Journal of Environmental Science and Health Part A

ISSN: 1093-4529 (Print) 1532-4117 (Online) Journal homepage: http://www.tandfonline.com/loi/lesa20

Comparison of HSPF and SWAT models

performance for runoff and sediment yield
prediction

Sangjun Im , Kevin M. Brannan , Saied Mostaghimi & Sang Min Kim

To cite this article: Sangjun Im , Kevin M. Brannan , Saied Mostaghimi & Sang Min Kim
(2007) Comparison of HSPF and SWAT models performance for runoff and sediment yield
prediction, Journal of Environmental Science and Health Part A, 42:11, 1561-1570, DOI:
10.1080/10934520701513456

To link to this article: https://doi.org/10.1080/10934520701513456

Published online: 19 Oct 2007.

Submit your article to this journal

Article views: 274

Citing articles: 14 View citing articles

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=lesa20
Journal of Environmental Science and Health Part A (2007) 42, 1561–1570
Copyright C Taylor & Francis Group, LLC

ISSN: 1093-4529 (Print); 1532-4117 (Online)

DOI: 10.1080/10934520701513456

Comparison of HSPF and SWAT models performance

for runoff and sediment yield prediction

SANGJUN IM1 , KEVIN M. BRANNAN2 , SAIED MOSTAGHIMI2 and SANG MIN KIM3
1
Department of Forest Sciences, Seoul National University, Seoul, Republic of Korea
2
Department of Biological Systems Engineering, Virginia Polytechnic Institute and State University, Blacksburg, Virginia, USA
3
Department of Agricultural Engineering, Gyeongsang National University, Jinju, Gyeongsangnam-do, Republic of korea

A watershed model can be used to better understand the relationship between land use activities and hydrologic/water quality
processes that occur within a watershed. The physically based, distributed parameter model (SWAT) and a conceptual, lumped
parameter model (HSPF), were selected and their performance were compared in simulating runoff and sediment yields from the
Polecat Creek watershed in Virginia, which is 12,048 ha in size. A monitoring project was conducted in Polecat Creek watershed
during the period of October 1994 to June 2000. The observed data (stream flow and sediment yield) from the monitoring project
was used in the calibration/validations of the models. The period of September 1996 to June 2000 was used for the calibration and
October 1994 to December 1995 was used for the validation of the models. The outputs from the models were compared to the
observed data at several sub-watershed outlets and at the watershed outlet of the Polecat Creek watershed. The results indicated that
both models were generally able to simulate stream flow and sediment yields well during both the calibration/validation periods. For
annual and monthly loads, HSPF simulated hydrologic and sediment yield more accurately than SWAT at all monitoring sites within
the watershed. The results of this study indicate that both the SWAT and HSPF watershed models performed sufficiently well in the
simulation of stream flow and sediment yield with HSPF performing moderately better than SWAT for simulation time-steps greater
than a month.
Keywords: Watershed model, HSPF, SWAT, runoff, sediment yield.

