You are on page 1of 17

Journal of Hydrology (2008) 355, 131– 147

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/jhydrol

The influence of impoundments on riverine nutrient


transport: An evaluation using the Soil and Water
Assessment Tool
Nathan S. Bosch *

University of Michigan, School of Natural Resources and Environment, Dana Building, 440 Church Street, Ann Arbor,
MI 48109-1041, United States

Received 11 December 2007; received in revised form 11 March 2008; accepted 14 March 2008

KEYWORDS Summary Studies of nutrient dynamics within stream reaches and in lakes and impound-
Catchment; ments have resulted in important advances in our understanding of each system. However,
Impoundment; less effort has been directed at understanding how linkages between stream and impound-
SWAT; ment systems interact to determine nutrient dynamics at the watershed scale, and the pos-
Nitrogen; sible influence of position of lakes and impoundments within river systems. The Soil and
Phosphorus; Water Assessment Tool (SWAT) was applied to the Huron and Raisin watersheds in southeast-
River ern Michigan to better understand the effect of impoundments on riverine nitrogen (N) and
phosphorus (P) exports. The two watersheds were calibrated and validated for stream dis-
charge and water quality parameters using data from 1995 to 2005 by using time series plots
and statistical measures to verify model predictions. Simulated hydrology and water quality
parameters closely resembled observed data except for daily streamflow in the Huron
watershed and monthly nitrate loads in the Raisin watershed. The presence of impoundments
had a marked effect on nutrient export from both watersheds. Modeled total phosphorus (TP)
and total nitrogen (TN) export loads approximately doubled when all impoundments were
removed from the Huron watershed model. The Raisin watershed showed a greater absolute
increase in TP loads and a similar absolute increase in TN loads compared to the Huron, but
because nutrient loads were several times larger in the Raisin, the proportional change was
less. Impoundments placed near river mouths or in N and P source areas were most effective
at reducing export, and many smaller reservoirs caused a greater reduction in nutrient loads
than did a single large reservoir. In addition, impoundments increased the interannual vari-
ability in nutrient loads. Based on simulations using SWAT, impoundments have a substantial
effect on riverine nutrient export, and that effect varies with their size and location.
ª 2008 Elsevier B.V. All rights reserved.

* Tel.: +1 734 741 2147; fax: +1 734 741 2055.


E-mail address: boschn@umich.edu

0022-1694/$ - see front matter ª 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.jhydrol.2008.03.012
132 N.S. Bosch

Introduction determined by sediment–water interactions, because ni-


trate is the most common form of N and can be lost from
River systems, with their associated impoundment and the system through denitrification in anoxic impoundment
stream components, provide much opportunity for nutrients sediments. The sediment retention argument is echoed by
to be transformed and removed as they are transported an extensive study of the world’s 633 largest reservoirs
through the watershed. This nutrient transport through (Vorosmarty et al., 2003). This study reported that 25–
the river network and eventually to a receiving body of 30% of the world’s annual sediment flux in rivers is trapped
water has broad implications for eutrophication, drinking in artificial basins each year. Following removal of a dam,
water suitability, recreational activities, and fishery har- short-term sediment transport downstream has been shown
vests. Nutrient processing in stream reaches, especially in to increase, the expected consequence of sediment accu-
headwater streams, has been shown to have an important mulation behind dams (Hart et al., 2002). More importantly,
influence on river nitrogen (N) (Alexander et al., 2000; Bern- however, elevated, long-term sediment transport has also
hardt et al., 2005; Dodds et al., 2002; Peterson et al., 2001; been observed by Hart et al. (2002), highlighting the func-
Webster et al., 2003) and phosphorus (P) dynamics and tion of impoundments as sites of sediment and associated
export (Barlow et al., 2004; Evans and Johnes, 2004; Evans P accumulation. In a study that integrated a broad range
et al., 2004; Smith et al., 2005) in many studies. In contrast, of Australian research pertinent to watershed nutrient pro-
limited research has been directed toward understanding cessing and transport, Harris (2001) concluded that
the impact of natural and artificial impoundments on nutri- impoundments play a large role in watershed nutrient
ent transport and processing in river systems. A few studies dynamics. Dissolved organic N and total Kjeldahl N (organic
have investigated how river impoundments interact with N and ammonia) were consumed in Australian impound-
stream reaches and the potential alteration of stream nutri- ments, whereas dissolved inorganic N (mostly nitrate) was
ent dynamics in the river system. The principal expectation produced and exported downstream. Increases in impound-
is that impoundments will act as net nutrient sinks in river ment residence time resulted in more N losses from water-
systems (Alexander et al., 2002; Dixit et al., 2000; Ejsmont- sheds due to increases in denitrification.
karabin et al., 1993; Kling et al., 2000; Maurer et al., 1995). Other research has reported seemingly contradictory re-
A paleolimnological analysis within a lake chain in Saskatch- sults. A study of nine North American lake chains represent-
ewan, Canada, found strong evidence that lakes were effec- ing a range of landscape types including alpine, prairie,
tive in removing N and P from urban point-source nutrient tundra, and glacial plains (Soranno et al., 1999) found few
loading (Dixit et al., 2000). Maurer et al. (1995) found that consistent patterns in changes in lake constituents along
a large, eutrophic natural lake in Minnesota was still an lake chains. Surprisingly, two of the surface-connected lake
active net annual sink for both N and P despite consistent chains showed increases in total phosphorus (TP) and total
sedimentation over the last century. In a study by Ejsmont- nitrogen (TN) in downstream lakes. Modeling of nitrogen ex-
karabin et al. (1993), increased water residence time in port from 16 watersheds in the northeastern US found little
smaller impoundments was found to increase P retention. effect of reservoirs (Seitzinger et al., 2002). Although the
Alexander et al. (2002) used the Spatially Referenced number of reservoirs per watershed varied from 0 to above
Regression on Watershed Attributes (SPARROW) model to 20, differences in the proportion of N removed with or with-
estimate N and P removal in 75 lakes and reservoirs in out the reservoirs was estimated to be very small, less than
New Zealand and found median removals by individual 2 percentage points. This surprising result was attributed to
impoundments were 10% and 6% for TP and TN, respec- the model structure, in which the proportion of N removed
tively. Kling et al. (2000) demonstrated how a chain of small depended on the ratio of depth to water residence time. Be-
lakes interspersed between river segments affected biolog- cause impoundments increase both depth and residence
ical and chemical variables in a river-lake system draining times compared to pre-dam conditions, their ratio showed
into Toolik Lake, Alaska. Several longitudinal and river-lake little change, so that the Seitzinger et al. model predicted
trends were observed, including a trend for particulate that impoundments would have little effect on N removal.
nutrients to decrease predictably downslope. Dissolved River nutrient export is influenced not only by watershed
nutrient concentrations were extremely low and difficult loading, stream-reach processing, and impoundment of
to measure accurately in this Arctic river system, but ni- water and its constituents; in addition, the landscape
trate was shown to be consumed by lakes and produced in position of impoundments and the linkages between these
stream reaches. Kling et al. argued for more research inte- features also merit attention. In a recent report by the
grating material processing in both lakes and rivers from a American Society of Limnology and Oceanography (ASLO,
landscape perspective. 2003) resulting from a workshop on emerging research ques-
In a BioScience issue dedicated to the ecology of dam re- tions for limnology, the authors concluded that one emerg-
moval, Stanley and Doyle (2002) discussed the potential ing research need centered around landscape positioning
merits of impoundments with regard to nutrient removal and linkages of groundwater, wetlands, streams, and lakes.
from the river systems. They argued that P dynamics in They specifically encouraged research at the watershed
impoundments are primarily governed by sediment storage scale that seeks to understand linkages among these compo-
and movement, because impoundments allow sediments nents. The report further argued that an improved under-
with associated P (inorganic P adsorbs to mineral sediment standing of these linkages between hydrologic components
and soil particles because of natural affinity) to settle out with long residence times (e.g., lakes) and short residence
of the water column and eventually become buried and times (e.g., streams) would offer insight into transport
unavailable. In contrast, N removal from the water is mostly and retention of N and P.
The influence of impoundments on riverine nutrient transport 133

The goal of the present study was to better understand Barnwell, 1987), giving the model a thorough representa-
nutrient processing and transport through river systems by tion of in-stream N and P processing and removal mecha-
studying impoundment and stream-reach nutrient dynamics nisms. SWAT also models N and P routing through two
at the watershed scale within the context of their landscape types of impoundments included in the model, reservoirs
position and linkages. Because it is not feasible to actually and ponds. Nutrient dynamics are determined identically
create or remove impoundments at multiple locations with- in ponds and reservoirs and are limited to removal of N
in the study watersheds, a watershed hydrologic model was and P by settling (Chapra, 1997). SWAT estimates N and
used to simulate these changes and predict results. The Soil P removal in a particular impoundment based on the
and Water Assessment Tool (SWAT) was chosen for its effec- incoming nutrient concentrations, the user-defined set-
tiveness in previous watershed applications with hydrologic tling velocity, and the area of sediment–water interface
and water quality simulation (Green et al., 2006; Qi and estimated from user-defined impoundment dimensions.
Grunwald, 2005; Veith et al., 2005) and for its explicit inclu- Ponds and reservoirs differ in SWAT, however, in that
sion of spatially defined impoundments (Neitsch et al., ponds receive water from a portion of a single subwater-
2002). The Huron and Raisin watersheds in southeastern shed while reservoirs are located on the main channel net-
Michigan were chosen as the study watersheds because they work and thus receive water from all upstream
represent typical Midwestern watersheds and had the nec- subwatersheds. Previous SWAT applications have specifi-
essary data for setting up the model and performing calibra- cally evaluated model N and P predictions for watersheds
tion procedures. In addition, nutrient budgets in these two in Finland (Francos et al., 2001; Grizzetti et al., 2003),
watersheds have shown that impoundments may be greatly Pennsylvania (Veith et al., 2005), Texas (Saleh et al.,
altering river exports of nutrients such that more study of 2000; Santhi et al., 2001), Maryland (Chu et al., 2004),
these impoundments is warranted (Bosch and Allan, 2008). and Indiana (Engel et al., 1993). These studies each con-
cluded that measured P and N annual and monthly loads
were reasonably predicted by SWAT, except in the case
Methods of monthly predictions in the Maryland study.

