Estimating landslide trigger factors using distributed lag nonlinear models

Earthquake events that are often accompanied by prolonged rainfall before, during, or after the mainshock, usually result in thousands of landslides. To estimate landslide trigger factors in such scenarios, we propose a hybrid model combining a statistical model for cumulative rainfall with a physical model for coseismic landslide displacement. The statistical model is a Distributed Lag Nonlinear Model (DLNM) and the physical model is a rigorous Newmark's analysis. The chain of events that led to landsliding following the 2011 Sikkim earthquake is used as a case study. Trigger information of 164 landslide points from field investigations were used to train the model and predict the trigger for 1196 satellite-based landslide points. The hybrid model significantly improves predictions over generalized additive models. Cumulative rainfall shows a significant spatial correlation with trigger factors and heavy rainfall three weeks before the earthquake played a key role in preparing the ground for landslides.

 

https://www.sciencedirect.com/science/article/abs/pii/S1364815224003207