Enhancing landslide susceptibility mapping through advanced hybridization of bootstrap aggregating based decision tree algorithms

Landslide phenomena annually cause irreparable financial and human losses, predominantly occurring in mountainous regions characterized by specific topographic and climatic conditions. Consequently, this study aims to prepare landslide susceptibility maps for the Gollojeh Watershed in Zanjan province, Iran, using a combination of the bootstrap aggregating (BA) data mining method with three algorithms: random forest (BA-RF), logistic model tree (BA-LMT), and classification and regression tree (BA-CART). Initially, 140 landslide locations were identified; 98 (70% of these locations) were randomly selected for model training and the remaining 42 (35% of these locations) were used for model validation. In the next step, 13 landslide-affecting factors including elevation, ground slope, slope aspect, plan curvature, profile curvature, topographic wetness index, stream power index, distance from faults, lithology, distance from canals, distance from roads, land use, and precipitation were considered. To quantitatively evaluate the accuracy of the hybridized models, the root mean square error (RMSE) index, along with the receiver operating characteristic (ROC) curve and the area under the ROC curve (AUC), were used. The results showed that the BA-RF algorithm demonstrated the highest prediction power, with RMSE and AUC values of 0.293 and 0.903 during the validation phases, respectively. This was followed by the BA-LMT (RMSE = 0.480 and AUC = 0.889) and BA-CART algorithms (RMSE = 0.492 and AUC = 0.847). Overall, the southern part of the study area was found to be low prone to landslides, while the middle part was more susceptible.