calibration, and scenario runs. The wetness
The upslope contributing area, ˛, and the slope, tan(ˇ), classes determined for the two watersheds used in this
are both obtained from a DEM, whereas the lateral study are shown in Figure 1 for the Ethiopian watershed
transmissivity of the soil profile, T0 , when water table and Figure 2 for the New York watershed. This wetness
intersects the soil surface (Beven, 1986) is a function class map is then substituted for the soils map in the HRU
of the soil layer depth, D, and soil layer saturated definition process. Although the wetness classes can be
hydraulic conductivity, Ks , (e.g. T0 D Ks0 ð D0 ), and are used in HRU delineation instead of a soil map, SWAT still
obtained from soil information. We utilized the D-infinity requires specific soil properties associated with the soils
algorithm in TauDEM (Tarboton, 1997) to determine the map (e.g. SSURGO database). Thus, in SWAT-WB, soil
˛ parameter. We assume that the STI values relate to properties required by SWAT were areally weighted and
a location’s likelihood of saturation and therefore the averaged for each wetness class. This practice will not
likelihood to contribute surface runoff. Higher STI values drastically affect model results for two reasons. First, in

Copyright  2010 John Wiley & Sons, Ltd. Hydrol. Process. 25, 915–925 (2011)
 SWAT LANDSCAPE WATER BALANCE
Figure 3. Elevation of the Gumera basin, located east of Lake Tana in
the Ethiopian Highlands
Ethiopia, soil survey information is rare or nonexistent,
and, to our knowledge, no defined database exists that
would contain the parameters needed by SWAT. Thus,
SWAT-WB utilized the UNFAO’s World’s Soil map (sup-
plemented by literature values) as the base map (FAO-
AGL, 2003), which classifies only five distinct soil types
in all of the 1270 km2 Blue Nile sub-catchment that was
modelled. We assume that these soils data, with such a
course spatial resolution, will not be adversely affected
by the averaging process used in SWAT-WB. Second,
in New York, where information on soils is more read-
ily available, soil formation (in glaciated areas) is at least
partially driven by topography (Page et al., 2005; Sharma
et al., 2006; Thompson et al., 2006; Easton et al., 2008).
Therefore, averaging across topographic features with the
wetness index should not pose any problems.
Watershed descriptions
Gumera watershed, Blue Nile basin, Ethiopia. SWAT-
WB was tested on the Gumera River watershed, a
heavily cultivated region in the Ethiopian Highlands.
Located approximately 30 km northeast of Bahir Dar
(11Ð83 ° N, 37Ð63 ° E), this 1270 km2 watershed drains
into Lake Tana, the headwaters of the Blue Nile River
(Figure 3). Land use in the Gumera watershed consists
of 96% mixed agriculture and 4% brush (or pasture).
Elevation determined from a 75 m DEM ranged from
1797 to 3708 m above sea level with slopes ranging
from 0 to 79%. Temperatures in the basin show large
elevation (9–25 ° C) and diurnal variation, but small
seasonal changes. The annual average temperature is
18 ° C (Conway, 2000). The climate of the basin is tropical
highland monsoonal with the majority of the rain falling
between June and October. Rainfall amounts decrease
from the southwest to the northeast with approximately
80% occurring between June and October. Predominant
soils were gathered from the FAO World Soil map and
were classified as haplic and chromic luvisols (58 and
22%, respectively). Other soils present in the basin were
eutric fluvisols (8%), eutric leptosols (8%), and eutric
Copyright  2010 John Wiley & Sons, Ltd.
919
vertisols (3%), with minimal areas classified as urban
(>1%) (FAO-AGL, 2003).
Precipitation and temperature data were gathered from
the National Meteorological Agency of Ethiopia for
the Debre Tabor, Bahir Dar, Addis Zeman, and Nefas
Mewcha stations, the closest gauges to the Gumera
basin. The average annual precipitation from 1994 to
2005 was 1470 mm (data courtesy of the Ethiopian
Ministry of Water Resources; MoWIR, 2002), with
average potential evapotranspiration losses of 1220 mm.
Daily precipitation and temperature data from 1992 to
2003 was used for model calibration and validation. Other
required climatic data included relative humidity, wind
speed, and solar radiation. These data were obtained
for the nearby city of Bahir Dar through the United
States National Climatic Data Center (NCDC, 2007). The
SWAT-WB model included 656 HRUs in 25 sub-basins,
whereas the SWAT-CN model included 117 HRUs in 24
sub-basins.
Town Brook watershed, Catskills, New York. SWAT-
WB was also tested on the Town Brook watershed
(Figure 2) in the USA, a 37 km2 sub-catchment of
the Cannonsville reservoir basin. The region is typified
by steep-to-moderate hillslopes of glacial origins with
shallow permeable soils, underlain by a restrictive layer.
The climate is humid with an average annual temperature
of 8 ° C and average annual precipitation of 1123 mm.
Elevation in the watershed ranges from 493 to 989 m
above mean sea level. The slopes are quite steep with a
maximum of 91%, and a mean of 21%. Soils are mainly
silt loam or silty clay loam with soil hydrological group
C ratings (USDA–NRCS, 2000). Soil depth ranges from
less than 50 cm to greater than 1 m and is underlain
by a fragipan restricting layer (e.g. coarse-loamy, mixed,
active, mesic, to frigid Typic Fragiudepts, Lytic or Typic
Dystrudepts common to glacial tills) (Schneiderman
et al., 2002). The lowland portion of the watershed
is predominantly agricultural, consisting of pasture and
row crops (20%) or shrub land (18%), whereas the
upper slopes are forested (60%). Water and wetlands
comprise 2%. Impervious surfaces occupy <1% of the
watershed and were thus excluded from consideration in
the model. Several studies in this watershed or nearby
watersheds have shown that VSAs control overland
flow generation (Frankenberger et al., 1999; Mehta et al.,
2004; Lyon et al., 2006; Schneiderman et al., 2007;
Easton et al., 2008) and that infiltration-excess runoff
is rare (Walter et al., 2003). The SWAT-WB model
included 180 HRUs in three sub-basins, whereas the
SWAT-CN model included 172 HRUs in the three
