Orientation-aware seismic landslide hazard assessment utilizing deep transfer learning under pulse-like ground motions

Abstract

Pulse-like ground motions (GMs) induced by the near-fault directivity are characterized by large-amplitude coherent velocity pulses, which have been demonstrated to cause significantly greater damage to buildings and slopes during earthquakes than ordinary (non-pulse-like) GMs. However, due to the deficiency of the pulse-like GM records, the consideration of their effects in the sliding displacement-based seismic landslide hazard assessment has been limited. This study is the first to address this challenge by employing the deep transfer learning technique to develop an orientation-aware prediction model of Newmark slope sliding displacements for pulse-like GMs. In the model, the ground-motion orientation is also considered via the maximum, median, and minimum displacements in all directionalities. Earthquake source parameters, site parameters, ground-motion intensity measures, and critical acceleration (a_c) were used as predictive variables for the model. Furthermore, the developed model is validated by comparing it with other predictive models. The results indicate that the proposed model generates a higher prediction accuracy and better generalization capability. Finally, the proposed prediction model is applied to the orientation-aware seismic landslide hazard assessment for a near-fault region. It is validated by using the actual landslide data from the 1994 Northridge earthquake (M_w 6.7) in California. The validation results indicate that the proposed model performs exceptionally well in predicting near-fault seismic landslides, providing a solid basis for reducing earthquake-induced landslide risks in near-fault areas.