Model description
Study area
SWAT is a process-based watershed model with spatially ex-
plicit parameterization at the subwatershed spatial scale, The Huron (2377 km2) and Raisin (2737 km2) are two pre-
with partial lumping of parameters within each subwater- dominately agricultural watersheds in southeastern Michi-
shed (Arnold and Fohrer, 2005; Arnold et al., 1998). The gan that drain into the western end of Lake Erie (Fig. 1).
hierarchy of spatial organization within SWAT is important They are characterized by hilly to moderately undulating
to understand. The watershed is defined by the main wa- topography in their upper watershed underlain with mor-
tershed outlet as chosen by the user, which is the mouth aines, till deposits, and outwash plains. The lower portions
of each of the rivers into Lake Erie in this study. The wa- of the Huron and Raisin are underlain by sands and clays
tershed is then subdivided into subwatersheds. The modeler from glacial Lake Erie (Knutilla and Allen, 1975). Both
can define as many or as few subwatersheds as desired watersheds originate in ‘‘lake district’’ areas, but their mid-
according to the level of spatial resolution that is reason- dle and lower sections differ notably in their degree of river
able. Each subwatershed is then further divided into a num- impoundment. The Huron watershed has a total of 88
ber of Hydrologic Response Units (HRU) based on unique impoundments that have surface areas of at least 10 ha,
combinations of land use and land cover (LULC) and soil whereas the Raisin has only 14 impoundments. These river
types within the subwatershed. These HRUs are not spatially impoundments include artificial reservoirs, natural lakes,
defined within the subwatershed; they are simply account- and smaller ponds, and most are controlled through dam
ing categories which represent the total area of the unique operations or water level control structures.
LULC and soil type they represent within a subwatershed.
HRU-scale processes are simulated separately for each Model data sources
HRU and then aggregated up to the subwatershed scale by
a weighted average. SWAT is a time-continuous simulation The Geographic Information System (GIS) interface created
model that operates on a daily time step. Major model com- for SWAT, called AVSWAT (Di Luzio et al., 2004a; Di Luzio
ponents include weather, soil temperature, hydrology, agri- et al., 2004b), was used to support GIS data input into the
cultural management, sediment erosion, nutrient cycling (N Huron and Raisin models. GIS data used in the model in-
and P), crop growth, and pesticides. cluded topography, stream channel network, LULC, soil
N and P dynamics are modeled explicitly by SWAT in type, climatic data, point-source discharges, agricultural
both the terrestrial and water routing phases. In the ter- practices, and impoundment characteristics. Sources and
restrial phase, nutrients can be taken up by plants based scale of data are as follows:
on the supply of the nutrients and demand by the plants,
or the nutrients can be transported via the mass flow of (a) Digital elevation model (DEM) – from USGS National
water (Neitsch et al., 2002). This transport of N and P Elevation Dataset; grid cell: 30 m · 30 m.
can eventually lead to the N and P entering a stream (b) Stream Network – from USGS National Hydrography
channel through surface runoff or lateral sub-surface flow. Dataset; 1:100,000-scale data.
Once nutrients enter the stream channel, SWAT uses nutri- (c) LULC – from NOAA Coastal Change Analysis Program
ent routing relationships adapted from QUAL2E (Brown and data; grid cell: 30 m · 30 m.
134 N.S. Bosch

Figure 1 The Huron and Raisin watersheds in southeastern Michigan. The location of reservoirs incorporated into models are
shown. Locations of USGS gage stations used for flow calibration and sampling sites used for water quality calibration are shown as
well.

(d) Soil types – from US EPA State Soil Geographic Data- model and 5400 ha for the Raisin model. This threshold area
base; grid cell: 90 m · 90 m. sets the minimum subwatershed area because it defines the
(e) Climatic data – from NOAA National Climatic Data minimum source area for the formation of a stream chan-
Center and Michigan State University’s Michigan Cli- nel. Choosing these areas resulted in 31 subwatersheds in
matological Resources Program; daily precipitation, the Huron and 35 subwatersheds in the Raisin which was
temperature, windspeed, humidity, and solar radia- the desired level of spatial detail for this project. There
tion for 36 local weather stations. were 392 and 242 HRUs distributed in the Huron and Raisin
(f) Agricultural practices data – from USDA National Agri- watershed models, respectively.
cultural Statistics Service (NASS) and various local Weather and point-source discharge data were added to
experts; generalized land management schedules both models. Actual measured data were used for all weath-
(operations include fertilizer application, tillage, crop er parameters including daily rainfall, air temperature, so-
planting/harvesting, and manure application) applied lar radiation, wind speed, and relative humidity. The
to different areas of watersheds. period of records used in the model was January 1, 1995
(g) Point-source dischargers – from EPA Permit Compli- to December 31, 2005. When data records for certain sta-
ance System (PCS) database and local dischargers; tions were incomplete, data from the nearest station were
average daily flow, total suspended solids, total used to fill in missing data. Average daily loadings were en-
phosphorus, ammonia, chemical/biological oxygen tered into the models for point-source dischargers accord-
demand, and dissolved oxygen data were collected. ing to collected loading records from the EPA Permit
(h) Impoundment characteristics – from Great Lakes Compliance System database (PCS, 2006). Discharge loading
Flow Project (Allan and Hinz, 2004), the Michigan information was input for 26 Huron dischargers and 18 Raisin
Department of Natural Resources, and local officials; dischargers. To balance water discharge from point-source
surface area and volume measurements for typical dischargers, water withdrawals were added to the effected
and flood stage conditions were collected. subwatersheds using the same average daily water volumes.
Impoundments located on mainstream and tributary
channels were defined as SWAT reservoirs if their surface
Model initial setup and parameterization area exceeded 50 ha and were near subwatershed outlets.
All others that were at least 10 ha and located on a stream
Models for the Huron and Raisin watersheds were first setup channel were included in the models as ponds. This designa-
by delineating the subwatersheds and distributing the HRUs. tion also realistically differentiated the roles of these two
For the stream definition in the delineation process, the types of impoundments – ponds intercept a proportion of
threshold area was set to 4700 ha for the Huron watershed a particular subwatershed’s water and nutrients whereas
The influence of impoundments on riverine nutrient transport 135

