Effects of distribution-based parameter aggregation on a spatially distributed agricultural nonpoint source pollution model

Accurate predictions of sediment yield from distributed models of runoff and sediment yield depends in part of how well matched the model structure is to input data spatial representation. This study investigated how model structure and input data representation affect sediment predictions made using the Soil and Water Assessment Tool (SWAT). The study focused on the integration of two specific components of SWAT: the Modified Universal Soil Loss Equation (MUSLE) and the hydrologic response unit (HRU). MUSLE, a watershed erosion model, was applied to different levels of watershed partitioning and alternative HRU schemes for a watershed and its two subwatersheds over a 4-year period of measured stream flow and sediment yield. The results show that across different levels of watershed partitioning, HRUs do not conserve sediment. Instead, HRUs introduce almost half of the variability in sediment generation, which has previously been attributed solely to input data aggregation. This occurs for two reasons. First, MUSLE defines a nonlinear relationship between sediment generation and HRU area, but the sediment load is scaled linearly from the HRU level to the subwatershed level. Second, HRUs aggregate land areas without regard for the surface connectivity assumptions, which are implicit in MUSLE. These conflicts caused by the integration of HRU and MUSLE makes it difficult to determine the effect of different land use on soil erosion. This study indicates that greater attention should be made to structuring the data inputs to match the underlying assumptions of sub-models within SWAT.