Introduction The most direct and accurate method for acquiring this is
by monitoring streamflow and water quality constituents.
Since the early 1980s, nonpoint source pollution from However, monitoring can be is very expensive; furthermore,
agricultural has long been recognized as the major threat to monitored data is often misinterpreted, due to the lack of
water quality.[1] Runoff from agricultural lands transports understanding of hydrologic and sediment transport pro-
pollutants such as sediment and nutrients that may be in cesses that occur within the watershed.[2] For these reasons,
solution (suspended particles) or absorbed on to particles. a properly selected watershed model that is calibrated for
Both pollutants in solution and attached to particles the conditions being investigated can be used to predicting
increase the potential of eutrophication of water bodies. runoff, soil loss, and transport of pollutants from a water-
Estimation of runoff and soil erosion is major challenge shed, along with the tasks involved in the evaluation of the
for watershed managers and other stakeholder groups effects that different land use practices could have on water
who attempt to protect the water quality of streams, rivers quality.
and other water resources. When attempting to ameliorate Although there are many watershed models avail-
problems caused by nonpoint source pollution, watershed able, model selection depends on the objectives of the
managers need extensive amounts of data in order to make application, available resources, and the data availabil-
sound decisions about management changes that could ity. Some examples of models for simulating nonpoint
reduce levels of pollution leaving the land and entering the source pollution include AGNPS (Agricultural Nonpoint
water bodies being protected. Source)[3] GWLF (Generalized Watershed Loading Func-
tion Model),[4] HSPF (Hydrologic Simulation Program-
Address correspondence to Sang Min Kim, Department of Agri- Fortran),[5] and SWAT (Soil and Water Assessment Tool).[6]
culturel Engineering, Gyeongsang National University, Jinju,
The degree of complexity used in these models varies greatly
Gyeongsang nam-do, Republic of korea. E-mail: bbot2001@
freechal.com
in the representation of the physical processes occurring in
Received March 14, 2007. a watershed.
1562 Im et al.
The U.S. Environmental Protection Agency (US EPA) sumed to be unidirectional, and routed using the kinematic
commissioned the development of Better Assessment Sci- wave equation. The migration of each sediment fraction be-
ence Integrating Point and Nonpoint Sources (BASINS) tween suspension in water and the bed is modeled in a reach
for supporting the development of Total Daily Maximum by balancing deposition and scour of sediment, along with
Loads (TMDLs).[7] The current version of BASINS in- bed-material. Bicknell et al.[5] provides the detailed descrip-
cludes two models; HSPF and SWAT model. HSPF is a tion of the PERLND, IMPLND, and RCHRES modules.
lumped-parameter model that simulates runoff and pollu-
tant loadings from a watershed, integrates these with point
Overview of SWAT model
source contributions into the receiving waters, and then
models hydraulic and water quality processes in reaches.[5] SWAT is a physically based, continuous model developed to
SWAT is a “physically based,” watershed scaled model assess the impacts of management practices on hydrology,
used to predict the effects of land management on wa- sediment, and water quality for ungaged watersheds.[6] The
ter, sediment, and agricultural chemical yields in a com- model operates on a daily time step and allows a basin to
plex watershed.[6] In parallel with the EPA supported be subdivided into homogeneous hydrologic response units
HSPF, SWAT was developed by the Department of Agri- (HRU), having unique soil and land use characteristics.
culture’s (USDA) Agricultural Research Service (ARS) for Major components of the hydrologic cycle and their in-
assessing the nonpoint source pollution from agricultural teractions considered in SWAT are surface runoff, lateral
watersheds. flow in the soil profile, groundwater flow, evapotranspira-
In this study, we compared the accuracy of SWAT and tion, channel routing, and pond and reservoir storage.[6]
HSPF in simulating runoff and sediment yield for the Pole- SWAT uses the NRCS curve number method to compute
cat Creek watershed; located in Virginia. Both models were surface runoff from daily rainfall amount. Peak runoff rate
calibrated and validated using observed flow and water is estimated by a modified rational method and NRCS
quality data obtained at several sub-watershed outlets and TR-55 method. Three methods are available for simu-
at the watershed outlet of the Polecat Creek watershed. lating evapotranspiration: Penman-Monteith, Hargreaves,
and Priestley–Taylor methods. Infiltrated water is redis-
tributed in the soil profile by the SWAT percolation com-
Materials and methods
ponent, which uses a storage routing technique combined
with a crack-flow method to predict flow through each soil
Overview of HSPF model
layer.[6] Water that percolates below the soil profile is as-
HSPF is a lumped-parameter model that simulates hydrol- sumed to recharge the shallow aquifer. This water can either
ogy and water quality processes on a continuous basis in flow laterally towards a stream or percolate into the deeper
natural and man-made water systems.[5] It can represent aquifer and is considered lost from the simulated system.
spatial variability by dividing the watershed into land seg- In SWAT, soil erosion is calculated for each hydrologic
ments that are assumed to produce a homogeneous hydro- response unit (HRU) with the Modified Universal Soil Loss
logic and water quality response. equation (MUSLE) developed by Williams and Berndt.[8]
There are three modules in HSPF that represent the ma- Sediment in the HRU is routed through the reaches us-
jor watershed processes. Pervious land is simulated in the ing a stream power equation; where the maximum amount
PERLND module; impervious land in the IMPLND mod- of sediment transported from a reach is a function of the
ule; and the RCHRES module simulates the process in channel velocity at peak flow.
streams or completely mixed reservoirs. The hydrologic pro- Reach routing operates on a daily time step. Flow rate and
cesses in PERLND module are represented mathematically average velocity are calculated using Manning’s equation.
as fluxes and storages. Water movement in a PERLND seg- The transport of sediment in the channel is controlled by de-
ment is modeled along three pathways: overland flow, inter- position and degradation (or scour) processes. Deposition
flow, and groundwater flow. Erosion and sediment transport in the channel is based on sediment particle fall velocity cal-
in the land surface are simulated as wash-off of detached culated with Stokes Law. SWAT uses a modified Bagnold’s
sediment in storage and scour of the soil matrix. The sed- equation to predict degradation of the stream lining. Bed
iment outflow from the land surface can be classified into degradation is also adjusted for streambed erodibility and
sand, silt, and clay on the basis of the soil texture of each cover. A detailed description of the model and the equation
land surface.[5] IMPLND segments are used to simulate ar- used can be found in Neitsch et al.[6]
eas where no or negligible infiltration occurs.
The hydrologic process includes runoff and evaporation.
Description of study watershed
Sediment is considered as the accumulation of solids on the
impervious surface that can be washed-off during a runoff The Polecat Creek watershed was selected in this study for
event.[5] The RCHRES module routes runoff and water the comparison of HSPF and SWAT models abilities to
quality constituents through a single reach or a completely simulate hydrologic and sediment transport processes. The
mixed reservoir. Flow through a RCHRES segment is as- Polecat Creek watershed is 12,048 ha in size and is located
Watershed model comparison 1563

