Abstract:Agriculture has been identified as a leading source of Nonpoint Source (NPS) pollution as a result of its intensive fertilizer applications and crop management practices, and has led to ecological and human health concerns in the Three Gorges Reservoir Area (TGRA), China. Watershed models, Annualized Agricultural Nonpoint Source (AnnAGNPS), are considered a cost-effective and time-efficient approach for the assessment of pollutant loads and management practices, are coupled with GIS to facilitate watershed planning and management, which offer an unprecedented flexibility in the organization of spatial data, the spatial extent of input parameter aggregation has previously been shown to have a substantial impact on model output. However, the aggregation of spatial information in the GIS, the effects of data aggregation on model input parameters, such as land use and cover and soil type, are not well studied in watershed of TGRA condition. For the study watershed, values of critical source area (CSA) and minimum source channel length (MSCL) was set from 0.5 to 15 hm2 and the MSCL from 7.5 to 200 m. The study focused on the impact of the size or the number of cells and spatial discretization unit (SDU) used to partition a watershed (10 watershed delineations) on the input parameter spatial aggregation and output of model, the results show that across levels of watershed partitioning, there are different extent of parameter spatial aggregation of land use and soil types. The results indicate that input parameter spatial aggregation had little effect on runoff yields predictions, and more effect on sediment, total nitrogen and phosphorous loads; within the range of acceptable accuracy, in this paper, the applicability and predictive capacity of AnnAGNPS model in TGRA conditions is examined. The optimal threshold values of SDU, relative to the scale of Heigou watershed and TGRA conditions, required to adequately predicting runoff, sediment yields and nitrogen, phosphorous were found to be around 0.5-18, 2-6, 0.5-6 hm2, respectively. The model can be applied for estimation of runoff and sediment losses with acceptable accuracy in TGRA, the CSA and MSCL values determine the hydrographic segmentation of the watershed and the cell size, especially in heterogeneous areas of the watershed, which can increase the accuracy of model results, which indicates that greater attention should be made to partition a watershed to match the underlying assumptions of sub-models within AnnAGNPS.