A Random Forest Machine Learning Approach for the Identification and Quantification of Erosive Events

Predicting the occurrence of erosive rain events and quantifying corresponding soil loss is extremely useful in all applications where assessing phenomenon impacts required. These problems, addressed literature at different spatial temporal scales according to most diverse approaches, are here by implementing random forest (RF) machine learning models. For this purpose, we used datasets built through many years observations plot-scale experimental site SERLAB (central Italy). Based on 32 features describing rainfall characteristics, RF classifier has achieved a global accuracy 84.8% recognizing non-erosive events, thus demonstrating slightly higher performances than previously (non-machine learning) methodologies. A critical performance percentage correctly recognized observed total (72.3%). However, since relevant identified, found only slight underestimation erosivity (91%). The regression model for estimating event loss, based three (runoff coefficient, erosivity, period occurrence), demonstrates better (RMSE = 2.30 Mg ha−1) traditional models 3.34 ha−1).