Uncertainty modelling and error propagation in a spatial soil-erosion model

A general approach for handling uncertainty in spatial models is presented and illustrated using a simple soil-erosion model. The method is based on developing quality models for input data that characterise spatial uncertainty. The estimated errors of the input parameters of the erosion model (slope, vegetation cover, erodibility and overland flow) are then propagated through the soil-erosion model using analytical and simulation methods. Comparison of these methods shows that the uncertainty produced using the simulation method is less than the uncertainty produced using the analytical method. Analysis of both methods indicates that this difference is due to the fact that the simulation method considers the interactions between the input parameters caused by the non-linearity of the soil-erosion model, while the analytical method fails to consider these interactions. The results show that errors associated with vegetation cover and overland flow affect the predicted erosion more than the errors in the other parameters and thus these two parameters should be targeted for error-minimisation strategies. Interpretation of the spatial distribution of the errors associated with the model and its parameters sometimes permits their cause to be determined, thus allowing error mitigation strategies to be developed. For the soil-erosion model studied here we identify errors associated with vegetation cover and topography and recommend methods that can be used to minimise them