National Scale Earthquake Susceptibility Mapping Utilizing Explainable Artificial Intelligence in The Nepal Himalaya

Abstract

Nepal faces high earthquake risk due to its active tectonics and rapid development on vulnerable terrain. This study presents a national-scale earthquake susceptibility assessment using explainable artificial intelligence. Historical earthquake records were combined with key geophysical and geomorphic factors, including fault proximity, fault density, tectonic zones, topography, and seismic event density. Random Forest and Extreme Gradient Boosting models were developed to estimate spatial earthquake probability, and explainable AI techniques were applied to interpret variable importance and model behavior. The Random Forest model achieved higher accuracy and lower uncertainty compared to XGB. The resulting maps highlight clusters of elevated seismic probability along major fault corridors. The approach supports transparent, data-driven seismic risk evaluation suitable for planning and disaster preparedness in Nepal.