{"success":true,"data":{"author":"Gerard, Zhao, Sullivan (KTH Royal Institute of Technology); TACO repackaging by ISP, Universitat de Valencia","bucketAccessType":"public","bucketName":"isp-uv-es/wildfirespreadts-taco","bucketPrefix":"","category":"Earth Observation & Geospatial","cloudProvider":"huggingface","createdAt":"2026-11-16T14:11:25.489Z","description":"WildfireSpreadTS (NeurIPS 2023 Datasets and Benchmarks) is a benchmark for next-day active fire prediction from multi-modal daily time series. It covers 607 wildfire events in the western United States from 2018 to 2021 at 375 m, with 9 to 94 co-registered days per event. Each day carries VIIRS reflectance (M11, I2, I1), VNP13A1 NDVI and EVI2, gridMET surface weather (precipitation, wind speed and direction, minimum and maximum temperature, energy release component, specific humidity, Palmer drought severity index), a NOAA GFS forecast of the same variables, MODIS IGBP land cover and VIIRS active-fire detections stored as HHMM detection time; slope, aspect and SRTM elevation are stored once per event. The authors prescribe leave-one-year-out cross-validation; this copy encodes one fold (2018 validation, 2020 test, 2019 and 2021 train) and keeps the year so others can be rebuilt. About 8.4 GB in TACO format, read with tacoreader. Original on Zenodo under CC BY 4.0; cite Gerard et al. 2023.","displayTitle":"WildfireSpreadTS: Multi-Modal Daily Time Series for Wildfire Spread Prediction (TACO)","doi":"10.52202/075280-3258","id":"844aa419-71fb-5148-bcaf-f59988c6b960","license":"CC-BY-4.0","licenseEvidenceUrl":"https://zenodo.org/records/8006177","objectCountEstimate":12,"probeStatus":"ok","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"isp-uv-es/wildfirespreadts-taco\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"isp-uv-es/wildfirespreadts-taco\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=isp-uv-es, license:cc-by-4.0, size_categories:1K<n<10K, format:imagefolder, downloads=333"],"format":"mixed","generatedAt":"2026-11-16T14:11:30.936Z","layout":{"example":"assets/logos.png","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"89504e470d0a1a0a0000000d49484452000002220000008c0806000000413dc29d00009fbb49444154789cecfd77986c5775e78d7fd7dafb84ca9d6e0e0a57010524a1608288064cce2667f03018637bb0fd9b796dbf1ef0cc388dc7f1c7181b838d8d6dd2788c49c6649011c2422040a09ca51bfa76aaae74c2de6bbd7fecaa9baf7415407dafcee779eae9eeea3aa74eed3a55fb7bd65eebbb808a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8a8ac32002981944742fffff311f58454545454545c589cda1e2e34041424460e67b7c7c45454545454545c53d32110f3c1616fbfe1e8b8c288a70f1a3ce6da449b26f1b73800089e318173feadc461c47076d37112acc74d0f35454545454543c5ca866be7b8188a0aab0c6c0797fd8ffb76fdd6c5ff7ca175e7afa8e931e79f7ce3db77cfd9b575df5f52bbe3dbfd2ed69a35ec3a58fb970ee89975e72d1f66d9b4ebbe996db7ff8c10f7fe2df6ebdedaef2d0fd1863e0bddff77c1515151515150f072a2132e6a068842a1480310cef05975c785ef3a79fffcc9fbce5b63b6eb9f196db77de78cb6dabbbf72cb8c7fdc485532f7ff1339f17c751321c64fd24896b499ad4f7ec5db8637ecfddf3b3731b66b76cda78525114f928cf87f57ad274ce97fff8cf5ff8f417bf72f9c2bab9693e","encoding":"hex","key":"assets/logos.png","truncatedAt":512}],"sampleObjects":["assets/logos.png","assets/sample_0.png","assets/sample_1.png","assets/sample_2.png","assets/sample_3.png"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\npretty_name: \"WildfireSpreadTS\"\nsize_categories:\n- n<1K\ntask_categories:\n- image-segmentation\ntags:\n- earth-observation\n- remote-sensing\n- taco\n---\n\n<p align=\"center\"><img src=\"assets/logos.png\" width=\"546\" alt=\"Image and Signal Processing Group, Universitat de València · ELLIOT\"></p>\n\n# WildfireSpreadTS\n\n> **This is a repackaging, not a new dataset.** It is **WildfireSpreadTS** by KTH Royal Institute of Technology (Gerard, Zhao, Sullivan), NASA/NOAA VIIRS, USGS SRTM, NASA MODIS, gridMET, NOAA GFS, converted to TACO. Pixel values and labels are kept as released except where the description below says otherwise. All credit belongs to the original authors: if you use it, please cite them and follow their licence.