A Physics-Informed Framework for Earthquake-Induced Landslide Hazard Assessment Considering Faulting Style and Ground Motion Directionality

Accurate assessment of earthquake-induced landslide (EIL) hazard remains challenging because landslide occurrence is jointly controlled by faulting style, pulse-like ground motions, ground motion directionality, and slope response. Based on a global landslide database from 10 earthquakes, including 5 strike-slip and 5 non-strike-slip events, this study identifies a systematic azimuthal asymmetry in landslide distribution. The results show that the preferred directions of landslide occurrence are consistent with the dominant directions of pulse effects, but that the controlling patterns differ markedly between the two faulting styles. To incorporate these characteristics into hazard assessment, we develop a physics-informed framework that explicitly embeds faulting style and ground motion directionality into the regional EIL hazard assessment chain. A pulse probability model conditioned by faulting style and the angle between slope aspect and fault strike is combined with observed or GMPE-derived intensity measures to estimate pulse-aware peak ground velocity fields. These fields are then used in Newmark analysis to calculate permanent displacement, which is further integrated with 11 conventional conditioning factors in a convolutional neural network for landslide hazard assessment. Application to the 2017 Mw 7.0 Jiuzhaigou earthquake in China gives an AUC of 0.906, representing a 6.6% improvement over the physics-based model alone. The predicted hazard pattern is also consistent with the observed concentration of landslides in high-hazard zones. These results demonstrate that incorporating faulting style and ground motion directionality can improve regional EIL hazard assessment.