{"success":true,"data":{"author":"IBM Research, DLR, ESA Phi-lab","bucketAccessType":"public","bucketName":"ibm-esa-geospatial/ImpactMesh-Flood","bucketPrefix":"","category":"Earth Observation & Geospatial","cloudProvider":"huggingface","createdAt":"2026-11-14T14:15:15.730Z","description":"Flood subset of ImpactMesh, a multimodal, multitemporal benchmark for disaster mapping released by IBM, DLR and ESA Phi-lab. It pairs Sentinel-2 L2A (all bands), Sentinel-1 RTC SAR and Copernicus DEM with flood extent annotations from the Copernicus Emergency Management Service (EMS), each sample having four time steps: a month before, just before, during and after the event. Version 1 holds 24,578 samples over 206 flood events worldwide (19,448 train, 2,171 validation, 2,959 test), after quality cleaning to reduce label noise, and a 562-sample test_holdout of events unseen in training for measuring generalisation. The listing (measured 2026-11-14) shows train/, val/, test/ and split/ folders; each split ships tar archives per modality (S2L2A, S2RGB, S1RTC, DEM, MASK) and the repository uses about 361 GB. Untar the archives as the card shows and read the samples with rasterio or xarray. Suited to semantic segmentation of flood water and change detection with optical and radar data. Released under CC-BY-4.0, with Copernicus Sentinel attribution.","displayTitle":"ImpactMesh-Flood: Multimodal Multitemporal Flood Mapping Dataset","id":"a0147bd5-c7b6-553a-a693-7e7c0bb22149","license":"CC-BY-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/ibm-esa-geospatial/ImpactMesh-Flood","objectCountEstimate":22,"probeStatus":"ok","referenceUrl":"https://github.com/IBM/ImpactMesh","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"ibm-esa-geospatial/ImpactMesh-Flood\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"ibm-esa-geospatial/ImpactMesh-Flood\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=ibm-esa-geospatial, license:cc-by-4.0, size_categories:10K<n<100K, downloads=873"],"format":"mixed","generatedAt":"2026-11-14T14:15:17.471Z","layout":{"example":"test/DEM.tar","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"44454d2f454d53523230305f31315f35364a4e505f783533383937355f79363834323936355f44454d2e74696600000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000303030303634340032353331343337003232343537353300303030303030333631313200313532303132353231373100303137373635002030000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000007573746172202000626c75000000000000000000000000000000000000000000000000000000000070726f6a5f696d706163746d6573680000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000","encoding":"hex","key":"test/DEM.tar","truncatedAt":512}],"sampleObjects":["split/impactmesh_flood_test.txt","split/impactmesh_flood_test_holdout.txt","split/impactmesh_flood_test_seen_events.txt","split/impactmesh_flood_train.txt","split/impactmesh_flood_val.txt"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\npretty_name: ImpactMesh-Flood\nsize_categories:\n- 10K<n<100K\nviewer: false\ntask_categories:\n- image-feature-extraction\n---\n\n[![arXiv](https://img.shields.io/badge/arXiv-comming_soon-b31b1b?logo=arxiv)](https://arxiv.org/abs/todo)\n[![Code](https://img.shields.io/badge/GitHub-ImpactMesh-EE4B2B?logo=github)](https://github.com/IBM/ImpactMesh)\n[![IBMblog](https://img.shields.io/badge/Blog-IBM-0F62FE)](https://research.ibm.com/blog/geospatial-flood-and-wildfire-mapping)\n\n\n# ImpactMesh-Flood\n\nImpactMesh is a large-scale multimodal, multitemporal dataset for flood and wildfire mapping, released by IBM, DLR, and the ESA Φ-lab. \nIt integrates **Sentinel-1 SAR**, **Sentinel-2 optical**, **Copernicus DEM**, and high-quality annotations from Copernicus EMS.\nThe technical report is released soon. You find the wildfire subset here: https://huggingface.co/datasets/ibm-esa-geospatial/ImpactMesh-Fire.