Presentation of RFFR New Ensemble Model for Landslide Susceptibility Assessment in Iran

The current study is focused on landslide susceptibility mapping over a critical mountainous watershed, Chehel-Chai, located in the Golestan Province, Iran. An integrated data mining new ensemble model, comprised of random forest and frequency ratio (RFFR), was proposed and employed as a robust computational algorithm in the study area. Landslide inventory map was prepared in Geographic Information System (GIS) by using several field surveys, local information, and available organizational resources. In this study, using different literature review and data availability, 12 landslide conditioning factors including proximity from fault (PFF), proximity from stream/river (PFS), proximity from road (PFR), lithological units, soil texture, land use/land cover (LU/LC), slope degree, slope aspect, altitude, plan curvature (PlanC), profile curvature (ProfC), and topographic wetness index were chosen and the corresponding maps were produced in the ArcGIS 10.2. For modeling, the FR values were calculated and then used for implementing RF in R 3.0.2 statistical software by “randomForest” package. In order to validate the built model, the receiver operating characteristic (ROC) curve using 30% of cast-off landslide was considered. The results revealed that the RFFR new ensemble model with the AUC value of 0.831 had a good performance (AUC = 83.10%) for landslide susceptibility zonation over the study area. Based on the RFFR model, about 42.27% of the Chehel-Chai Watershed has high (24.18%) and very high (18.09%) susceptibility to landslide occurrence. Hence, the proposed new algorithm was found to be suitable for landslide susceptibility modeling in the study area and, accordingly, for land use planning and landslide hazard management.