A Machine-Learning-Enhanced Geospatial Framework for Sustainable and Disaster-Resilient Infrastructure: Multi-Hazard Societal Impact Assessment in Sudan

Abstract

Sudan faces riverine flooding along the Blue Nile and chronic drought across Darfur–Kordofan, yet no national assessment integrates both hazards with social vulnerability to support sustainable and climate-resilient development. This study develops a Societal Impact Index (SII) for Sudan’s eighteen states using a terrain-based flood-susceptibility surface, a drought-frequency indicator (SPEI-12), and thirteen social-vulnerability indicators. These are combined into four weighted pillars following the Intergovernmental Panel on Climate Change (IPCC) risk architecture and validated against independent humanitarian-needs assessments, with convergent checks based on displacement and malnutrition. An unsupervised machine-learning audit, combining k-means clustering with principal component analysis, tests whether the data’s structure supports the composite ranking. The audit shows that the five High-impact states follow two distinct pathways: hazard-and-exposure dominance in Al Qadarif and Al Jazirah, and sensitivity dominance in the remaining three Darfur states. This distinction enables risk-reduction and infrastructure measures to be tailored to the dominant pathway in each state. The first two principal components correlate only weakly with the SII (r=0.02 and r=0.36), indicating that the ranking reflects the assigned weights as well as the data structure. Each score is exactly decomposed into pillar contributions, improving transparency, while a prototype scenario tool illustrates practical use. Flood-exposed population increased by 7.4 percent between 2017 and 2020, highlighting the need for continuous updating. The reproducible, open-data framework can support equitable resource allocation, sustainable infrastructure planning, long-term vulnerability reduction, and future disaster-resilience digital twins in data-scarce Sahelian settings.