Today we are releasing 𝗚𝗘𝗢𝗜𝗗-𝗙𝗹𝗼𝗼𝗱, a large-scale, multi-modal benchmark dataset for flood segmentation.</p>\n<p>Most flood datasets inherit noisy labels from automated pipelines and offer limited modality or format choice. We attempted to improve on both fronts.</p>\n<p>Every tile was manually filtered and validated to select the best acquisitions, and we specifically reworked the permanent water layer rather than taking it as given, exploiting recent advances in geospatial embeddings. To make the data more usable across different contexts, splits and modalities are selectable at download time: pre/post Sentinel-1 (GRD + RTC), Sentinel-2 L2A, and DEM at 10m. A held-out, cross-dataset test set ships alongside the main split, for evaluation without having to build your own.</p>\n<p>The benchmark spans 10 years: 219 CEMS Rapid Mapping activations, 65 countries, from 2016 to 2026. Data, training/eval code, and training configs are available on GitHub, released under CC BY 4.0.</p>\n","updatedAt":"2026-08-04T12:45:14.003Z","author":{"_id":"6544b9b0fc182f71e194bf89","avatarUrl":"/avatars/afa1ddb38e30f805bff7ad4a45e85132.svg","fullname":"Gaetano Chiriaco","name":"gtano","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8959006071090698},"editors":["gtano"],"editorAvatarUrls":["/avatars/afa1ddb38e30f805bff7ad4a45e85132.svg"],"reactions":[{"reaction":"🚀","users":["edornd"],"count":1}],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.02315","authors":[{"_id":"6a7194f0ec5082b9f872cfdb","user":{"_id":"6544b9b0fc182f71e194bf89","avatarUrl":"/avatars/afa1ddb38e30f805bff7ad4a45e85132.svg","isPro":false,"fullname":"Gaetano Chiriaco","user":"gtano","type":"user","name":"gtano"},"name":"Gaetano Chiriaco","status":"claimed_verified","statusLastChangedAt":"2026-08-04T09:59:34.090Z","hidden":false},{"_id":"6a7194f0ec5082b9f872cfdc","name":"Luca Barco","hidden":false},{"_id":"6a7194f0ec5082b9f872cfdd","name":"Andrea Bragagnolo","hidden":false},{"_id":"6a7194f0ec5082b9f872cfde","name":"Claudio Rossi","hidden":false},{"_id":"6a7194f0ec5082b9f872cfdf","name":"Edoardo Arnaudo","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/6544b9b0fc182f71e194bf89/5Vq1YdNXtrAIPT5tuRf4-.png"],"publishedAt":"2026-08-03T00:00:00.000Z","submittedOnDailyAt":"2026-08-04T00:00:00.000Z","title":"GEOID-Flood: A Large-Scale Multi-Modal Benchmark Dataset for Flood Segmentation","submittedOnDailyBy":{"_id":"6544b9b0fc182f71e194bf89","avatarUrl":"/avatars/afa1ddb38e30f805bff7ad4a45e85132.svg","isPro":false,"fullname":"Gaetano Chiriaco","user":"gtano","type":"user","name":"gtano"},"summary":"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.","upvotes":1,"discussionId":"6a7194f0ec5082b9f872cfe0","projectPage":"https://huggingface.co/datasets/links-ads/geoid-flood","githubRepo":"https://github.com/links-ads/geoid-flood","githubRepoAddedBy":"user","githubStars":1,"organization":{"_id":"649e8973d07383fba8d765d6","name":"links-ads","fullname":"LINKS - AI, Data & Space","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6271277f7b9f120adb37a8b4/bbHKnXwd0-KvLyYHXbFqm.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6271277f7b9f120adb37a8b4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6271277f7b9f120adb37a8b4/oKyiyyxn_aP3MOGhUQUYb.png","isPro":false,"fullname":"Edoardo Arnaudo","user":"edornd","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"649e8973d07383fba8d765d6","name":"links-ads","fullname":"LINKS - AI, Data & Space","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6271277f7b9f120adb37a8b4/bbHKnXwd0-KvLyYHXbFqm.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.02315.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2608.02315 in a model README.md to link it from this page.

Cite arxiv.org/abs/2608.02315 in a Space README.md to link it from this page.

Add this paper to a collection to link it from this page.