Earthquake-induced landslides are a major hazard in mountainous regions, where complex topography and near-surface conditions jointly control ground-motion amplification and slope instability. In this context, ground-motion models used as triggering inputs for landslide analyses must accurately represent site effects in complex terrain. This study develops a geomorphometry-informed ground-motion model based on predictors derived from global remote sensing Digital Elevation Models (DEMs), conceived as a triggering component for earthquake-induced landslide applications. The model is based on the eXtreme Gradient Boosting (XGBoost) regression algorithm and predicts peak ground acceleration, peak ground velocity, and spectral accelerations by integrating seismic source parameters, finite-fault source-to-site metrics, and geomorphometric site proxies derived from global DEMs. The model is trained on an extended Italian strong-motion dataset comprising about 8300 recordings from 90 earthquakes with finite-fault rupture models and is evaluated using a strict leave-one-event-out validation scheme. Results show that finite-fault parameterization reduces prediction errors by about 11% compared to point-source formulations, while DEM-derived site proxies improve predictive performance by approximately 5% relative to VS30 and 12% relative to the fundamental frequency f0. Residual analysis yields inter-event variability of 0.19–0.22 and intra-event variability of 0.23–0.26. The proposed framework demonstrates how global remote sensing products provide value-added predictors for ground-motion triggering in complex terrain, suitable for integration with earthquake-induced landslide susceptibility models.
Mori et al. (Tue,) studied this question.