FloodSENS

FloodSENS is a machine-learning flood mapping application from Luxembourg-based RSS-Hydro. It addresses a recurring gap in satellite flood monitoring: optical images captured during floods are often partly covered by cloud, which leaves flooded areas unmapped and the real impact of an event underestimated.

The model is based on the U-Net architecture and uses Sentinel-2 flood images together with derivative layers from digital elevation models relating to topography and water flow, which lets it reconstruct flooded areas below partial cloud cover. It was trained on expertly labelled Sentinel-2 flood images across different biomes, events and locations, and in internal testing it performs on average at least as well as a calibrated band-ratio index (>90% correct prediction). Outputs are flood-extent rasters in GeoTIFF, which the provider states are delivered in under 30 minutes from image availability; 3D visualisation of the results is available on request, and the provider describes local containerised deployment and options such as flood depth measurement and commercial imagery integration.

FloodSENS was developed with ESA's InCubed programme, which named the UN World Food Programme, the National Disasters Management Institute of Mozambique (INGC) and the reinsurance broker Willis Re as initial partners and testing customers. It is aimed at disaster response agencies, humanitarian organisations and the insurance and reinsurance market.