sub-basins.
Model calibration
To calibrate both the SWAT-WB and SWAT-CN mod-
els for Gumera and Town Brook, we utilized the dynam-
ically dimensioned search (DDS) algorithm (Tolson and
Shoemaker, 2005). The DDS calibration routine allows
Hydrol. Process. 25, 915– 925 (2011)
920
for parameters to be calibrated at the watershed, sub-
basin, HRU, or wetness class level, which in turn allowed
for EDC to be calibrated separately for each wetness
class. Because there is no information at the wetness class
level on runoff generation, we forced the calibration to
maintain a pre-imposed distribution of EDC values that
roughly correlated with the STI (e.g. higher STI wet-
ness classes had lower EDC values, whereas lower STI
wetness classes had higher EDC values). For SWAT-CN,
the CN was calibrated for each unique soil/land use type
instead of the EDC. Eleven other parameters were cal-
ibrated in both models. These 11 parameters relate to
surface and groundwater interaction, soil water, evapo-
ration and plant uptake, and soil water holding capacity.
Models were calibrated to minimize the root mean square
error (RMSE).
Streamflow at the Gumera watershed outlet was cali-
brated over a period of 8 years, from 1996 to 2003, and
streamflow in Town Brook was calibrated from 1998 to
2002.
Model validation
Streamflow data from 1992 to 1995 was used to val-
idate the Gumera model. For Town Brook, streamflow
data from 2002 to 2004 was used to validate the model.
To test SWAT-CN’s and SWAT-WB’s abilities to cor-
rectly predict distributed hydrology, we used measure-
ments by Lyon et al. (2006) of height of water table above
the restricting layer for a section of the Town Brook
watershed. Briefly, 44 pieziometric data loggers, installed
to depths of 50 cm, recorded the water table depth in
15 min intervals from April 2004 to September 2004.
The field site encompassed five wetness index classes
(Figure 3) and three land use types: pasture (PAST),
shrub (RNGB), and mixed forest (FRST). To compare
the measured and SWAT–VSA water table heights, the
piezometer data were averaged across index classes; there
were between 2 and 32 piezometers per index (i.e. two
(a) 30
25
Discharge (mm)
20
15
10
5
0
(b) 30
25
Discharge (mm)
20
15
10
5 r = 0.64
0 0
Jan-95 Jan-96 Jan-97 Jan-98 Jan-99 Jan-00 Jan-01 Jan-02 Jan-03
Figure 4. Observed and modelled streamflow for Gumera using (A) SWAT-WB and (B) SWAT-CN
Copyright  2010 John Wiley & Sons, Ltd.
E. D. WHITE ET AL.
piezometers on index class six, 32 on index class 10, etc).
To compare measured water table heights with SWAT-
CN water table heights, we averaged across land use;
there were 4–32 measurements per land use. SWAT (and
SWAT-WB) reports soil water in millimetre of water
integrated over the soil profile (i.e. cumulative water
depth for all soil layers). Thus, we converted the model-
predicted soil water to an equivalent depth by dividing
by the SSURGO reported porosity and assuming the
SSURGO reported soil depth represented the depth to the
restricting layer. According to the SSURGO data base,
the depth of the local restricting layer is 1Ð2–1Ð4 m.
Model evaluation
Criteria used to assess the ability of the models to
predict discharge in Gumera and Town Brook included
a visual comparison between the modelled and the
observed hydrographs, Nash-Sutcliffe model efficiencies
(NSE) (Nash and Sutcliffe, 1970) and the RMSE.
RESULTS
Both the SWAT-WB and SWAT-CN models were ini-
tialized with the same input data, and calibrated and
validated using the same automatic procedure for both
the Gumera and Town Brook watersheds. Thus, a com-
parison of the model outputs is warranted.
Gumera basin
The SWAT-WB-predicted streamflow was significantly
different (based on paired t-test) from the SWAT-CN-
predicted streamflow, with SWAT-WB returning sig-
nificantly more accurate results than the SWAT-CN
model of the Gumera watershed (Figure 4). SWAT-
WB predicted streamflow with a daily NSE D 0Ð77 and
RMSE D 2Ð42 for the calibration period and NSE D 0Ð81
and RMSE D 2Ð47 for the validation period (Table I
30
Observed y = 0.86x + 0.084
R2 = 0.74
Water Balance
Observed
20
10
0
0 10 20 30
SWAT-WB
Observed 30
Curve Number
Observed
20
10
y = 0.87x -0.065
2
0
10 20 30
SWAT-CN
Hydrol. Process. 25, 915–925 (2011)
 SWAT LANDSCAPE WATER BALANCE
Table I. Model statistics for daily streamflow in Gumera basin
SWAT-WB
Calibration
NSE 0Ð77
RMSE 2Ð42
Percent difference
6Ð7
a Same input as SWAT-WB.
Figure 5. Spatial distribution of surface runoff in Gumera modelled by (A) SWAT-CN and (B) SWAT-WB
and Figure 4). SWAT-CN was less accurate predicting
streamflow, with a daily NSE D 0Ð64 and a RMSE D
2Ð67 for the calibration period and NSE D 0Ð63 and
a RMSE D 2Ð72, for validation (Table I and Figure 4).
The percent difference between measured and mod-
elled flows was substantially smaller for SWAT-WB
(
6Ð7%) than for SWAT-CN (
16Ð5%). These SWAT-
CN results are similar to those of Setegn et al. (2008)
who modelled streamflow of the Gumera watershed with
SWAT-CN. Their model predicted streamflow with an
NSE of 0Ð61. Validation results for their model returned
an NSE of 0Ð61 similar to our results for SWAT-
CN.
Perhaps more interesting is how SWAT-WB and
SWAT-CN differ in the predicted distribution of runoff in
the watershed. For one storm in October 1997 (104 mm
of rain), SWAT-CN predicted that all HRUs within the
watershed would contribute runoff, with a minimum
depth of 17 mm of runoff and a maximum of 71 mm
(Figure 5A). For the same storm, SWAT-WB predicted
that some HRUs would produce no runoff, whereas oth-
ers produced as much as 97 mm of runoff (Figure 5B).
Both models predicted some surface runoff for some
upland areas, but SWAT-CN predicted much less runoff
being generated in the low-lying, flatter areas near the
watershed outlet, where SWAT-WB predicted the most
runoff. At the watershed outlet, 16Ð5 mm of flow were
measured. SWAT-WB predicted the outlet flows to be
15Ð9 mm, whereas SWAT-CN substantially under pre-
dicted the flows (9Ð42 mm).
Copyright  2010 John Wiley & Sons, Ltd.
921
SWAT-CNa
Validation Calibration Validation
0Ð81 0Ð64 0Ð63
2Ð47 2Ð67 2Ð72

