Load and query

Metadata is a Parquet index, imagery is cloud-optimized. Read the index, filter what you need, then fetch only those pixels by byte range.

# pip install tacoreader
import tacoreader

ds = tacoreader.load(
  "https://huggingface.co/datasets/tacofoundation/methaneset/resolve/main/methaneset-emit/"
)
df = ds.data.to_pandas()
df[df["site:country"] == "Turkmenistan"]

Explore and reproduce

Each example is a small recipe: load one dataset, do one thing, and keep the result.

Multispectral

Sentinel-2: catalog to a pixel baseline

Open the two subsets, reproduce the MBMP enhancement and train a minimal pixel classifier.

Open →

Multispectral

Landsat 8/9: the same tour

The Landsat counterpart: schema, EDA, B6/B7 enhancement and the baseline.

Open →

Hyperspectral

EMIT: radiance, matched filters, masks

Stream the 285-band cube, decode the uint64 bitmask plumes and measure IMEO vs Carbon Mapper IoU.

Open →

Bank

Plume bank and synthetic injection

Query the bank by solar geometry and wind, scale a map to any rate and inspect the LES cubes.

Open →

Metadata

Query the catalog

Filter samples by country, date or flux with SQL over the Parquet index, before downloading imagery.

Open →

Install the clients

Language	Install
Python	`pip install tacoreader`
R	`install.packages("taco", repos = "https://asterisk-labs.r-universe.dev")`
Julia	`Pkg.add("Taco")`
JavaScript	`npm install @asterisk-labs/taco`

The Python client is the reference implementation and the only one that writes. See tacoreader on PyPI.

Notebooks and clients from the MethaneSET dataset card .