Validation of field-scale soil erosion models using common datasets

As a first step toward evaluating the suitability of erosion models for global change studies, common datasets (representing 73 site-years of data from seven sites in three countries) were prepared for use with six field-scale erosion models. Five of these are continuous-simulation types (GLEAMS, EPIC, CSEP, MEDRUSH and WEPP); the other is event-based (EUROSEM). Each dataset was split into a 'training set' and a 'testing set'. Measured values for runoff and erosion from the testing set were withheld from the modellers. An analysis of results from the continuous simulation models is presented here. A comparison of simulated and measured results for runoff and erosion for the training and testing sets, and at several timescales, reveals several model-specific responses as well as a number of more general conclusions: calibration is desirable for many models runoff is always better simulated than soil loss long-term average results are generally best simulated there is some evidence that relative results are more reliable than absolute.