Logistic Regression Assessment in the Investigation of the Landslide Potential (Case Study: From Nasirabad to Sattar Khan Dam)

IntroductionIranian territory has the main prerequisites for the occurrence of a wide range of landslides due to its mountainous topography, tectonic activities, high seismicity, and different geological and climatic conditions. Therefore, reducing the effects of natural disasters, particularly landslides, is one of the key challenges for land-use planners and policymakers in this field. In this study, the southern side of the Ahar Chai basin from Nasirabad Village to Sattarkhan Dam is evaluated for the probability of the landslide occurrence. This region is highly susceptible to landslide occurrence because of the extensive manipulation and its natural conditions. Indeed, the occurrence of the large shallow landslides in this region is an indication of this susceptibility. In this study, Linear Regression Model has been used to prepare the landslide zonation.
MethodologyThe study area was the southern sides of the Ahar Chai River, from Nasirabad village in Varzaghan to the Sattarkhan Dam, with an area of 128 km2. In order to study the potential of the landslide occurrence in this region, nine main factors including slope, slope direction, lithology, land use, precipitation, distance from the fault, distance from the river, distance from the road, and vegetation were identified. The model which was used in this study was Logistic Regression. This model is one of the predictive statistical methods for dependent variables in which zero and one respectively indicate the occurrence and non-occurrence of landslides. In addition, instead of being linear, the regression of the variables is S-shaped or logistic curve and the estimations are in the range of zero-one. Indeed, values close to zero indicate the low probability of the occurrence and values close to one indicate the high probability of the occurrence.
DiscussionIn Logistic Regression model, after entering the data into the Logistic Regression model and using the effective parameters in Idrisi software, the coefficients of the model were extracted. A value of 965, which represents a very high correlation between the independent and dependent variables, was obtained for the ROC index. After determining the validity of the Logistic Regression model, using the above indicators, landslide sensitivity zonation map was prepared. In the present model, the land use factor with the highest coefficient was the best predictive variable in determining the probability of the landslide occurrence in this region. In addition, the SPI index and the distance from the fault had respectively the second and third highest coefficients. After zoning the landslide, the slip area was calculated for each class and its results showed that zones with highest risk had the lowest area percentage and these areas were located in the western slopes.
ConclusionThe results showed that while land use, lithology factors, and SPI index with positive coefficients had higher correlation, the other factors with negative coefficients had lower correlation. Based on the map, the western, southern, and the north-eastern parts of the region have the highest potential for landslide occurrence. Furthermore, the high value of the ROC index and its proximity to number one indicates that landslides in the study area have a strong correlation with the probability values derived from the Logistic Regression Model. In addition, the assessment of the SCAI scaling hazard zonation map shows that there is a high correlation between the hazard map with the existing slip points and the field observations of the area. It can be said that, in addition to the natural factors, some human factors including unstructured road construction may play an important role in the occurrence of the landslides. It is also necessary to avoid making changes in the ecosystems and land use. Finally, any policies to construct structures should be commensurate with the geomorphologic and geological conditions.