Landslide Susceptibility Modelling Using Statistical Approaches in Mantsa River Catchment, Dawro Zone, South Western Ethiopia

Abstract

This study focuses on modelling landslide susceptibility in the Mantsa river catchment Dawro Zone, southwestern Ethiopia, utilizing statistical approaches, specifically frequency ratio and logistic regression. A comprehensive landslide inventory map was created through google earth images and field surveys, identifying 280 landslide locations, which were divided into a training set (70% or 196 landslides) and a validation set (30%, or 84 landslides). Thirteen possible predictive factors were identified for the susceptibility maps, including slope, elevation, and curvature, and aspect, proximity to stream, roads, lineament land use land cover, soil type, lithological units, rainfall, stream power index and topographic wetness index. These factors were mapped using various data sources, such as digital elevation models (DEM), geological maps, and rainfall data, all transformed into raster format for analysis. The study employed both FR and LR models to explore significant correlations between landslide occurrences and the identified factors. Results indicated that lulc, lithology, rainfall, and proximity to lineaments, have more landslide susceptibility. The resulting susceptibility maps showed that an extensive portion of the study area is prone to landslide. Validation via the ROC curve demonstrated that the area under the curve (AUC) for the FR and LR models were 87.4% and 85%, respectively indicating that the FR model offers slightly better predictive accuracy. Both models revealed estimable accuracy in predicting landslide susceptibility. The findings provide valuable understandings for landslide susceptibility for effective planning tools, facilitating practical measures to address potential landslide hazard and enhance environmental protection efforts in the region