Comparative Study on Hazard Assessment of Earthquake-triggered Landslides Based on Coupling Model
The study takes extreme earthquake area of Jiuzhaigou earthquake as the research area.Based on data from 4834 earthquake landslides,we select 9 factors as the risk evaluation factors of earthquake landslide:slope gradient,slope direction,topographic relief,elevation,distance to fault,epicenter distance,stratum lithology,distance to river and distance to road.4 coupling models of certainty factor:logistic regression model(CF-LR),information value-logistic regression model(I-LR),certainty factor-random forest model(CF-RF)and information value-random forest model(I-RF),are used to evaluate the seismic landslide risk in the study area.The accuracy of the four combined models is compared and analyzed by using frequency ratio and receiver operating characteristic curve(ROC).The results show that the frequency ratio of seismic landslide risk grade of the 4 combined models increases significantly with the increase of landslide risk grade,and the frequency ratio of extremely high risk and high risk area accounts for more than 85%of the total frequency ratio.The AUC values of CF-LR model,I-LR model,CF-RF model and I-RF model are 0.877,0.879,0.903 and 0.905,respectively,indicating that the 4 combined models can accurately evaluate the risk of Jiuzhaigou earthquake landslide.Coupling CF model or I model with LR model and RF model respectively,the accuracy of RF coupling model is improved by 2.6%compared with LR coupling model,indicating that RF model has more advantages in solving the problem of nonlinear relationship.This result can provide reference for disaster risk assessment and disaster prevention and reduction planning in this area.