16Ð5
Town Brook watershed
The SWAT-WB results for Town Brook were compared
with the results from SWAT-CN. SWAT-WB-predicted
streamflow agreed well with measured flow, the cali-
bration NSE was 0Ð77 and the RMSE was 2Ð09, the
validation NSE was 0Ð79 and RMSE was 2Ð07. Pre-
dicted streamflow for the SWAT-CN Town Brook model
resulted in a daily NSE of 0Ð75 and a RMSE of 2Ð12
for calibration, and a NSE of 0Ð78 and a RMSE of 2Ð14
for validation. The SWAT-WB and SWAT-CN-predicted
streamflows were not statistically different (based on a
paired t-test). The percent difference between measured
and modelled flows was similar for both SWAT-WB
(5Ð6%) and SWAT-CN (6Ð2%). Visual comparison of
SWAT-WB’s hydrograph with the measured hydrograph
(Figures 4 and 6) indicates that the model performs well
for the Town Brook watershed, a fact supported by the
reasonably high daily NSE values of 0Ð77 and 0Ð79 for the
calibration and validation periods, respectively (Table II).
Comparing the SWAT-CN and SWAT-WB-predicted
water table heights with those measured by Lyon
et al. (2006) shows that the SWAT-WB-predicted soil
water table height agreed with measurements across the
monitored hill side in the watershed with r 2 D 0Ð68
(Figure 7A). There was a slight tendency for SWAT-
WB to under predict water table height for large water
table heights (Figure 7A), and slightly under predict
small water table heights. SWAT-CN, however, systemat-
ically under-predicted water table height for all conditions
(Figure 7B).
Hydrol. Process. 25, 915– 925 (2011)
922 E. D. WHITE ET AL.