in Caroline County, Virginia. Polecat Creek flows into the the precipitation data, nine rain gages that were uniformly
Mattaponi River, which is a main tributary of the York distributed in and around the watershed[10] were used in the
River. The York River flows into the Chesapeake Bay.[9] HSPF and SWAT simulations.
The land use information required by both of the models Stream flow and water quality were monitored at several
was extracted from GIS data developed by the Chesapeake locations in the Polecat Creek watershed. Stream flow was
Bay Local Assistance Department (CBLAD).[10] The domi- measured at the each sub-watershed outlets (Fig. 1). Wa-
nate land uses in the Polecat Creek are forest (73%), pasture ter samples were automatically collected changes in time-
(13%), urban or developed land (10%), cropland (2%), and integrated volumes[10] at five monitoring stations (Fig. 1).
then water or reservoir (2%). Sediment concentrations were analyzed according to the
Soils data were obtained from the National Resources standard methods.[11] Sediment loads were calculated by
Conservation Service (NRCS) STATSGO database.[10] The multiplying the measured concentrations in samples by wa-
major soil series occurring in the watershed are Suffolk, ter discharge rate. These loads were summed at monthly
Rumford, Cecil and Appling soil series. The Suffolk soil intervals for the model comparisons.
series are common in the uplands of Coastal Plain with Based on the location of each monitoring station, sub-
slopes range from 0 to 50%. They are very deep and well- watershed boundaries were delineated in the Polecat Creek
drained soils and cover about 64% of the watershed. The watershed as shown Figure 1. The sub-watersheds drain-
Rumford series are also deep and well drained soils that ing to stations QPB, QPC, QPD, and QPE were used in
were formed in the sandy and loamy marine sediment on the model comparisons (Fig. 1). The upper sub-watershed
Coastal Plain. The Cecil and Appling series are very deep, (QPA) was not used for assessing model performance be-
well drained moderately permeable soils on ridges and side cause the watershed is relatively small in size and is entirely
slopes of the Piedmont upland. The Rumford, Cecil, and contained within sub-watershed QPD. Sub-watershed QPB
Appling series comprise more than 30% of the watershed has a drainage area of 2,658 ha and contains the most signif-
area. icant developed area in the watershed. Sub-watershed QPC
Climate data were obtained from a weather station lo- is located in the northwestern part of the Polecat Creek wa-
cated in the center of the watershed. The climate parameters tershed and is 888 ha in size. Sub-watershed QPD received
include evaporation, solar radiation, wind speed, air tem- inflow from sub-watersheds QPA and QPC. The outlet of
perature and relative humidity. Dewpoint temperature data the watershed, QPE has a drainage area of 12,048 ha. Land
were obtained from the National Weather Service station at use for QPE includes 74% forest, 13% pasture, 2% crop-
the Richmond International Airport (WBAN 13740). For land, and 10% urban or developed land. The drainage areas

Fig. 1. The Polecat Creek watershed.

1564 Im et al.
Table 1. Land use data for selected sub-watersheds in the Polecat where n represents the number of observations in the time
Creek watershed series, Si is the simulated values, Oi is the corresponding
observed values, and Ō and S̄ are the observation and sim-
Land use QPB QPB QPD QPE
ulation averages, respectively.
Area (ha) 2, 658 888 2, 605 12, 048 The RMSE expresses the degree to which the simulated
Forest (%) 56.5 71.8 77.9 74.4 value differs from the observed whereas the ME evaluates
Cropland (%) 11.6 11.9 13.0 12.8 the extent to which errors are around the zero value. The
Pasture (%) 0.3 3.8 1.4 1.5 correlation coefficient (r ) measures the tendency of the sim-
Developed area (%) 26.4 13.2 7.7 10.2 ulated and observed values to vary together linearly and
Lake (%) 5.2 0.0 0.0 1.1 ranges from −1 to 1, with negative value indicating that the
observed and simulated values tend to vary inversely. The
index of agreement (d) was proposed by Willmot[12] to re-
and land use data for each sub-watershed are presented in flect the degree to which the observed variable is estimated
Table 1. accurately by the simulated variable and d ranges from 0
(no correlation) and 1 (perfect fit).
A monthly time-step was used for all observed and sim-
ulated data. For sediment load, these values were log-
Calibration and validation of the models transformed before calculating the statistics, so differences
between simulated and observed values would not be highly
Evaluation of simulation performance skewed and dominated by a few large values.
After the input files were prepared, the HSPF and SWAT
models were calibrated for streamflow and sediment yield HSPF calibration
for the period of September 1996 to June 2000. Calibration
was done for each sub-watershed (QPB, QPC, and QPD) The HSPF model was run on an hourly time step and
and for the watershed outlet (QPE). Hydrology was the output was average at a daily time-step prior to the com-
first model component calibrated. The hydrology calibra- parison. Initial parameter estimates were specified from
tion involved a comparison of model results to stream flow HSPFParm[13] based on the watershed characteristics and
observations at selected locations, and the subsequent ad- the published information. The HSPF Expert System for
justment of hydrologic parameters. Sediment calibration Calibration (HSPEXP)[14] was used to assist the hydrologic
was performed next. calibration of HSPF. The default error criteria within the
The calibrated parameters of both HSPF and SWAT HSPEXP can be used in the calibration of HSPF for the
were then validated using independent data set. Validation Polecat Creek watershed. For HSPF sediment calibration,
for both models was conducted for the period of September a stepwise approach was used. First, an acceptable match
1994 to December 1995. In order to stabilize state-variables was obtained between long-term sediment balances. Then,
of the models, an initialization period of one year from sediment parameters in HSPF were then further adjusted
September 1995 to August 1996 was inserted at the begin- to obtain a satisfactory agreement between observed and
ning of each simulation. simulated monthly sediment yields.
The simulated runoff volumes and sediment yields were Multiple parameters controlling different sub-processes
evaluated by visual inspection and using quantitative statis- of the hydrologic and sediment transport modules of HSPF
tics. For the visual inspection, the scatter plots were used. were manipulated during the calibration. Parameters AG-
The quantitative statistics used included: mean error (ME), WRC and KVARY, related to groundwater flow, were man-
root mean square error (RMSE), correlation coefficient (r ) ually adjusted for reduction of streamflow. Parameters IN-
and the index of agreement (d) as follows: FILT, UZSN, LZSN, and LZETP were then adjusted man-
ually for monthly runoff volume. Finally, to simulate the
1 n shape of storm hydrographs, IRC, INTFW, INFEXP, and
ME = (Si − Oi ) (1) INFILD were finely adjusted. Sediment transport is highly
n i=1
 sensitive to the exponents JRES, JSER, and JGER.[15]
 n
1  These parameters were adjusted carefully to improve the
RMSE =  (Si − Oi )2 (2) agreement between observed and simulated long-term and
n i=1 monthly sediment yields.
n
(Oi − O)(Si − S)
r =  i=1 n (3)
n SWAT calibration
i=1 (Oi − O) i=1 (Si − S)
2 2
n SWAT was also calibrated using a similar stepwise
i=1 (Oi − Si )
2
d = 1 − n (4) procedure used for HSPF. Initial parameter values in
i=1 (|Si − O| + |Oi − O|) this study were approximated based on the quantitative
Watershed model comparison 1565
Table 2. Observed and simulated annual runoff volumes on the Polecat Creek watershed

