--- license: cc-by-4.0 pretty_name: "BFC-Bench: Burnt & Flooded Crop Benchmark Data" tags: - remote-sensing - earth-observation - sentinel-1 - sentinel-2 - agriculture - floods - wildfires - phenology - geospatial task_categories: - image-segmentation size_categories: - 1K Burnt & Flooded Crop Benchmark Data (BFC-Bench) **Global, phenology-timed agricultural hazard benchmark data for geospatial foundation models.** 5,102 chips of flooded cropland (1,297 events) and 4,786 chips of burnt cropland (1,463 events), 2020–2026. Each chip is 512 × 512 px at 10 m and comes with its damage label, Sentinel-2 and Sentinel-1 images from before and after the event, the crop growth stage at the event date, and a quality-control verdict. ![Events with damaged cropland: 1,297 flood events and 1,463 burn events](https://huggingface.co/datasets/eadrah/AgDamage_Benchmark_v1/resolve/main/figures/global_map.png) Each hazard folder holds `shards/*.tar`, with five GeoTIFFs (label, Sentinel-2 and Sentinel-1 before and after) and a JSON sidecar per chip, and `manifest.csv` / `manifest.parquet`, one row per chip. ## Overview and Code access - **Overvew and code:** [GitHub](https://github.com/Esmaeel-A/Burnt-and-Flooded-Crop-Benchmark-Data): the pipelines, overview, chip format and statistics - **Publication:** *title, authors, venue, year* (to be added): [https://doi.org/10.xxxx/axxxxxxxx](https://doi.org/10.xxxx/axxxxxxxx) (in preperation) - **Authors:** *Esmaeel Adrah¹, Julina Maharjan¹, Jamon Van Den Hoek², He Yin^1,3,4*
¹ Department of Geography, Kent State University
² Department of Geography and Environmental Sciences, Oregon State University
³ School of Environment, Society & Sustainability, University of Utah
⁴ Scientific Computing and Imaging Institute, University of Utah - **Contact:** *Esmaeel Adrah (eadrah@kent.edu | esmaeelad@gmail.com)*, *He Yin (he.yin@utah.edu)* If you use this dataset or code, please cite: ```bibtex @article{bfcbench, title = {< Burnt & Flooded Crop Benchmark Data (BFC-Bench): Global, Phenology-timed, Agriculutral Hazard Benchmark Data for Geosppatial Foundation Models>}, author = {>}, journal = {}, year = {<2026>}, doi = {10.xxxx/axxxxxxxx} } ``` ## Examples ### Flooded crop ![Flood events in six countries: Sentinel-2 and Sentinel-1 before and after, label and flood persistence](figures/example_flooded.png) Six chips from Niger, Pakistan, China, Brazil, Thailand and Greece, one per event. Columns: Sentinel-2 before and after, Sentinel-1 VV before and after, the label, and flood persistence, the share of valid Sentinel-1 overpasses in which each pixel was flooded. fc is the chip's flooded-cropland share. ### Burnt crop ![Burn events in six countries: Sentinel-2 and Sentinel-1 before and after, label and dNBR severity](figures/example_burnt.png) Six chips from Ethiopia, India, China, Brazil, Vietnam and Turkey. Columns: Sentinel-2 before and after, Sentinel-1 before and after, the label, and dNBR burn severity, from 0 (unburned) to 4 (high). ## Acknowledgments **Affiliated institutions:**

**Funding:** *This research was funded by NASA Disaster Grant number#* (to be added).