This report describes the calibration and validation of a spatially distributed watershed model of the Cannonsville Reservoir Basin. The Soil and Water Assessment Tool 2000 (SWAT2000) was selected as the watershed model. A set of SWAT2000 inputs representative of the watershed conditions was derived from a wide array of data sources. Important methods were developed for converting available information to SWAT2000 inputs for groundwater soluble phosphorus concentrations, initial soil phosphorus levels and daily manure application. The Cannonsville Reservoir is a New York City water supply reservoir located in upstate New York that has historically experienced water quality problems associated with phosphorus loading. As a result, the watershed has been subjected to multiple water quality regulations including a recent Total Maximum Daily Load (TMDL) assessment for phosphorus. The reservoir watershed covers an 1178 km2 area and is dominated by agriculture, particularly dairy farming. The SWAT2000 model of the Cannonsville Reservoir Watershed is a valuable tool that can be used to help identify and evaluate quantitatively the long-term effects of various phosphorus management options for mitigating loading to the reservoir. SWAT2000 was developed by the Agricultural Research Service of the United States Department of Agriculture.
SWAT2000 simulates through time the daily soil water balance, growth of plants, build-up and subsequent transport of soil nutrients to surface waters in response to agricultural management practices. The simulated mass balance of soil phosphorus in SWAT2000 is an important aspect of any watershed model that is to be used for regulatory purposes. The authors modified a few of the SWAT model equations to better simulate measured flows, sediment loading and phosphorus loading during the winter. The model was calibrated and validated for the prediction of dissolved and particulate phosphorus transport, and therefore also flow and sediment transport, against a large set of monitoring data. Extensive continuous flow and water quality data over a 10-year period from multiple locations within the basin were used for model calibration and validation. Sensitive model parameters were adjusted within their feasible ranges during calibration to minimize model prediction errors for daily flows and monthly sediment and phosphorus loading. At the main flow gauging station in the basin (Walton), draining almost 80% of the watershed, daily calibration resulted in model predictions of average flow within 1.0% of the measured average flow while the daily Nash-Sutcliffe (NS) measure was 0.79. Daily validation results at Walton showed the model predicted average flow within 4.5% of the measured average flow with a NS of 0.78. At the main water quality gauging station in the basin (Beerston), just downstream of Walton, the calibration results showed the model predicted the average monthly sediment and total phosphorus loading within 3% and 6% of their respective measured average monthly loadings. The monthly calibration NS values at Beerston for sediment and total phosphorus loading were 0.66 and 0.68, respectively. Validation results at Beerston showed the model predicted the average monthly sediment and total phosphorus loading within 27% and 9% of their respective measured average monthly loadings. The monthly validation NS values at Beerston for sediment and total phosphorus loading were 0.51 and 0.61, respectively. The largest errors in model predictions for phosphorus and sediment loading were always associated with peak flow prediction errors. Model predictions were also shown to qualitatively replicate bi-weekly sampling of total phosphorus concentrations taken from 10 different locations across the watershed. Model simulation results over the calibration and validation period (1990-2000) highlighted a number of useful findings. The model predicted that 68% of the total phosphorus loading to surface waters in the watershed originates from active agricultural lands. Corn land use was simulated as the major source of agricultural phosphorus loading even though it covered only 1.2% of the watershed area. Areas North and East of the Town of Delhi tended to have the largest rates of phosphorus loading per unit area. Areas immediately surrounding the Cannonsville Reservoir that are not monitored were simulated to have substantially lower non-point source phosphorus (NPS) unit area loading rates than the monitored portion of the watershed.