Comparison of artificial neural network and multivariate regression methods in landslide hazard zonation, case study: Vanak Basin, Isfahan province

Landslides are major natural hazards which not only cause damages to human life but also provide economic losses on infrastructures. In order to determination of the most important method of estimation recognizing appropriate method to estimate landslide, in this research, the efficiency of two methods of landslide hazard zonation including methods of Artificial Neural Network and Multivariate regression were compared. Therefore, in this research, first, landslide inventory map was obtained using aerial photos interpretation, satellite images processing, geology maps review and field surveying. Also, the 9 important effective factors are in occurrence of landslide including lithology, land use, slope angle, slope aspect, elevation, precipitation, distance to fault, distance to road, density of drainage were determined using inspect of field and literature review. After producing of layers and weighting to effective factors using inventory map, landslide hazard zonation was made by Artificial Neural Network and Multivariate regression models. From 200 landslides identified, 140 (≈70%) locations were used for the landslide susceptibility maps, while the remaining 60 (≈30%) cases were used for the model validation. The quality sum (Qs) and precision (P) indices for Artificial Neural Network model are 0.15, 0.08 and for Multivariate regression model are 0.14, 0.05 respectively. This results show that artificial Neural Network is the better model in landslide hazard zonation in this area, therefore an accurate landslide hazard zonation map can be prepared by selecting and applying the proper method.