Recurrent neural network-based seismic slope displacement hazard assessment considering the effects of slope aspects

Directionality of ground motion with respect to faulting is crucial if directivity pulse-like effects are observed; as a result, ground motion along slope sliding directions with pulse-like effects has a significant impact on slope stability during earthquakes. Ignoring the effects of slope aspects will possibly lead to underestimating regional coseismic landslide hazards. However, few studies have quantitatively investigated the effect of slope aspects and ground motion directionality on co-seismic landslide hazards. Hence, this study develops Newmark slope displacement prediction models by applying recurrent neural network (RNN) techniques on 3042 pairs of seismic ground motion records, focusing on 12 important input parameters (including seismological information, geological conditions, and slope characteristics). Seismic slope displacements considering the effect of slope aspects and pulse-like ground-motion were directly predicted by incorporating the relevant inputs into the proposed RNN model. The newly proposed RNN models are then incorporated into a probabilistic seismic slope displacement hazard analysis (PSSDHA) to demonstrate their effectiveness in geotechnical engineering applications.