Head-cut gully erosion susceptibility mapping in semi-arid region using machine learning methods: insight from the high atlas, Morocco

Gully erosion has been identified in recent decades as a global threat to people and property. This problem also affects the socioeconomic stability of societies therefore limits their sustainable development, it impacts nonrenewable resource on human scale, namely, soil. The focus this study is evaluate prediction performance four machine learning (ML) models: Logistic Regression (LR), classification regression tree (CART), Linear Discriminate Analysis (LDA), k-Nearest Neighbors (kNN), which are novel approaches gully modeling research, particularly semi-arid regions with mountainous character. 204 samples areas non-erosion were collected through field surveys high-resolution satellite images, 17 significant factors considered. dataset cells (70% for training 30% testing) randomly prepared assess robustness different models. functional relevance between soil effective was computed using ML models evaluated metrics, including accuracy, kappa coefficient. kNN ideal model study. value AUC from ROC considering testing datasets KNN 0.93; remaining associated similar terms values. values GLM, LDA, CART 0.90, 0.91, 0.84, respectively. accuracy validation CART, KNN, GLM 0.85, 0.82, 0.89, 0.84 Kappa 0.70, 0.65, 0.68, models, particular have achieved outstanding results creating susceptibility maps. maps created most reliable could be useful tool management, watershed conservation prevention water losses.