Cloud-native Mirror (Cloud-Optimized GeoTIFFs, organized with Portolan) of the Forest Data Partnership commodity maps produced by Google.
Per-pixel probability (0 to 1) that land is planted with cocoa, coffee, oil palm or rubber, at 10 m resolution, from the community model release 2026a. The first release of this mirror covers 2024 for the West African cocoa belt (224 one-degree tiles from Sierra Leone to Gabon, all four commodities). More years and regions will follow. Each commodity-year also carries a pan-tropical overview (about 320 m) for fast low-zoom viewing.
Google publishes the source in a requester-pays Cloud Storage bucket, so every read is billed to the reader and browsers cannot load it. This mirror serves the same files unchanged over plain HTTPS with range requests and CORS: no account, no billing, readable from QGIS, GDAL, Python and web maps.
License: CC-BY-4.0. Attribution: "Produced by Google for the Forest Data Partnership."
The Forest Data Partnership commodity maps, mirrored on Source Cooperative.
Per-pixel probability, from 0 to 1, that land is planted with cocoa, coffee, oil palm or rubber, at 10 m resolution, from the Forest Data Partnership community model release 2026a, produced by Google.
The West African cocoa belt for 2018 and 2024: 224 one-degree tiles per commodity and year, from Sierra Leone to Gabon (15°W–17°E, 1°S–11°N, land south of 10°N). One collection per commodity and year, eight in total:
Each item is one Cloud-Optimized GeoTIFF of about 11133 × 11133 Float32 pixels in
EPSG:4326, with a 347 px thumbnail. Item ids (lng_<west>_lat_<south>) are the same
in every collection, so a tile can be compared across commodities and years.
Each collection also has a pan-tropical overview: one COG of all 5,516 tiles
of the release for that commodity and year (24°S–24°N) at about 320 m, 2.5 to 2.6 GB.
It serves maps below zoom 9 and coarse statistics. Every collection publishes its
items as STAC GeoParquet (items.parquet).
CC-BY-4.0. Attribute as: "Produced by Google for the Forest Data Partnership."
This catalog is a mirror, not the official publication. Google publishes the files
in a requester-pays Google Cloud Storage bucket,
gs://earth-engine-public-requester-pays/forestdatapartnership, documented at
https://developers.google.com/earth-engine/guides/forest_data_partnership_gcs_readme.
There, every read is billed to the reader and browsers cannot read the files.
The tiles were copied unchanged on 2026-10-10 (2024 at 17:10 UTC, 2018 at
18:12 UTC). Each item's file:checksum is Google's MD5 for the file, and random
anonymous downloads were checked byte for byte against it. The overviews are
derived files, built from the 347 px level that Google publishes inside each tile.
The catalog's source, generator and issue tracker are at
https://github.com/portolan-mirrors/forest-data-partnership-catalog.
Model method and limitations: https://github.com/google/forest-data-partnership/tree/main/models/model_2026a
Find the tiles that cover a point, with one request, in DuckDB:
Read one tile without downloading it, in QGIS or GDAL: