Transactions of the ASABE
Vol. 49(2): 367−374 2006 American Society of Agricultural and Biological Engineers ISSN 0001−2351 367
M
ODELING LONG-TERM WATER QUALITY
IMPACT OF STRUCTURAL BMPS
K. S. Bracmort, M. Arabi, J. R. Frankenberger, B. A. Engel, J. G. Arnold
ABSTRACT. Structural best management practices (BMPs) that reduce soil erosion and nutrient losses have been recommended
and installed on agricultural land for years. A structural BMP is expected to be fully functional only for a limited period after
installation, after which degradation of the BMP is likely to lead to a reduction in the water quality improvement provided
by the BMP. Assessing the impact of BMPs on water quality is of widespread interest, but no standard methods exist to
determine the water quality impact of structural BMPs, particularly as the impact changes through time. The objective of this
study was to determine the long-term (~20 year) impact of structural BMPs in two subwatersheds of Black Creek on sediment
and phosphorus loads using the Soil and Water Assessment Tool (SWAT) model. The BMPs were represented by modifying
SWAT parameters to reflect the impact the practice has on the processes simulated within SWAT, both when practices are fully
functional and as their condition deteriorates. The current condition of the BMPs was determined using field evaluation
results from a previously developed BMP condition evaluation tool. Based on simulations in the two subwatersheds, BMPs
in good condition reduced the average annual sediment yield by 16% to 32% and the average annual phosphorus yield by
10% to 24%. BMPs in their current condition reduced sediment yield by only 7% to 10% and phosphorus yield by 7% to 17%.
Keywords. Conservation effects, Phosphorus, Sediment, SWAT, Watershed modeling.
est management practices (BMPs) are routinely
used to reduce nonpoint-source pollution resulting
from agricultural activities and improve water
quality. Studies have demonstrated how structural
management practices can improve water quality, but the
duration of their effectiveness and performance is largely un-
known. Although a “design life” has been established by the
USDA Natural Resources Conservation Service for most
structural BMPs (e.g., USDA-NRCS, 2004), the ability of a
BMP to function effectively throughout its assigned design
life is uncertain, and its continued effectiveness past the de-
sign life is unknown. Quantifying the impact some 25 years
after BMPs are installed would provide insight into water
quality improvement taking place over time due to BMP im-
plementation.
Watershed modeling is one approach to analyzing the
water quality impact, both short and long term, of BMP
implementation. Watershed models have been used for
decades to study nonpoint-source pollution and the impact of
non-structural BMPs, but a limited number of studies have
Submitted for review in December 2004 as manuscript number SW
5684; approved for publication by the Soil & Water Division of ASABE in
March 2006.
The authors are Kelsi S. Bracmort, ASABE Member, Engineering
Program Specialist, USDA-NRCS, Washington, D.C.; Mazdak Arabi,
ASABE Member Engineer, Research Associate, Jane R. Frankenberger,
ASABE Member Engineer, Associate Professor, and Bernard A. Engel,
ASABE Member, Professor, Department of Agricultural and Biological
Engineering, Purdue University, West Lafayette, Indiana; and Jeff G.
Arnold, ASABE Member Engineer, Hydraulic Engineer, USDA-ARS
Grasslands Soil and Water Research Laboratory, Temple, Texas.
Corresponding author: B. A. Engel, Department of Agricultural and
Biological Engineering, Purdue University, West Lafayette, IN 47907;
phone: 765-494-1162; fax: 765-496-1115; e-mail: engelb@purdue.edu.
examined the application of structural BMP simulation in
watershed hydrology models.
The Soil and Water Assessment Tool (SWAT) predicts the
impact of land management practices on water, sediment,
and agricultural chemical yields in watersheds with varying
soils, land use, and management conditions over time
(Arnold et al., 1998). The continuous-time, process-based
model requires specific information about weather, soil
properties, topography, vegetation, presence of ponds or
reservoirs, groundwater, the main channel, and land manage-
ment practices.
