Evaluating the effectiveness and robustness of machine learning models with varied geo-environmental factors for determining vulnerability to water flow-induced gully erosion

Abstract

Assessing and mapping the vulnerability of gully erosion in mountainous and semi-arid areas is a crucial field of research due to the significant environmental degradation observed in such regions. In order to tackle this problem, the present study aims to evaluate the effectiveness of three commonly used machine learning models: Random Forest, Support Vector Machine, and Logistic Regression. Several geographic and environmental factors including topographic, geomorphological, environmental, and hydrologic factors that can contribute to gully erosion were considered as predictor variables of gully erosion susceptibility. Based on an existing differential GPS survey inventory of gully erosion, a total of 191 eroded gullies were spatially randomly split in a 70:30 ratio for use in model calibration and validation, respectively. The models’ performance was assessed by calculating the area under the ROC curve (AUC). The findings indicate that the RF model exhibited the highest performance (AUC = 89%), followed by the SVM (AUC = 87%) and LR (AUC = 87%) models. Furthermore, the results highlight those factors such as NDVI, lithology, drainage, and density were the most influential, as determined by the RF, SVM, and LR methods. This study provides a valuable tool for enhancing the mapping of soil erosion and identifying the most important influencing factors that primarily cause soil deterioration in mountainous and semi-arid regions.