Identifying landslide prone slopes and classification of its types using logistic regression model and fuzzy logic(Case study: Ghahremanlou Catchment, North Khorasan Province)

Landslides are among the most destructive natural phenomena, and their harmful effects have also increased as a result of increased human occupation and activities in nature. When the resistance force is less than the force of the material weight on the slope due to the sheer force of the soil, landslides would occur with the least intervention of secondary factors including heavy rain, earthquake, etc. Due to the special conditions of geology, climate, tectonics, etc., many landslides occurred in Ghahremanlou catchment area located in North Khorasan province, resulted in a lot of damage to the farmlands, communication lines and rural settlements.The study aimed to identify landslide-prone slopes and classify landslide types in Ghahremanlou catchment using a logistic regression model and fuzzy logic to investigate and identify the factors affecting landslides in this catchment and to draw a landslide sensitivity zoning map. Landslide mapping and identification map can be used as an efficient method in planning to reduce the risks and damages caused by landslides.Ghahremanlou catchment area is located in the southwest of Farooj city, located between longitudes 57 degrees and 55 minutes to 58 degrees and 4 minutes east and between latitudes 37 degrees and 2 minutes to 37 degrees and 13 minutes.The villages of Mayvan, Chokanlu, Ostad, Khosravieh, Hasht Sorkh, and Ghahremanlou are located within this catchment.The area is located between the geological formations of Kopehdagh in the north and Aladagh in the south and has Sandstone, granite and shale rocks. The slopes are composed of Orbitolina limestones from the Cretaceous geological period (with the Tirgan Formation) and red marls and Neogene sandstones. In this study, landslides in the area were determined by field surveys and Google Earth images. Then, using a digital model of a 30 m altitude SRTM sensor, slope map, slope direction, topography, and hydrographic network of the catchment were calculated.A road network map has been prepared from OSM and Google Earth data by digitizing the main road lines and side roads in the study area. The average annual rainfall map is extracted from meteorological data of North Khorasan province and its map is prepared based on the IDW interpolation model. To determine the vegetation of the region, the normalized vegetation difference index algorithm was used. Using the Wetness index, from the Landsat 8 satellite image for 1399, the OLI sensor of the humidity index of Ghahramanloo catchment has been calculated.Then, a fuzzy model and logistic regression were used to determine the effect of each of these parameters and effective criteria in landslides.In this study, 28 slopes in which landslides occurred were identified. Then 12 independent variables have been calculated including slope, slope direction, etc. After preparation using the fuzzy logic model (linear normalization functions (straight and inverse), norm and Gaussian between zero and one) entered the logistic regression model. Then the position of landslides in the region as a dependent variable, zero or one (boolean) was prepared and entered into the logistic regression model. After introducing independent variables related to the logistic regression model, 70% of the pixels with a landslide (in 28 identified positions) were introduced to the model as a training sample and 30% of it as a model for checking the accuracy of the model in the Pseudo index. R2 and ROC were used. The results are obtained by identifying areas with landslide potential using the Natural Breaks function and considering the same variance and scatter of data in each class. These results show that the most effective causes of landslides in the region are distance from the waterway, land use, rainfall, and distance from the road.On the other hand, the farther they are from the communication lines, the lower the impact on landslides. Also, in the land-use variable (according to how it is normalized), barren lands have less effect than rangelands, and also increasing rainfall has a positive effect on the occurrence of landslides in the catchment. In a way that with increasing rainfall, the occurrence of landslides also increases. To evaluate the accuracy of the regression model in this research, the value of Pseudo R2 equal to 0.2311 and ROC equal to 0.9134 indicates a good fit of logistic regression and its appropriate descriptive capability. Based on the landslide classification models by Kruden and Warrens, 1996, the prediction of landslides is classified using a logistic regression model, and landslides in the region into rotational, shallow, and transitional types.The results of the existing landslides indicates that out of the total area of 10241 hectares in Ghahremanlou catchment, 310 hectares, i.e., 3%, and 216.3 hectares, i.e., 2.11%, have very high landslide potential, respectively. And 950.50 hectares of the area, i.e., 9.28%, has a medium landslide potential. On the other hand, 66.5% of the catchment is located in the western direction (west, northwest, and southwest) due to its location at high altitudes and steep slopes, with very little potential for landslides.