Evaluation of Three Watershed‐Scale Pesticide Environmental Transport and Fate Models1

Abstract: The U.S. Environmental Protection Agency (USEPA) Office of Pesticide Programs (OPP) has completed an evaluation of three watershed‐scale simulation models for potential use in Food Quality Protection Act pesticide drinking water exposure assessments. The evaluation may also guide OPP in identifying computer simulation tools that can be used in performing aquatic ecological exposure assessments. Models selected for evaluation were the Soil Water Assessment Tool (SWAT), the Nonpoint Source Model (NPSM), a modified version of the Hydrologic Simulation Program‐Fortran (HSPF), and the Pesticide Root Zone Model‐Riverine Water Quality (PRZM‐RIVWQ) model. Simulated concentrations of the pesticides atrazine, metolachlor, and trifluralin in surface water were compared with field data monitored in the Sugar Creek watershed of Indiana’s White River basin by the National Water Quality Assessment (NAWQA) program. The evaluation not only provided USEPA with experience in using watershed models for estimating pesticide concentration in flowing water but also led to the development of improved statistical techniques for assessing model accuracy. Further, it demonstrated the difficulty of representing spatially and temporally variable soil, weather, and pesticide applications with relatively infrequent, spatially fixed, point estimates. It also demonstrated the value of using monitoring and modeling as mutually supporting tools and pointed to the need to design monitoring programs that support modeling.     

HYDROLOGIC MODELING AT THE WATERSHED SCALE USING NPSM1

ABSTRACT: The Nonpoint Source Model (NPSM) was chosen for nonpoint source pollutant modeling within three different watersheds. The first step in using NPSM, hydrologic calibration, is discussed here for three 8‐digit Hydrologic Unit Codes (HUCs) from the White River Basin in Indiana (Driftwood HUC), the Albemarle‐Pamlico River Basin in Virginia and North Carolina (Contentnea HUC), and the Apalachicola‐Chattahoochee‐Flint River Basin in Alabama, Georgia, and Florida (Ichawaynochaway HUC). Model predicted flows were compared statistically with USGS gauge data at the HUC outflow points for an uncalibrated and calibrated model run for the period from January 1, 1990, through December 31, 1992, and a validation run for the period from January 1, 1993, through December 31, 1995. Least squares regression of NPSM predicted flows versus USGS gauge data were 0.75, 0.44, and 0.69 for the calibration runs and 0.71, 0.69, and 0.64 for the validation runs in the Driftwood, Contentnea, and Ichawaynochaway HUCs, respectively. Nash Sutcliffe coefficient values were not as strong, ranging from ?0.66 to 0.45 for the calibration runs and 0.31 to 0.37 for the validation runs of the model. The Ichawaynochaway HUC proved the most difficult to calibrate indicating that the model may not be as useful in some geographic locations.     

APPLICATION OF THE BASINS DATABASE AND NPSM MODEL ON A SMALL OHIO WATERSHED1

ABSTRACT: The purpose of this study was to evaluate the Better Assessment Science Integrating Point and Nonpoint Sources (BASINS) watershed management system. BASINS data were used with the NPSM model to predict discharge and sediment concentrations at the outlet of a 103 km² Ohio watershed. It was concluded that the NPSM model should always be calibrated but only a few of the parameters provided with BASINS needed to be calibrated. For a three‐year study period, there was a 2 percent underestimation of discharge using area weighted precipitation values and a 25 percent overestimation using the single station data in BASINS. A comparison of observed and predicted monthly discharge resulted in an r² of 0.86 with area‐weighted precipitation and an r² of 0.74 with the single station data. Calibrating the model to substantially improve sediment predictions was unsuccessful and we concluded that a calibration period of one year was too short. For the three‐year study period, the r² for sediment was 0.36 with a slope of 0.37 and an intercept of 18.8 mg/l. The mean observed and predicted sediment concentrations were 27.1 mg/l and 22.6 mg/l, respectively.