Geospatial assessment of landslide susceptibility map using information value and weight of evidence methods in the Sego catchment area, Southern Ethiopia

This study evaluates landslide susceptibility in the Sego catchment, southern Ethiopia, by integrating a detailed landslide inventory with bivariate statistical modelling in a GIS environment. This study addresses a key gap in Ethiopian Rift catchments by providing the first quantitatively validated landslide susceptibility assessment for the Sego catchment, and by explicitly comparing Information Value (IV) and Weight of Evidence (WoE) using a high-resolution inventory and multi-source conditioning factors in a data-scarce setting. Landslide locations (125 events) were identified through field surveys and visual interpretation of high-resolution Google Earth imagery; 94 landslides (75%) were used for model training and 31 (25%) for independent validation. Nine conditioning factors—elevation, slope, aspect, curvature, land use/land cover, proximity to streams, normalized difference vegetation index (NDVI), lithology and rainfall—were derived from a digital elevation model, Landsat-8 OLI and Sentinel-2 MSI imagery, a 1:25,000 geological map and PERSIANN (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks) rainfall data. Landslide susceptibility maps were produced using two bivariate statistical methods, Information Value (IV) and Weight of Evidence (WoE), by computing class-based weights for each factor and combining them in ArcGIS to obtain a landslide susceptibility index. The landslide-susceptible map, produced using information values, is categorized into five levels in the study area: very low, low, moderate, high, and very high. The areal coverage is 13.5%, 24.7%, 26.5%, 23.8%, and 11.5%, respectively, and 103.62 km2 of the study area falls in the very high and high susceptible zones. By employing the weight-evidence method, the landslide-susceptible map resulted in the following categories: very low (16.5%), low (27.8%), moderate (25.3%), high (17.3%), and very high (13.1%). 88.88 km2 falls in the very high and high susceptible zones. The validation results demonstrated that the weight-of-evidence model yielded better results than the information-value model, with area under the curve values of 80.8% and 75.4%, respectively. These AUC values indicate fair-to-good discriminatory performance, meaning the models can reliably distinguish more landslide-prone slopes from relatively stable areas, with WoE providing the stronger screening map. The susceptibility maps delineate priority zones for land-use planning and disaster-risk management (safer siting of roads/settlements, prioritised field inspections, and targeted drainage/toe-protection measures) to support kebele- and woreda-level risk reduction and preparedness. Key uncertainties relate to potential incompleteness/temporal representativeness of the mapped landslide inventory, the spatial resolution/quality of some inputs (notably satellite rainfall), and the conditional-independence/linearity assumptions of bivariate models.