Predicting post-fire sediment yields with RULSE, WEPP, and ERMiT: accuracy and limitations

High severity wildfires can increase hillslope- and watershed-scale sediment yields by two or more orders of magnitude. Resource managers need models to predict the potential effects of burning on site productivity and downstream aquatic resources, and to determine whether post-fire rehabilitation treatments should be applied. RUSLE and Disturbed WEPP are commonly used to predict annual post-fire sediment yields, whereas the Erosion Risk Management Tool (ERMiT) predicts storm-based sediment yields on a probabilistic basis. The goals of this study were to: 1) test these models against 2-5 years of annual and storm-based sediment yields from 88 plots in 10 Colorado Front Range fires; and 2) identify potential improvements to each model. Both RUSLE and WEPP poorly predicted annual post-fire sediment yields, as the R2 values ranged from 0.16 to 0.26 and the root mean square error for sediment yields greater than 1 Mg ha-1 yr-1 was comparable to the mean measured value of 9-10 Mg ha-1 yr-1. The variability in sediment yields among plots within the same fire was about twice the predicted variability, and both models were much more successful at predicting mean values from plots grouped by burn severity or fire (R2=0.53-0.60). These results indicate that the poor performance was at least partly due to the inability of the models to adequately account for the plot-scale factors that affect post-fire sediment yields. Field measurements show that variations in rilling, soil water repellency, and micro-topography are important controls on post-fire sediment yields, but neither model directly simulates these components, particularly on a spatially explicit basis. ERMiT was designed to overcome some of these limitations by varying the hydraulic conductivity, rill erodibility, and burn severity across a hillslope to generate a probability distribution of predicted post-fire sediment yields. A comparison of ERMiT predictions against 290 storm-based sediment yields shows that half of the measured sediment yields were larger than the 10% exceedence probability for a 1.33-year rain storm, even though the measured storms were only 1.0-1.33-year events. This under-prediction is at least partly due to the fact that ERMiT assigns a relatively low probability for an entire hillslope to burn at high severity. Both ERMiT and Disturbed WEPP also assume a more rapid decline in post-fire sediment yields than shown by our field data. In conclusion, the current limitations on model accuracy include the inability of the models to represent all of the factors affecting post-fire sediment yields, our inability to adequately characterize these factors in the field, and the limited data for model parameterization and calibration. Only some of these limitations can be addressed by model improvements.