reservoirs intercept all of the water and nutrients from the the location of that subwatershed and predominant agricul-
subwatershed as well as upstream subwatersheds. Parame- tural practices in that area. Some adjustment of the sche-
ter values for surface area and volume under regular and dule assignments was then needed in order to match the
flood conditions for each pond and reservoir were input local USDA NASS statistics. In addition, each of these gen-
according to collected data as referenced above. When mul- eral schedules had distinct versions based on timing of the
tiple ponds or reservoirs were located in a single subwater- different crops in the rotation pattern to ensure that all
shed, the surface area and volume were summed since agricultural land in a particular year was not in the same
SWAT only accepts one pond and one reservoir per subwa- crop type, but rather that different HRUs were in different
tershed. The impoundment outflows were estimated based years of the rotation schedule. Without this staggering of
on the targeted release option which allows extra filling of management schedules, all of the cropland would have
impoundments to the emergency spillway depth during started as corn in 1995. These versions were distributed ran-
flood events, followed by a return to volume at the principal domly among HRUs and then adjusted to obtain correct per-
spillway depth within 5 days. Lake level control operations centages for each crop type in each year.
or outflow management were not known for all impound- Several other land-use types had land management prac-
ments, so this outflow simulation method was chosen to tices that differed from SWAT defaults. Range grasslands in
best approximate hydrology in the majority of impound- the Huron watershed are mostly managed for hay produc-
ments in these two watersheds. tion, so a management schedule was created for this LULC
Impoundment water quality parameters phosphorus (P) class that included planting, three partial harvests, and
and nitrogen (N) settling were changed from default values two light fertilizer applications. Range grasslands in the Rai-
to more closely approximate impoundment characteristics. sin typically are in a conservation reserve program with no
Settling rates for the Ford-Belleville Lake complex (located management, so the beginning growing season operation
immediately upstream of French Landing site in Huron wa- was scheduled for January 1 and an end of growing season
tershed; see Fig. 1) and Kent Lake (located about 10 km operation was scheduled for December 31 each year. To
downstream from Millford gage in Huron watershed) were correspond with leaf out in the spring and leaf senescence
calculated using the formulas outlined by Loehr et al. in the fall, operations for forest and wetland land covers
(1980), which are based on total impoundment volume, were scheduled with the growing season beginning on May
impoundment surface area, inlet and outlet nutrient con- 1 and ending on October 10 of each year.
centrations, along with inlet and outlet nutrient loads.
These calculated rates were 6.5 and 9.5 m/yr for Ford- Model calibration and validation
Belleville P and N settling velocities, respectively, and
5.5 m/yr for both P and N settling in Kent Lake. Estimation The Huron and Raisin SWAT models were calibrated and val-
of these rates was possible for these reservoirs because of idated for hydrology and water quality using available
associated field sampling efforts (Bosch, 2007). For 2004 stream discharge measurements and water quality sampling
and 2005 at the sampling sites shown in Fig. 1, average TP data. Daily mean stream discharge measurements were
concentrations were 39 and 117 lg P/L for the Huron and available for multiple gage sites from the USGS in both
Raisin respectively, while TN concentrations averaged 1.44 watersheds over the time 1995–2005. This included three
and 3.90 mg N/L respectively. A survey of 27 Midwestern gage stations for the Raisin (Manchester, Adrian, and Mon-
reservoirs (Walker and Kuhner, 1978) found a median P set- roe) and four gage stations for the Huron (Milford, Hamburg,
tling rate of 12.7 m/yr, but no studies were found which ci- Mill Creek, and Ann Arbor). In the Raisin watershed, near-
ted average N settling rates for the region. Since there was daily water chemistry data were available near Monroe
no data detailing P removal for ponds and reservoirs other (Fig. 1) provided by the National Center for Water Quality
than Ford-Belleville and Kent, all other impoundments were Research at Heidelberg College for the entire time period
set to P settling rates of 12.7 m/yr. The N settling rate cal- of 1995–2005. Monitoring data used in model development
culated for Kent Lake (5.5 m/yr) was used for all ponds and included suspended sediment (SS), total phosphorus (TP),
reservoirs other than Ford-Belleville Lake. soluble reactive phosphorus (SRP), nitrate plus nitrite
Current agricultural crop management practices were reported here as nitrate (NO3), ammonia (NH4), and total
generalized into management schedules based on 2 and 5- Kjeldahl nitrogen (TKN). Other phosphorus (OthP) was cal-
year crop rotations for the Huron watershed and on a 5-year culated as the difference between TP and SRP. Total nitro-
crop rotation in the Raisin watershed. The management gen (TN) was calculated as the sum of NO3 and TKN. Organic
schedules contained all common operations including fertil- nitrogen (OrgN) was calculated as the difference between
izer application, tillage events, planting, and harvesting. TKN and NH4. Using SS, TP, SRP, OthP, TN, OrgN, NO3,
Because of more intensive livestock operations in the south- and NH4 concentration data and stream discharge measure-
western portion of the Huron watershed and in the western ments, annual and monthly loads for these water quality
portion of the Raisin watershed, additional management parameters were estimated using the USGS estimator proto-
schedules were developed to capture the additional man- col (Richards, 1998). Fewer water quality data were avail-
agement operations of manure application and corn silage able for the Huron watershed. Approximately, biweekly
production for these areas. The resulting schedules (6 for TP, SRP, total particulate nitrogen (TPN), total dissolved
the Huron and 3 for the Raisin) thus took into consideration nitrogen (TDN), NO3, and NH4 concentration data were col-
variations between corn for grain and corn for silage crops lected during 2003–2005 at the French Landing site (Fig. 1)
as well as timing of manure applications that vary spatially near the mouth of the river (Bosch, 2007). No suspended
across the watershed. Each subwatershed was randomly as- sediments data were available for the Huron watershed.
signed one or more of the management schedules based on OthP was calculated once again as the difference between
136 N.S. Bosch

TP and SRP. TN was calculated as the sum of TPN and TDN. most sensitive 15 parameters for each model were then se-
OrgN was defined as OrgN = TN NO3 NH4. The limited lected to be used in the autocalibration process. Reasonable
frequency of the data only allowed annual load estimates parameter ranges were prescribed, and the autocalibration
using the AutoBeale protocol (Richards, 1998) for 2003– tool applied a shuffled complex evolution optimization
2005 for the Huron River. This estimation procedure first scheme to select the optimal parameter value set for the
computes an average daily load based on dates with concen- 15 hydrologic parameters after several thousand model
tration data and then adjusts this average daily load accord- runs. The calibration was based on measured daily mean
ing to the proportion of the annual flow included within the stream discharge data (1998–2001) at the most downstream
subset of dates with corresponding concentration data. This gage station in each watershed. Model parameter values
adjusted average daily load is then multiplied by 365 to ob- were then adjusted to reflect the optimal value set dis-
tain an annual load estimate. cerned by the autocalibration process.
Measured data were then compared to simulated SWAT Sediment model parameters were then manually cali-
outputs. The first 3 years (1995–1997) were used for model brated (Santhi et al., 2001) for the Raisin watershed at
spin-up, during which the model came to rely less on initial the Monroe sampling site for 1998–2001 using measured
condition prescribed by initial parameter values, and this monthly total SS loads. Since no measured sediment data
model output was discarded. For stream discharge in both were available for the Huron watershed, optimal parameter
rivers and Raisin water quality, the next four years were values from calibration of the Raisin model were used in the
used for calibration (1998–2001), and the remaining four Huron model by assuming that these values more closely
years (2002–2005) for model validation. For Huron wa- approximated best Huron values than did model default sed-
tershed water quality, 2003 and 2004 annual loads were iment parameter values. Individual nutrient species loads
used for calibration, and only 2005 annual loads were used (TP, SRP, OthP, TN, OrgN, NO3, NH4) were then similarly
for validation. Four statistical measures, root mean square manually calibrated for the Raisin (1998–2001) and the Hur-
error (RMSE), average relative bias, r-squared (R2), Nash– on (2003–2004) based on estimated loads calculated from
Sutcliffe Efficiency (ENS) (Nash and Sutcliffe, 1970), and water discharge and concentration measurements.
comparison of measured and simulated means, were used
in addition to visual graphical comparison to evaluate model
performance during calibration and validation. Average rel- Reservoir alteration scenarios
ative bias was calculated as the average of all of the abso-
lute differences between observed and simulated data, Using the calibrated models, a series of scenarios were run
divided by the observed mean, for each data type. with each watershed model to test how reservoirs may influ-
Nearly all SWAT applications in the literature utilize R2 ence the magnitude and variability of annual river nutrient
and ENS to evaluate model performance. R2 is a commonly export loads. The first set of scenarios tested the impact
used statistical measure that indicates the percentage of of the presence or absence of reservoirs in the watersheds.
variance in measured data accounted for by the variance Scenarios were run with no ponds or reservoirs, with ponds
in the simulated data. ENS is more common in the modeling but no reservoirs, with both ponds and reservoirs, and with
literature and indicates how closely the plot of measured the reservoirs in each watershed swapped between water-
versus simulated data points resembles the 1:1 line. Values sheds. In the Swapped reservoirs scenario, only the reser-
that are less than or near zero for R2 and ENS indicate model voirs from the other watershed were included in the
performance is poor and thus unacceptable, and values of specified watershed (i.e. Huron watershed model was run
unity would indicate perfect model prediction (Santhi with only the Raisin reservoirs present). The next set of sce-
et al., 2001). It is generally understood in SWAT applications narios tested how reservoir position in the watershed af-
that R2 and ENS values greater than 0.5 indicate the model fected nutrient export. In these scenarios all reservoirs
prediction is satisfactory or acceptable (Moriasi et al., were first removed from the watersheds. Then five reser-
2007). voirs were added back into the watersheds and positioned
Model hydrology was first roughly calibrated manually randomly, all upstream, all downstream, or near high nutri-
(Santhi et al., 2001) starting with the most upstream gage ent source areas. The five added reservoirs were all identi-
stations (Fig. 1). A baseflow filter program (Arnold and Al- cal and based on Ford Lake characteristics in the Huron
len, 1999; Arnold et al., 1995) was used to determine rela- watershed with a surface area of 3.9 km2 and an average
tive proportions of annual flow at each gage station that was depth of 8.1 m. The third set of scenarios was meant to test
due to surface runoff versus baseflow contributions. This the impact of reservoir size on nutrient export. All reser-
information was then used to adjust SWAT parameters to voirs were first taken out of the Huron and Raisin water-
correctly reflect observed water balances at multiple gage sheds. Then the model was rerun for each watershed with
station locations in the watershed. Additional gage stations one large reservoir (all reservoir dimensions five times lar-
were then calibrated in downstream order until the most ger than Ford Lake), 5 mid-sized reservoirs (same size as
downstream gage location was finally calibrated for hydrol- Ford Lake), or 25 small reservoirs (all reservoir dimensions
ogy. After this initial stage of hydrologic calibration, an 1/5 the size of Ford Lake) all located near the mouth of
automated calibration tool (van Griensven and Bauwens, the watershed. Because SWAT limits the user to only one
2001, 2003; van Griensven et al., 2002) interfaced with reservoir per subwatershed, the one large reservoir scenario
SWAT was utilized for more detailed calibration at the daily simulated an impoundment at the outlet of the entire
temporal resolution. This tool first provided a sensitivity watershed, but the 5 mid-sized and 25 small reservoir sce-
analysis that identified the most sensitive hydrologic model narios required placement in additional upstream subwater-
parameters in rank order for each watershed model. The sheds as close as possible to the river mouth.
The influence of impoundments on riverine nutrient transport 137