Annual runoff volume (mm yr−1 ) Relative error (%)

Sub-watershed Observed HSPF SWAT HSPF SWAT

(a) Calibration period

QPB 332 333 334 −0.3 0.6
QPC 453 476 432 −5.0 4.8
QPD 389 339 388 12.8 0.2
QPE 381 394 373 −3.4 2.1
(b) Validation period
QPB 178 175 197 1.7 10.7
QPC 191 225 267 −17.5 −39.5
QPD 164 179 204 9.1 24.4
QPE 211 180 183 14.7 13.3

characteristics of each sub-watershed and calibration puted as a function of maximum and initial sediment con-
guidelines provided in Neithch et al.[6] The hydrology cal- centrations by a given water volume, channel erodibility
ibration involved the balancing of surface water, ground- factor, and channel cover factor (CH COV). The channel
water, and evapotranspiration to the corresponding mea- erodibility factor assumed to have one order of magnitude
sured values. Amounts of groundwater and surface flow smaller than the values of soil erodibility used in MUSLE
were estimated using the hydrograph separation program equation.[6] The CH COV, which is defined as the ratio of
(HYSEP)[16] from the observed data and these values where erosion from channel with a specified vegetative cover to the
used in the SWAT calibration. corresponding channel erosion with no vegetation cover,
Surface runoff is extremely sensitive to the NRCS Curve was also calibrated to obtain the best results in performance
Number (CN2). Adjusting the value of CN2 was first done criteria.
to obtain a reasonable match between surface runoff vol-
umes. Parameters ESCO (soil evaporation compensation
Results and discussion
factor) and EPCO (soil evaporation compensation fac-
tor) were then calibrated by minimizing the differences in
Hydrology simulation
groundwater and total runoff volumes.
Sediment calibration was performed after the comple- The long-term balances of runoff were first evaluated by
tion of the hydrology calibration. The MUSLE parameters comparing mean of the simulated of the annual values for
were set within the recommended ranges of USDA based both HSPF and SWAT against observed values for both the
on watershed characteristics and land use. Parameters re- calibration and validation periods. The simulated annual
lated to sediment transport equation such as SPCON and runoff for both models agreed well with the observed for
SPEXP were adjusted to match simulated sediment yields the calibration period. The relative error (RE) of mean an-
with observed values. Channel erosion in a reach was com- nual value for both HSPF and SWAT predictions were less

Table 3. Statistical comparison of observed and simulated monthly runoff for the Polecat Creek watershed

ME RMSE
(mm month−1 ) (mm month−1 ) r∗ d∗
Sub-watershed HSPF SWAT HSPF SWAT HSPF SWAT HSPF SWAT

