Mapping Disaster Resilience: GeoAI Best Practices from the UN-SPIDER Network
Second Edition 2026
Geospatial Artificial Intelligence (GeoAI), the convergence of satellite-derived data with modern machine-learning techniques, has become a decisive factor in disaster-risk governance. The active Earth-observation fleet continues to expand, and open models now convert large volumes of imagery into decision-ready layers in minutes rather than days. This Compendium consolidates proven GeoAI practice from across the UN-SPIDER Regional Support Office (RSO) network into a single practical reference, documenting the methods, data, partnerships and measurable outcomes behind each case so that other offices and national agencies can replicate or adapt them. Its scope spans the end-to-end workflow, from data acquisition and model development through validation, deployment and governance, and covers rapid-onset hazards such as wildfire and flood alongside slow-onset stresses including landslide, coastal erosion and water scarcity.
For disaster risk reduction and emergency response, the significance of GeoAI lies less in any single technical advance than in who can now use it. Capability that was once the preserve of the largest space agencies is increasingly available to offices with modest resources, open data and sponsored cloud credits. Hazards can be detected in near-real time, their impacts forecast in advance, and policy responses grounded in defensible evidence rather than retrospective estimate. Crucially, the practices in this edition show that computationally efficient and transparent methods can match the performance of far heavier models while remaining reproducible in data-scarce settings, which is precisely what allows GeoAI to serve the least-resourced and most hazard-exposed States that lie at the centre of the UN-SPIDER mandate.
Highlights of RSO Contributions
The case studies in this edition illustrate the breadth of operational GeoAI now in use across the network:
- Wildfire fuel mapping (Cyprus). A Random Forest framework in Google Earth Engine fused Sentinel-1 and Sentinel-2 data with terrain and vegetation variables to produce fuel-model maps at 10 to 30 metre resolution, a substantial improvement over the coarser regional datasets previously available for local fire-behaviour planning.
- Flood susceptibility in data-scarce basins (Cyprus). Comparing four machine-learning models across eight watersheds, the work showed that a simplified Random Forest model using only land use, slope, elevation and flow accumulation achieved around 95 per cent agreement with flood-inventory data, offering authorities a rapid, low-cost screening tool in place of fully detailed hydrodynamic models.
- AI-enhanced landslide susceptibility (Cyprus). Climate Hazards Group InfraRed Precipitation with Station (data-CHIRPS) were fused with sparse rain-gauge observations using machine learning, then combined with geomorphological factors in a multi-criteria framework, producing high-resolution susceptibility maps that captured more than 30 per cent of known landslide occurrences within the highest-risk classes.
- Coastal erosion monitoring (Cyprus). An unsupervised Pulse Coupled Neural Network extracted shorelines from Sentinel-2 imagery (validated to an RMSE of 9.21 metres) and quantified the downdrift impact of a coastal breakwater, establishing a repeatable, low-cost workflow transferable to other Mediterranean coasts and small island developing States.
- Water reuse planning across the Middle East and North Africa (MENA) region. The Water-REPEAT framework combined Earth Observation, computer vision and large language models to map wastewater infrastructure and demand across Egypt, Saudi Arabia and the United Arab Emirates. In Egypt alone, AI-based detection identified approximately 164 treatment plants beyond the 552 already documented, a thirty per cent increase in known infrastructure, supporting evidence-based, climate-resilient water planning.
- Coastal water-quality assessment (Cyprus). A Random Forest classifier applied to Sentinel-2 imagery in Google Earth Engine generated turbidity-probability maps along the coastline, enabling near-real-time detection of rainfall-driven water-quality changes for environmental monitoring and coastal management.