Results Comparison of measured versus simulated streamflow af-


firms the overall predictive power of SWAT for the hydrol-
Model performance ogy of the Huron and Raisin rivers and indicates where
model predictions deviate from observations. For the Rai-
Overall, SWAT predicted measured stream discharge rea- sin, the monthly validation plot (Fig. 2a) shows consistent
sonably well for both watersheds at the monthly and daily overprediction of stream discharge during summer months
temporal resolutions for the validation years 2002–2005 when flows are low. The monthly validation plot for the Hur-
(Table 1). These evaluative statistics were consistent for on (Fig. 2b) shows similar overprediction during summer
both the calibration and validation time periods. The model months. Both watershed models simulate a slower response
more accurately predicted river flow at the monthly tempo- of discharge to baseflow conditions than is observed.
ral resolution. On average, the Raisin model underpredicts Though not shown here, this trend is seen in daily plots as
stream discharge at daily and monthly resolutions, and the well.
Huron model overpredicts monthly and underpredicts daily Modeled water quality was generally a close fit to actual
resolution. The Raisin model easily exceeds the guidelines values for the Raisin and was less satisfactory for the Huron
of R2 > 0.6 and ENS > 0.5 for monthly temporal resolution watershed. Owing to the high frequency and long duration
(Moriasi et al., 2007). Performance of the Huron model is of water quality sampling, the Raisin model was able to be
satisfactory based on its ENS and marginal for R2. finely calibrated using monthly loads, resulting in accurate
prediction for SS and P species (Table 2) during both calibra-
tion and validation time periods. SWAT predicted TP
monthly loads in the Raisin very well during minimal loading
months (summer and fall) and predicted increases in TP
Table 1 Calibration and validation results for monthly and loads during winter and spring months of high loading,
daily stream discharge for the Raisin River at Monroe and though not always of the correct magnitude (Fig. 2c). OrgN
Huron River at Ann Arbor and NH4 prediction was satisfactory (Table 2), but TN and
Calibration Validation NO3 prediction was poor according to statistical measures
(1998–2001) (2002–2005) and as seen in time series plots. For NO3, simulated loads
consistently lagged 3 months behind measured loads
Raisin River (Fig. 2d). This was also apparent in TN monthly loads, which
Monthly was to be expected because NO3 made up about 80% of TN
Observed mean (m3/s) 23.5 19.3 on average (Fig. 2e).
Simulated mean (m3/s) 22.6 21.5 Model evaluation for the Huron was more difficult
RMSE (m3/s) 9.3 7.5 because measurement data were available only for 2003–
Average relative bias 0.29 0.33 2005 and only at a biweekly sampling frequency. As a conse-
ENS 0.79 0.83 quence, only annual loads for N and P species over these 3
R2 0.82 0.85 years were examined. Model prediction for all N species
Daily during the calibration period (2003–2004) was satisfactory
Observed mean (m3/s) 23.3 19.2 (Fig. 3), with model results underpredicting by only 12%
Simulated mean (m3/s) 22.5 21.4 on average. During the validation period (2005), however,
RMSE (m3/s) 16.6 14.2 the Huron model grossly overpredicted OrgN and NH4, by
Average relative bias 0.41 0.50 89% and 129% respectively. Annual loads for P species be-
ENS 0.69 0.69 haved similarly, showing close correspondence to actual
R2 0.73 0.73 data during the calibration period (Fig. 4), with an average
underprediction of only 8%. However, simulated loads dur-
Huron River ing the validation period were once again too high. It was
Monthly decided to continue with the Huron model to test various
Observed mean (m3/s) 14.1 11.4 reservoir scenarios despite poor predictions of OrgN, NH4,
Simulated mean (m3/s) 15.4 12.5 and P species with the understanding that scenario results
RMSE (m3/s) 5.8 5.0 needed to be interpreted by focusing comparisons on
Average relative bias 0.32 0.35 proportional rather than absolute changes in river export
ENS 0.57 0.59 loads.
R2 0.58 0.60

Daily
Observed mean (m3/s) 14.1 11.4 Reservoir scenarios
Simulated mean (m3/s) 13.9 11.1
RMSE (m3/s) 6.7 6.1 Changes in the presence and placement of reservoirs in each
Average relative bias 0.37 0.40 watershed model produced large impacts on magnitudes
ENS 0.61 0.55 and interannual variability of average annual nutrient loads.
R2 0.61 0.55 Presence/absence scenarios clearly supported the hypothe-
sis that the many reservoirs in the Huron watershed greatly
Root mean square error (RMSE), Nash–Sutcliffe simulation effi-
reduce N and P river exports (Fig. 5). Results also showed
ciency (ENS), average relative bias, and coefficient of determi-
that the 33 reservoirs in the Huron watershed caused
nation (R2) are used as evaluators of model performance.
greater reductions in nutrient export than did the 55 ponds.
138 N.S. Bosch

(a) 70
Observed
60

Discharge (m3/s)
Simulated
50
40
30
20
10
0

(b) 35
30
Discharge (m3/s)

25
20
15
10
5
0

(c) 70
60
TP load (Mg P)

50
40
30
20
10
0

(d) 2100
Nitrate load (kg N)

1800
1500
1200
900
600
300
0

(e) 1800
1500
TN load (Mg N)

1200

900
600
300
0
Fe

Fe

M
Ja

Ap

Ju

No

Se

De

Ap

Ju

Se

De
ay

ay
ct
b-

b-
n-

l-0

l-0
r- 0

p-

r- 0

p-
v-

c-

c-
-0
-0

-0
02

03

05
02

03

05
2

03

05
2

4
3

Figure 2 Monthly plots of observed and simulated hydrologic and nutrient data for the validation time period (2002–2005).
Monthly average stream discharge (m3/s) for the Raisin River at Monroe (a) and the Huron River at Ann Arbor (b). Monthly total
phosphorus load (Mg P) (c), nitrate plus nitrite load (Mg N) (d), and total nitrogen load (Mg N) (e) for the Raisin River at Monroe.

Average annual river exports of TP and TN over the time ervoirs from the Raisin watershed, there was a 4% decrease
period 1998–2005 were lower by 41% and 54%, respectively, in TP export and a 13% decrease in TN when compared to
in the presence of the 88 impoundments compared with the scenario with no ponds or reservoirs.
their simulated removal. Removal of dissolved nutrients in The presence or absence of 14 impoundments in the Rai-
the forms of SRP and NO3 accounted for a majority of the sin watershed resulted in less pronounced proportional
TP and TN removal when impoundments were present. changes in annual nutrient loads. The annual absolute con-
When reservoirs were swapped between watersheds such sumption of TP from the water by impoundments in the Rai-
that the Huron watershed was simulated with only the res- sin (kg P/yr) was greater than in the Huron, but due to
The influence of impoundments on riverine nutrient transport 139

Table 2 Calibration and validation results for monthly Table 2 (continued)


sediment and nutrient loads for the Raisin River at Monroe Calibration Validation
Calibration Validation (1998–2001) (2002–2005)
(1998–2001) (2002–2005) Ammonia
Suspended sediment Observed mean (Mg N) 7 4
Observed mean (Mg) 10639 7165 Simulated mean (Mg N) 6 6
Simulated mean (Mg) 10568 9698 RMSE (Mg N) 4 3
RMSE (Mg) 7143 7614 Average relative bias 0.20 0.24
Average relative bias 0.46 0.73 ENS 0.51 0.43
ENS 0.77 0.55 R2 0.54 0.59
R2 0.78 0.62 Root mean square error (RMSE), Nash–Sutcliffe simulation effi-
ciency (ENS), average relative bias, and coefficient of determi-
Total phosphorous nation (R2) are used as evaluators of model performance.
Observed mean (Mg P) 14.2 11.1
Simulated mean (Mg P) 13.9 12.8
RMSE (Mg P) 8.7 9.2
Average relative bias 0.39 0.50
ENS 0.77 0.64 higher loads in the Raisin, the reductions were proportion-
R2 0.76 0.65 ally much less. Total absolute removal of TN (kg N/yr) due
to impoundments was similar for both the Raisin and Huron,
Soluble reactive phosphorous
but again the proportional consumption was less in the Rai-
Observed mean (Mg P) 2.7 2.8
sin. Both TP and TN loads were reduced by only 13% when all
Simulated mean (Mg P) 3.0 2.7
the ponds and reservoirs were included in the Raisin sce-
RMSE (Mg P) 1.8 2.1
nario. Unlike the Huron, Raisin ponds reduced nutrient loads
Average relative bias 0.09 0.12
more than Raisin reservoirs, especially with NO3. When the
ENS 0.74 0.71
Raisin watershed was simulated with the reservoirs of the
R2 0.74 0.79
Huron watershed, TP and TN river export decreased by
45% and 47%, respectively, in comparison to the simulation
with no ponds or reservoirs.
Other phosphorous
The location and size of reservoirs also affected annual
Observed mean (Mg P) 12.3 8.6
nutrient export. As expected, reservoirs located near the
Simulated mean (Mg P) 10.8 10.2
mouth of the river usually resulted in lower nutrient exports
RMSE (Mg P) 8.2 7.6
compared to reservoirs placed in the upper watershed
Average relative bias 0.36 0.40
(Fig. 6). However, P export from the Raisin watershed did
ENS 0.74 0.59
not follow this pattern, showing instead that reservoirs in
R2 0.74 0.63
the upper watershed lowered average annual TP export
more than did reservoirs in the lower watershed. Reservoirs
Total nitrogen concentrated in nutrient source areas more effectively re-
Observed mean (Mg N) 405 403 duced loads than did reservoirs scattered throughout the
Simulated mean (Mg N) 419 428 watershed. Scenarios that tested size of reservoirs showed
RMSE (Mg N) 429 421 that 25 small reservoirs were typically more effective at
Average relative bias 23 31 reducing N and P loads than 5 mid-sized or 1 large reservoir
ENS 0.10 0.27 (Fig. 7).
R2 0.12 0.26 Not only did the presence or absence and placement of
reservoirs change the magnitude of annual nutrient ex-
Organic nitrogen ports, but interannual variability changed as well. Using
Observed mean (Mg N) 77 65 the coefficient of variation (CV) of annual export loads
Simulated mean (Mg N) 75 71 to compare interannual variability, presence/absence sce-
RMSE (Mg N) 45 44 narios produced the most striking changes (Fig. 8). Annual
Average relative bias 2.0 2.2 variability was generally lowest for all nutrient species
ENS 0.70 0.68 when no ponds or reservoirs were present, and CVs were
R2 0.78 0.75 highest when reservoirs were added. However, NH4 showed
the opposite trend. As with magnitude changes in the Hur-
Nitrate on, reservoirs contributed much more to increased vari-
Observed mean (Mg N) 330 385 ability than did ponds. The CV increased by 12% for TP
Simulated mean (Mg N) 338 351 and 5% for TN when reservoirs were added. The presence
RMSE (Mg N) 436 499 of reservoirs in the Huron watershed increased year-to-
Average relative bias 24 35 year variation in annual river export of nutrients. This
ENS 0.42 0.08 trend was also seen to a lesser degree in the Raisin wa-
R2 0.00 0.02 tershed, but was less obvious because of fewer
impoundments.
140 N.S. Bosch