(a) 60 60
y = 0.79x + 0.511
Observed r2 = 0.72
Discharge (mm) 50
Water Balance

Observed
40
40

30
20
20

10 0
0 20 40 60
0
SWAT-WB
(b) 60
60
Observed
50 y = 0.72x + 0.543
Discharge (mm)

Curve Number r2 = 0.66

40

Observed
40
30

20 20

10
0
0 0 20 40 60
Oct-98 Apr-99 Oct-99 Apr-00 Oct-00 Apr-01 Oct-01 Apr-02 Oct-02 Apr-03 Oct-03 Apr-04
SWAT-CN

Figure 6. Observed and modelled streamflow in the Town Brook watershed using (A) SWAT-WB and (B) SWAT-CN

Table II. Model statistics for daily streamflow in Town Brook

SWAT-WB SWAT-CNa

Calibration Validation Calibration Validation

NSE 0Ð77 0Ð79 0Ð75 0Ð78

RMSE 2Ð09 2Ð07 2Ð12 2Ð14
Percent difference 5Ð6 6Ð2
a Same input as SWAT-WB.

(a) (b)
1500 1500
SWAT-WB Water Table (mm)

SWAT-CN Water Table (mm)

1300 1300
10 Mixed Forest
9 Pasture
1100 8 1100 Shrub
7
900 6 900 R2 = 0.57
5
R2 = 0.68
700 700

500 500
500 700 900 1100 1300 1500 500 700 900 1100 1300 1500
Measured Water Table (mm) Measured Water Table (mm)
Figure 7. Relationship between (A) SWAT-WB and (B) SWAT-CN predicted water table heights above the restricting layer by index class (SWAT-WB)
or land use (SWAT-CN) and the measured water table heights for March–September 2004 from Lyon et al. (2006). Individual measured points within
an index class or land use represent the average of the pieziometic measurement within the respective classes for a single day

Similar to the Gumera results, differences in the of the wetness classes in the low lying areas of the
spatial distribution of runoff are evident when an event watershed would be saturated at the start of this event,
from November 2003 is compared between the Town leading to these low-lying wet areas producing nearly
Brook models (Figure 8A and B). As expected, SWAT- identical volumes of runoff (i.e. almost the entire volume
CN predicts some surface runoff from the majority of of precipitation).
the watershed and it is clearly driven by differences
in land use, whereas SWAT-WB predicts substantial
portions of the watershed producing no surface runoff, DISCUSSION
not surprising considering the emphasis the model places
on topographic position as it pertains to runoff generation. Clear improvements were made to SWAT in the Ethiopian
For this particular storm, SWAT-WB predicted that most watershed by removal of the CN; however, the results are

