Distributed erosion and sediment yield models are being increasingly used for predicting soil erosion and sediment yields in agricultural catchments. In most applications, validation of such models has commonly been restricted to comparison of the predicted and measured sediment output from a catchment, because spatially distributed information on rates and patterns of soil redistribution within the catchment has been lacking. However, such spatially distributed data are needed for rigorous model testing, in order to validate the internal functioning of a model and its applicability at different spatial scales. The study reported in this paper uses two approaches to test the performance of the agricultural non-point source pollution (AGNPS) and areal non-point source watershed environmental response simulation (ANSWERS) erosion and sediment yield models in two small catchments in Devon, UK. These involve, firstly, comparison of observed and predicted runoff and sediment output data for individual storm events monitored at the basin outlets and, secondly, information on the spatial pattern of soil redistribution within, the catchments derived from Cs-137 measurements. The results obtained indicate that catchment outputs simulated by both models are reasonably consistent with the recorded values, although the AGNPS model appears to provide closer agreement between observed and predicted values. However, the spatial patterns of soil redistribution and the sediment delivery ratios predicted for the two catchments by the AGNPS and ANSWERS models differ significantly. Comparison of the catchment sediment delivery ratios and the pattern of soil redistribution in individual fields predicted by the models with equivalent information derived from Cs-137 measurements indicates that the AGNPS model provides more meaningful predictions of erosion and sediment yield under UK conditions than the ANSWERS model and emphasises the importance of using information on both catchment output and sediment redistribution within the catchment for model validation. (C) 2002 Elsevier Science B.V. All rights reserved.
