Modeling of Gully Erosion Based on Random Forest Using GIS and R

Abstract

Generally, gully erosion and its areal extension is a natural process fully controlled by external forces and shaped by internal settings. It adversely impacts soil productivity, eco-system function, and quality of environment as it affects land and water quality. For the development of sustainable land utilization strategy, it is initially required to develop an effective management process. This study aims to develop a gully erosion potentiality map using a well-known machine learning algorithm, that is, random forest (RF) model in the River Bakulla basin area, Jharkhand, India. In this work, 12 gully erosion predisposing factors (i.e., altitude, plan curvature, slope length, land use, soil types, slope gradient, topographical wetness index, distance from river, drainage density, distance from road, and distance from lineament) were selected based on available data and literature review. Finally, the gully erosion susceptibility map (GESM) generated by the RF model and the output was validated by employing the unused gully locations with ROC curve. The predicted results reveal that RF model has high prediction accuracy; the AUC value was 91% at end of the analysis. RF-generated GESM can be a very useful tool for the management action and land improvement measures in initial stages of gully development, to protect the development of gully erosion.