Assessment of Landslide Susceptibility Zoning Map Using Confidence Factor-Logistic Regression Hybrid Method By Means of Geomorphometric Indices

Landslide is a natural disaster resulted from geomorphologic, hydrologic, and geologic conditions along with human factors. As of present, many pieces of research have been performed to achieve such a purpose based on various models. However, many of such works have failed to provide satisfactory results due to failure to consider surface landform, so that limitations have been encountered when it came to the application of their findings. This highlights the necessity of using novel methods based on quantitative criteria of landform to identify the zones susceptible to landslide, so as to conduct proper planning for such zones. In this respect, morphometric parameters can represent landform characteristics of the hillsides susceptible to landslide. Moreover, in tectonically active areas, instability of the hillsides can be observed in the form of various types of landslide. This shows that, when it comes to the assessment of landslides, one should further consider neotectonic activities by means of geomorphologic indices.
In the present research, a total of 20 effective factors were used to assess accuracy of a landslide susceptibility zoning map; the factors include 14 conventional factors along with 6 geomorphometric factors. For this purpose, firstly, independence of the factors affecting the landslide was analyzed by performing Multicollinearity reach tests. To this end, two important indices, namely tolerance (TOL) and variance inflation factor (VIF), were used to undertake multi-directional reach test. These two indices are commonly used when running generalized linear models for analyzing the relationship between independent variables or conducting multi-directional reach test. Even though no consistent principle is presented for determining thresholds of these two indices (VIF and TOL) in multi-directional analysis and estimation of the factors affecting landslide, but records of the research works performed on this topic show that, multiple-correlation problem will arise (i.e. the data or independent variables are not correlated at all) if the value of VIF is smaller than 5 or 10 and the value of TOL is higher than 0.1 or 0.2. Moreover, in order to assess the correlation between the landslide and the selected factors, the weights obtained from the confidence factor (CF) model were used in the form of bivariate statistical analyses. The weights calculated by this model were further employed to prepare factorial maps and convert them into binary maps (wherein the levels with negative and positive weights were represented by 0 and 1, respectively) to be introduced into the conditional independence test. Continuing with the research, once finished with entering the weights obtained from the CF to the logistic regression model, the model coefficients were extracted. Using the obtained weights, the model was run to prepare a landslide susceptibility zoning map following either of two approaches, i.e. with and without considering geomorphologic indices along with other effective factors. Finally, Receiver Operating Characteristic (ROC) curve was used to validate, assess, and compare the maps obtained via either of the two approaches.
Based on the results of multiple correlation test, no correlation was obtained between the independent factors with VIF value of at most 3.559 and TOL values of at least 0.253. All of the values of VIF of the independent factors were lower than the critical value (5 or 10), while all of the values of TOL of the independent factors were higher than the calculated value by the critical theory. Maximum and minimum values of VIF were found to be 3.559 and 1.101, respectively, and the corresponding values to TOL were 0.253 and 0.895, respectively, which referred to the land roughness and slope direction, respectively.
The results obtained from the CF model showed that, the land roughness values exceeding 14 followed by slopes exceeding 40% possess the largest CF weights, among other levels of the considered factors, while waterway densities ranging between 0 and 10 and precipitations lower than 550 mm exhibited the lowest CF weights. According to the results obtained from the logistic regression model, the slope, NDVI, and slope direction exhibited the highest correlation coefficients, making them the best predictors of landslide occurrence in the region. Combing the weighted maps, zoning maps were prepared via the two approaches. Accordingly, 17.06% and 8.27% of the area understudy were identified as very highly susceptible to landslide, for the cases with and without considering the effects of geomorphologic factors, respectively. In addition, based on the results of ROC, the area under the curve was evaluated as 0.88 and 0.82 for the cases with and without considering the effects of geomorphologic factors, respectively. This confirms higher efficiency of the models into which geomorphologic factors are incorporated along with other common parameters considered in landslide susceptibility zoning.