MethaneSET: Unified Multi-Sensor Datasets for Satellite-Based Methane Plume Detection

Authors: Cesar Aybar, Julio Contreras, David Montero, Miguel D. Mahecha, Luis Gómez-Chova

Paper: Scientific Data (under review)

Methane is the second-largest driver of anthropogenic warming, and a disproportionate share of emissions comes from a small number of super-emitters detectable by satellite. MethaneSET provides analysis-ready datasets for methane plume detection spanning three sensors with expert-verified segmentation masks from two independent monitoring systems (IMEO MARS and Carbon Mapper). For multispectral sensors (Sentinel-2, Landsat-8/9), labeled scenes with plume masks and confirmed plume-free references are provided for change-detection workflows. For the EMIT imaging spectrometer, calibrated radiance cubes and precomputed matched-filter products support both end-to-end and retrieval-based detection. A synthetic plume bank of 238,545 projected WRF-LES enhancements, plus the 1,647 raw simulation cubes it is built from, enables physics-based data augmentation. All datasets follow the TACO specification and are distributed as Parquet catalogs with Cloud-Optimized GeoTIFFs.

Dataset	Samples	Size	Datacard
methaneset-s2-pretraining	56,344	36.91 GB	view
methaneset-s2-finetune	3,552	6.4 GB	view
methaneset-l89-pretraining	21,919	8.88 GB	view
methaneset-l89-finetune	1,353	1.2 GB	view
methaneset-emit	721	1,005 GB	view
methaneset-bank	238,545	5.1 GB	view
methaneset-bank-les	1,647	3.1 GB	view

What each dataset contains

• methaneset-s2-pretraining / methaneset-l89-pretraining: plume-free scenes for self-supervised pretraining: 13-band (Sentinel-2) or 6-band (Landsat-8/9) chips at 10 m, 200$\times$200 px, each with a plume-free reference, Copernicus DEM elevation and observation metadata.
• methaneset-s2-finetune / methaneset-l89-finetune: plume scenes with expert annotations and four clean backgrounds: target, bg0 (the MARS reference), bg1--bg3 (curated here), the MBMP ch4 enhancement, the binary plume mask and the DEM.
• methaneset-emit: 721 EMIT granules (700 with plumes + 21 confirmed plume-free): the 285-band L1B radiance cube (COG with Zstandard and SNR-adaptive bit discarding), dual plume masks from IMEO and Carbon Mapper, three matched-filter retrievals (mf, rmf, mag1c), an ERA5-Land 10 m wind field, Copernicus DEM elevation, and the per-pixel lat/lon and GLT geometry layers.
• methaneset-bank / methaneset-bank-les: 238,545 projected methane enhancements from WRF-LES simulations across a grid of wind speeds, solar geometries and source configurations, together with the 1,647 source 3D tracer cubes so enhancements can be regenerated for any observation geometry.

Usage

# pip install tacoreader
import tacoreader

# load any dataset straight from the Hub
ds = tacoreader.load(
    "https://huggingface.co/datasets/tacofoundation/methaneset/resolve/main/methaneset-s2-finetune/.tacocat/"
)
print(ds.id, len(ds.data))

# methaneset-emit ships as a folder TACO
ds = tacoreader.load(
    "https://huggingface.co/datasets/tacofoundation/methaneset/resolve/main/methaneset-emit/"
)

Each datacard (linked above) documents the full column list of its dataset.

Citation

@article{aybar2026methaneset,
  title   = {MethaneSET: Unified Multi-Sensor Datasets for Satellite-Based Methane Plume Detection},
  author  = {Aybar, Cesar and Contreras, Julio and Montero, David and Mahecha, Miguel D. and G{\'o}mez-Chova, Luis},
  journal = {Scientific Data},
  year    = {2026}
}

Acknowledgements

This project was funded by the Climate Change AI (CCAI) Innovation Grants program, hosted by CCAI with the support of the Global Methane Hub (GMH), and by the Spanish Ministry of Science, Innovation and Universities (grants PID2023-148485OB-C21 / C22 funded by MCIU / AEI / 10.13039 / 501100011033, ERDF, EU). We thank UNEP's International Methane Emissions Observatory (IMEO) for providing access to the MARS plume annotations and Carbon Mapper for making their plume detections publicly available. EMIT Level 1B radiance data were obtained from NASA's Land Processes Distributed Active Archive Center (LP DAAC).