\n\n![events_world](https://github.com/IBM/ImpactMesh/raw/main/assets/events_world_light.png)\n\n---\n## Features\n- Multimodal: SAR, optical, DEM\n- Multitemporal: Four time steps (pre-month, pre-event, event, post-event)\n- Global coverage: 200+ flood events\n- Scale: 24K samples\n- License: CC-BY 4.0\n\n## Modalities\n\n| File | Content |\n|---|---|\n| `S2L2A.tar` | Sentinel-2 Level-2A, all bands, four time steps |\n| `S2RGB.tar` | Sentinel-2 RGB quicklook of the event time step |\n| `S1RTC.tar` | Sentinel-1 RTC SAR, four time steps |\n| `DEM.tar` | Copernicus DEM |\n| `MASK.tar` | Flood annotations from Copernicus EMS |\n\n## Splits\n\n| Split | Samples |\n|---|---|\n| train | 19,448 |\n| val | 2,171 |\n| test | 2,959 |\n| **total** | **24,578** |\n\nCovering 206 flood events.\n\n`test_holdout` (562 samples) is a subset of `test`, listing the samples whose\nevent does not appear in `train` or `val`. Use it to evaluate generalisation to\nentirely unseen events.\n\n## Versions\n\nThe current release is **v1**, which we generally recommend using.\n\nCompared to the initial v0, we performed extensive quality cleaning to improve\ntraining and reduce label noise. This removed a substantial number of noisy\nsamples, so v1 is smaller but of higher quality. The v0 version is still\navailable on request, please open a discussion in the Community tab.\n\n## Quick Start\n\nDownload the dataset:\n```shell\nhf download ibm-esa-geospatial/ImpactMesh-Flood --repo-type dataset --local-dir data/ImpactMesh-Flood\n\n# Only download a single modality (e.g., S2L2A)\nhf download ibm-esa-geospatial/ImpactMesh-Flood --repo-type dataset --include \"*/S2L2A.tar\" --local-dir data/ImpactMesh-Flood\n\n# Only download a single split (e.g., validation)\nhf download ibm-esa-geospatial/ImpactMesh-Flood --repo-type dataset --include \"val/*\" --local-dir data/ImpactMesh-Flood\n```\n\nUntar the samples:\n```shell\nmkdir data/ImpactMesh-Flood/data\nfor f in data/ImpactMesh-Flood/*/*.tar; do\n  echo \"Extracting $f\"\n  tar -xf \"$f\" -C data/ImpactMesh-Flood/data\ndone\n```\n\nThe samples from all splits are saved in shared folders `data/ImpactMesh-Flood/data/{modality}`. After extracting, you can delete the tars:\n```shell\nrm -r data/ImpactMesh-Flood/train\nrm -r data/ImpactMesh-Flood/val\nrm -r data/ImpactMesh-Flood/test\n```\n\nWe use [TerraTorch](https://terrastackai.github.io/terratorch/stable/) for the model fine-tuning and provide data modules for ImpactMesh. You can download the code and configs for the fine-tuning from https://github.com/IBM/ImpactMesh.\n\nAlternatively, you can install the data loading code with:\n\n```shell\npip install impactmesh\n```\n\n```shell\nterratorch fit --config configs/terramind_v1_tiny_impactmesh_flood.yaml\n```\n\n## Citation\n\nOur technical report is released soon!\n\n## Acknowledgement\n\nImpactMesh was developed as part of the FAST‑EO project funded by the European Space Agency Φ‑Lab (contract #4000143501/23/I‑DT).\n\nSentinel-2 Level-2A data were downloaded from Microsoft Planetary Computer and are provided under Copernicus Sentinel license conditions (© European Union 2015–2025, ESA) (https://planetarycomputer.microsoft.com/dataset/sentinel-2-l2a).\n\nSentinel-1 Radiometrically Terrain Corrected ","encoding":"utf8","key":"README.md","name":"README.md","size":4713,"truncated":true}],"sourceObjectKey":"test/DEM.tar","starterCell":"from huggingface_hub import hf_hub_download\npath = hf_hub_download(repo_id=\"ibm-esa-geospatial/ImpactMesh-Flood\", filename=\"test/DEM.tar\", repo_type=\"dataset\")\nprint(path)","starterSamplePath":"split/impactmesh_flood_test.txt","totalObjects":22},"summary":"Flood subset of ImpactMesh, a multimodal, multitemporal benchmark for disaster mapping released by IBM, DLR and ESA Phi-lab.","title":"ImpactMesh-Flood: Multimodal Multitemporal Flood Mapping Dataset","updatedAt":"2026-11-14T17:34:58.202Z","visibility":"public","websiteUrl":"https://huggingface.co/datasets/ibm-esa-geospatial/ImpactMesh-Flood"}}