关键词: 滑坡;易发性分区;ROC曲线;诱发因子;频率密度;盒维数;志丹县;陕西省;

Abstract: Landslide susceptibility mapping is an effective method to predict landslides,but the current modeling data is relatively single,moreover the modeling data may affect the results of the mapping. Therefore,it is necessary to dig deeper into the modeling data sources. In view of this,we selected 10 landslide triggering factors in Zhidan County,Yan'an City,Shaanxi Province. Based on 289 landslide samples,the frequency ratio and box counting dimension of each triggering factor were calculated,and two sets of differentiated modeling data were constructed. Then,the landslide susceptibility maps of the study area were made by using the kernel function logistic regression model. Finally,the average absolute error(MAE) and the area under the ROC curve(AUC) were used to evaluate and compare the results of landslide susceptibility mapping. The results show that using the box dimension as the modeling input data can effectively improve the accuracy of the partition results and the classification and generalization capabilities of the model,which is worthy of promotion in the study area.

Key words: landslide; susceptibility mapping; ROC curve; landslide triggering factors; frequency ratio; box counting dimension; Zhidan County; Shaanxi Province;