The Black Creek Project in northeastern Indiana (1973 to
1984) implemented several structural management practices
to identify BMPs that reduce sediment and phosphorus levels
entering Lake Erie from agricultural activities (Morrison and
Lake, 1983). The U.S. EPA provided about one million
dollars to install and evaluate non-structural and structural
BMPs and their impact on agricultural nonpoint-source
pollution leaving the Black Creek watershed (Lake and
Morrison, 1977b). Flow and water quality data were recorded
for four years, encompassing the time period when practices
were installed and continuing for a short time after all
practices were installed.
A representative sample of grassed waterways, grade
stabilization structures, field borders, and parallel terraces
installed during the Black Creek Project had previously been
inspected and assigned a condition score using evaluation
tools developed for that purpose (Bracmort et al., 2004).
Evaluation of the current condition of the BMPs found that
one-third of the practices no longer exist today and that the
two-thirds that still exist are in fair condition and are partially
functional. However, water quality impacts of BMPs under
good and existing conditions were not quantified in that
study.
B
368 TRANSACTIONS OF THE ASABE
Figure 1. Smith Fry and Dreisbach location map and BMPs implemented.
The objective of this study was to analyze the long-term
water quality impact of structural BMPs implemented during
the Black Creek Project by developing a method to represent
the functionality of structural BMPs in varying conditions in
the SWAT model and then applying that method to the Black
Creek watershed. Water quality studies conducted during the
Black Creek Project, and the recent evaluation of a select
group of practices, provide the opportunity to model the
Black Creek watershed and to simulate BMP effects on the
agricultural watershed close to 30 years after the initiation of
the project.
METHODS AND MATERIALS
WATERSHED DESCRIPTION
The Black Creek watershed, roughly 50 km
2
in size, is
located in northeast Allen County, Indiana, about 24 km
northeast of the city of Fort Wayne (fig. 1). The watershed is
a tributary to the Maumee River, which flows northeast from
Fort Wayne to Lake Erie, at Toledo, Ohio. In general, the soils
are deep, ranging from moderately well drained to very
poorly drained and medium to fine textured. The dominant
hydrological soil group of soil series in the watershed is
type C. Average annual rainfall for Fort Wayne is 928 mm
(Indiana State Climatology Office, 2005). Topography for
the watershed consists of mostly gently sloping land with an
elevation of 256 m in the north and 228 m at the outlet. Row
crop agriculture and pasture are the primary land uses in the
watershed. Soybeans and corn are the main crops grown in
the watershed, followed by winter wheat and other small
grains (table 1).
Two drainage areas, Smith-Fry and Dreisbach, were used
for monitoring and modeling (fig. 1). The Dreisbach
watershed (6.23 km
2
) is located along the western border of
the Black Creek watershed and contains more BMPs per unit
area than the Smith Fry watershed (7.30 km
2
), which is
located along the eastern border of the Black Creek
watershed.
Watershed characteristics, management information, and
BMP design information needed for BMP representation
were collected for the mid to late 1970s for both watersheds.
Observed water quality data for model calibration and
validation purposes came from published Black Creek
reports (Lake and Morrison, 1977a; Christensen and Wilson,
1981). Flow data were obtained from unpublished project
literature. Data utilized for SWAT simulations are shown in
table 2 and are available at http://pasture.ecn.purdue.
edu/~blkcreek.
BMP REPRESENTATION IN SWAT
SWAT has previously been used to model the impact of
structural BMPs in good condition. Vache et al. (2002)
simulated riparian buffers, grassed waterways, filter strips,
and field borders by modifying the channel cover factor and
channel erodibility factor in SWAT to model the cover
density and erosion resistance of the structures. Santhi et al.
(2003) simulated grade stabilization structures in SWAT by
modifying the slope and soil erodibility factor. The impact of
filter strips on sediment and nutrient reduction was simulated
Table 1. Land use in the Dreisbach and Smith Fry watersheds.
Land Use
Dreisbach
(% area)
Smith Fry
(% area)
Pasture 37.5 8.7
Corn 23.4 33.6
Winter wheat 17 14.3
Soybean 7.2 31.8
Forest 5.8 8.9
Residential 9.1 2.7
369Vol. 49(2): 367−374
Table 2. Model input data information (available at http://pasture.ecn.purdue.edu/~blkcreek).
