{"success":true,"data":{"author":"EPFL ECEO lab (Porta, Dalsasso, McCarty, Tuia)","bucketAccessType":"public","bucketName":"EPFL-ECEO/CanadaFireSat-Raw","bucketPrefix":"","category":"Environmental Science","cloudProvider":"huggingface","createdAt":"2026-11-17T14:12:18.034Z","description":"The raw, compressed source archives of CanadaFireSat, a benchmark from the EPFL Environmental Computational Science and Earth Observation (ECEO) lab for high-resolution forecasting of where wildfires will burn across Canada. Labels come from National Burned Area Composite (NBAC) fire polygons since 2015, aggregated to a 2.8 x 2.8 km grid and 8-day windows. For each grid cell the inputs are a Sentinel-2 L1C image time series (at least 3 cloud-free images over the 64 days before prediction), hydrometeorological drivers from ERA5-Land and MODIS11 land surface temperature, MODIS13 and MODIS15 vegetation indices (NDVI, EVI, LAI, FPAR), and Fire Weather Index and Drought Code from the Copernicus Emergency Management Service, with 2020 NALCMS land cover for adversarial sampling and analysis. The repository is about 775 GB in six folders of split zip archives for positive and negative samples in spatial and tabular forms, plus a hard test set, with metadata/ holding split definitions and variable column lists. Use it with the CanadaFireSat-Data and CanadaFireSat-Model GitHub repositories; a cleaned Parquet version is published separately as EPFL-ECEO/CanadaFireSat. Released under the MIT license; cite Porta, Dalsasso, McCarty and Tuia, ISPRS Journal of Photogrammetry and Remote Sensing 2026.","displayTitle":"CanadaFireSat Raw: Sentinel-2, ERA5-Land and MODIS Predictor Archives for High-Resolution Wildfire Forecasting in Canada","huggingFaceUrl":"https://huggingface.co/datasets/EPFL-ECEO/CanadaFireSat-Raw","id":"7b16c5c5-2d68-5a1f-916e-37c7e17c4287","license":"MIT","licenseEvidenceUrl":"https://huggingface.co/datasets/EPFL-ECEO/CanadaFireSat-Raw","objectCountEstimate":51,"probeStatus":"ok","referenceUrl":"https://arxiv.org/abs/2506.08690","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"EPFL-ECEO/CanadaFireSat-Raw\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"EPFL-ECEO/CanadaFireSat-Raw\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=EPFL-ECEO, license:mit, size_categories:100K<n<1M, downloads=550"],"format":"mixed","generatedAt":"2026-11-17T14:23:36.188Z","layout":{"example":"ENV/neg_spatial.z01","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"ENV/neg_spatial.z01","truncatedAt":512}],"sampleObjects":["ENV/neg_spatial.z01","ENV/neg_spatial.z02","ENV/neg_spatial.z03","ENV/neg_spatial.z04","ENV/neg_spatial.zip"],"sidecars":[{"bytes":"---\nlicense: mit\ntask_categories:\n- image-segmentation\ntags:\n- environment\n- wildfire\n- satellite-imagery\n- remote-sensing\n- multimodal\nsize_categories:\n- 100K<n<1M\n---\n\n# Dataset Card for CanadaFireSat 🔥🛰️\n\nIn this benchmark, we investigate the potential of deep learning with multiple modalities for high-resolution wildfire forecasting. Leveraging different data settings across two types of model architectures: CNN-based and ViT-based.\n\n- 📝 Published paper from [ISPRS](https://www.sciencedirect.com/science/article/pii/S0924271626002972) ([ArXiv Version](https://arxiv.org/abs/2506.08690))<br>\n- 💿 Dataset repository on [GitHub](https://github.com/eceo-epfl/CanadaFireSat-Data) <br>\n- 🤖 Model repository on [GitHub](https://github.com/eceo-epfl/CanadaFireSat-Model) & Weights on [Hugging Face](https://huggingface.co/datasets/EPFL-ECEO)\n- 🟰 Another \"Clean\" version of the data with PARQUET files can be found at [CanadaFireSat](https://huggingface.co/datasets/EPFL-ECEO/CanadaFireSat)\n\n**Disclaimer:** [23/06/2026] CanadaFireSat does not contain yet the ignition proxy predictors used in Appendix E.