1000
Observed
800 Simulated

Annual load (Mg N)


600

400

200

0
Total N

Total N

Total N
Organic N

Organic N

Organic N
Ammonia

Ammonia

Ammonia
Nitrate/

Nitrate/

Nitrate/
Nitrite

Nitrite

Nitrite
2003 - Calibration 2004- Calibration 2005 - Validation

Figure 3 Annual nitrogen species loads (Mg N) for the Huron River at French Landing (watershed outlet sampling site) for the
calibration (2003–2004) and validation time period (2005).

Discussion of recreational users and homeowners along shorelines, out-


flows often are managed seasonally to alter lake levels, and
Model performance also may be temporarily lowered in preparation for large
precipitation events for flood control purposes. There are
Model calibration and validation results showed SWAT to be two active hydroelectric dams in the Huron River that may
a promising tool to better understand nutrient dynamics of allow minor lake level fluctuation in order to optimize elec-
the Huron and Raisin watersheds, including the role of tric power generation. Impoundment hydrology, whether
impoundments. Modeled stream discharge was adequately natural lake or man-made impoundment, is thus more com-
accurate at both monthly and daily scales for both water- plex than SWAT can simulate. The model was set up to allow
sheds according to a review of other SWAT applications increased storage volume during high flow times and then a
(Moriasi et al., 2007). However, simulated hydrology was return to its original, or target, storage volume within 5
more accurate for the Raisin than the Huron. Difference be- days. This was considered a reasonable compromise be-
tween the two watersheds with respect to number of tween widely different dam operations in the two water-
impoundments may explain differences in model perfor- sheds, but it likely reduced the predictive power of the
mance. Many actively managed dams in Michigan are li- Huron model.
censed to be managed as ‘‘instantaneous run-of-river Model overprediction of stream flow during summer
mode’’ as recommended by the Michigan Department of months as well as the model’s slow return to baseflow after
Natural Resources (C. Freiburger, personal communication), peak flow was apparent for the 1998–2005 calibration and
which means that instantaneous outflows from impound- validation period in both models, although more pro-
ments equal instantaneous inflows and water level fluctua- nounced in the Huron. After peak streamflow months, the
tion is minimized. However, it is not clear that this observed hydrograph returns to baseflow conditions more
requirement is closely adhered to. Owing to the concerns rapidly than does the simulated hydrograph, which falls

25
Observed
Simulated
20
Annual load (Mg P)

15

10

0
Total P SRP Other P Total P SRP Other P Total P SRP Other P

2003- Calibration 2004- Calibration 2005- Validation

Figure 4 Annual phosphorus species loads (Mg P) for the Huron River at French Landing (watershed outlet sampling site) for the
calibration (2003–2004) and validation time period (2005).
The influence of impoundments on riverine nutrient transport 141

180 6000
160

Average annual load (Mg P)

Average annual load (Mg N)


5000

Raisin watershed
140
120 4000
100
3000
80
60 2000
40
1000
20
0 0
Total P SRP Other P Total N Organic N Nitrate Ammonia
1600
Without ponds and reservoirs
With ponds, without reservoirs

Average annual load (Mg N)


1400
25
Average annual load (Mg P)

With ponds and reservoirs


Swapped reservoirs
Huron watershed

1200
20
1000

15 800

600
10
400
5
200

0 0
Total P SRP Other P Total N Organic N Nitrate Ammonia

Figure 5 Simulated average annual watershed export loads of P and N for years 1998–2005 based on four impoundment presence/
absence scenarios. In the Swapped reservoirs scenario, only the reservoirs from the other watershed were included in the specified
watershed (i.e. Huron watershed model was run with only the Raisin reservoirs). Top panels, Raisin watershed; bottom panels,
Huron watershed.

180 6000
160
Average annual load (Mg P)
Raisin watershed

Average annual load (Mg N)

5000
140
120 4000
100
3000
80
60 2000
40
1000
20
0 0
Total P SRP Other P Total N Organic N Nitrate Ammonia

30 1600
Without ponds and reservoirs
1400 Upper watershed
Average annual load (Mg P)

Average annual load (Mg N)

25 Near nutrient sources


Huron watershed

1200 Throughout watershed


Lower watershed
20
1000

15 800

600
10
400
5
200

0 0
Total P SRP Other P Total N Organic N Nitrate Ammonia

Figure 6 Simulated average annual watershed exported loads of P and N for years 1998–2005 based on four reservoir location
scenarios. The ‘‘Without ponds and reservoirs’’ scenario results are shown again for reference. Top panels, Raisin watershed;
bottom panels, Huron watershed.
142 N.S. Bosch

180 6000
160

Average annual load (Mg P)

Average annual load (Mg N)


Raisin watershed
5000
140
120 4000
100
3000
80
60 2000
40
1000
20
0 0
Total P SRP Other P Total N Organic N Nitrate Ammonia

30 1600
Without ponds and reservoirs
Average annual load (Mg P)

1 Large reservoir

Average annual load (Mg N)


1400
25 5 Mid-sized reservoirs
Huron watershed

1200 25 Small reservoirs


20
1000

15 800

600
10
400
5
200

0 0
Total P SRP Other P Total N Organic N Nitrate Ammonia

Figure 7 Simulated average annual watershed exported loads of P and N for years 1998–2005 based on three reservoir size
scenarios. The ‘‘Without ponds and reservoirs’’ scenario results are shown again for reference. Top panels, Raisin watershed;
bottom panels, Huron watershed.

50
Without ponds and reservoirs
Annual load Coefficent of Variation (%)

With ponds, without reservoirs


40 With ponds and reservoirs

30

20

10

0
Total P SRP Other P Total N Organic N Nitrate Ammonia

Figure 8 Coefficient of variation (%) for average annual load estimates from three presence/absence scenarios with Huron
watershed model.

more gradually and often does not return to the observed not adequately simulated by SWAT. It is also likely that inac-
baseflow before the next winter’s high flows begin. Several curate impoundment hydrology contributed to these re-
factors may contribute to the slower response time of the sults. Some impoundments actually may have had little or
model. Runoff from urban areas can be extremely fast (Paul no extra storage capacity to dampen hydrograph peaks as
and Meyer, 2001; Walsh et al., 2005), and although the modeled. The target of 5 days for return to conditions of
model does account for quicker surface runoff in urban or equal impoundment inflow and outflow was likely too long
other impervious LULCs, it does not account for stormwater as well.
conveyances that increase the speed of transport of water Other studies using SWAT have found similar results in
from land to stream channels and impoundments. In addi- stream discharge calibration and validation efforts. Abu
tion, similarly rapid drainage of agricultural lands after pre- El-Nasr et al. (2005) applied SWAT to the agricultural Jeker
cipitation events due to field tiles and drainage ditches is catchment located in eastern Belgium. They reported an ENS
The influence of impoundments on riverine nutrient transport 143