Data Type (date) Source Description
DEM (2001) National elevation data 30 m resolution, U.S. Geological Survey (USGS)
Soils (2002) Soil Survey Geographic Database Digital representation of county soil survey maps by the USDA-NRCS
Land use (1974-1978) Black Creek Project Digitized into GIS from aerial photos
Weather (1974-1977) Black Creek Project documents
[a]
Daily precipitation graphs
Weather (1970-2002) Purdue Applied Meteorology Group Minimum and maximum daily temperature and daily precipitation
Crop management Engel and Lim (2001) Management scenarios for crops
Streamflow (1975-1978) Black Creek Project documents
[b]
Daily streamflow
Water quality (1974-1977) Black Creek Project documents
[c]
Daily suspended solids, sediment P, soluble organic P, soluble inorganic P
[a]
Lake and Morrison (1977a).
[b]
Lake and Morrison (1977b).
[c]
Morrison and Lake (1983).
as a function of filter strip width. However, these studies did
not consider the change in BMP effectiveness as the condi-
tion deteriorates.
For this study, a method was developed to represent the
ability of grassed waterways, grade stabilization structures,
field borders, and parallel terraces in SWAT to reduce
sediment occurring from non-gully erosion. The method was
based on published literature pertaining to BMP simulation
in hydrological models and considering the hydrologic and
water quality processes simulated in SWAT.
Parallel terraces and field borders are implemented to
reduce sheet and rill erosion from fields and other non-chan-
nel areas. SWAT utilizes the Modified Universal Soil Loss
Equation (MUSLE) (Williams, 1975) to estimate sheet
erosion for each hydrologic response unit (HRU). HRUs are
portions of subwatersheds with unique soil, land use, and
management attributes. Appropriate model parameters for
representation of the effect of parallel terraces are the curve
number (CN2) and USLE support practice factor (USLE_P),
along with slope length (SLSUBBSN). FILTERW (width of
edge-of-field filter strip) was recognized to be the appropri-
ate parameter for representation of field borders.
Implementation of grassed waterways and grade stabiliza-
tion structures result in reduction of water flow and channel
erosion in the channel network. Sediment deposition and
degradation are two important channel processes that affect
sediment yield at the outlet of the watershed. Key parameters
for representing this effect are the channel Manning’s coeffi-
cient (CH_N2), channel slope (CH_S2), channel erodibility
factor (CH_EROD), and channel cover factor (CH_COV).
Parameters selected to represent the impact of each BMP, along
with their estimated values in good condition and without the
BMP, are listed in table 3. A sensitivity analysis was performed
to ascertain the sensitivity of SWAT flow and water quality
computations to the selected parameters.
After appropriate parameters were selected and sensitivity
analyses validated their impact on SWAT predictions, the
parameters were modified to represent BMPs in good and
existing conditions. Parameter values for fair and poor
condition were interpolated between values corresponding to
BMPs in good condition and values with no BMP in place.
In running the model with BMPs represented as described
above, the critical source area was selected to be small
enough that most of the BMPs could be represented by a
single sub-basin outlet. Critical source area refers to the
minimum area required by the model for initiation of channel
processes. For the five cases where multiple BMPs existed in
a single sub-basin, practices were simulated by assigning one
practice in the sub-basin to a certain land use or HRU and the
other practice to another land use or HRU.
The current condition of a practice was simulated based
on condition scores assigned to a subset of practices in the
Black Creek watershed by using the evaluation tools
described in Bracmort et al. (2004). The current condition of
BMPs that were not physically inspected was assumed to be
the average current condition score for that practice type.
Table 3. Representation of BMPs in SWAT.