\n\n## 📝 Summary Representation\n\nThe main use of this dataset is to push for the development of algorithms towards high-resolution wildfire forecasting via multi-modal learning. Indeed, we show the potential through our experiments of models trained on satellite image time series (Sentinel-2) and with environmental predictors (ERA5, MODIS, FWI). We hope to emulate the community to benchmark their EO and climate foundation models on CanadaFireSat to investigate their downstream fine-tuning capabilities on this complex extreme event forecasting task.\n\n<p align=\"center\">\n  <img src=\"images/summary-canadafiresat.png\"/>\n</p>\n\n## Sources\n\nWe describe below the different sources necessary to build the CanadaFireSat benchmark.\n\n### 🔥📍 Fire Polygons Source\n\n- 💻 National Burned Area Composite (NBAC 🇨🇦): Polygons Shapefile downloaded from [CWFIS Datamart](https://cwfis.cfs.nrcan.gc.ca/home) <br>\n- 📅 Filter fires since 2015 aligning with Sentinel-2 imagery availability <br>\n- 🛑 No restrictions are applied on ignition source or other metadata <br>\n- ➕ Spatial aggregation: Fires are mapped to a 2.8 km × 2.8 km grid | Temporal aggregation into 8-day windows\n\n### 🛰️🗺️ Satellite Image Time Series Source\n\n- 🛰️ Sentinel-2 (S2) Level-1C Satellite Imagery (2015–2023) from [Google Earth Engine](https://developers.google.com/earth-engine/datasets/catalog/COPERNICUS_S2_HARMONIZED) <br>\n- 🗺️ For each grid cell (2.8 km × 2.8 km): Collect cloud-free S2 images (≤ 40% cloud cover) over a 64-day period before prediction <br>\n- ⚠️ We discard samples with: Fewer than 3 valid images | Less than 40 days of coverage <br>\n\n### 🌦️🌲 Environmental Predictors\n- 🌡️ Hydrometeorological Drivers: Key variables like temperature, precipitation, soil moisture, and humidity from ERA5-Land (11 km, available on [Google Earth Engine](https://developers.google.com/earth-engine/datasets/catalog/ECMWF_ERA5_LAND_DAILY_AGGR)) and MODIS11 (1 km, available on [Google Earth Engine](https://developers.google.com/earth-engine/datasets/catalog/MODIS_061_MOD11A1)), aggregated over 8-day windows using mean, max, and min values.\n- 🌿 Vegetation Indices ([MODIS13](https://developers.google.com/earth-engine/datasets/catalog/MODIS_061_MOD13A1) and [MODIS15](https://developers.google.com/earth-engine/datasets/catalog/MODIS_061_MOD15A2H)): NDVI, EVI, LAI, and FPAR (500 m) captured in 8 or 16-day composites, informing on vegetation state.\n- 🔥 Fire Danger Metrics ([CEMS](https://ewds.climate.copernicus.eu/datasets/cems-fire-historical-v1?tab=overview) previously on CDS): Fire Weather Index and Drought Code from the Canadian FWI system (0.25° resolution).\n- 🕒 For each sample, we gather predictor data from 64 days prior to reflect pre-fire conditions.\n\n### 🏞️ Land Cover\n- ⛔️ Exclusively used for adversarial sampling and post-training analysis.\n- 💾 Data extracted is the 2020 North American Land Cover ","encoding":"utf8","key":"README.md","name":"README.md","size":14298,"truncated":true}],"sourceObjectKey":"ENV/neg_spatial.z01","starterCell":"from huggingface_hub import hf_hub_download\npath = hf_hub_download(repo_id=\"EPFL-ECEO/CanadaFireSat-Raw\", filename=\"ENV/neg_spatial.z01\", repo_type=\"dataset\")\nprint(path)","starterSamplePath":"ENV/neg_spatial.z01","totalObjects":51},"summary":"The raw, compressed source archives of CanadaFireSat, a benchmark from the EPFL Environmental Computational Science and Earth Observation (ECEO) lab for high-resolution…","title":"CanadaFireSat Raw: Sentinel-2, ERA5-Land and MODIS Predictor Archives for High-Resolution Wildfire Forecasting in Canada","updatedAt":"2026-11-17T20:09:31.844Z","visibility":"public","websiteUrl":"https://github.com/eceo-epfl/CanadaFireSat-Data"}}