Landslide Susceptibility Mapping Using Statistical Approach: A Case study in Burji - Maddale Area, Southern Ethiopia

Abstract

Landslides are one of the most widespread geohazards in hilly and mountainous terrains, which often cause huge injury and loss of human life and damage to property. The present study was conducted in the Burji to Maddale area. The main objective of this study is to prepare a landslide susceptibility map for the Burji to Maddale area using the Information value and logistic regression method. A comprehensive landslide inventory of 265 landslides were created from the study area using Google Earth image interpretation. To develop the model and validate, the landside inventory map was randomly split into two sets i.e., 70% for training and 30% was used for testing. Eight landslide causative factors, including slope, aspect, lithology, distance to stream, distance to lineament, rainfall, curvature, and land use/land cover, were integrated with training landslide to determine the weight of each landslide causative factor using the models. The landslide susceptibility index maps were then produced by summing the weights of all the landslide factors using the raster calculator of the spatial analyst tool in ArcGIS. The final landslide susceptibility map was reclassified into very low, low, moderate, high, very high in both models based on the natural break method. The information value model indicates that the landslide susceptibility map was divided into very low, low, moderate, high and very high susceptibility classes accounting for 14.5%, 15.56%, 19.74%, 19.71% and 30.37% of the area respectively, similarly, the logistic regression model result accounts for, 19.90%, 7.98%, 16.04%, 20.40% and 35.65% of the area respectively. The relative landslide density index (R-index), landslide susceptibitlity index(LDI) and area under the curve (AUC) of the receiver operating characteristic curves were calculated using the training and testing of landslide data sets in order to assess the performance of the information value and logistic regression models for landslide susceptibility modelling. The information value model has a success rate of 88.6% and a prediction rate of 86.4%, whereas the logistic regression model has a success rate of 89.4% and a predictive rate of 87.1%.