(a) Calibration period

QPB 0.04 0.11 12.20 16.77 0.87 0.81 0.97 0.94
QPC 1.88 −1.79 12.38 16.96 0.92 0.86 0.98 0.95
QPD −4.05 −0.01 14.88 14.89 0.86 0.84 0.96 0.96
QPE 1.03 −0.66 11.05 14.89 0.89 0.84 0.98 0.95
(b) Validation period
QPB −3.41 −3.37 7.45 9.05 0.83 0.73 0.88 0.75
QPC 2.78 6.29 7.53 11.34 0.79 0.54 0.87 0.68
QPD 1.19 3.31 4.84 8.35 0.89 0.67 0.94 0.77
QPE −2.51 −2.32 4.42 7.73 0.94 0.73 0.95 0.80
1566 Im et al.
than 15% for the calibration period (Table 2). The difference from the four monitoring sites within the Polecat Creek
between observed and simulated annual runoff volumes for watershed. Table 4 lists the simulated and observed annual
HSPF ranged from −17.5% (station QPC) to 14.7% (sta- sediment yields for the calibration and validation periods.
tion QPE) for the validation period. For SWAT, the RE in The simulated annual sediment yields by HSPF were rea-
the annual runoff volumes ranged from −39.5% (station sonable with RE of 8.1% error for station QPB, 9.5% for
QPC) to 24.4% (station QPD) for the validation period. station QPC, and 0.4% for QPE. HSPF model, however,
The accuracy of the two models was similar for the calibra- overestimated sediment yields on QPD during the calibra-
tion period, but HSPF predicted annual runoff better for tion by 374%. This discrepancy may be due to atypically
the validation period. high values simulated at flood events for the September
For the monthly runoff volumes, HSPF and SWAT both of 1996 and 1999. The differences between observed and
performed well for the calibration period, but HSPF was simulated sediment yields for SWAT ranged from −82.3%
more accurate for the validation period. The statistical per- to 13.9% during the calibration period. Excluding station
formance measures for monthly runoff are summarized in QPD, HSPF results were much more accurate than SWAT
Table 3. For the calibration period, the mean error (ME) of for the calibration period.
monthly runoff for HSPF ranged from −4.05 mm (QPD) The calibrated sediment parameter values were validated
to 1.88 mm (QPC). The range for RMSE was from 11.05 using measured data from the period of October 1994 to De-
mm to 14.88 mm for HPSF during the calibration period. cember 1995. The results of sediment simulation during the
HSPF also had a higher correlation coefficient during the validation period are also presented in Table 4. The simu-
calibration period (Table 3). For example, the r-values were lated annual sediment yield by HSPF during the validation
0.92 (QPC) and 0.89 (QPE) for HSPF; where the r-values period ranged from 19.3 kg ha−1 yr−1 (station QPC) to 72.4
were 0.86 (QPC) and 0.84 (QPE) for SWAT. kg ha−1 yr−1 (station QPB) and resulted in RE that ranged
The accuracy in simulated monthly runoff volumes was from 17.9% (station QPE) to, 84.7% (station QPD). Table
similar for HSPF and SWAT during calibration; however, 4 also shows that SWAT over-estimated annual sediment
HSPF performed well for the validation period. For the yield at QPC (43.1%), QPD (109.8%), and QPE (167.8%)
validation period at the watershed outlet (QPE), the ME for the validation period, but under-estimated annual sed-
between observed and simulated monthly values for SWAT iment yield by 51.8% at QPB. The sediment yield bias be-
was −0.66 mm and the r value was 0.84, respectively. Ac- tween observed and simulated is partly due to the improper
curacy of the simulated values for the validation period is application of SWAT for predicting sediment in water body,
important because it is a good indication of how well the such as Caroline lake.
calibrated model represents watershed processes. Table 5 shows the results of statistical comparison be-
A comparison of observed and simulated monthly runoff tween observed and simulated monthly sediment yields at
for the calibration and validation periods is presented in all four monitoring stations in the Polecat Creek watershed.
Figure 2. Comparison of simulated monthly runoff volume The RMSE at the watershed outlet (QPE) for the calibra-
clearly shows minimal differences through the simulation tion period were 3.06 kg ha−1 for HSPF and 7.17 kg ha−1
period, despite the different methods used in HSPF and for SWAT. The correlation coefficients between observed
SWAT models to represent watershed hydrologic processes. and simulated monthly sediment yields during the calibra-
tion period ranged from 0.57 to 0.86 for HSPF and 0.63
Sediment simulation to 0.68 for SWAT. For the index of agreement (d); which
The simulated sediment yields from HSPF and SWAT were measures how well the simulated and observed values cor-
both reasonably accurate as compared to the observed data respond to each other, the range of d lay between 0.75 and

Table 4. Observed and simulated annual sediment yields on the Polecat Creek watershed

Annual sediment yield (kg ha−1 yr−1 ) Relative error (%)

Sub-watershed Observed HSPF SWAT HSPF SWAT

(a) Calibration period

QPB 327.5 301.0 58.0 −8.1 −82.3
QPC 77.7 70.3 75.8 9.5 2.4
QPD 64.8 307.1 73.8 373.9 13.9
QPE 214.4 215.3 132.6 0.4 −38.1
(b) Validation period
QPB 56.0 72.4 27.0 29.3 −51.8
QPC 35.7 19.3 51.0 45.8 43.1
QPD 29.5 54.5 61.9 84.7 109.8
QPE 43.5 51.3 116.5 17.9 167.8
Watershed model comparison 1567

Fig. 2. Observed and simulated monthly runoff of the Polecat Creek watershed for each monitoring station.
1568 Im et al.

