Landslide susceptibility mapping with GIS and Comparison of logistic regression, frequency ratio and AHP models in. Case study: the watershed Kurdistan, Chashmidar

The aim of this research is to evaluate the efficiency of a statistical model a logistic regression and a probabilistic model- a frequency ratio - in landslide hazard mapping in Chashmidar, Kurdistan. Firstly with study of 60 landslide point cells and 60no-landslide point cells in the study area using field works, local interviews and the literature review, 17 effective factors including the following; slope gradient, slope aspect, slope curvature, plan curvature, profile curvature, elevation above the sea, distance from road, distance from faults, distance from stream, fault density, drainage density, lithology, stream power index, Sediment transport capacity index, wetness index, precipitation and land use are recognized on occurring landslides. Logistic regression was conducted by SPSS 17. Result showed that in the landslide susceptibility map by logistic regression the percentage of areas of very low zone, low, moderate, high and very high zone are 44.57%, 11.48%, 12.51%, 14.38% respectively. The landslide susceptibility map by frequency ratio ,the percentage of areas of very low zone, low, moderate , high and very high zone are 37.15%, 33.40%, 17.55%, 9.11% and 3% respectively. Results also in comparing tow methods showed that the performance of both models are reasonably well for landslide susceptibility maps; however, the logistic regression model had an accuracy of 82.2 percent of the area under the curve (AUC) in ROC in comparing with the accuracy of 80 percent by the probabilistic frequency ratio. Therefore, the logistic regression model can be used to produce landslide susceptibility map in the study area.