Comparison of landslide susceptibility mapping rapid assessment models in Jiuzhaigou County, Sichuan province, China

The main purpose of this study is to compare the five rapid assessment methods of landslide susceptibility mapping including information value(I), certainty factor(CF), logistic regression(LR), LR-I and LR-CF based on GIS technique. The five models are applied in Jiuzhaigou County, Sichuan, China, which is part of the area affected by the August 8, 2017 Jiuzhaigou earthquake. 6205 landslides were selected as the database based on history records, remote sensing interpretation and field investigation, and fifteen influence factors including elevation, slope gradient, slope aspect, terrain curvature, profile curvature, plane curvature, PGA, stratum lithology, rainfall, topographic relief, surface roughness, surface cutting degree, distance to fault, distance to highway and distance to river were chosen as the evaluation indices. The I, CF and LR models were used to build the landslide susceptibility mapping evaluation systems based on 80% of total(6205) landslide database, and the landslide susceptibility was set as extremely low, low, middle, high and extremely high in Jiuzhaigou County. The LR-I and LR-CF coupling models were proposed to optimally perform landslide susceptibility at Jiuzhaigou County. A susceptibility evaluation index system was secondly calculated by logistic regression and susceptibility level prediction for landslides was performed based on GIS platform. The validations of the resulting susceptibility maps were performed and compared by the frequency ratio and the area under curve(AUC) in receiver operating characteristic curve(ROC), which represents the respective success rate. The results indicate that the frequency ratio of high and extremely high susceptibility in the Jiuzhaigou County is more than 85% for all I, CF, LR, LR-I and LR-CF models and the AUC assessment accuracy of the five models is respectively 0.762, 0.756, 0.788, 0.838 and 0.836. The LR-I and LR-CF models can improve evaluation accuracy approximately 8% compared to I and CF models and approximately equal to5% compared to LR model, proving that the LR-I and LR-CF are better than I, CF and LR for predicting landslide susceptibility. These models can provide reliable approaches for rapidly building up evaluation index system and regional landslide susceptibility. © 2020, Science Press. All right reserved.