Scenario-based physically informed assessment of rainfall- and earthquake-induced landslide failure probability across China

Abstract

Landslides are among the most widespread and destructive geological hazards in China, driven by complex mechanisms involving multiple interacting factors. Their impacts are particularly severe when intense rainfall and strong ground motion occur in combination. At the national scale, existing landslide hazard assessments often rely on empirical or machine learning approaches. While effective for regional-scale applications, these models are constrained by incomplete inventories and often lack sufficient physical interpretability, which restricts their use for physically interpretable, scenario-based slope-stability assessment. To address these limitations, we propose a scenario-based physically informed framework for national-scale landslide slope-stability assessment in China. This framework integrates the TRIGRS and Newmark models to simulate slope stability under 15 rainfall intensity–duration scenarios and 10 seismic ground motion levels. Failure-probability estimation is supported by incorporating the Rosenblueth point estimate method and empirical failure probability curves to translate model outputs into probabilistic metrics. Furthermore, a sequential rainfall-conditioned seismic scenario involving extreme rainfall followed by earthquake shaking is introduced to evaluate nonlinear amplification effects from multi-source interactions. The results indicate that comparatively high rainfall-induced failure probabilities occur in the southwestern mountainous regions and the Loess Plateau. In contrast, areas with comparatively high earthquake-induced failure probability expand significantly with increasing PGA, with comparatively unstable areas primarily located along the western Sichuan area, northern Shaanxi, and eastern Tibet, where tectonic activity is intense. Under the sequential rainfall–earthquake scenarios, the predicted unstable area is substantially larger than that under the corresponding single-trigger scenarios, indicating pronounced within-model scenario amplification. These findings enhance the understanding of spatial distribution patterns of landslides under multi-source triggering conditions and provide a useful reference for national-scale scenario-based landslide assessment and hazard-informed planning.