Fig. 3. Observed and simulated monthly sediment yields of the Polecat Creek watershed for each monitoring station.

0.83 for HSPF and between 0.71 and 0.93 for SWAT for the monthly sediment yields from SWAT were closer to the ob-
calibration period. For the calibration period both mod- served values.
els predicted sediment yield from the Polecat Creek water- The observed sediment yields for four monitoring sta-
shed with a moderate level of accuracy. However, the overall tions during the validation period were compared to those
Watershed model comparison 1569
Table 5. Statistical comparison of observed and simulated monthly sediment yields for the Polecat Creek watershed

ME RMSE
(kg ha−1 month−1 ) (kg ha−1 month−1 ) r∗ d∗
Sub-watershed HSPF SWAT HSPF SWAT HSPF SWAT HSPF SWAT

(a) Calibration period

QPB 1.88 2.53 9.58 7.84 0.57 0.63 0.77 0.71
QPC −2.01 1.05 5.58 2.34 0.65 0.86 0.76 0.93
QPD 1.07 1.57 7.31 3.82 0.58 0.66 0.75 0.81
QPE −1.18 1.10 5.70 5.21 0.68 0.68 0.83 0.83
(b) Validation period
QPB −3.09 1.07 14.44 3.45 0.50 0.87 0.64 0.71
QPC −3.09 1.58 9.58 5.53 0.35 0.19 0.50 0.42
QPD −1.97 2.77 7.39 6.75 0.69 0.39 0.75 0.59
QPE −1.21 4.52 3.06 7.17 0.83 0.55 0.89 0.62
∗
The r- and d-values were estimated from log transformed monthly sediment yield.

