The influence of both process descriptions and runoff patterns on predictions from a spatially distributed soil erosion model

Spatially distributed, process-based erosion models have been built to predict both erosion and deposition rates and patterns. However, predictions by these models often do not meet our expectations. There are several possible reasons for this, which can be subdivided into three groups: (i) process descriptions within the models may be incomplete or incorrect, leading to structural errors (model error); (ii) parameterization is difficult leading to a high degree of uncertainty regarding model inputs (input error); and (iii) the spatial implementation of the model (the spatial discretization and the routing methods used) may be inadequate. Hitherto, most studies of erosion model performance have focused on input parameter uncertainty: the effects of process description and spatial implementation of the model have often been neglected. In this study an attempt is made to compare some aspects of the effects of process description, parameter uncertainty and model implementation on model performance. A series of model runs was carried out to evaluate to what extent the predictions of the spatially distributed erosion model LISEM are affected by the description of the hydraulic resistance, the parameter values used to characterize hydraulic resistance as well as the routing algorithm used to generate the runoff pattern. The process description used for overland flow resistance and velocity has a clear effect on the erosion and deposition rates predicted by the model. However, relatively minor variations in a single parameter value (hydraulic roughness) appear to affect model output more strongly than variations in the process description itself. The use of different routing algorithms leads to notably different erosion and deposition patterns, but total erosion and deposition rates are much less affected. Copyright (C) 2005 John Wiley Sons, Ltd.