The gs://earth-engine-public-requester-pays/forestdatapartnership GCS bucket
contains COG (Cloud Optimized GeoTIFF) files that together make up the
Forest Data Partnership commodity datasets.
For the 2026a model release, it contains commodity presence probability rasters for the years 2018, 2020, 2022, and 2024.
For the 2025b model release, it contains commodity presence probability rasters for the years 2017 to 2025.
License
This data is made freely available for use under the Creative Commons Attribution 4.0 (CC-BY 4.0) License.
If you use this dataset in your work, include the following attribution: "Produced by Google for the Forest Data Partnership"
This bucket is set up as "requester pays" — see this GCS documentation for details. In short this means that while the data itself is free, you will be responsible for any costs associated with downloading the files (egress charges).
Directory structure
The top level of the directory is divided by release (e.g. 2025b). Within each
release, data are subdivided by commodity and then year. Within each leaf
directory are COGs that together cover the full pantropical range (from 24
degrees South to 24 degrees North) of model outputs. Negative values are used
for Southern and Western latitudes and longitudes, respectively. Note that the
latitude-longitude pair indicated on the filename specifies the SW corner of the
tile.
File structure
Each file is a Cloud optimized GeoTIFF that is 11133x11133 pixels covering a
tile that measures one degree of latitude by one degree of longitude. Files have
a single band labeled probability and use projection EPSG:4326 (WGS84).
Each file contains overview layers for 5567x5567, 2784x2784, on down to
348x348 pixels for the top level overview. Overview layer resampling method
depends on the model release. For the 2026a release, overview layers are
calculated with mean sampling; for the 2025b release, they are calculated with
nearest neighbor or cubic sampling (for newer assets). To avoid ambiguity,
either use the data at 10 meter resolution or rebuild the overviews as
necessary for your use case.
Each pixel is an unquantized 32-bit floating point number between 0 and 1 indicating the model's inferred probability that the area under the pixel is part of a planted area for the commodity indicated in the full path.
Each path indicates the release, commodity, year, and latitude-longitude
pair it is for. For example:
gs://earth-engine-public-requester-pays/forestdatapartnership/2025b/coffee/2024/lng_-10_lat_9.tif
indicates a tile showing the 2025b coffee presence model's inferences for 2024
for the 1-degree by 1-degree tile at 10 degrees West and 9 degrees North. As
mentioned, the latitude-longitude value indicates the location of the lower
left-hand (SW) corner of the tile.
File Indexes
An index of the files in this dataset is provided for each release (e.g.
2025b) in three forms (GeoParquet, GeoPackage, and CSV) in the files:
gs://earth-engine-public-requester-pays/forestdatapartnership/{release}/fdap_index.parquetgs://earth-engine-public-requester-pays/forestdatapartnership/{release}/fdap_index.gpkggs://earth-engine-public-requester-pays/forestdatapartnership/{release}/fdap_index.csv
Replace {release} with the specific release version you are using (for
example, 2025b).
This index contains one entry for each file in the dataset. The information provided for each file is:
release: The release of the dataset (e.g.,2025b).commodity: The commodity (e.g.,cocoa,coffee,palm,rubber).year: The year that the image covers.lng,lat: The longitude and latitude of the SW corner of the tile.path: The GCS path of the file.geometry: The geometry of the tile as a WGS84 (i.e., EPSG:4326) polygon.wgs84_west,wgs84_south,wgs84_east,wgs84_north: The min and max longitude and latitude of the tile.