# CaBuAr: California Burned Areas, Pre- and Post-Fire Sentinel-2 Scenes with Cal Fire Masks

## Facts

- Provider: Daniele Rege Cambrin, Luca Colomba and Paolo Garza, Politecnico di Torino
- License: CDLA-Permissive-2.0 (https://huggingface.co/datasets/DarthReca/california_burned_areas)
- Size: 45 files
- Format: Parquet, HDF5 (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/DarthReca/california_burned_areas
- Access: public, readable without an account
- Category: Earth Observation & Geospatial
- Source: https://arxiv.org/abs/2401.11519
- Page: https://scigantic.com/archive/376d6c75-ff8a-5117-b5b2-22874adc613e
- Keywords: Earth Observation & Geospatial, image segmentation, satellite imagery, remote sensing, Parquet, HDF5, Hugging Face

## About

CaBuAr (California Burned Areas) is a burned-area segmentation dataset of Sentinel-2 scenes acquired before and after wildfires in California, with binary ground-truth masks derived from the California Department of Forestry and Fire Protection (Cal Fire) fire perimeters mapped onto each scene. Full scenes are 5490 x 5490 pixels with all 12 spectral bands for the pre-fire and post-fire images; a pre-patched variant cuts them into 512 x 512 patches, keeping only patches with at least one burned pixel. Data are HDF5 files compressed with BZip2 through hdf5plugin (required to read them), grouped by fold and wildfire uid with pre_fire, post_fire and mask arrays, and provided in raw and normalized (0 to 255) versions under raw/ and normalized/; metadata.parquet gives coordinates, CRS and timestamps. There are 5 random folds (0 to 4), and the repo also includes the ChaBuD challenge test file. Load with h5py, the Hugging Face loading script (post-fire and pre-post-fire configurations) or TorchGeo's CaBuAr dataset class. Released under CDLA-Permissive-2.0 by the authors; cite Cambrin, Colomba and Garza, 'CaBuAr: California burned areas dataset for delineation', IEEE Geoscience and Remote Sensing Magazine 2023.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("DarthReca/california_burned_areas")
print(ds)
```

## Data structure

File types in sampled files: Parquet, HDF5

Example file: `metadata.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/376d6c75-ff8a-5117-b5b2-22874adc613e
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/376d6c75-ff8a-5117-b5b2-22874adc613e
- Over MCP: `get_schema_card` with id `376d6c75-ff8a-5117-b5b2-22874adc613e` at https://api.scigantic.com/mcp

## Related topics

- [Earth Observation & Geospatial datasets](https://scigantic.com/topics/earth-observation-and-geospatial)
- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Satellite imagery datasets](https://scigantic.com/topics/satellite-imagery)
- [Image segmentation datasets](https://scigantic.com/topics/image-segmentation)
- [Remote sensing datasets](https://scigantic.com/topics/remote-sensing)
- [HDF5 datasets](https://scigantic.com/topics/hdf5)
