Abstract
The inhabited zone of the Ugandan Rwenzori Mountains is affected by landslides, frequently causing loss of life, damage to infrastructure and loss of livelihood. This area of ca. 1230 km(2) is characterized by contrasting geomorphologic, climatic and lithological patterns, resulting in different landslide types. In this study, the spatial pattern of landslide susceptibility is investigated based on an extensive field inventory constructed for five representative areas within the region (153 km(2)) and containing over 450 landslides. To achieve a reliable susceptibility assessment, the effects of (1) using different topographic data sources and spatial resolutions and (2) changing the scale of assessment by comparing local and regional susceptibility models on the susceptibility model performances are investigated using a pixel-based logistic regression approach. Topographic data are extracted from different digital elevation models (DEMs) based on radar interferometry (SRTM and TanDEM-X) and optical stereophotogrammetry (ASTER DEM). Susceptibility models using the radar-based DEMs tend to outperform the ones using the ASTER DEM. The model spatial resolution is varied between 10, 20, 30 and 90 m. The optimal resolution depends on the location of the investigated area within the region but the lowest model resolution (90 m) rarely yields the best model performances while the highest model resolution (10 m) never results in significant increases in performance compared to the 20 m resolution. Models built for the local case studies generally have similar or better performances than the regional model and better reflect site-specific controlling factors. At the regional level the effect of distinguishing landslide types between shallow and deep-seated landslides is investigated. The separation of landslide types allows us to improve model performances for the prediction of deep-seated landslides and to better understand factors influencing the occurrence of shallow landslides such as tangent curvature and total rainfall. Finally, the landslide susceptibility assessment is overlaid with a population density map in order to identify potential landslide risk hotspots, which could direct research and policy action towards reduced landslide risk in this under-researched, landslide-prone region.