Modelling soil erosion in a Mediterranean watershed: Comparison between SWAT and AnnAGNPS models

In this study, the simulations generated by two of the most widely used hydrological basin-scale models, the Annualized Agricultural Non-Point Source (AnnAGNPS) and the Soil and Water Assessment Tool (SWAT), were compared in a Mediterranean watershed, the Carapelle (Apulia, Southern Italy). Input data requirements, time and efforts needed for input preparation, strength and weakness points of each model, ease of use and limitations were evaluated in order to give information to users. Models were calibrated and validated at monthly time scale for hydrology and sediment load using a four year period of observations (streamflow and suspended sediment concentrations). In the driest year, the specific sediment load measured at the outlet was 0.89 t ha^-1 yr^-1, while the simulated values were 0.83 t ha^-1 yr^-1 and 1.99 t ha^-1 yr^-1 for SWAT and AnnAGNPS, respectively. In the wettest year, the specific measured sediment load was 7.45 t ha^-1 yr^-1, and the simulated values were 8.27 t ha^-1 yr^-1 and 6.23 t ha^-1 yr^-1 for SWAT and AnnAGNPS, respectively. Both models showed from fair to a very good correlation between observed and simulated streamflow and satisfactory for sediment load. Results showed that most of the basin is under moderate (1.4-10 t ha^-1 yr^-1) and high-risk erosion (> 10 t ha^-1 yr^-1). The sediment yield predicted by the SWAT and AnnAGNPS models were compared with estimates of soil erosion simulated by models for Europe (PESERA and RUSLE2015). The average gross erosion estimated by the RUSLE2015 model (12.5 t ha^-1 yr^-1) resulted comparable with the average specific sediment yield estimated by SWAT (8.8 t ha^-1 yr^-1) and AnnAGNPS (5.6 t ha^-1 yr^-1), while it was found that the average soil erosion estimated by PESERA is lower than the other estimates (1.2 t ha^-1 yr^-1).