Copyright  2010 John Wiley & Sons, Ltd. Hydrol. Process. 25, 915–925 (2011)
 SWAT LANDSCAPE WATER BALANCE
not as definitive for the Town Brook watershed in New
York State. Although SWAT-WB has substantially higher
model accuracy for the calibration period, it has similar
accuracy during validation as SWAT-CN. By comparing
the hydrograph from the Town Brook outlet (Figure 6)
and the model statistics, it is clear that SWAT-WB gen-
erally performs as well as SWAT-CN. Thus, in cases
where rainfall is evenly distributed throughout the year
both models perform equally well in predicting discharge.
However, we would argue that the water balance method
captures the processes controlling runoff generation real-
istically. In Town Brook, Easton et al. (2008) showed that
the inclusion of the STI-based wetness index better cap-
tured the spatial distribution of water table depths and, by
extension, runoff producing areas. SWAT-WB provided a
similar level of accuracy in predicting water table heights
(Figure 7). While we have no distributed runoff data for
either of the watersheds, including the wetness index in
SWAT-WB resulted in what appears to be more a realis-
tic distribution of runoff generating areas than SWAT-CN.
Indeed, in VSA-dominated watersheds, runoff generation
is closely related to soil moisture levels (as controlled by
perched water table levels), which in turn is governed, to
a large extent, by topographic position. In many SWAT-
CN applications, the location of an HRU within each
sub-basin is not a concern and thus, any locations that
share land use and soil are treated identically, regardless
of its topographic position and the corresponding likeli-
hood to produce runoff. In SWAT-WB, STIs were used
to link HRUs by similar topographic position, provid-
ing model users the capability to examine intra-watershed
runoff dynamics.
This difference in spatial distribution of runoff-
generating areas predicted by SWAT-WB and SWAT-CN
is clearly demonstrated for both Gumera (Figure 5) and
Town Brook (Figure 8). For the same large storm event in
the Gumera basin (Figure 5), SWAT-WB did not generate
surface runoff for all HRUs, whereas SWAT-CN pre-
dicted that the entire watershed would contribute surface
runoff more or less evenly. Due to the imposed topo-
graphical controls, SWAT-WB predicted that the wettest
portions of the watershed would contribute more runoff
Figure 8. Spatial distribution of surface runoff in Town Brook modelled by (A) SWAT-WB and (B) SWAT-CN
Copyright  2010 John Wiley & Sons, Ltd.
923
than drier areas. In addition to the fact that SWAT-CN
predicts a nearly uniform runoff volume for the entire
watershed, there are two other points of interest that
should be discussed. First is the fact that SWAT-CN
predicts that the area nearest Gumera’s outlet produces
the least amount of surface runoff, exactly opposite of
SWAT-WB’s results, which predict that this area pro-
duces high runoff volumes. These differences between
the models can be explained by the inclusion of slope
in the HRU delineation (and therefore EDCi calibration).
Again, holding with VSA principles, SWAT-WB assumes
that these flat, near-stream regions will wet up and con-
tribute the most runoff, because of reduced lateral flow
(due either to shallower slope reducing the hydraulic gra-
dient, or perhaps more importantly the accumulation of
interflow from upslope areas), whereas SWAT-CN treats
these HRUs the same as any upland region with the same
soil and land cover. The second interesting point is that
both models predict that certain upland regions generate
a significant portion of surface runoff from the test storm,
but for different reasons. These soils have a low saturated
conductivity; for SWAT-CN, this results in a high CN and
for SWAT-WB an increase in STI values, both of which
increase runoff.
SWAT-CN and most other watershed models have
been developed for temperate climates where rainfall is
generally well distributed throughout the year. Utilizing
models developed during a temperate climate for Ethiopia
conditions, with a monsoonal climate, is problematic.
Temperate models assume that there is a nearly unique
relationship between precipitation amounts or intensity
and runoff generated. This is not the case for Ethiopia as
demonstrated by the results of Liu et al. (2008) where for
three watersheds with more than 16 years of record, the
rainfall relationship was far from unique. The first rains
after the dry season all infiltrate and nearly no runoff
is generated. As the rainfall season progresses more and
more rainfall becomes runoff. Because the intensity of the
rain did not affect the runoff amounts for a given storm,
Liu et al. (2008) concluded that the runoff mechanism
was dominated by excess saturation processes.
Hydrol. Process. 25, 915– 925 (2011)
