Climate-Driven Landslide Hazard: Enhancing Community Resilience through Physics-Based Regional Simulations

Landslides remain a major global hazard, with thousands of events each year posing significant impacts to human life, infrastructure, critical lifelines, and ecosystems. Global climate change and evolving land-use practices are expected to increase the frequency, severity, and spatial extent of landslides, further amplifying the socio-economic risks. While advanced constitutive models enable detailed slope-stability analyses at the single-hillslope scale, a critical gap persists in assessing landslide hazards at the community scale. Addressing this gap is essential for improving resilience and informing effective mitigation strategies. Existing regional-scale models often struggle to balance physical rigor, predictive accuracy, and computational efficiency. They rely either on empirical approaches with limited physical basis or on mechanistic formulations that are highly sensitive to data availability and quality. This study presents CRISIS (Coupled Regional Rainfall-Induced and Seismic Slope Instability Simulations), a physics-based regional landslide prediction framework. It couples a pseudo-3D slope-stability formulation with a hydrological model capable of simulating transient three-dimensional groundwater flow. CRISIS operates in complementary back-analysis and forward-prediction modes, integrating multi-scale topographic, hydraulic, and hydrological data from remote sensing, geophysical surveys, and field and laboratory testing. Back-analysis of both failed and stable slopes generates high-resolution spatial estimates of shear strength. These estimates are iteratively refined and used for forward prediction of landslide location, size, depth, and timing. Results reveal a strong geospatial connection between hillslope location and dominant failure mechanisms. This emphasizes that landslide initiation is inherently geospatial and governed by coupled surface - subsurface processes. Application of the framework to Hurricane Maria in Puerto Rico, where over 70,000 landslides were triggered, demonstrates close agreement between predicted and observed landslide locations, sizes, and timing. Additional validation against historical storm events further confirms the robustness of the approach. The broader vision of this work is to advance regional characterization, assessment, and mitigation of cascading hazards through data-model integration. This enables climate-driven risk quantification and supports early warning systems for resilient communities.

Speaker: Mirna Kassem, UC Berkeley