of 0.55 for daily stream discharges during the model valida- in Texas resulted in monthly ENS statistics of 0.43, 0.39,
tion period, the same value found for the Huron. Time series 0.64, and 0.81 for OrgN, OthP, mineral N (sum of NH4 and
plots of simulated and observed daily discharge also were NO3), and SRP (Santhi et al., 2001). For a small agricultural
similar, and showed the same tendency of overestimation watershed in Maryland, Chu et al. (2004) reported validation
during baseflow period and underestimation during peak ENS statistics of 0.36, 0.05, 0.08, and 0.64 for monthly
flow events. The simulated hydrograph in the Jeker also NO3, NH4, TP, and SRP, respectively. Time series plots of
exhibited a slower than observed return to baseline monthly simulated and observed N and P loads in the pres-
following a peak flow event. The authors postulated that ent study show that occasional poor statistical measures
these performance discrepancies were in part because the are not the result of a time lag phenomenon as seen for
model did not account for two large urban water conduits NO3 in the Raisin, but rather are due to underprediction or
used to extract drinking water. It is likely that similar weak- overprediction of loads in high loading months. Unlike the
nesses in hydrologic simulations of the Huron and Raisin are Raisin, these other study basins do not have extensive tile
also due to agricultural and urban conveyances not ac- drainage, so the proposed explanation for this time lag in
counted for in the model. Qi and Grunwald (2005) applied NO3 export in the Raisin remains plausible.
SWAT for the agricultural Sandusky watershed in Ohio. They Simulated water quality parameters for the Huron wa-
calibrated and validated for stream discharge at a monthly tershed deviated more strongly from measured values.
time scale and reported an ENS value of 0.58 for validation SWAT performance for the Huron is difficult to evaluate
at the most downstream USGS gage site. Monthly validation due to the lack of high resolution, long-term data. No mea-
results from other SWAT studies include ENS values of 0.62 surement data were available for SS, but it is a reasonable
for a large primarily forested and rangeland catchment in assumption that parameter values related to upland and
Texas (Santhi et al., 2001), 0.52 for a small agricultural wa- stream channel sediment transport used in the Raisin model
tershed in Maryland (Chu et al., 2004), 0.75 for a small should be applicable to the Huron model. If, however, this
mostly agricultural catchment in Pennsylvania (Veith assumption is invalid and SS simulation is poor, that would
et al., 2005), and 0.76 for a mid-sized forested catchment result in poor prediction for OthP and OrgN because of their
in upstate New York (Benaman et al., 2005). These statisti- association with sediment particles in the model. No sys-
cal measures associated with model validation reported in tematic trend of underprediction or overprediction was
other SWAT applications are similar to those of the Huron found for OthP or OrgN, so there is nothing to suggest this
and Raisin. assumption was faulty. In general, SWAT slightly overpre-
Water quality simulation with SWAT was less accurate dicted N and P loads for the 2003–2004 calibration period
than hydrology for these two watersheds, but was still sat- in the Huron and more severely underpredicted for the
isfactory overall. This is as expected since inaccuracy in 2005 validation period (Figs. 3 and 4). Total precipitation
the simulation of hydrology by the model can be magnified for 2005 (83 cm) was slightly lower than average (89 cm),
in the simulation of water quality parameters which are but simulated surface runoff was 50% higher in 2005 than
dependant on hydrologic processes. Calibration and valida- it was in 2003 or 2004. This increase in surface runoff seems
tion for the Raisin watershed resulted in monthly estimates to match the observed hydrograph (Fig. 2b), but the magni-
of SS, TP, SRP, OthP, OrgN, and NH4 that were reasonable tude of the increase would be expected to result in much
representations of measured values. However, the timing higher nutrient loads as predicted by SWAT. The observed
of NO3 (and thus TN) delivery was predicted poorly by SWAT. nutrient data, surprisingly, do not show such an increase
This poor agreement between observed and simulated val- in nutrient load for 2005 (Figs. 3 and 4), so some other
ues over the calibration and validation time period was unknown factors presumably are not adequately taken into
due primarily to a 3 month time lag in simulated NO3 loads. account by SWAT, leading to this unexpected result. Huron
This lag may be due to tile drainage in agricultural areas and watershed water quality simulation would be much im-
other runoff conveyances, which SWAT was unable to proved with more data for calibration and validation,
represent. It seems likely that NO3 rapidly entered into soil similar to what was available for the Raisin watershed
water and tile drains for quick transport to stream channels model.
and export from the watershed, whereas SWAT allowed
more delayed entry into groundwater and later transfer into Reservoir scenarios
the stream channel. Despite TN and NO3 delivery timing
being poorly simulated, the average annual loads for the As expected, the presence of reservoirs resulted in substan-
calibration and validation time periods were accurate (Ta- tial reductions in annual loads of N and P in the Huron wa-
ble 2). Thus, conclusions based on model simulations at tershed, relative to predicted export in the absence of
the scale of average annual statistics are valid, even for N reservoirs. Though reservoirs reduced proportionally less
species. of the total nutrient export in the Raisin watershed, the
Other studies using SWAT for water quality simulation model finds that the 14 impoundments in the Raisin surpris-
showed a similar range of monthly validation statistical ingly reduced absolute TP loads even more than did the 88
measures as those observed for the Raisin, but none showed impoundments in the Huron, and caused roughly the same
the same time lag in nitrogen loads (Chu et al., 2004; Griz- absolute reduction in TN loads. These observations indicate
zetti et al., 2003; Santhi et al., 2001). Grizzetti et al. (2003) that simulated impoundments in the Huron watershed did
reported ENS statistics of 0.30 and 0.64 for monthly TN and not reduce nutrient loads as efficiently as did those in the
TP, respectively, in their study of the moderately-sized and Raisin. This is likely due to lower N and P concentrations
mostly forested Vantaanjoki basin in southern Finland. Mod- in the Huron which would cause SWAT to predict less N
el application in a large rangeland and forested watershed and P removal in Huron impoundments. It is also interesting
144 N.S. Bosch

that in relation to reservoirs, ponds were more important in higher nutrient concentrations coupled with lower flows
reducing nutrient loads in the Raisin compared to the typical to lower order stream reaches would increase nutri-
Huron. This may be because ponds in the Raisin have higher ent removal by local impoundments. Having many more
nutrient concentrations than the reservoirs, since they re- impoundments stretching downstream to the river mouth
ceive water from high nutrient source areas according to would allow further processing of these nutrients as well
fertilizer and manure application patterns. as nutrients from new sources.
Comparison of the modeled load reduction for individual Interannual variation in nutrient export loads was also
N and P species in the two watersheds allows further infer- affected by reservoirs. The 33 reservoirs in the Huron wa-
ences. In the Raisin watershed, SRP load reduction was tershed reduced nutrient export loads from the watershed,
similar for ponds and reservoirs, but reservoirs in the Huron but they also increased the variability of annual export
removed more than 10 times the amount of SRP as did from year to year. This was also observed for the Raisin wa-
ponds. This may be a locational effect as most major reser- tershed and for ponds, although to a lesser extent. The
voirs in the Huron are near urban areas with associated only exception to this finding was NH4 export loads, which
wastewater treatment plants and thus high SRP loading. were less variable when impoundments were present. It is
Ponds typically are located in more remote areas that re- unclear why NH4 loads did not resemble trends in all other
ceive little SRP loading or are too small to adequately pro- loads, but it is important to note that NH4 contributes less
cess and reduce SRP loads that pass through them. Though than 10% of TN in Huron and Raisin export. These results
total absolute TN reductions are similar for the Huron and agree with an earlier survey study of 141 Missouri reservoirs
Raisin watersheds, the N species contributing to this that found a threefold year-to-year variation in seasonal
decrease are different in each watershed. More OrgN is re- averages of TP (Knowlton and Jones, 2006) for many of
moved in the Huron impoundments, even though OrgN loads the reservoirs.
are smaller in the Huron river. This may be because OrgN There are at least two possible explanations for in-
makes up a much larger percentage of TN in the Huron creased variability of river loads in the presence of
watershed, so the Huron impoundments could be propor- impoundments. First, the effectiveness of impoundments
tionally more effective. Nitrate accounts for more of the in reducing export loads could vary from year to year as
TN in the Raisin than in the Huron, and the loads are much has been shown in other studies (Harris, 2001; David
higher. It thus seems reasonable that the absolute NO3 et al., 2006). In both of these studies, the variation was re-
load reduction would be greater in the Raisin, which is the lated to water residence time such that low flow conditions
case. allowed for more nutrient removal. Given that impound-
Scenarios to test the influence of reservoir location and ments in the Huron cut export loads in half, any change in
size showed that both could affect nutrient export. The their load reducing capacity would have a large impact on
location scenarios mostly confirmed that reservoirs near the annual nutrient load exported. A year with high concen-
the mouth of the river reduced nutrient loads more than trations at low flow conditions would allow more efficient
the same reservoirs located in the upper watershed. This processing by impoundments because of longer water resi-
is expected because a reservoir at the mouth of a river dence times and lead to greater export load reductions
can process the total nutrient load remaining in the river for that year. If, however, high concentrations coincide
after any reductions due to in-stream channel processes with extreme runoff events, nutrients would quickly pass
immediately prior to export from the watershed. An through impoundments with little processing. This phenom-
exception to this generalization was TP in the Raisin wa- enon could be occurring in the SWAT model. These flow and
tershed, where river export was slightly higher when reser- concentration dynamics are supported by different patterns
voirs were at the river mouth. It is unclear why this of variation in concentrations among the different N and P
occurred. Reservoir placement near nutrient source areas fractions. TP, TN, and NO3 concentrations have been found
(higher fertilizer or manure application rates) usually re- to increase substantially with stream discharge in two adja-
sulted in greater export load reductions than random cent catchments (Richards et al., 2001), which coincides
placement throughout the watershed. One plausible expla- with increased variation with impoundments for these three
nation for both higher TP loads in the Raisin watershed fractions. Richards et al. demonstrate little change in SRP
when reservoirs were in the lower watershed as well as concentrations with increased streamflow, and modeling re-
lower TP loads when reservoirs were near source areas sults in less of an increase in variation of SRP with
would be different inflow nutrient concentrations. Stream impoundments.
sections near nutrient source areas are likely to have the A second possibility is that impoundments create a vari-
highest nutrient concentrations in the watershed, and so able delay in nutrient load transport out of the watershed.
reservoirs at these points should allow for the greater load Vorosmarty et al. (1997) reported that reservoirs have
reduction. caused a 700% increase in the standing stock of water glob-
Another interesting scenario outcome was the finding ally, and argued that the mean age of river water likely has
that many smaller reservoirs were most effective at reduc- increased threefold. In the Huron, impoundments increase
ing most P and N parameter export loads. Because SWAT al- the water residence time on average 2400%, while in the
lows only one reservoir per subwatershed, the 25 small Raisin, impoundments increase the time by 160%. Without
reservoirs could not all be placed near the outlet of the wa- impoundments, nutrient loading to stream channels during
tershed, so most of the subwatersheds contained one of runoff events would result in rapid transport of these loads
these impoundments. Near the river mouth, nutrient con- through the stream network and out of the watershed with
centrations would likely decrease due to cumulative dilu- only stream channel processes available for consumption of
tion and stream channel processing. Near source areas, N and P from the water. Typically, the majority of annual
The influence of impoundments on riverine nutrient transport 145

