GIS-based evolution and comparisons of landslide susceptibility mapping of the East Sikkim Himalaya

ABSTRACT The main sought of this study is to assess the landslide susceptibility map (LSM) of the East Sikkim Himalaya based on the comparative model analysis using frequency ratio (FR), logistic regression (LR), random forest (RF) and integration of analytical hierarchy process (AHP) with FR (AHP-FR). The models were trained by 166 landslides (70% training) and 12 landslide causative factors whilst tested with the help of 71 landslides (30% testing). Their spatial correlation between the landslides and the causative factors was analysed by using a multicollinearity test. The generated LSM was classified into five classes, i.e. very low, low, moderate, high and very high. In East Sikkim, very high classes of the AHP-FR, LR and FR models cover the area of 11.97%, 11.99% and 7.13%, respectively. The accuracy of prepared LSM was evaluated by using the success rate curve (SRC), prediction rate curve (PRC) and seed calculation area index (SCAI). The area under the curve (AUC) of the success rate curve is 0.88 for the RF model, 0.85 for AHP-FR, 0.78 for LR and 0.79 for FR, whilst the prediction rate curve is 0.86% for the RF model, 0.81 for AHP-FR, 0.79 for LR and 0.78 for FR. The SCAI values of very high susceptibility classes are 0.14, 0.17, 0.18 and 0.19 for the RF, AHP-FR, LR and FR models, respectively. The RF and integrated AHP-FR methods indicate better results as compared to other statistical models.