River systems are a major source of water for agricultural and urban water needs. Water quality assessments of river systems are becoming critical throughout the country and there is a real concern about the sustainable supply of quality water and the health of the water bodies. River systems should be continuously monitored to assess the effects of different land management practices on water quality. But, long-term continuous monitoring is not currently being conducted due to high costs. Therefore, there is a need for an alternate tool such as a basin-scale hydrologic/water quality model that is capable of predicting the effects of land management with reasonable level of accuracy.

SWAT (Arnold et al., 1993) is a basin-scale hydrologic/water quality model developed to predict the effects of alternative river basin land use management decisions on water, sediment, and chemical yields. SWAT operates on a daily time step and is capable of simulating 100 or more years. Major components of the model include hydrology, weather, erosion, soil temperature, crop growth, nutrients, pesticides, subsurface flow, and agricultural management. SWAT offers distributed parameter and continuous time simulation with flexible watershed configuration, automatic irrigation and fertilization, inter-basin water transfer, and lake water quality simulation capabilities.

Until now the instream nutrient dynamics were not considered in the SWAT model. In order to simulate the instream dynamics, the kinetic routines from an instream water quality model, QUAL2E (Brown and Barnwell, 1987), were modified and incorporated in SWAT. In this paper we have described the instream water quality component of SWAT and presented the preliminary results from the application of the model to Wister Lake watershed situated in eastern Oklahoma and western Arkansas.

Since recent past, GIS has been playing an important role in natural resources modeling and proving to be an effective tool for non-point source (NPS) pollution models (Pelletier, 1985; Hession and Shanholtz, 1988; Srinivasan and Arnold, 1994). A continuous time, distributed parameter model like SWAT overcomes some of the limitations of single-event models (Arnold et al., 1995). SWAT considers a basin or watershed divided into subbasins based on topography, soil, and land use and thus preserves the spatially-distributed parameters of the entire basin and homogeneous characteristics within a subbasin. But manual collection of inputs for such models is often difficult and tedious due to the level of aggregation and the nature of spatial distribution. For this a GIS has been proven to be an excellent tool to aggregate and organize input data for distributed parameter hydrologic/water quality models (Tim et al., 1991; Rewerts and Engel, 1991; Srinivasan et al., 1993; Rosenthal et al., 1995).

The SWAT model has been integrated with a raster-based GIS, GRASS (Shapiro et al., 1992), designed and developed by the Environmental Division of the U.S. Army Construction Engineering Research Laboratory. The SWAT/GRASS interface (Srinivasan and Arnold, 1994) consists of three modules: (a) project manager, (b) input extractor and aggregator, and (c) input editor. The project manager interacts with the user to collect, prepare, edit, and store the basin and subbasin information to be formatted into SWAT input files. The input extractor and aggregator uses a variety of hydrologic tools (Srinivasan and Arnold, 1993) to derive SWAT input information from GRASS raster/site map layers such as basin boundary map with subbasin delineation, digital elevation map (DEM), soils map, land use/land cover map and weather generator/station location map. In addition the reservoirs, inflow, pond and lake data are collected directly from the user. The input editor is used to either view, edit or check the data collected from the previous phase, which are arranged in different data forms. Rosenthal et al. (1995) used this interface to aggregate SWAT input data for the Lower Colorado River basin of Texas and found that the SWAT/GRASS interface reduced the data collection and manipulation time by several folds, and allowed the user to modify and analyze various alternative management practices rather easily. Further details about the interface are given by Srinivasan and Arnold (1994).

The water quality parameters simulated by the instream water quality component are algae as Chlorophyll-A (Chl-a), dissolved oxygen (DO), carbonaceous biochemical oxygen demand (CBOD), organic nitrogen (OrgN), ammonium nitrogen (NH4-N), nitrite nitrogen (NO2-N), nitrate nitrogen (NO3-N), organic phosphorus as P (OrgP), and soluble phosphorus as P (SolP). The basic mass transport equation used in QUAL2E is given by

... 1

and the general first-order rate equation for every constituent is given by:

...2

where C is the constituent concentration (ML-3), t is time (T), `i' is the spatial element, `n' is the time segment, `r' is the first-order rate constant (T-1), and `p' is the internal constituent source/sink (ML-3T-1). In QUAL2E each reach is divided into several computational elements and for each time step the resultant differential equation is solved numerically by implicit backward difference technique. In the instream water quality component of SWAT, the diffusion is ignored assuming complete mixing within a reach. The advection component is considered in the SWAT's existing routing component. Thus, in the instream water quality component only the first-order rate decay is considered, which depends on the travel time of the constituent within a reach by considering a reach as a single computational element. First-order decay relationships for Chl-A, nutrients, CBOD, and DO used in QUAL2E were adopted in SWAT with necessary modifications.