BMP Function
Representative SWAT Parameter
Variable (input file)
Value with no BMP
(from calibration)
Value with BMPs
in good condition
Grassed waterway
Increase channel cover CH_COV (.rch) 0.2 0.0
Reduce channel erodibility CH_EROD (.rch) 0.2 0.0
Increasing channel roughness CH_N(2) (.rch) 0.04 0.24
Parallel terrace
Reduce overland flow CN(2) (.mgt) Assigned by SWAT
[a]
−−
[b]
Reduce sheet erosion USLE_P (.mgt) 0.3 0.2
Reduce slope length SLSUBBSN (.hru) Assigned by SWAT
[c]
−−
[d]
Field border Increase sediment trapping FILTERW (.hru) 0 5 (m)
Grade stabilization structure Reduce gully erosion CH_EROD (.rch) 0.2 0.0
Reduce slope steepness CH_S(2) (.rch) Assigned by SWAT
[c]
−−
[e]
[a]
Assigned by SWAT based on hydrologic soil group, land use, and antecedent moisture condition.
[b]
Obtained from SWAT user’s manual (version 2000) for contoured and terraced condition (based on land treatment and hydrologic soil group).
[c]
Assigned by SWAT based on the digital elevation model (DEM).
[d]
Estimated for each parallel terrace based on its features and SWAT-assigned overland slope of the HRU where it is installed:
SLSUBBSN = (A × S + B) × 100/S, where S is average slope of the HRU, A = 0.21, and B = 0.9 (ASAE Standards, 2003).
[e]
Estimated for each grade stabilization structure based on its features and SWAT-assigned slope and length of the channel segment where it is installed:
CH_S(2) = CH_S(2)
SWAT
assigned
− D/CH_L(2), where D is height of the structure (1.2 m), and CH_L(2) is length of the channel segment.
370 TRANSACTIONS OF THE ASABE
SENSITIVITY ANALYSIS
The SWAT model outputs depend on many input parame-
ters related to the soil, land use, management, weather,
channels, aquifer, and reservoirs. Therefore, modeling BMPs
with SWAT necessitates evaluation of the sensitivity of
SWAT flow, sediment, and nutrient outputs to the selected
parameters for representation of BMPs. Table 4 summarizes
30 SWAT parameters selected for sensitivity analysis in this
study. These parameters were chosen based on the results of
previous studies by Arnold et al. (2000), Eckhardt and Arnold
(2001), Santhi et al. (2001), and Santhi et al. (2003).
Sensitivity of streamflow, sediment, and phosphorus outputs
of the SWAT model to the selected parameters was sought by
perturbing model parameters “one at a time” and determining
a linear sensitivity parameter (S
i
), defined as (adapted from
Gu and Li, 2002):
)/()(
)/()(
1,2,1,2,
1212
iiii
i
yyyy
S
α+αα−α
+−
=
(1)
where y
1
and y
2
are model outputs corresponding to perturba-
tion of the ith element of the parameter vector from a
i,1
to a
i,2
,
while other parameters were kept constant. In equation 1, it
is assumed that the response of model outputs to parameter
perturbation is linear. S
i
is essentially a normalized estimate
of sensitivity of design variables (streamflow, sediment
yield, etc.) to a parameter perturbation, with higher values in-
dicating higher sensitivity. Parameters with high sensitivity
were also chosen for model calibration.
M
ODEL CALIBRATION AND VALIDATION
Model calibration and validation were performed for
monthly streamflow, surface runoff, sediment, mineral
phosphorus, and total phosphorus using rainfall, flow, and
water quality input data collected for the time when the BMPs
were installed and for a few years after implementation. The
Dreisbach watershed was calibrated with the inclusion of
26 BMPs (5 grassed waterways, 10 grade stabilization
structures, 7 field borders, and 4 parallel terraces) using the
representation method for BMPs in good condition, since
documentation indicated these were installed in the first
phase of the project, the early to mid-1970s, prior to
collection of observed water quality data (Morrison and
Lake, 1983). The Smith Fry watershed was calibrated
without BMPs because these 10 BMPs (1 grassed waterway,
4 grade stabilization structures, 3 field borders, and 2 parallel
terraces) were implemented in the late 1970s (fig. 1).
Calibration involved comparing average monthly simu-
lated and observed values, and computing the Nash-Sutcliffe
efficiency (E
N−S
) (Nash and Sutcliffe, 1970) and coefficient
of determination (R
2
) between observed and simulated
values. Model calibration was considered satisfactory if the
simulated quantity was within 20% of observed data, R
2
was
greater than 0.6, and E
N−S
was greater than 0.5.
