Landslide Susceptibility Simulation Using Data Mining Models in Rabor Area

Landslide, as one of the most important natural hazards, causes financial losses and destruction of natural resources, every year. Rabar area, located upstream of Halilrood watershed, is prone to landslides due to the presence of marl formations and hence, a high amount of sediment has entered the Safa reservoir in Rabar city. Therefore, the purpose of this study is to zone this environmental hazard using convolutional neural network (CNN), support vector machine (SVM) and evidential belief function (EBF) models in Rabor region. To achieve this purpose, the parameters of altitude, slope, and distance from the fault, geology, land use, soil type, Normalized Difference Vegetation Index, and distance from the river were used. Then, using the data of the Geological Survey of Iran and field observations using GPS, a landslide distribution map was prepared as a dependent variable. There were 70 landslides, 49 (70%) of which were used for simulating and 21 (30%) for model validation. The results obtained from the validation of the models using the ROC showed that the AUC values for CNN, SVM and EBF models were 0.987, 0.958 and 0.899, respectively. Overall, the results showed a satisfactory correlation between the landslide data available in the area and the landslide susceptibility maps and the deep learning model of  the convolutional neural network had a higher performance compared with the other two models. Finally, the landslide susceptibility map was classified into four classes: low, medium, high, and very high susceptibility. According to the results of all models; the central, southeastern, and southwestern parts of the study area have a high and very high landslide risk. Carrying out appropriate designs such as retaining walls, preventing water infiltration, appropriate drainage, planting vegetation suitable for the environment, and landslide-prone slopes and etc can be appropriate in preventing and controlling this hazard.