Temperature

In the instream water quality component of SWAT, water temperature simulation is not physically based, but is estimated from the air temperature based on a relationship developed by Stefan and Preud'homme (1993) through regression analysis of many river observations. The relationship is given by

... 3

where Tw and Ta are temperature of the water and air (C), respectively. It can be seen from the above equation that the water will be warmer than air if the air temperature is below 20C, which may be consistent with most rivers. Exceptions to this will include small streams under heavy shading, places where the stream temperature is influenced by anthropological activity (e.g. discharge of waste heat from a power generation plant).

Temperature Effects on Rate Coefficients

The rate coefficients (ri) depend on the water temperature. The temperature correction for the rate coefficients is given by:

... 4

where XT is the temperature-corrected coefficient at the local water temperature T, X20 is the value of the coefficient at 20C, and is an empirical constant for each reaction coefficient. The default values for given by Brown and Barnwell (1987) are used in SWAT. Every temperature dependent first-order reaction rate coefficient is corrected using the above equation.

Prior to the addition of the instream kinetics , SWAT did not predict Chl-A, CBOD, and DO loads from sub-basins to the streams. Appropriate loading estimations for these were established from the available literature. Our aim is to define the loading of these variables as a function of flow, nitrogen, phosphorous, organic matter, and temperature, whose dynamics are already defined in SWAT. A brief description of these functions are given below.

Algae/Chlorophyll-A

Chl-A is assumed to be directly proportional to suspended algal biomass concentration in water. Therefore the algal biomass loading to the stream is estimated as Chl-A loading from the sub-basins. We used the relationship developed by Cluis et al. (1988). They developed relationships between nutrient enrichment index (TN:TP), Chl-a and algal growth potential in the North Yamaska river, Canada. This lead to the functional relationship of the form:

... 5

where AGP is the algal growth potential (mg/l), Chl_A is Chl-A (g/l), Q is flow (m3/s), TN is the total kjeldhal nitrogen (TKN = OrgN + NH4-N) load (kg) and TP is total phosphorous load (kg of P), and `a' and `b' are coefficient and exponent, respectively. Cluis et al. (1988) presented the values of `a' and `b' for different seasons, and in SWAT the summer values (a = 7.25 and b = -4.68) were used. In addition, through their analysis they established that 1 mg/l of AGP is equivalent to 1g/l of Chl-A.

This relationship is based on regression analyses and not physically based. We resolved to using this type of expression because, (a) the actual physical relationships are very complex, and (b) studies conducted to define these actual physical relationships are very limited.

Carbonaceous Biochemical Oxygen Demand (CBOD)

The SWAT loading function for the ultimate CBOD is based on a relationship given by Thomann and Mueller (1987):

... 6

where Lu is ultimate CBOD (mg/l), Corg is organic carbon load (mg), and q is the flow (l). The organic carbon concentration is calculated as

... 7

where OC1 is the fraction of organic carbon content in surface layer of the soil profile (g-C/g-soil), ER is the enrichment ratio (g-C/g-sediment), and Yd is the sediment yield (kg).

Dissolved Oxygen (DO)

Assuming initially that the water from rainfall is saturated with oxygen (100% saturation level), the dissolved oxygen loading from a sub-basin is calculated by subtracting the oxygen uptake by the oxygen demanding substance in the runoff. The amount of oxygen withdrawn from the water depends on the average time of overland flow. The DO loading from a sub-basin is estimated by

... 8

where Od is dissolved oxygen concentration (mg/l), Os is the oxygen saturation concentration at temperature Tw (mg/l), Kl is CBOD deoxygenation rate (day-1), 'alpha_n' is the oxygen uptake rate of organic nitrogen (mg-O/mg-N), 'beta_n' is the oxidation rate coefficient of organic nitrogen (day-1), N is organic nitrogen concentration (mg/l) and tov is the overland travel time (day). For simplicity Kl and 'alpha_n' are assumed to be 1.047 and 'beta_n' is 4.6 (Brown and Barnwell, 1987). The dissolved oxygen saturation concentration is given by

... 9

where T is the water temperature in K (K = C + 273.15), and this equation is valid only if T is between 273.15 and 313.15 K (Brown and Barnwell, 1987).