Once calibration of the model was completed, validation
was performed to evaluate the accuracy of the model to
predict values from an observational data set different from
the calibration data (Wilson, 2002). Model validation used
optimal parameter values selected during model calibration
Table 4. List of SWAT parameters considered in sensitivity analysis.
Parameter Short Description Minimum Maximum Units
CN2 Initial SCS runoff curve number 35 98 −−
SLOPE Average slope steepness 0 0.6 m/m
SLSUBBSN Average slope length 10 150 m
ESCO Soil evaporation compensation factor 0 1 −−
CH-N1 Manning’s “n” value for tributary channels 0.008 30 −−
CH-S1 Average slope of tributary channels 0 10 m/m
CH-K1 Effective hydraulic conductivity in tributary channel alluvium 0 150 mm/h
CH-N2 Manning’s “n” value for the main channel 0.008 0.3 −−
CH-S2 Average slope of the main channel 0 10 m/m
CH-K2 Effective hydraulic conductivity in main channel 0 150 mm/h
GWQMN Threshold depth of water in shallow aquifer 0 5000 mm
ALPHA-BF Baseflow alpha factor 0 1 days
GW-DELAY Groundwater delay time 0 500 days
GW-REVAP Groundwater “revap” time 0.02 0.2 −−
SOL-AWC Available water capacity of the soil layer 0 1 mm/mm
CH_EROD Channel erodibility factor 0 0.6 cm/h/Pa
CH_COV Channel cover factor 0 1 −−
SPCON Linear coefficient for sediment routing 0.001 0.01 −−
SPEXP Exponent coefficient for sediment routing 1 1.5 −−
PRF Peak rate adjustment factor for sediment routing 0 2 −−
USLE_P USLE equation support practice factor 0.1 1 −−
USLE_C Maximum value of USLE equation cover factor 0.001 0.5 −−
SOL_LABP Initial soluble P concentration in soil layer 0 100 mg/kg
SOL_ORGP Initial organic P concentration in soil layer 0 4000 mg/kg
RS1 Local algae settling rate at 20°C 0 2 m/day
RS2 Benthic (sediment) source rate for dissolved P 0.001 0.1 mg/m
2
⋅day
RS5 Organic P settling rate in the reach at 20°C 0.001 0.1 1/day
BC4 Rate constant for mineralization of P to dissolved P 0 1 1/day
AI0 Ratio of chlorophyll-a to algae biomass 0.001 0.01 µg/mg
AI2 Fraction of algal biomass that is phosphorus 0.01 0.02 mg P/mg al
RHOQ Algal respiration rate at 20°C 0.05 0.5 1/day
371Vol. 49(2): 367−374
for a number of months following the calibration time period.
Predicted and observed data were compared using E
N−S
and
R
2
to test model validity.
M
ODEL RUNS
Model simulations were performed to quantify the
long-term impact of the BMPs in good and current conditions
on water quality over a 25-year period (1975-2000). Each
watershed was simulated for 25 years with no BMPs
(baseline case), BMPs in good condition, and BMPs in
varying condition (representing their existing condition).
Predicted sediment and total P for these scenarios were
compared to analyze the impact of BMP presence and BMP
condition on water quality. All inputs were held constant
except for the presence and condition of BMPs.
RESULTS
Results indicated that SWAT outputs, computed at the
outlet of the study watersheds, were sensitive to the
parameters selected for representation of BMPs. Figure 2
depicts the sensitivity of streamflow, sediment yield, and
total P outputs of the SWAT model to the parameters listed in
table 4. The magnitude of the sensitivity index, S
i
(eq. 1),
corresponding to each model parameter is subject to the
initial set of parameters that are used in the analysis.