924 E. D. WHITE ET AL.
Water balance models are consistent with saturation
excess runoff processes because the runoff is related to
the available watershed storage capacity and the amount
of precipitation. The implementation of water balances
into runoff calculations in the Blue Nile basin is not a
novel concept and others have shown that water balance
type models often perform better than more complicated
models in Ethiopian type landscapes (Johnson and Curtis,
1994; Conway, 1997; Ayenew and Gebreegziabher, 2006;
Liu et al., 2008). However, these water balance models
are typically run on monthly or yearly time steps because
the models are generally not capable of separating base-,
inter-, and surface-runoff flows. To truly model erosion
and sediment transport (of great interest in the Blue Nile
basin), large events must be captured by the model and
daily simulations are required to do so. Thus, SWAT-WB
not only maintains a water balance but also calculates
the interflow and the base flow component, and gives
a reasonable prediction of peak flows. SWAT-WB is
therefore more likely to be capable of predicting erosion
source areas and sediment transport than either SWAT-
CN or water budget models with monthly time steps.
Indeed, Tebebu et al. (2010) found gully formation and
erosion in the Ethiopian Highlands to be related to water
table levels and saturation dynamics, which SWAT-WB
reliably predicts (Figure 7 for Town Brook).
During initial development of SWAT-WB, we encoun-
tered the same issue of fractionating excess moisture into
runoff and base flow components. When the entire soil
profile depths were used to calculate the available stor-
age, i , almost all precipitation became base flow. If only
the uppermost soil layer was used nearly all precipitation
became surface runoff. To address these issues, and to
allow for SWAT-WB to be applied at a daily time step, it
became evident that a new parameter, in addition to soil
depth and moisture content, would need to be added to the
equation used to calculate i . Although this new parame-
ter, EDC, is calibrated, it should be noted that, in theory,
this parameter is physically measureable in the field or
perhaps from base flow-separated runoff (Easton et al.,
2010), whereas the CN remains largely a theoretical cal-
ibration parameter in saturation-excess runoff-dominated
areas.
Interestingly, the EDCi solution to these issues is
similar to a recent water balance model developed for
the Blue Nile by Kim and Kaluarachchi (2008) that
combines a water balance with a traditional tank model.
To differentiate between surface and various subsurface
flows, Kim and Kaluarachchi (2008) developed a model
using two ‘tanks’. The upper tank, described by an upper
zone soil moisture term, was used to calculate surface
runoff, and a lower zone term was used to model base
flow. The upper layer produced no surface runoff until
a ‘runoff orifice’ depth was filled by rainfall. This upper
zone soil layer with its runoff orifice depth is analogous to
SWAT-WB’s EDCi term; both parameters acknowledge
that in saturation-excess-dominated areas only a certain
portion of the soil profile plays a role in runoff generation.
Copyright  2010 John Wiley & Sons, Ltd.
CONCLUSION
Daily modelling of stream flow and surface runoff in a
monsoonal watershed was improved by the replacement
of the CN method with a simplified water balance
routine in the SWAT model. SWAT-WB uses calculated
saturation-deficit values with an EDC to partition rainfall
into surface runoff and infiltrating water. This EDC-based
water balance method is analogous to other tank models
that have been successfully applied in monsoonal regions
SWAT-WB was as accurate in predicting discharge
at the outlet as the CN method in a watershed that
experiences evenly distributed rainfall throughout the
year (e.g. for conditions the CN was developed and tested
under). However, SWAT-WB predicted the distribution
of water table heights on a hillslope in the watershed
significantly better than SWAT-CN, giving us increased
confidence that the spatial distribution of runoff dynamics
are more realistically captured.
These results indicate that SWAT performs better
in saturation-excess-controlled areas when a simple
saturation-deficit water balance model is used to calculate
runoff volumes. With this physically based and easy-to-
use model, effective water and land management schemes
will be easier to successfully implement in watersheds
dominated by saturation-excess runoff generation, partic-
ularly data-poor regions where use of the CN method-
ology has not been validated. For instance, the model
has direct application to identification and delineation for
hydrologically sensitive areas in watersheds dominated
by VSA hydrology.
ACKNOWLEDGEMENTS
We thank the Amhara Regional Agricultural Research