We applied the SWAT model with the instream water quality component to simulate the hydrology and water quality in Wister Lake watershed (Figure 1). A detailed description of the watershed and the input data sets for the watershed was presented by Storm et al. (1994). The watershed covers approximately 260,000 ha (640,000 acres) streatching between Oklahoma and Arkansas, is situated in the Arkansas river basin. The outlet of the watershed in Wister Lake, situated on the Oklahoma side. Results from a EPA lake survey conducted in 1974 have declared Wister Lake as eutrophic and excessively turbid. A preliminary Total Maximum Daily Load (TMDL) analysis for the Wister Lake watershed was conducted by Smolen et al. (1993), which analyzes the sources of point and non-point source pollution.

Figure 1: Wister Lake Watershed and Subwatersheds

The four primary stream systems flowing into Wister Lake are Poteau river, Black Fork, Fourche Maline, and Holson Creek. There are four continuous stream gage monitoring stations, two on Poteau river, one on Fourche Maline and one on Black Fork. In addition there are three miscellaneous gages having stream flow measurements at approximately six weeks interval. Stream flow data is available from all these stations from December 1991. Along with stream flow data, water quality data is also available from all the seven stations at approximately six weeks interval. Data from all these stations have not yet been consolidated and updated. Therefore, for this work we used the data from only two stations (locations shown in Figure 1). Both of them have continuos stream flow data and six-week water quality data.

The entire watershed was divided into subwatersheds in such a way that the outlet of each subwatershed coincides with one of the stream monitoring stations. Rainfall data is available from five rain gage stations situated in the vicinity of the watershed. Given the locations of the rain-gages, SWAT/GRASS interface picks up the closest rain-gage to each subwatershed. In the process the interface picked up only two of the available rain-gage stations. For the land use map, USGS land use/land cover map of the area was used, and soils map was obtained by digitizing the county soil maps of Leflore and Latimer counties in Oklahoma, and Scott and Polk Counties in Arkansas. The soil properties were derived from the STATSGO database (USDA-NRCS, 1994). The soils and land use maps are shown in Figures 2 and 3, respectively. The land use map shows that 75 % of the watershed is covered with forest, 23 % with pasture, and the rest of the area is urban, agricultural and rangeland. For this work we used a 100 m resolution for all the data layers though some of the data layers are actually at a lower resolution.

Figure 2: Soils Map of Wister Lake Watershed

Figure 3: Land Use Map of Wister Lake Watershed

SWAT simulations could be conducted based on two subwatershed configurations namely, dominant and virtual basin approach. The details of both approaches are given by Mamillapalli et al. (1996). Due to the low spatial variation in land use and soils we used the dominant approach for this work.

Using the input data accumulated by the SWAT/GRASS interface, the SWAT model was used to simulate the hydrology and water quality in the Wister lake watershed from 1991 to 1994. All the model runs were made starting from 1989 using synthetic weather data for 1989 and 1990, in order to `prime' the model. A minimal calibration of the model for stream flow was conducted. The initial uncalibrated run of the model showed that the model was under estimating stream flow. Therefore, the curve number for all the watersheds was increased by 10 % of the original value and the available water holding capacity of all the soils were reduced by a value of 0.05 mm/mm.

Stream flow

Figure 4a shows the time series plot of cumulative stream flow from 1991 to 1994 for the station at Fourche Maline River at Red Oak (Station 1). In general SWAT over estimated the stream flow, but the simulated trend matches well with the observed. Except for stream flow during November, 1994, all the predicted monthly flows match very well with the observed data. The coefficient of determination between observed and predicted monthly stream flow data is 0.64.

Four-year time series plot of cumulative flow at Poteau River at Cauthron (Station 2) is shown in Figure 4b. At this station the prediction is not as good as Station 1. The predicted total flow for each year is reasonably close to the observed values, but some of the individual monthly values do not match very well. This may be due to lack of good distribution of rain gage locations. Both rain gages that are being used in our simulations are in Oklahoma side of the watershed. But, the stream gage and the subwatersheds flowing through the stream gage are in Arkansas side. Storm et al. (1994) noted significant variation in precipitation within the existing four rain gages in Oklahoma side. Therefore, we feel that for a better prediction of stream flow from Arkansas side, we need weather data from locations closer to the subwatersheds than the existing ones.

