Assessing landslide susceptibility using combination models

Assessing and mapping landslide susceptibility is a powerful approach to decrease the cost of landslide disasters. The aim of this paper is to design combination models by combining the Hoeffding tree and forest by penalizing attributes with MultiBoosting (MB), Random SubSpace (RS) and Rotation Forest (RF) to analyse the results of each combination model for modelling landslide susceptibilities. For this purpose, a case study was conducted in Yanshan County, Jiangxi Province, China. Then, 380 landslide polygons and eleven environmental variables were collected and processed, and they were the input data for the six combination models. The results demonstrated that the combined models performed well. Using validation data, the FPA-RF model obtained a maximum AUC value (0.794), followed by the FPA-RS (0.793), FPA-MB (0.788), VFDT-RF (0.746), VFDT-MB (0.741) and VFDT-RS (0.740) models. The FPA-RF model exhibited the most stable and accurate performance in this paper. Among the eleven environmental variables, aspect, land use, and altitude were the most important variables in the combination models. Therefore, the combination models we developed are useful tools that can decrease losses from landslide disasters.