Soil erosion susceptibility assessment using logistic regression, decision tree and random forest: study on the Mayurakshi river basin of Eastern India

Soil erosion is one of the major environmental hazards causing severe land degradation in the sub-tropical monsoon dominated Mayurakshi river basin (MRB) of Eastern India. Hence, this study aims to delineate the areas with severe soil erosion probability (SEP) using logistic regression (LR), decision tree (DT), and Random forest (RF). A soil erosion inventory map was prepared using 150 rill and gully erosion prone sites, out of which 70% sample points were randomly chosen for modelling and remaining 30% sites were used for model validation. Alongside, 12 conditioning factors including elevation, curvature, aspect, runoff, TWI, slope, geology, stream frequency, rainfall erosivity, NDVI, LS-factor, and LULC were selected as spatial data base for model building. Multicollinearity among conditioning factors were performed using tolerance (TOL) and variance inflation factor (VIF). The analysis concludes that the possibility of soil erosion is very high in the undulating western parts of Mayurakshi river basin as compared to other sectors. The validation results obtained from ROC curve and kappa statistics is showing that DT and RF reached a higher prediction accuracy than LR model.