predicted by HSPF and SWAT and are presented in Table 5. Overall annual runoff volumes predicted by HSPF and
For the validation period, the ME for monthly sediment SWAT agreed well with the observed data from the Polecat
yield ranged from −3.09 kg ha−1 to −1.21 kg ha−1 for Creek watershed. The mean error (ME) of monthly stream
HSPF and 1.07 kg ha−1 to 4.52 kg ha−1 for SWAT. With flow by HSPF ranged from −4.05 mm to 2.78 mm for the
exception of QPE, RMSE values for HSPF were greater entire simulation period, while the range of ME by SWAT
than those for SWAT for the validation period. The RMSE was from −3.37 mm to 6.29 mm. HSPF also had a higher
values for monthly sediment yield were 3.06 kg ha−1 for correlation coefficient and the index of agreement during
HSPF and 7.17 kg ha−1 for SWAT at the watershed outlet the calibration and validation periods. The results for this
(QPE) during the validation period. The correlation coeffi- study indicate that the simulated monthly stream flows from
cients ranged from 0.35 to 0.83 for HSPF and 0.19 to 0.87 HSPF were more accurate than those predicted by SWAT
for SWAT. Except for QPB, the overall trend of simulated during the calibration and validation periods.
monthly sediment yield by HSPF was generally closer to Sediment yields for the Polecat Creek watershed were ad-
observed values for the validation period than those pre- equately simulated by both HSPF and SWAT models. Com-
dicted by SWAT. paring simulated monthly sediment yields with measured
Figure 3 shows the scatter plot of monthly sediment yield data during the calibration period, it can be concluded that
for the calibration and validation periods. Overall the sim- SWAT predicted slightly better than HSPF with a lower
ulated monthly sediment yields from HSPF and SWAT ME and RMSE values and a higher correlation coefficient.
showed good agreement with observed for both of cal- However, similar results between HSPF and SWAT were
ibration and validation periods. However, HSPF under- obtained for monthly sediment yield during the validation
predicted the low values below 1 kg ha−1 month−1 for all period.
the sub-watersheds. Based on the modeling results of this study, HSPF would
be recommended for the hydrologic and sediment yield sim-
ulations, but other factors should also be considered when
Conclusion selecting a model. HSPF did simulate hydrology and sedi-
ment more accurately than SWAT for the Polecat Creek wa-
Accuracy of the HSPF and SWAT models in simulating
tershed. However, HSPF is less user-friendly than SWAT.
hydrology and sediment yields was compared in this study.
HSPF includes numerous parameters to represent the hy-
Data from a 12,048 ha Polecat Creek watershed in north-
drologic cycle, and erosion loss and transport. The calibra-
eastern Virginia was used for this comparison. The observed
tion of these parameters in HSPF is strenuous and time-
data from Polecat Creek[10] was used to assess the perfor-
consuming process. In SWAT, most of the parameters are
mance of each model in simulating hydrologic and water
automatically generated from GIS data or other informa-
quality responses. The hydrology and sediment components
tion and relatively easy to adjust with proper instruction.
of HSPF were calibrated against the observed data collected
at the Polecat Creek watershed outlet. Although SWAT is
designed for use in the ungaged watershed, the model was References
also calibrated in this study. Data collected from Septem-
ber 1996 to June 2000 were used for the calibrations of both [1] Novotny, V.; Olem, H. Water Quality; Prevention, Identification, and
HSPF and SWAT. Model validations for two models were Management of Diffuse Pollution. Van Nostrand Reinhold, New
done for the period of October 1994 to December 1995. York, 1994.
1570 Im et al.
[2] Niemi, R.M.; Niemi, J.S. Bacteria pollution of waters in pristine and [10] Mostaghimi, S.; Wynn, J.; Car, J. The Polecat Creek Watershed Water
agricultural lands. J. Environ. Qual. 1991, 20, 620–627. Quality Monitoring Project, Report No. PC0902; Chesapeake Bay
[3] Young, R.A.; Onstad, C.A.; Bosch, D.D.; Anderson, W.P. AGNPS: Local Assistance Department, Richmond, VA, 2001.
A nonpoint source pollution model for evaluating agricultural wa- [11] Kopp, J.F. Methods of Chemical Analysis of Water and Wastes, EPA-
tersheds. J. Soil Water Conserv. 1989, 44 (2), 168–173. 600/4-79-020; U.S. Environmental Monitoring and Support Labo-
[4] Haith, D.A.; Shoemaker, L.L. Generalized watershed loading func- ratory, Cincinnati, OH, 1979.
tions for stream flow nutrients. Water Resour. Bull. 1987, 23 (3), [12] Willmot, C.J. On the validation of models. Phys. Geogr. 1981, 2,
471–478. 184–194.
[5] Bicknell, B.R.; Imhoff, J.C.; Kittle, J.L.; Donigian, A.S.; Johanson, [13] Donigian, A.S. Jr.; Imhoff, J.C.; Kittle, J.L. HSPFParm-An Inter-
R.C. Hydrologic Simulation Program-FORTRAN User’s Manual, v. active Database of HSPF Model Parameters, V. 1.0, EPA-823-R-
11, U.S. Environmental Protection Agency, Athens, GA, 1996. 99-004; U.S. Environmental Protection Agency, Washington, DC,
[6] Neitsch, S.L.; Arnold, J.G.; Kiniry, J.R.; Williams, J.R. Soil and Wa- 1999.
ter Assessment Tool; the Theoretical Cocumentation (version 2005), [14] Lumb, A.M.; McCammon, R.B.; Kittle, J.L. User Manual for an
U.S. Agricultural Research Service, Temple, TX, 2005. Expert System (HSPEXP) for Calibration of the Hydrological Sim-
[7] USEPA. Better Assessment Science Integrating Point and Nonpoint ulation Program-Fortran, Water Resources Investigation Report 94-
Sources (BASINS) v. 3.0 User’s Manual, EPA-823-B01-001; U.S. 4168; U.S. Geological Survey, Reston, VA, 1994.
Environmental Protection Agency, Office of Water: Washington, [15] Fontaine, T.A.; Jacomino, V.M.F. Sensitivity analysis of simulated
DC, 2001. contaminated sediment transport. J. Am. Water Resour. Assoc.
[8] Williams, J.R.; Berndt, H.D. Sediment yield prediction based on 1997, 33 (2), 313–326.
watershed hydrology. Trans. ASAE 1977, 20 (6), 1100–1104. [16] Sloto, R.A.; Crouse, M.Y. HYSEP: A Computer Program for
[9] Im, S.; Brannan, K.M.; Mostaghimi, S. Simulating hydrologic and Streamflow Hydrograph Separation and Analysis, Water Resources
water quality impacts on an urbanizing watershed. J. Am. Water Investigation Report 96-4040; U.S. Geological Survey, Lemoyne,
Resour. Assoc. 2003, 39 (6), 1465–1479. PA, 1996.