ABSTRACT. Sediment and nutrient loadings in the Little

Sediment and nutrient loadings in the Little River Research Watershed in south central Georgia were modeled using the continuous simulation Annualized Agricultural Nonpoint-Source Pollution (AnnAGNPS) model, part of the AGNPS suite of modeling components. Specifically, nitrogen, phosphorus, sediment, and runoff were predicted over a seven-year period. Land under cultivation makes up approximately 25% of the 333 km 2 watershed. Livestock facilities include swine, poultry, dairy cows, and beef cattle. Results from the simulation were compared to seven years of monitoring data at the outlet of five nested subwatersheds and at the outlet of the Little River Research Watershed (LRRW). The average annual predicted runoff in the upper part of the watershed was one-third to half of observed runoff. In contrast, predicted runoff in the lower part of the watershed was close to observed, and was 100% of observed at the outlet of the watershed. Runoff underprediction was attributed to the method of landcover discretization. The extent of forest land in the upper watershed (55% to 63%) and the fragmented landscape that has relatively small fields surrounded by riparian forests and tracts of forest resulted in overestimation of forested area in the watershed. In addition to runoff, sediment and nutrient loads were also underpredicted in the upper part of the LRRW. Two factors are most likely responsible for underprediction. Runoff is underpredicted at these sites, which reduces the carrying capacity of sediment loads. In addition, the overestimation of forested areas at these sites coincides with underestimation of sediment-producing areas, such as cropland. In contrast to the upper part of the watershed, sediment and nutrient loads were overpredicted in the lower part of the watershed. This may have resulted from inadequately simulating nonpoint-source pollution attenuation by the extensive riparian forests and forested in-stream wetland areas found in these watersheds. Prediction results can be improved through better input into the model, as well as modification of the processes within the model to account for forest and riparian conditions.