Floods are the most frequently recorded type of disaster event. Between 2000 and 2019 the UN Office for Disaster Risk Reduction counted 3,254 major floods worldwide, more than twice the 1,389 recorded in the two decades before. Each one raises the same operational question, and it has to be answered while the water is still moving rather than in hindsight: which areas are under water right now? Global Flood Monitoring (GFM), a component of the Copernicus Emergency Management Service coordinated by the Joint Research Centre of the European Commission, exists to answer that question continuously. For every flood, everywhere, without anyone having to ask for it.
Fully automated, worldwide
GFM has no manual trigger. Every Sentinel-1 scene that comes down is picked up and processed automatically, anywhere in the world, 24/7, with whole dataset of flood extent available roughly six hours after the scene arrives at the processing system. There is no activation request, no analyst in the loop, no waiting for office hours.
Synthetic Aperture Radar is what makes this possible. Optical sensors depend on daylight and a clear line of sight to the ground, and the meteorological conditions that produce a flood are precisely the ones that put cloud in the way. Radar is independent of both illumination and cloud cover, which is why a radar-based service can observe an event while it is still developing rather than after the sky has cleared.
An ensemble, not a single classifier
Flood detection in SAR imagery is not a solved problem. Water usually appears dark because a smooth surface reflects the radar pulse away from the sensor, but so do dry sand, asphalt, and radar shadow behind terrain; wind-roughened water can look bright; and in dense urban fabric or forest the signal is dominated by the structures above the ground. GFM therefore does not rely on one classifier. Three independently developed flood-mapping algorithms contribute to the product, and auxiliary layers. A reference water mask separating permanent and seasonal water from genuine flooding, and an exclusion mask marking areas where SAR-based detection is not reliable, are delivered alongside the flood extent so that users can judge what they are looking at.
Validated against 55,212 reference points
Between December 2022 and February 2023, the service was validated against a stratified sample drawn from an archive of more than 4.2 million processed Sentinel-1 scenes. Of over 87 million candidate points, 55,212 were selected and distributed across all continents, each one a specific location on a specific date. Each point was inspected manually against Sentinel-1, Sentinel-2 and very high resolution imagery within a window of ten days either side of the reference date, and classified as permanent water, seasonal water, flood, or other. The result was an overall accuracy of 95.51 %, with a 95 % confidence interval of ±0.18 %. GeoVille developed the browser-based GFM Validator used to carry out this assessment, and it now serves as the baseline against which future product versions are measured.
GeoVille's role in the consortium
GFM is delivered by a consortium led by the Earth Observation Data Centre (EODC), with GeoVille, TU Wien, the Luxembourg Institute of Science and Technology, CIMA Foundation and the German Aerospace Center (DLR).
GeoVille operates the cloud and High Performance Computing operating layer, including process monitoring, the data access and dissemination system with its Web Map Service, and quality assurance. In a service with no manual trigger, that infrastructure is not a supporting detail: a flood map that is correct but arrives late, or that cannot be reached when a national authority needs it, has no operational value. Continuous monitoring of the processing chain and of product quality is what makes the one-hour figure a service commitment rather than a best case.
Open access
GFM products are free and openly available through the Global Flood Awareness System portal and the dedicated GFM web application, which also issues notifications for user-defined areas of interest. The user community includes national civil protection and water authorities, humanitarian organisations, research institutions and commercial downstream services.
Find the numbers here: https://www.undrr.org/news/un-report-dramatic-rise-climate-disaster-over-last-twenty-years
Contact us!