AI summaryⓘ
The authors created a big dataset called GEOID-Flood to help test how well computer models can find flood areas using satellite images taken before and after floods. Their dataset includes images from radar and optical satellites along with elevation data and carefully checked flood labels from many flood events worldwide. They used this dataset to compare new foundation models with older methods and found that the new models do a bit better, especially when they combine radar and optical data and learn from flood changes over time. Also, models trained on this dataset work better on new flood events than those trained on older datasets. This helps understand how to improve flood detection from satellite data.
Geospatial foundation modelsFlood mappingSynthetic Aperture Radar (SAR)Optical imagerySentinel-1Sentinel-2Digital Elevation Model (DEM)Multi-modal data fusionBenchmark datasetTransfer learning
Authors
Gaetano Chiriaco, Luca Barco, Andrea Bragagnolo, Claudio Rossi, Edoardo Arnaudo
Abstract
Geospatial foundation models aim to learn representations that transfer across regions and sensors, yet evaluating them on specific tasks requires large, high-quality, multi-modal benchmarks that measure how well such models extract value from data. Concerning flood mapping, existing datasets rarely combine bi-temporal SAR and co-registered optical imagery at scale, leaving the value of foundation models for this downstream task largely untested. We introduce GEOID-Flood, a large-scale multi-modal flood segmentation benchmark, derived from Copernicus Emergency Management Service activations, spanning 219 events across 65 countries over ten years. The dataset provides more than 14,000 tiles with co-registered pre- and post-event Sentinel-1, in GRD and RTC format, pre-event Sentinel-2 composite, and DEM, including manually validated labels that separate background from permanent water and flooded water. Using this benchmark, we evaluate foundation models against conventional encoders across single-image, multi-temporal, and multi-modal protocols. We report three main findings: foundation models offer a consistent but modest advantage; optical-SAR fusion with finetuning best resolves transient flooding; and models trained on GEOID-Flood transfer to unseen events better than those trained on existing datasets. Dataset and code available at https://github.com/links-ads/geoid-flood.