Assessment of co-seismic landslide hazard using the Newmark model and statistical analyses: a case study of the 2013 Lushan, China, Mw6.6 earthquake

Abstract

The April 20, 2013 Mw6.6 earthquake of Lushan County, Sichuan Province, China, has triggered 4540 landslides (> 1000 m2). Exploring a more effective method to assess landslide hazard in the affected area of this event is of great significance for disaster prevention and mitigation. By applying the Newmark model and two statistical analysis models (logic regression and support vector machine, LR and SVM), this study addressed this issue. In the Newmark model, we used the landslide point density, the average gradient (mean slope) and the mean peak ground acceleration to group the lithology and created a critical acceleration (ac) map. The Newmark displacements and the probability of the slope instability are mapped by combining the ac map and PGA map. In the statistical analysis models of LR and SVM, 7040 samples (4540 landslide sites and 2500 random non-landslide sites) were randomly divided into the training set (5000 samples) and validation set (2040 samples). Based on the relationship between landslide distribution and influence factors, we selected the critical acceleration (ac) value, topographic relief, PGA, and distance to rivers as the independent variables for LR and SVM. Finally, the ROC curves for three landslide hazard models were drawn and the AUC values were calculated. The landslide hazard maps produced by LR are similar to those by applying SVM. The AUC values indicate that these two models combined with ac data perform better than the simplified Newmark model. In this study, a new method of integrating statistical analysis models (LR and SVM) with critical acceleration (ac) for earthquake landslide hazard assessment is presented, which can be used to carry out seismic landslide hazard assessment more effectively than the simplified Newmark model.