Improved Landslide Susceptibility mapping using statistical MLR model

In this study, the accuracy of the Landslide Susceptibility Maps (LSMs), prepared using the GIS-based statistical multiple linear regression (MLR) model, is improved by incorporating time-dependent factors. The LSM is prepared to identify landslide-prone areas based on the 108 training landslide points and 10 causative factors. The accuracy is evaluated with the area under the curve (AUC) using 56 testing landslide points. The results showed that landslides were more likely to occur at closer distances to the road, in areas having severely fragmented and jointed strata. The prepared LSM shows that, when annual mean NDVI (Normalized Difference Vegetation Index) is considered as a causative factor, areas with dense and sparse vegetation were less likely to experience landslides, whereas locations that were classified as barren land were more prone to landslides. The accuracy (AUC) of the LSM produced using the MLR model was 86% when NDVI is not considered and it increases to 92% when NDVI is considered as the causative factor. We, therefore, suggest using mean NDVI as a landslide causative factor in the landslide predictive model while working in dense vegetation and tropical areas.