Figure 4a: Observed and Simulated Cumulative Stream Flow at Station 1

Figure 4b: Observed and Simulated Cumulative Stream Flow at Station 2

Water Temperature

Figures 5a and 5b show the daily time series plot of simulated water temperature (C) and six-week observed water temperature at Stations 1 and 2, respectively. The breaks in the daily simulated water temperatures are due to the days when SWAT did not predict any flow. The simulated water temperature values are reasonably close to observed data and the trends of the simulations match closely with the observed.

Figure 5a: Observed and Simulated Water Temperature at Station 1

Figure 5b: Observed and Simulated Water Temperature at Station 2

Total Nitrogen

Time series plot of daily simulated total nitrogen concentration and the six-week observed concentration from Stations 1 and 2 are shown in Figures 6a and 6b. The simulated total nitrogen concentrations are unreasonably high compared to the observed concentrations. There may be several reasons for this:

· Looking at the SWAT daily output file (data not presented in this paper) we find that nitrate loading to the stream is high. Since crop production activities in this area is negligible, the main source of nitrogen is biomass residue. The over estimation of total nitrogen concentration in stream water may be due to the over estimation of nitrogen mineralization in the soil.

· SWAT considers default values for observed values of initial soil nitrate and organic nitrogen content in the absence of extensive field observed values. This default value may not be accurate enough.

Researchers agree that modeling nitrogen is one of the most challenging tasks even at field scale. We are trying to find out some more reasons for inaccurate predictions of stream nitrogen concentrations and will try to improve the predictions of SWAT.

Figure 6a: Observed and Simulated Total Nitrogen in Stream Flow at Station 1

Figure 6b: Observed and Simulated Total Nitrogen in Stream Flow at Station 2

Total Phosphorous

Figures 7a and 7b show the simulated daily total phosphorous concentrations and the six-week observed concentrations. We find that SWAT consistently predicts lower total P concentration than the observed values, but the predictions are not very far off from the observed concentrations. Most of the phosphorous in runoff is carried by sediments. Therefore, prediction of sediment in runoff has to be good for reasonable prediction of phosphorus in runoff. In this paper we did not look into the sediment prediction capability of SWAT due to lack of sufficient observed data.

Figure 7a: Observed and Simulated Total Phosphorous in Stream Flow at Station 1

Figure 7b: Observed and Simulated Total Phosphorous in Stream Flow at Station 2

Dissolved Oxygen

Figure 8a shows the daily dissolved oxygen concentration and the six-week observed concentration for Station 1. The breaks in the predicted concentration data correspond to the no-flow days. SWAT predicted lower oxygen concentrations during the high flow periods (October to February) and higher concentration during low-flow periods (April to August). The inaccuracy in reduction of oxygen concentration may be an effect of inaccurate prediction of nutrients. Figure 8b shows the comparison of simulated and predicted oxygen concentration at Station 2. Here the predictions during high flows are reasonably close to the observed, but during low flow periods SWAT tends to over estimate dissolve oxygen concentration in the stream. In general, the dissolved oxygen prediction of SWAT is reasonably good.

Figure 8a: Observed and Simulated Dissolved Oxygen Concentration at Station 1

Figure 8b: Observed and Simulated Dissolved Oxygen Concentration at Station 2

The instream kinetics of QUAL2E was incorporated in a basin scale hydrologic/water quality model, SWAT. We have described the instream water quality component of SWAT and presented the preliminary results from the application of the model to Wister Lake watershed situated in eastern Oklahoma and western Arkansas. The input data for the model were aggregated using a GIS interface. The GIS used by the SWAT interface is GRASS.

SWAT predicted monthly stream at Station 1, situated in Oklahoma side of the watershed with reasonable accuracy. But, the stream flow predictions at Station 2, Which is in Arkansas side is not as good. This may be due to lack of a good distribution of weather stations. The water temperature and dissolved oxygen predictions by SWAT are satisfactory. The nutrient predictions are inaccurate due to many speculative reasons. Efforts for improving the water quality prediction capability of SWAT are under way.

There are only a very few models that are linked with a GIS and capable of simulating both hydrology and water quality on a river basin scale. By adding the instream water quality component for SWAT we have added a new dimension to basin scale modeling. The results from this study are from a `first-cut' preliminary analysis. The GIS-linked SWAT model shows a good potential for being used by a tool to predict the effects of land use activities on surface water bodies.

