Identification the most important predictors in landslide susceptibility mapping using Maximum Entropy Model

Slope instability and landslides are important hazards to human activities that often result in the loss of economic resources, property damage and facilities. These hazards occur in the natural or man-made slopes. In the current study, the maximum entropy model was used which is one of the progressive data mining models, in order to modelling landslide susceptibility map for Gorganrood watershed. In the first step, the landslide inventory map was prepared consiste of 351 landslides. 18 geo-environmental factors were selected as predictors, such as: Digital elevation model, slope percent, aspect, distance from fault, distance from river, distance from road, rainfall, landuse, drainage density, lithology, soil texture, plan curvature, profil curvature, lithological formation, Topographic wetness index, LS factor, stream power index, Relative Slope Position and Surface roughness index. Three different sample data sets (S1, S2, and S3) including 70% for training and 30% for validation were randomly prepared to evaluate the robustness of the model. The accuracy of the predictive model was evaluated by drawing receiver operating characteristic (ROC) curves and by calculating the area under the ROC curve (AUC). The ME model performed excellently both in the degree of fitting and in predictive performance (AUC values well above 0.8), which resulted in accurate predictions. Furthermore, In this study the importance of variables was evaluated by the model. Dem (digital elevation model) (32.4% importance), lithology (22.9% importance) and distance from fault (14.8% importance) were identified respectively the main controlling factor among all other variables.