Multiscale Comparison of Regional Rainfall-Induced Landslide Predictive Models: The Example of Hurricane Maria in Puerto Rico

Severe storms can trigger hundreds to thousands of landslides in mountainous regions, yet predicting rainfall-induced landslides across spatial scales remains challenging. Most existing hazard models operate either at single-hillslope scales using high-resolution mechanistic formulations or at regional to global scales using simplified empirical approaches. However, structured workflows linking these scales are few. This study establishes a quantitative cross-scale evaluation framework using a hindcast of Hurricane Maria in Puerto Rico, by comparing two fundamentally different models: the global, probabilistic LHASA (Landslide Hazard Assessment for Situational Awareness) v2.0 model and a high-resolution, mechanistic, deterministic Transient-Flow Pseudo-3D model developed herein. LHASA was applied island-wide at 1 km resolution to identify high-risk areas, including a 2.5 km 2 region in Utuado municipality, which was subsequently analyzed at 1 m resolution using the mechanistic model. Results demonstrate the complementary strengths of both approaches. LHASA provides rapid regional screening with minimal computational cost, whereas the mechanistic model, at greater computational cost, produces detailed predictions of landslide location, area, volume, and timing that closely match observed landslides. Rather than identifying a single superior model, this study advances a strategically integrated, multi-scale framework that links probabilistic screening with targeted mechanistic analysis for operational landslide hazard assessment.