Mapping rain-induced landslides in data-scarce conflict-affected regions using deep learning model: A study from the Jebel Marra Volcanic Massif, Sudan

Catastrophic landslides in the Jebel Marra volcanic massif of western Sudan have recently resulted in multiple fatalities and extensive destruction in Tarsin village and its surrounding agricultural lands following intense rainfall on 1 September 2025. Although these events pose a recurring threat to vulnerable communities, no prior studies have evaluated landslide susceptibility in this region. This lack of research is largely due to the prolonged armed conflict in Darfur, which has made ground-based investigations impossible since 2003. In response to these challenges, the present study provides the first comprehensive landslide susceptibility assessment for the Jebel Marra volcanic massif. The analysis integrates geospatial, geological, and geophysical data from multiple sources with deep convolutional neural networks (CNN). A landslide inventory comprising 350 events was developed using multi-temporal satellite imagery. Key conditioning factors, including topography, hydrology, structural lineaments, vegetation cover as measured by the Normalized Difference Vegetation Index (NDVI), and anthropogenic influences, were incorporated. The CNN model, trained and validated with stratified k-fold cross-validation, demonstrated higher performance (precision: 0.975, recall: 0.992, area under the curve (AUC): 1.000) than a benchmark Random Forest model. Feature importance analysis identified elevation, curvature, and lineament density as the primary controlling factors. The resulting susceptibility map delineates high-risk zones concentrated in the central highlands and along drainage corridors, representing 15 to 20 percent of the study area. These findings provide a critical evidence base for disaster risk reduction, humanitarian response, and land-use planning in this conflict-affected region, where natural hazards exacerbate existing vulnerabilities.