Markov–Switching Spatio–Temporal Generalized Additive Model for Landslide Susceptibility

Statistical susceptibility models predict the occurrence of landslides, which is the first step toward landslide hazard estimation. However, most of the models in the literature do not consider spatiotemporal dependencies among landslide occurrences. This work introduces a novel Markov-Switching Spatio-Temporal Generalized Additive Model (MSST-GAM) for landslide susceptibility. This model predicts an unobserved sequence of risk states using nonlinear functions of time-dependent covariates. Spatial dependencies are modeled by a neighborhood structure. The model was applied to a multi-temporal inventory of post-seismic debris flow in a region affected by the 2008 Wenchuan earthquake. A five-fold spatiotemporal cross-validation is used to evaluate the model performance. It is observed that the MSST-GAM improved the performance significantly over GAM and Logistic Regression model in terms of AUC-ROC. Additionally, MSST-GAM improves the mean log-likelihood by 24.1% compared to GAM. The results show that the newly proposed model is a viable alternative for landslide susceptibility mapping. •Introduced a novel Markov-switching GAM approach for landslide susceptibility.•This model predicts risk state sequence using nonlinear functions of covariates.•Spatial dependencies are modeled by a neighborhood structure.•Spatiotemporal cross validation shows the model generalizes well over space and time.•AUC-ROC and log-likelihood show that the model outperforms LR and GAM.