Format
What TACO is
TACO (Transparent Access to Cloud-Optimized datasets) packages Earth observation data, metadata and sample structure as one portable dataset. You query the catalog by region, time, cloud cover or split and only then load pixels, so a multi-hundred-gigabyte collection is usable without downloading it.
Three levels
- Collection: identity and contract. Title, version, license, providers and the schema every sample must follow.
- Catalog: the index. A Parquet table with one row per sample, queryable with SQL.
- Sample: the data. Imagery as Cloud-Optimized GeoTIFFs, fetched by byte range.
Containers
A TACO ships as a plain folder, a Cloud-Optimized ZIP, or a partitioned catalog. The same contract is validated on write and used on read, so folders, ZIPs and partitioned catalogs behave identically.
Clients
| Language | Install | Role |
|---|---|---|
| Python | pip install taco-eo | read + write |
| R | install.packages("taco") | reader |
| Julia | Pkg.add("Taco") | reader |
| JavaScript | npm install @asterisk-labs/taco | reader |
How MethaneSET uses it
The seven datasets expose their metadata as Parquet indexes and their imagery as cloud-optimized rasters. This site reads the EMIT index straight from Hugging Face in the browser, with no server in between.
TACO specification v3.0 · spec · source