Dynamics of population exposure to landslides under climate change scenarios

Mountain regions are experiencing more frequent disasters, likely driven by increased landslide activity under climate change and rapid population growth. However, the interaction between population dynamics and landslide processes remains insufficiently studied. We address this gap by proposing a spatial, dynamic framework to estimate current and future populations exposed to landslides. The methodological framework comprises: (i) construction of a landslide susceptibility index; (ii) projection of future landslide scenarios using state-of-the-art climate change projections; and (iii) estimation of exposed population. We generate landslide susceptibility scenarios by integrating a Convolutional Neural Network (CNN) with Explainable Artificial Intelligence (XAI). Four landslide scenarios are projected for 2050 and 2100 using rainfall under Shared Socioeconomic Pathways (SSP) SSP2-4.5 and SSP5-8.5. We then produce fifty exposure scenarios by combining the landslide susceptibility with five SSP-based demographic trajectories for 2050 and 2100. The framework is applied to the Darjeeling Himalayas, India, a recognized landslide hotspot. Results indicate substantial spatial shifts in landslide susceptibility under different rainfall scenarios within the same modeling setup. Overall, exposed population tends to increase, although not linearly across scenarios. Scenarios aligned with SSP3 and SSP4 show marked growth of exposed population in landslide-sensitive zones. Notably, the combination of SSP3 (2100) population with the SSP2-4.5 (2050) susceptibility scenario yields the highest number of people at risk. These findings provide actionable evidence for planning, risk-informed land-use, and assessing the region’s carrying capacity, while the framework itself is scalable to other mountain areas with comparable geohydrological settings. • Future landslide-susceptibility scenarios developed by integrating Explainable Artificial Intelligence (XAI) with SSP–RCP climate projections • Population exposure to future landslides estimated for 2050 and 2100 • 50 village-level scenarios of landslide susceptibility and population exposure produced • Spatio-temporal dynamics of landslide susceptibility and exposure mapped