Explainable AI and Data-Driven Approaches for Landslide Forecasting

Landslides are among the most devastating natural hazards, causing severe impacts on humans and damage to natural features and infrastructure. Recent years have witnessed increased landslide activity in Kerala, in southern India, due to significant variations in rainfall patterns. There is a growing global demand for tools and techniques to provide early warning to people living in the landslide-susceptible zones. The advent of artificial intelligence (AI)- and machine learning (ML)-based computational models has led to significant breakthroughs in the development of accurate landslide prediction models. The Meenachil River Basin in Kerala, southern India, has been studied to develop an early warning system for landslides in the landslide-susceptible zones. A total of 15 landslide conditioning factors were utilised to develop a landslide prediction model based on remote sensing data (DEM and Sentinel 2) and other secondary sources. The support vector machine (SVM) model achieved high accuracy in predicting landslide susceptibility zones in the study area. Further, the ML-based explainable algorithm, Shapley Additive exPlanations (SHAP), analysed feature interactions for both landslide and non-landslide locations, thus providing greater clarity on the real driving factors of landslides. This explainable ML model reveals new aspects of each feature's explainability in developing an accurate model for forecasting landslide occurrences.