GIS-based comparison of the GA-LR ensemble method and statistical models at Sefiedrood Basin, Iran

Landslide is a natural phenomenon which occurs on mountainous areas and causes a lot of damages every year. Decision makers, engineers, and urban planners can mitigate or prevent loss of lives as well as potential economic losses in the future through using the landslide susceptibility map. This study aims to identify the areas susceptible to landslide occurrence in Sefidrood basin located in Alborz Mountains in the north of Iran using bivariate models, i.e., frequency ratio (FR), weights of evidence (WoE), and Dempster-Shafer theory (DST), and compare them with ensemble method of multivariate logistic regression (LR) and genetic algorithm (GA). In the first step, 265 landslide locations were identified as the inventory map. From these identified landslides, 70% (186 landslide locations) was randomly selected as the training data and the remaining 30% (79 landslide locations) was used for validation purposes. In the next step, considering the region’s condition and the experts’ opinion, twelve discrete and continuous conditioning factors were prepared including slope angle, slope aspect, altitude, distance from streams, distance from roads, distance from faults, land use, rainfall, NDVI, lithology, profile curvature, plan curvature, and rainfall. After establishing the database, the proposed models were analyzed using conditional factors and inventory map. Finally, after preparing landslide susceptibility maps, the performance of each method was investigated and compared using ROC curve. The results from validation process showed that the area under the curve (AUC) for the models of FR, WoE, DST, and GA-LR was 0.741, 0.814, 0.826, and 0.938, respectively. According to the results, GA-LR model exhibited higher accuracy in prediction of landslide susceptibility in the study area compared with other models. Therefore, the map produced by this model can help engineers, land use planners, and crisis management personnel make more accurate decisions.