Institute (ARARI), the Ethiopain Ministry of Water
Resources (MoWR), the International Water manage-
ment Institute (IWMI), and the Eastern Nile Technical
Regional Office (ENTRO) for providing data.
REFERENCES
Arnold JG, Muttiah RS, Srinivasan R, Allen PM. 2000. Regional
estimation of base flow and groundwater recharge in the upper
Mississippi basin. Journal of Hydrology 227(1– 4): 21–40.
Arnold JG, Srinivasan R, Muttiah RS, Williams JR. 1998. Large area
hydrologic modeling and assessment part I: model development.
Journal of the American Water Resources Association 34(1): 73–89.
Ashagre BB. 2009. Formulation of best management option for a
watershed using SWAT (Anjeni watershed, Blue Nile Basin, Ethiopia).
MS thesis, Cornell University, Ithaca, NY.
Ayenew T, Gebreegziabher Y. 2006. Application of a spreadsheet
hydrological model for computing long-term water balance of Lake
Awassa, Ethiopia. Hydrological Sciences Journal 51(3): 418– 431.
Beven KJ. 1986. Runoff production and flood frequency in catchments
of order n: an alternative approach. In Scale Problems in Hy-
drology, Gupta VK, Rodriguez-Iturbe I (eds). Reidel, Dordrecht: The
Netherlands; 107–131.
Beven KJ, Kirkby MJ. 1979. Towards a simple physically based variable
contributing model of catchment hydrology. Hydrological Sciences
Bulletin 24(1): 43–69.
Bryant RD, Gburek WJ, Veith TL, Hively WD. 2006. Perspectives on
the potential for hydropedology to improve watershed modeling of
phosphorus loss. Geoderma 131: 299– 307.
Hydrol. Process. 25, 915–925 (2011)
 SWAT LANDSCAPE WATER BALANCE
Chanasyk DS, Mapfum E, Williams W. 2003. Quantification and
simulation of surface runoff from fescue grassland watersheds.
Agricultural Water Management 59(2): 137– 153.
Collick AS, Easton ZM, Adgo E, Awulachew SB, Gete Z, Steenhuis TS.
2009. Application of a physically based water balance model on four
watersheds throughout the upper Nile basin in Ethiopia. Hydrological
Processes 23: 3718– 372. DOI: 10.1002/hyp.7517.
Conway D. 1997. A water balance model of the upper Blue Nile in
Ethiopia. Hydrological Sciences Journal 42(2): 265–286.
Conway D. 2000. The climate and hydrology of the upper Blue Nile
River. Geographical Journal 166: 49–62.
Easton ZM, Fuka DR, Walter MT, Cowan DM, Schneiderman EM,
Steenhuis TS. 2008. Re-conceptualizing the Soil and Water Assessment
Tool (SWAT) model to predict runoff from variable source areas.
Journal of Hydrology 348(3– 4): 279– 291.
Easton ZM, Walter MT, Fuka DR, White ED, Steenhuis TS. 2010. A
simple concept for calibrating runoff thresholds in quasi-distributed
variable source area watershed models. Hydrological Processes. In
review.
FAO-AGL. 2003. World reference base: map of world soil resources.
United Nations, Food and Agriculture Organization.
Frankenberger JR, Brooks ES, Walter MT, Walter MF, Steenhuis TS.
1999. A GIS-based variable source area model. Hydrological Processes
13(6): 804– 822.
Gassman PW, Reyes MR, Green CH, Arnold JG. 2007. The soil and
water assessment tool: historical development, applications, and
future research directions. Transactions of the American Society of
Agricultural and Biological Engineers 50(4): 1211– 1250.
Green WH, Ampt G. 1911. Studies of soil physics: part I—the flow of
air and water through soils. Journal of Agricultural Science 4: 1–24.
Guswa AJ, Celia MA, Rodriguez-Iturbe I. 2002. Models of soil dynamics
in ecohydrology: a comparative study. Water Resources Research
38(9): 1166– 1181.
Hawkins RH. 1979. Runoff curve numbers from partial area watersheds.
Journal of Irrigation and Drainage Engineering-ASCE 105(4):
375– 389.
Johnson PA, Curtis PD. 1994. Water balance of Blue Nile river basin
in Ethiopia. Journal of Irrigation and Drainage Engineering-ASCE
120(3): 573–590.
Kim U, Kaluarachchi JJ. 2008. Application of parameter estimation and
regionalization methodologies to ungauged basins of the upper Blue
Nile river basin, Ethiopia. Journal of Hydrology 362: 39–56.
Kim NW, Lee J. 2008. Temporally weighted average curve number
method for daily runoff simulation. Hydrological Processes 22:
4936– 4948.
King KW, Arnold JG, Binger RL. 1999. Comparison of Green–Ampt
and curve number methods on Goodwin creek watershed using SWAT.
Transactions of the American Society of Agricultural and Biological
Engineers 42(4): 919– 925.
Liu BM, Collick AS, Zeleke G, Adgo E, Easton ZM, Steenhuis TS.
2008. Rainfall-discharge relationships for a monsoonal climate in the
Ethiopian highlands. Hydrological Processes 22(7): 1059– 1067.
Lyon SW, Seibert J, Lembo AJ, Walter MT, Steenhuis TS. 2006.
Geostatistical investigation into the temporal evolution of spatial
structure in a shallow water table. Hydrology and Earth System Sciences
10(1): 113– 125.
Lyon SW, Walter MT, Gerard-Marchant P, Steenhuis TS. 2004. Using a
