MACHINE LEARNING APPROACHES FOR LANDSLIDE SUSCEPTIBILITY MAPPING FROM SATELLITE DATA

Abstract

Landslides are among the most destructive natural hazards, causing severe damage to infrastructure, ecosystems, and human lives. Accurate prediction and mapping of landslide-prone areas are essential for effective disaster mitigation and land-use planning. Traditional geotechnical and statistical models often struggle to represent the complex nonlinear interactions between terrain, soil, vegetation, and climatic factors that influence slope instability. This study presents a machine learning-based framework for landslide susceptibility mapping using multi-source satellite imagery and geospatial data. The proposed methodology integrates remote sensing variables such as digital elevation models, land surface temperature, vegetation indices, and rainfall intensity to train supervised learning algorithms including Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosting. Feature selection and data preprocessing techniques are applied to enhance the reliability of the models. The trained models are validated using ground-truth landslide inventories and performance metrics such as the Area Under the Curve (AUC), accuracy, and F1-score. Experimental results demonstrate that the machine learning approach outperforms conventional analytical methods in identifying high-risk zones with improved spatial precision and reduced false-positive rates. The resulting susceptibility maps provide valuable insights for policymakers and disaster management authorities in developing early-warning systems and sustainable risk reduction strategies