Comparison of landslide susceptibility assessment models in Zhenkang County, Yunnan Province, China

Accurate landslide susceptibility evaluation is of great significance for disaster prevention and mitigation. In order to improve the accuracy of landslide susceptibility evaluation, based on geographic information system(GIS) platform, 150 landslide disaster points in Zhenkang County were converted into raster data as evaluation samples, and 12 evaluation factors including elevation, gradient, slope direction, relief amplitude, topographic curvature, profile curvature, stratum, fault, annual rainfall, river, land use type and road were selected and passed the independence test to construct the evaluation index system of landslide susceptibility in the study area. Using GIS to randomly extract 70% of landslide rasters as training samples, the single evaluation model(normalized frequency ratio(NFR), information(I), certainty factor(CF)) and the coupled evaluation model (normalized frequency ratio-logistic regression(NFR-LR), information-Logistic regression(I-LR), certainty factor-logistic regression(CF-LR)) were adopted to evaluate landslide susceptibility. The frequency ratio of the remaining 30% landslides was analyzed. The AUC value is used to express the evaluation success rate and the prediction rate for accuracy testing. The results show thatthe frequency ratio of high and extremely high susceptibility is more than 86% of the totle and both the success rate and the prediction rate are greater than 0.75. Compared with the single model, the models with NFR, I and CF respectively coupled with LR have higher success rate and prediction rate, which shows that the coupled LR model has higher evaluation accuracy than the single model. © 2022, Science Press. All right reserved.