nutrient export occurs within these short time periods careful approximation of impoundment hydrology before
(Bernhardt et al., 2005; Meyer and Likens, 1979). Thus, calibration of stream flow. An important step in this
the timing of nutrient export loads at the river mouth likely improvement would involve reducing the number of days
would closely resemble predictable nutrient loading events for the impoundment to return to equal inflow and outflow
such as the first rain after a January dairy manure applica- rates. Breaking the watershed into more subwatersheds
tion. With many impoundments in the river network, how- would have allowed less lumping of multiple ponds and
ever, a longer and more variable delay develops for reservoirs into single impoundments, and thus potentially
nutrients in transport from source to river mouth. For exam- improving hydrologic simulation as well as avoiding the
ple, the majority of the NO3 load caused by a November assumption that impoundment bathymetry is unimportant.
storm event in the less impounded Raisin watershed would Improvement of the Huron and Raisin SWAT models in
pass out of the river system by the end of the year consis- these three areas would allow for a more confident and de-
tently. The NO3 from the same event in the Huron, however, tailed analysis of impoundments and their impacts on N
might or might not pass through the system before the end and P export in these two watersheds.
of the year depending on the residence times of the
impoundments at that time.
Field-based studies confirm increased interannual vari-
Conclusion
ability associated with reservoirs as well as their substantial
The main objective of this study was to calibrate and vali-
effects on nutrient transport. Ahearn et al. (2005) con-
date the SWAT model for two watersheds to better under-
ducted a paired watershed study on two adjacent forested
stand the influence of impoundments over riverine
watersheds in California that differed only in their degree
nutrient export. Calibration and validation of models for
of impoundment. Owing to the combined effect of the
the Huron and Raisin watersheds resulted in satisfactory sta-
two study reservoirs, annual flow-weighted mean TP con-
tistical measures for monthly and daily stream discharge
centrations decreased from 10.3 to 75.1% for the 3 years
simulations. Further analysis showed that both models over-
of the study. TN results were only reported for one year
predicted stream discharge during summer baseflow condi-
and showed a 3.4% decrease in mean concentration. Unlike
tions because of a slow return of discharge to baseflow
the simulations of the Huron and Raisin, which demon-
conditions following high flow events during winter and
strated consistent NO3 removal by impoundments, NO3
spring. This indicated that actual runoff occurred more
loads were reduced or increased by the two California reser-
completely and quickly than was simulated. The effect
voirs based on precipitation patterns. In a review of N and P
was greatest in the Huron watershed, presumably due to
biogeochemistry in Australian catchments, Harris (2001) re-
its many impoundments and greater urban land area. Simu-
ported that Australian reservoirs show high interannual var-
lated water quality was satisfactory for some but not all
iation and can remove large proportions of TP (56%) and TN
parameters following calibration. Water quality modeling
(26%). In other studies, Harris found that reservoirs gener-
in the Raisin was more thoroughly calibrated due to better
ally acted as sources of NO3 and occasionally as sources
observed data, and statistical measures showed that all
for TP in wet years. David et al. (2006) found that a reser-
parameters were predicted well in the Raisin model except
voir in Illinois removed 58% of the NO3 passing through it
for TN and NO3. Water quality prediction for Huron wa-
over the 23-year period of record. Retention times for this
tershed was judged adequate based on the modest extent
impoundment varied from 0.36 to 0.84 years and were the
of calibration and validation that was possible due to rela-
primary driver for year-to-year changes in retention of
tively few data. This paucity of observed data may also have
NO3. They also demonstrated that this removal could totally
contributed to poor prediction of OthP and OrgN during the
be explained by denitrification.
validation year.
Collection of more and better water quality data, fur-
Impoundments had a marked effect on river nutrient
ther SWAT model development in tile drainage and
export from the watersheds, based on model results in
impoundment nutrient removal, and improved SWAT
which all impoundments were removed. TP and TN export
modeling of hydrology are all warranted based on this re-
loads approximately doubled when all impoundments were
search. Water quality monitoring for the Huron watershed
removed from the Huron watershed. The Raisin watershed
near the river mouth should include suspended sediments,
experienced an even greater absolute increase in TP loads
and samples need to be collected on a more frequent ba-
in the absence of impoundments, and an increase in TN
sis. A longer record of data would also be beneficial. Sam-
similar to the Huron. However, because nutrient loads
pling the inflow and outflow of more impoundments would
were several times larger in the Raisin, the proportional
allow for more accurate incorporation of N and P settling
change was less impressive. Reservoirs placed near river
rates by taking into account differences in impoundment
mouths or in high source areas were most effective in
biology and bathymetry. These improvements in water
reducing export loads, and many smaller reservoirs were
quality data would allow the Huron watershed model to
better than a single large reservoir. In addition, impound-
be more confidently calibrated and validated, as was the
ments increased interannual nutrient load variability.
Raisin. Further model development, especially the ability
to incorporate tile drainage and other water conveyances,
is warranted as well. Currently, SWAT includes settling as
the only removal mechanism for N and P in impoundments. Acknowledgements
The addition of denitrification as another removal mecha-
nism for N would be a welcomed improvement. Application This work was supported by grants and fellowships from the
of the SWAT model likely could be improved by a more University of Michigan School of Natural Resources and
146 N.S. Bosch

