Soil Erosion Quantification using Machine Learning in Sub-Watersheds of Northern Portugal

Protected areas (PA) play an important role in minimizing the effects of soil erosion watersheds. This study evaluated performance machine learning models, specifically support vector with linear kernel (SVMLinear), polynomial (SVMPoly), and random forest (RF), on identifying indicators 761 sub-watersheds PA northern Portugal, by using water Europe, according to revised universal loss equation (RUSLE2015), as target variable. The parameters analyzed were: Europe total burned area sub-watershed period 1975-2020, fire recurrence, topographic wetness index (TWI), morphometric factors, namely (A), perimeter (P), length (L), width (W), orientation (O), elongation ratio (Re), circularity (Rc), compactness coefficient (Cc), form factor (Ff), shape (Sf), DEM, slope, curvature. median determination (R2) for each model was RF (0.61), SVMpoly (0.68), SVMLinear (0.54). Regarding parameters, those that registered greatest importance were A, P, L, W, curvature, area, indicating analysis which considers together data affected moisture, is indicator