Landslide susceptibility mapping is essential for risk mitigation in geologically complex and landslide-prone regions like the Western Ghats, India. Frequent landslides in these terrains pose significant risks to infrastructure, ecosystems, and human settlements, necessitating advanced predictive modeling techniques for effective hazard management. This study aims to analyse and predict long-term landslide susceptibility in the basaltic terrain of the Savitri River Basin, Western Ghats, India using machine learning techniques, addressing the need for more precise hazard assessment in these hazardous landscapes. Geological features and environmental variables were integrated using R-Studio, incorporating satellite imagery, field data, and an inventory of 559 landslides, compiled from Google Earth images, Bhukosh-GSI, and NASA. The RF and SVM models were trained with a 70% training dataset and a 30% testing dataset, followed by cross-validation to ensure model robustness. Results confirm the effectiveness of machine learning models in landslide susceptibility mapping, with the RF model outperforming SVM, achieving an AUC-ROC of 0.95. Field validation, alongside receiver operating characteristic (ROC) curves, was used to assess the accuracy of the generated landslide susceptibility maps. Among the conditioning factors, lithology emerged as a key determinant of landslide susceptibility, underscoring the importance of geological controls in landslide-prone basaltic terrains. The outcome of this research may enhance the scientific understanding of landslide mechanisms in basaltic terrains providing valuable insights for hazard mitigation, land-use planning, environmental protection, slope stability assessment, and disaster risk management. These findings will serve as crucial resource for policymakers, urban planners, and environmental managers in developing proactive mitigation strategies and sustainable land-use planning frameworks.
Baile et al. (Sun,) studied this question.