Figure 2b illustrates the sensitivity of sediment output of
SWAT to various input parameters listed in table 4 for two
cases. In the first case, the default value was used for the
USLE practice factor, i.e., USLE_P = 1. It was observed that
in this case the parameters that affect the magnitude of
channel degradation, such as PRF, CH_COV, and CH_EROD
(see table 4), did not bear a high sensitivity for sediment
outputs. However, when the USLE practice factor was
altered to 0.3, i.e., USLE_P = 0.3, the parameters correspond-
ing to sediment transport in the channel network were among
the most sensitive parameters, as demonstrated in figure 2b.
The sensitivity index for parameter FILTERW, which was
used for representation of field borders (not shown in fig. 2),
was 0.35 and 0.72 for sediment and total P, respectively.
Therefore, sediment and total P outputs of the SWAT model
were sensitive to all of the parameters selected for representa-
tion of parallel terraces, field borders, grassed waterways,
and grade stabilization structures.
Satisfactory model calibration and validation results were
obtained for both watersheds, as evaluated by the Nash-Sut-
cliffe (E
N−S
) and coefficient of determination (R
2
) values
(tables 5 and 6, figs. 3 and 4). The calibrated model was able
to adequately predict both low and high streamflows and
sediment yields in both watersheds except for March 1978,
for which streamflows were underpredicted. While the
model slightly overpredicted mineral and total phosphorus
yields at the outlets for the months with low phosphorus
0.0
0.3
0.6
0.9
1.2
CN2
SOL−AWC
GWQMN
CH−K1
SLOPE
ALPHA−BF
GW−DELAY
GW−REVAP
CH−N2
CH−S2
CH−K2
SLSUBBSN
ESCO
CH−N1
CH−S1
0.0
1.0
2.0
3.0
4.0
CN2
USLE_P
CH−N2
USLE_C
SPCON
CH−S2
SOL−AWC
SLOPE
SLSUBBSN
CH−K1
PRF
CH−N1
SPEXP
GWQMN
CH−S1
GW−DELAY
ALPHA−BF
ESCO
CH−K2
GW−REVAP
CH_EROD
CH_COV
USLE_P = 1.0
USLE_P = 0.3
0.0
0.8
1.6
2.4
3.2
CN2
USLE_P
AI0
SOL_ORGP
AI2
RHOQ
USLE_C
RS1
SOL_LABP
SLOPE
SOL−AWC
RS5
CH−K1
SLSUBBSN
BC4
SOL_ORGN
CH−N1
GWQMN
CH−K2
CH−N2
CH−S1
ALPHA−BF
GW−DELAY
K−P
CH−S2
GW−REVAP
RS2
ESCO
SOL_NO3N
RS4
AI1
SWAT Parameter
(a) Streamflow
(b) Sediment yield
(c) Total P yield
Sensitivity Index (S
i
)Sensitivity Index (S
i
)Sensitivity Index (S
i
)
Figure 2. Sensitivity of SWAT parameters in table 4, determined based on
(a) streamflow, (b) sediment, and (c) total P.
yields, the high-yield months were underpredicted. The cali-
brated model was used to evaluate the impact of BMPs on
streamflow, sediment, and total P yields at the outlet of the
study watersheds.
STREAMFLOW
Runoff volume and streamflow at the outlet of the
Dreisbach and Smith Fry watersheds were not affected by
implementation of BMPs. This was anticipated, because
BMP selection was targeted at sediment and phosphorus
reduction. Parallel terraces, the only type of BMPs in the
study area that influence runoff prediction parameters
(overland slope and curve number), cover less than 2% of the
Table 5. Results of calibration of SWAT for average monthly streamflow, sediment, and nutrients.
Dreisbach (Site 6) Smith Fry (Site 2)
Variable
Observed Simulated R
2
E
N−S
Observed Simulated R
2
E
N−S
Streamflow (mm) 16 17 0.92 0.84 19 18 0.86 0.73
Surface runoff (mm) 15 15 0.91 0.80 16 18 0.84 0.62
Suspended solids (t/ha) 0.027 0.024 0.97 0.92 0.151 0.16 0.94 0.86
Mineral P (kg/ha) 0.070 0.070 0.92 0.84 0.071 0.071 0.9 0.78
Total P (kg/ha) 0.077 0.094 0.93 0.78 0.075 0.068 0.64 0.51