The SWAT model with instream water quality component can be used as a tool for EPA-recommended TMDL analysis. Having estimates of point source pollution loading from subbasins, SWAT model could be used to analyze the effects of both point and non-point source pollution on the stream body. SWAT can also to used to analyze the effects of alternative management strategies and aid regulatory agencies in decision making.

In summary the major strengths of the model are: (i) simple, yet complex enough to simulate the interactions between weather, crop growth, and land use management on a river basin scale for long periods, (ii) a GIS interface, which will save substantial amount of resources in aggregating input data for large-scale simulations, and (iii) a graphical output interface and analysis tool to visualize the simulation results.

We sincerely thank Dr. Daniel Storm and R. Lakshminarayanan, Biosystems and Agricultural Engineering Department, Oklahoma State University for providing us the GIS map layers.

Arnold, J.G., P.M. Allen, and G. Bernhardt, 1993. A comprehensive surface-groundwater flow model. J. Hydrol. 142:47-69.

Brown, L. C. and T. O. Barnwell, Jr., 1987. The enhanced water quality models QUAL2E and QUAL2E-UNCAS documentation and user manual. EPA document # EPA/600/3-87/007, Cooperative Agreement # 811883, Environmental Research Laboratory, U.S. Environmental Protection Agency, Athens, GA.

Cluis, D., P. Couture, R. Bégin, and S.A. Visser, 1988. Potential eutrophication assessment in rivers; relationship between produced and exported loads. Schweiz. Z. Hydrol. 50(2):166-181.

Hession, W. C., and V. O. Shanholtz, 1998. A geogrphic information system for targeting non-point source agricultural pollution. J. Soil Water Conserv. 43(3):264-266.

Mamillapalli, S., R. Srinivasan, J. G. Arnold, and B. A. Engel, 1996. Effect of saptial variability on basin scale modeling. To be presented at the Third International NCGIA Conference/Workshop on Integrating GIS and Environmental Modeling, Sante Fe, New Mexico, January, 21-25.

Pelletier, R. E., 1985. Evaluating nonpoint source pollution using remotely sensed data in soil erosion models. J. Soil Water Conserv. 40(4):332-335.

Rewerts, C. C., and B. A. Engel, 1991. ANSWERS on GRASS: Integrating a watershe simulation with a GIS. ASAE papre # 91-2621, ASAE, St. Joseph, MI.

Rosenthal, W. D., R. Srinivasan, and J. G. Arnold, 1995. Alternative river management using a linked GIS-hydrology model. Trans. ASAE, 38(3):783-790.

Smolen, M. D. , R. Lakshminarayanan, W. C. Hession, and D. E. Storm, 1993. Estimating total maximum daily load (TMDL) for wister lake. ASAE paper # 93-2072, ASAE, St. Joseph, MI.

Srinivasan, R., and J.G. Arnold, 1994. Integration of a basin-scale water quality model with GIS. Water Resources Bulletin 30(3):453-462.

Stefan, H. G. and E. B. Preud'homme, 1993. Stream temperature estimation from air temperature. Water Res. Bull. 29(1):27-45.

Storm, D. E., M. D. Smolen, R. Lakshminarayanan, W. C. Hession, 1994. Wister lake watershed project: Annual Report FY 93. Agricultural Experiment Stationand Cooperative Extension Service, Oklahoma State University, Stillwater, OK.

Thomann, R.V. and J.A. Mueller, 1987. Principles of surface water quality modeling and control pp 261-384. Harper & Row Publishers, New York.

Tim, U. S., S. Mostaghimi, V. O. Shanholtz, and N. Zhang, 1991. Identification fo critical nonpoint pollution source area using geographic information systems and simulation modeling. ASAE paper # 91-2114, ASAE, St. Joseph, MI.

USDA-NRCS, 1994. State Soil Geographic Data Base, US Department of Agriculture - Natural Resource Conservation Service, National Soil Survey Center, Publication # 1492, Washington, DC.

Raghavan Srinivasan
Associate Research Scientist
Blackland Research Center
Phone: (817) 770-6670
Email: srin@brcsun0.tamu.edu

Jeffrey G. Arnold
Hydraulic Engineer
USDA-ARS
Phone: (817) 770-6502
Email: arnold@brcsun0.tamu.edu

Address: 808 East Blackland Road, Temple, TX 76502
Fax: (817)-770-6561