\n>\n> [original dataset](https://zenodo.org/records/8006177) · [paper](https://doi.org/10.52202/075280-3258) · licence: CC-BY-4.0\n>\n> Repackaged into TACO by the [Image and Signal Processing Group (ISP)](https://isp.uv.es/), Universitat de València, within the [ELLIOT](https://elliot-ai.eu) project.\n\n## Citation\n\nPlease cite the original work:\n\n```bibtex\n@inproceedings{gerard2023wildfirespreadts,\n  title     = {WildfireSpreadTS: A dataset of multi-modal time series for wildfire spread prediction},\n  author    = {Gerard, Sebastian and Zhao, Yu and Sullivan, Josephine},\n  booktitle = {Advances in Neural Information Processing Systems 36 (Datasets and Benchmarks)},\n  year      = {2023}\n}\n```\n\n## About the data\n\nDaily multi-modal time series over 607 wildfire events in the western United States (2018, 2019, 2020, 2021), at 375 m nominal resolution: 9 to 94 co-registered days per event.\n\n**607 samples** · splits: `test` 201 · `train` 230 · `validation` 176 · tasks: `semantic-segmentation`\n\nPackaged as [TACO v3](https://asterisk.coop/taco/spec/).\n\n<details>\n<summary>Full description</summary>\n\n**Channels.** Per day: VIIRS reflectance (M11/I2/I1), the VNP13A1 vegetation indices NDVI and EVI2, gridMET surface meteorology (precipitation, wind speed and bearing, minimum and maximum temperature, energy release component, specific humidity, Palmer drought severity), a GFS forecast of the same quantities, MODIS IGBP land cover, and the VIIRS active-fire detections. Slope, aspect and elevation are stored once per event rather than once per day, being bit-identical across every day of every event; land cover stays per day because it varies in one event of the 607.\n\n**Active fire.** The raster stores the UTC granule time of each detection as an HHMM code, with 0 for no detection, rather than a binary mask. Published baselines binarise it with `fire > 0`, which is an exact function of what is stored; the reverse is not.\n\n**Splits.** The release publishes no train/val/test split: the authors prescribe leave-one-year-out cross-validation. `split` carries one of their folds, whole years at a time (2018 val, 2019 train, 2020 test, 2021 train), and `year` is carried so every other fold stays reconstructible. The fold is lopsided (230 train, 176 val, 201 test): four years over three parts is 2/1/1. It is not spatially disjoint either. An event raster is 111-128 km across and 4,129 of the 183,921 event pairs overlap on the ground; that overlap graph has six components with 595 of the 607 events in one, so no spatially disjoint partition exists.\n\n**Caveats.**\n- The GFS forecast precipitation channel is a daily sum in 5,131 of the 13,607 rasters and the last hourly value in the other 8,476, carried per frame as `forecast_precipitation_agg`.\n- The GFS forecast wind direction spans only -90..+90 degrees, the signature of an atan rather than an atan2, so its quadrant is unrecoverable and it is declared uncalibrated.\n- Every raster is tagged UTM zone 10N regardless of where its fire is, so the resolution is measured geodesically per event (361.3-374.1 m) rather than read off the transform.\n\n</details>\n\n## Getting started\n\n```bash\ngit clone --recursive https://github.com/OscarPellicer/taco\npip install -e \"taco/python[ml]\"     # builds the reader: C++23, CMake, Ninja, pkg-config, libcurl >= 7.83, OpenSSL >= 3\n```\n\nR","encoding":"utf8","key":"README.md","name":"README.md","size":6928,"truncated":true}],"sourceObjectKey":"assets/logos.png","starterCell":"from huggingface_hub import hf_hub_download\npath = hf_hub_download(repo_id=\"isp-uv-es/wildfirespreadts-taco\", filename=\"assets/logos.png\", repo_type=\"dataset\")\nprint(path)","starterSamplePath":"assets/logos.png","totalObjects":12},"summary":"WildfireSpreadTS (NeurIPS 2023 Datasets and Benchmarks) is a benchmark for next-day active fire prediction from multi-modal daily time series.","title":"WildfireSpreadTS: Multi-Modal Daily Time Series for Wildfire Spread Prediction (TACO)","updatedAt":"2026-11-16T15:14:18.546Z","visibility":"public","websiteUrl":"https://huggingface.co/datasets/isp-uv-es/wildfirespreadts-taco"}}