Evaluation of Seismic Landslide Susceptibility by Integrating Statistical Learning Model and Newmark Model—A Case Study of the Wenchuan Earthquake

Whether it can quickly and effectively predict the susceptibility of regional earthquake landslides to achieve rapid rescue, loss assessment and post-disaster reconstruction has always been a difficult problem. However, the traditional high-precision evaluation of seismic landslide susceptibility often relies heavily on the complete or incomplete landslide inventory, which is poor in timeliness and cannot effectively evaluate the target area before or shortly after the earthquake. In most cases, the Newmark model relies on experts’ experience to select model parameters, therefore the evaluation result of this method is unstable and it lacks strong generalization ability. A fused model is proposed to classify the positive and negative training samples of the study area through the evaluation results of the Newmark model under the slope units, and it applies a variety of statistical learning models to evaluate the landslide susceptibility of the Wenchuan earthquake based on the classification results of the Newmark model. The results show that the evaluation of the statistical learning model fused with the Newmark model has higher accuracy. This method can overcome the inherent shortcomings of a single Newmark model to obtain better evaluation results without relying on obtaining the complete landslide inventory. Meanwhile, the model can be applied to quickly obtain the evaluation results of regional landslide susceptibility before or shortly after the earthquake, thereby effectively reducing human and economic losses caused by earthquake landslides.