Search papers, labs, and topics across Lattice.
This paper introduces GEOID-Flood, a comprehensive multi-modal benchmark dataset designed for flood segmentation, integrating bi-temporal SAR and optical imagery across 219 events in 65 countries over a decade. The dataset, featuring over 14,000 tiles with validated labels, allows for a robust evaluation of foundation models against traditional encoders in various protocols. Key findings reveal that while foundation models provide a modest advantage, the optimal performance in transient flooding scenarios is achieved through optical-SAR fusion with finetuning, and models trained on GEOID-Flood demonstrate superior transferability to unseen events compared to those trained on existing datasets.
Foundation models trained on the new GEOID-Flood dataset outperform traditional encoders in flood segmentation, especially when leveraging optical-SAR fusion.
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.
