AI-ready · TACO · multi-sensor
Methane plumes,
ready for AI
MethaneSET unifies Sentinel-2, Landsat 8/9 and EMIT into queryable TACO datasets: expert-verified plume masks, calibrated radiance and a synthetic plume bank of 238,545 enhancements, distributed as Parquet catalogs with Cloud-Optimized GeoTIFFs.
One problem, three sensors
Every sensor, one schema
Multispectral instruments need a temporally paired, plume-free reference. Imaging spectrometers resolve the methane signature in a single acquisition. MethaneSET serves both under the same metadata contract.
EMIT
285 bands at 60 m. Calibrated radiance hypercubes, matched-filter products and two independent plume masks per granule, from IMEO and Carbon Mapper.
Sentinel-2
All 13 bands at 10 m. Expert-verified plume masks and confirmed plume-free references for change-detection workflows.
Landsat 8/9
9 OLI bands at 30 m. The same labels and plume-free pairing, extending the record and the surface diversity of the training set.
TACO format
Query before you download
Each dataset is a Collection of Catalogs of Samples. Metadata lives in Parquet, imagery in Cloud-Optimized GeoTIFFs, so you filter by region, date or sensor before pulling a single pixel.
Collection
Dataset-level identity: title, version, license, providers and the schema contract.
Catalog
The index. A Parquet table, one row per sample, queryable with SQL over HTTP.
Sample
The data itself: radiance, masks and ancillary layers as COGs, fetched by byte range.
Get started
One line to load
Read the metadata, filter the samples you need, then stream only those pixels. Works from Python, R, Julia and the browser.
Install: taco-eo.# pip install taco-eo import taco ds = taco.load( "hf://datasets/tacofoundation/methaneset/methaneset-emit" ) df = ds.metadata() # Parquet index as a DataFrame df[df["site:country"] == "Turkmenistan"]
Paper
MethaneSET
Unified multi-sensor AI-ready datasets for satellite-based methane plume detection. Submitted to Scientific Data.