Stima del deflusso parcellare e della corrispondente perdita di suolo tramite modelli ANFIS

This work presents an application of the Adaptive Network-based Fuzzy Inference System (ANFIS) tool to the simulation of runoff and soil erosion at the plot scale. Data representing the characteristics of precipitation were used for runoff estimation, while to simulate soil loss it was also used the observed runoff. In the study, the database composed of continuous climatic records and plot runoff and soil loss measurements at the experimental station “Le Masse” of the University of Perugia has been used. The performances of the implemented models have been evaluated with several statistical indexes, with a cross-validation technique. Although the small number of data in the database (61 erosive events in the period 2008-2012), encouraging results have been obtained for some divisions of the database (r=0.851 and E =0.703 for runoff; r=0.973 and E=0.808 for soil loss). The correlation coefficient r between observed and simulated data shows good mean values, especially for soil loss estimation (r = 0.804 for the best performing model), while the Nash-Sutcliffe efficiency index E shows variable mean values and in general is higher for the runoff estimation. The results are in general very variable depending on the testing set, proof that the available database is not yet sufficient to make the ANFIS networks acquire a good generalization capability.