topographic index to distribute variable source area runoff predicted
with the SCS curve-number equation. Hydrological Processes 18:
2757– 2771.
Mehta VK, Walter MT, Brooks ES, Steenhuis TS, Walter MF, John-
son M, Boll J, Thongs D. 2004. Evaluation and application of SMR
for watershed modeling in the Catskill Mountains of New York State.
Environmental Modeling & Assessment 9(2): 77– 89.
MoWR (Ministry of Water Resources). 2002. Ethiopian Water Sector
Strategy. MoWR: Addis Ababa.
Nash JE, Sutcliffe JV. 1970. River flow forecasting through conceptual
models: Part I a discussion of principles. Journal of Hydrology 10:
282– 290.
Neitsch SL, Arnold JG, Kiniry JR, Williams JR. 2005. Soil and
Water Assessment Tool: Theoretical Documentation—Version 2005 .
Copyright  2010 John Wiley & Sons, Ltd.
925
Grassland, Soil and Water Research Laboratory, Agricultural Research
Service: Temple, TX.
NCDC. 2007. Global summary of the day. National Climatic Data
Center’s Climate Data Online. 24 November 2007.
Olivera F, Valenzuela M, Srinivasan R, Choi J, Cho H, Koka S,
Agrawal A. 2006. ArcGIS-SWAT: a geodata model and GIS interface
for SWAT. Journal of the American Water Resources Association 42:
295– 309.
Page T, Haygarth PM, Beven KT, Joynes A, Butler T, Keeler C, Freer J,
Owens PN, Wood GA. 2005. Spatial variability of soil phosphorus in
relation to the topographic index and critical source areas: sampling
for assessing risk to water quality. Journal of Environmental Quality
34: 2263– 2277.
Setegn SG, Srinivasan R, Dargahi B. 2008. Hydrological modeling in
the Lake Tana basin, Ethiopia using SWAT model. Open Hydrology
Journal 2: 49–62.
Schneiderman EM, Pierson DC, Lounsbury DG, Zion MS. 2002. Model-
ing the hydrochemistry of the Cannonsville watershed with generalized
watershed loading functions (GWLF). Journal of the American Water
Resources Association 38(5): 1323– 1347.
Schneiderman EM, Steenhuis TS, Thongs DJ, Easton ZM, Zion MS,
Neal AL, Mendoza GF, Walter MT. 2007. Incorporating variable
source area hydrology into a curve-number-based watershed model.
Hydrological Processes 21: 3420– 3430.
Sharma SK, Mohanty BP, Zhu JT. 2006. Including topography and
vegetation attributes for developing pedotransfer functions. Soil
Science Society of America Journal 70: 1430– 1440.
Sheridan JM, Shirmohammadi A. 1986. Application of curve number
procedure on coastal plain watersheds. Proceedings from the 1986
Winter Meeting of the American Society of Agricultural Engineers.
Paper No. 862505 . ASAE: Chicago, IL.
Steenhuis TS, Collick AS, Easton ZM, Leggesse ES, Bayabil HK,
White ED, Awulachew SB, Adgo E, Ahmed AA. 2009. Predicting
discharge and erosion for the Abay (Blue Nile) with a simple model.
Hydrological Processes 23: 3728– 3737.
Steenhuis TS, Winchell M, Rossing J, Zollweg JA, Walter MF. 1995.
SCS runoff equation revisited for variable-source runoff areas. Journal
of Irrigation and Drainage Engineering-ASCE 121(3): 234– 238.
Tarboton DG. 1997. A new method for the determination of flow
directions and contributing areas in grid digital elevation models. Water
Resources Research 33(2): 309–319.
Tebebu TY, Abiy AZ, Dahlke HE, Easton ZM, Zegeye AD, Tilahun
SA, Collick AS, Kidnau S, Moges S, Dadgari F, Steenhuis TS. 2010.
Surface and subsurface flow effects on permanent gully formation and
upland erosion near Lake Tana in the northern Highlands of Ethiopia.
Hydrology and Earth System Sciences Discussions 7: 3837– 3878.
Thompson JA, Pena-Yewtukhiw EM, Grove JH. 2006. Soil landscape
modeling across a physiographic region: topographic patterns and
model transportability. Geoderma 133: 57–70.
Tolson BA, Shoemaker CA. 2005. Dynamically dimensioned search
algorithm for computationally efficient watershed model calibration.
Water Resource Research 43: W01413– W01429.
USDA-SCS. (Soil Conservation Service). 1972. Part 630 hydrology in
the national engineering handbook. Section 4, Chapter 10.
USDA-NRCS. 2000. Soil survey geographic (SSURGO) database for
Delaware county. New York.
USDA-NRCS. 2004. Estimation of direct runoff from storm rainfall.
National Engineering Handbook. Part 630: Hydrology.
Walter MT, Mehta K, Marrone AM, Boll J, Gérard-Merchant P, Steen-
huis TS, Walter MF. 2003. A Simple estimation of the prevalence of
Hortonian flow in New York city’s watersheds. Journal of Hydrologic
Engineering 8(4): 214–218.
Wang X, Shang S, Yang W, Melesse AM. 2008. Simulation of an
agricultural watershed using an improved curve number method in
SWAT. Transactions of the American Society of Agricultural and
Biological Engineers 51(4): 1323– 1339.
White ED, Feyereisen GW, Veith TL, Bosch DD. 2009. Improving daily
water yield estimates in the Little River watershed: SWAT adjustments.
Transactions of the American Society of Agricultural and Biological
Engineers 52(1): 69–79.
Hydrol. Process. 25, 915– 925 (2011)