Environment and a Rackham Dissertation Award. I am grate- David, M.B., Wall, L.G., Royer, T.V., Tank, J.L., 2006. Denitrifica-
ful to the following people for their insight related to this tion and the nitrogen budget of a reservoir in an agricultural
work: Dave Allan, Jeff Arnold, Tom Croley, Bill Currie, landscape. Ecological Applications 16 (6), 2177–2190.
Tom Johengen, George Kling, and Raghavan Srinivasan. Di Luzio, M., Arnold, J.G., Srinivasan, R., 2004a. Integration of
SSURGO maps and soil parameters within a geographic informa-
tion system and nonpoint source pollution model system.
References Journal of Soil and Water Conservation 59 (4), 123–133.
Di Luzio, M., Srinivasan, R., Arnold, J.G., 2004b. A GIS-coupled
Abu El-Nasr, A., Arnold, J.G., Feyen, J., Berlamont, J., 2005. hydrologic model system for the watershed assessment of
Modelling the hydrology of a catchment using a distributed and a agricultural nonpoint and point source pollution. Transaction
semi-distributed model. Hydrological Processes 19 (3), 573–587. in GIS 8 (1), 113–136.
Ahearn, D.S., Sheibley, R.W., Dahlgren, R.A., 2005. Effects of river Dixit, A.S., Hall, R.I., Leavitt, P.R., Quinlan, R., Smol, J.P., 2000.
regulation on water quality in the lower Mokelumne River, Effects of sequential depositional basins on lake response to
California. River Research and Applications 21 (6), 651–670. urban and agricultural pollution: a palaeoecological analysis of
Alexander, R.B., Smith, R.A., Schwarz, G.E., 2000. Effect of stream the Qu’Appelle Valley, Saskatchewan, Canada. Freshwater
channel size on the delivery of nitrogen to the Gulf of Mexico. Biology 43 (3), 319–337.
Nature 403 (6771), 758–761. Dodds, W.K. et al, 2002. N uptake as a function of concentration in
Alexander, R.B., Elliott, A.H., Shankar, U., McBride, G.B., 2002. streams. Journal of the North American Benthological Society 21
Estimating the sources and transport of nutrients in the Waikato (2), 206–220.
River Basin, New Zealand. Water Resources Research 38 (12), Ejsmontkarabin, J., Weglenska, T., Wisniewski, R.J., 1993. The
1268–1291. effect of water-flow rate on zooplankton and its role in
Allan, J.D., Hinz, L., 2004. An assessment of flows for rivers of the phosphorus cycling in small impoundments. Water Science and
Great Lakes Basin. Project report prepared under grant from Technology 28 (6), 35–43.
Great Lakes Protection Fund, Ann Arbor. Engel, B.A., Srinivasan, R., Arnold, J., Rewerts, C., Brown, S.J.,
ASLO, American Society of Limnology and Oceanography, 2003. 1993. Nonpoint-source (Nps) pollution modeling using models
Emerging Research Questions for Limnology: The Study of Inland integrated with geographic information-systems (Gis). Water
Waters. Boulder, Colorado. Science and Technology 28 (3–5), 685–690.
Arnold, J.G., Allen, P.M., 1999. Automated methods for estimating Evans, D.J., Johnes, P., 2004. Physico-chemical controls on
baseflow and ground water recharge from streamflow records. phosphorus cycling in two lowland streams. Part 1 – the
Journal of the American Water Resources Association 35 (2), water column. Science of the Total Environment 329 (1–3),
411–424. 145–163.
Arnold, J.G., Allen, P.M., Muttiah, R., Bernhardt, G., 1995. Evans, D.J., Johnes, P.J., Lawrence, D.S., 2004. Physico-chemical
Automated base-flow separation and recession analysis tech- controls on phosphorus cycling in two lowland streams. Part 2 –
niques. Ground Water 33 (6), 1010–1018. the sediment phase. Science of the Total Environment 329 (1-3),
Arnold, J.G., Fohrer, N., 2005. SWAT2000: current capabilities and 165–182.
research opportunities in applied watershed modelling. Hydro- Francos, A. et al, 2001. Hydrological and water quality modelling in
logical Processes 19 (3), 563–572. a medium-sized coastal basin. Physics and Chemistry of the
Arnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R., 1998. Earth Part B – Hydrology Oceans and Atmosphere 26 (1), 47–52.
Large area hydrologic modeling and assessment – Part 1: model Green, C.H., Tomer, M.D., Di Luzio, M., Arnold, J.G., 2006.
development. Journal of the American Water Resources Associ- Hydrologic evaluation of The Soil and Water Assessment Tool for
ation 34 (1), 73–89. a large tile-drained watershed in Iowa. Transactions of the ASAE
Barlow, K., Nash, D., Grayson, R., 2004. Investigating phosphorus 49 (2), 413–422.
interactions with bed sediments in a fluvial environment using a Grizzetti, B., Bouraoui, F., Granlund, K., Rekolainen, S., Bidoglio,
recirculating flume and intact soil cores. Water Research 38 G., 2003. Modelling diffuse emission and retention of nutrients
(14–15), 3420–3430. in the Vantaanjoki watershed (Finland) using the SWAT model.
Benaman, J., Shoemaker, C.A., Haith, D.A., 2005. Calibration and Ecological Modelling 169 (1), 25–38.
validation of soil and water assessment tool on an agricultural Harris, G.P., 2001. Biogeochemistry of nitrogen and phosphorus in
watershed in upstate New York. Journal of Hydrologic Engineer- Australian catchments, rivers and estuaries: effects of land use
ing 10 (5), 363–374. and flow regulation and comparisons with global patterns.
Bernhardt, E.S. et al, 2005. Can’t see the forest for the stream? – Marine and Freshwater Research 52 (1), 139–149.
In-stream processing and terrestrial nitrogen exports. Bioscience Hart, D.D. et al, 2002. Dam removal: challenges and opportunities
55 (3), 219–230. for ecological research and river restoration. Bioscience 52 (8),
Bosch, N.S., Allan, J.D., 2008. The influence of impoundments on 669–681.
nutrient budgets in two catchments of Southeastern Michigan. Kling, G.W., Kipphut, G.W., Miller, M.M., O’Brien, W.J., 2000.
Biogeochemistry 87 (3), 325–338. Integration of lakes and streams in a landscape perspective: the
Bosch, N.S., 2007. The Influence of Impoundments on the Transport importance of material processing on spacial patterns and
and Fate of Nitrogen and Phosphorus in Two Southeastern temporal coherence. Freshwater Biology 43, 477–497.
Michigan River Catchments. University of Michigan, Ann Arbor, Knowlton, M.F., Jones, J.R., 2006. Natural variability in lakes and
177pp. reservoirs should be recognized in setting nutrient criteria. Lake
Brown, L.C., Barnwell, T.O., 1987. The Enhanced Water Quality and Reservoir Management 22 (2), 161–166.
Models QUAL2E and QUAL2E-UNCAS Documentation and User Knutilla, R.L., Allen, W.B., 1975. Water Resources of the River
Manual. EPA/600/3-87/007, USEPA, Athens, GA. Raisin Basin, Southeastern Michigan. US Geological Survey,
Chapra, S.C., 1997. Surface Water Quality Modeling. McGraw-Hill, Reston, VA, United States (USA).
Boston. Loehr, R.C., Martin, C.S., Rast, W., 1980. Phosphorus Management
Chu, T.W., Shirmohammadi, A., Montas, H., Sadeghi, A., 2004. Strategies for Lakes. Ann Arbor Science Publishers Inc., Ann
Evaluation of the SWAT model’s sediment and nutrient compo- Arbor.
nents in the piedmont physiographic region of Maryland. Trans- Maurer, W.R., Claflin, T.O., Rada, R.G., Rogala, J.T., 1995. Volume
actions of the ASAE 47 (5), 1523–1538. loss and mass-balance for selected physicochemical constituents
The influence of impoundments on riverine nutrient transport 147

in Lake Pepin, Upper Mississippi River, USA. Regulated Rivers 11 Smith, D.R., Haggard, B.E., Warnemuende, E.A., Huang, C., 2005.
(2), 175–184. Sediment phosphorus dynamics for three tile fed drainage
Meyer, J.L., Likens, G.E., 1979. Transport and transformation of ditches in Northeast Indiana. Agricultural Water Management
phosphorus in a forest stream ecosystem. Ecology 60 (6), 1255– 71 (1), 19–32.
1269. Soranno, P.A. et al, 1999. Spatial variation among lakes within
Moriasi, D.N. et al, 2007. Model evaluation guidelines for system- landscapes: ecological organization along lake chains. Ecosys-
atic quantification of accuracy in watershed simulations. Trans- tems 2 (5), 395–410.
actions of the Asabe 50 (3), 885–900. Stanley, E.H., Doyle, M.W., 2002. A geomorphic perspective on
Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through nutrient retention following dam removal. Bioscience 52 (8),
conceptual models: Part 1 – a discussion of principles. Journal 693–701.
of Hydrology 10 (3), 282–290. van Griensven, A., Bauwens, W., 2001. Integral water quality
Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R. King, K.W., modelling of catchments. Water Science and Technology 43 (7),
2002. Soil and Water Assessment Tool Theoretical Documenta- 321–328.
tion: Version 2000. Agricultural Research Service, Temple, van Griensven, A., Bauwens, W., 2003. Multiobjective autocalibra-
Texas. tion for semidistributed water quality models. Water Resources
Paul, M.J., Meyer, J.L., 2001. Streams in the urban landscape. Research 39 (12).
Annual Review of Ecology and Systematics 32, 333–365. van Griensven, A., Francos, A., Bauwens, W., 2002. Sensitivity
PCS, Permit Compliance System, 2006. Environmental Protection analysis and auto-calibration of an integral dynamic model for
Agency. <http://www.epa.gov/enviro/html/pcs/pcs_query. river water quality. Water Science and Technology 45 (9), 325–
html>. 332.
Peterson, B.J. et al, 2001. Control of nitrogen export from Veith, T.L., Sharpley, A.N., Weld, J.L., Gburek, W.J., 2005.
watersheds by headwater streams. Science 292 (5514), 86–90. Comparison of measured and simulated phosphorus losses with
Qi, C., Grunwald, S., 2005. GIS-based hydrologic modeling in the indexed site vulnerability. Transactions of the ASAE 48 (2), 557–
Sandusky watershed using SWAT. Transactions of the ASAE 48 565.
(1), 169–180. Vorosmarty, C.J. et al, 2003. Anthropogenic sediment retention:
Richards, R.P., 1998. Estimation of Pollutant Loads in Rivers and major global impact from registered river impoundments. Global
Streams: A Guidance Document for NPS Programs. Project report and Planetary Change 39 (1–2), 169–190.
prepared under Grant X998397-01-0, US Environmental Protec- Vorosmarty, C.J. et al, 1997. The storage and aging of continental
tion Agency, Region VIII, Denver. runoff in large reservoir systems of the world. Ambio 26 (4),
Richards, R.P. et al, 2001. Storm discharge, loads, and average 210–219.
concentrations in northwest Ohio rivers, 1975–1995. Journal of Walker, W.W., Kuhner, J., 1978. An Empirical Analysis of Factors
the American Water Resources Association 37 (2), 423–438. Controlling Eutrophication in Midwestern Impoundments.
Saleh, A. et al, 2000. Application of SWAT for the Upper North In:International Symposium on the Environmental Effects of
Bosque River Watershed. Transactions of the ASAE 43 (5), 1077– Hydraulic Engineering Works, University of Tennessee,
1087. Knoxville.
Santhi, C. et al, 2001. Validation of the SWAT model on a large river Walsh, C.J. et al, 2005. The urban stream syndrome: current
basin with point and nonpoint sources. Journal of the American knowledge and the search for a cure. Journal of the North
Water Resources Association 37 (5), 1169–1188. American Benthological Society 24 (3), 706–723.
Seitzinger, S.P. et al, 2002. Nitrogen retention in rivers: model Webster, J.R. et al, 2003. Factors affecting ammonium uptake in
development and application to watersheds in the northeastern streams – an inter-biome perspective. Freshwater Biology 48
USA. Biogeochemistry 57 (1), 199–237. (8), 1329–1352.

You might also like