Contents

Details

Analytics

Home

All Products

Docs

Source Cooperative is a Radiant Earth project

WRI/Google DeepMind Global Drivers of Forest Loss (1km) · Epoch · Source Cooperative | Source Cooperative

WRI/Google DeepMind Global Drivers of Forest Loss (1km)

Name: WRI/Google DeepMind Global Drivers of Forest Loss (1km)
Creator: Epoch

This dataset maps the dominant driver of tree cover loss globally from 2001 to 2025 globally. Produced by the World Resources Institute (WRI) and Google Deepmind. Using satellite imagery and a neural network to classify seven categories of drivers.

Source & citation: Zenodo record 14163025; also on the WRI Data Explorer.

Contents

root

Details

Visibility: Public
Owner: Epoch
Created: 16 Jun 2026
Last Updated: 11 Sep 2026

Analytics

28 days

Downloads

Data served

113.4 KB

Countries

28-day downloads

21 Aug 2026

17 Sep 2026

Analytics

README

Global Drivers of Forest Loss at 1 km, v1.3

This publication is a mirror of Global drivers of forest loss at 1 km resolution (Sims et al. 2025), version 1.3, coverage 2001 to 2025. We added a virtual icechunk store that mirrors the .tif file.

Producer and source

Producer: World Resources Institute (WRI) and Google DeepMind.
Source: Zenodo record 19485190 (v1.3), file drivers_forest_loss_1km_2001_2025_v1_3.tif. Version DOI https://doi.org/10.5281/zenodo.19485190. Concept DOI https://doi.org/10.5281/zenodo.14162799.
Reference: Sims, M., et al. (2025). Global drivers of forest loss at 1 km resolution. Environmental Research Letters. https://doi.org/10.1088/1748-9326/add606

Licence

CC-BY 4.0. Attribution is required. Cite the paper and the Zenodo record. The mirror does not imply endorsement by WRI or Google DeepMind.

Changes to Original Publication

Nothing. The original file is a COG (Cloud-optimized GeoTIFF). We mirror it unchanged.

Coverage

Property	Value
Extent	Global, longitude -180 to 180, latitude -56 to 84
Resolution	0.01° (~1 km), grid 36,000 × 14,000
CRS	EPSG:4326
Years	2001 to 2025, one dominant driver per cell
Bands	8 × uint8, NoData 255

Bands

The data has 8 bands. Band 1 gives the classification. Bands 2 to 8 give the quantized pseudo-probabilities.

Band	Name	Meaning
1	`classification`	Dominant driver, values 1 to 7
2 to 8	`probability_1` to `probability_7`	Class probability, 0 to 250. Multiply by 1/250 for [0, 1]

Classes:

1 permanent agriculture
2 hard commodities
3 shifting cultivation
4 logging
5 wildfire
6 settlements and infrastructure
7 other natural disturbances

Files

How to read

Example of an anonymous read of the COG through the source.coop proxy:

Example of a GDAL-free read with icechunk and xarray. Import virtual_tiff first. This import registers the codec that the chunks use:

Known limitations

The product does not separate natural forest loss from the clearing of planted trees.
The product gives only one dominant driver per 1 km cell for the whole period. It does not record co-occurring drivers or their sequence.
Loss detection inherits the accuracy of Global Forest Change v1.13.
This mirror does not include the training and validation data from the original study.

README

Global Drivers of Forest Loss at 1 km, v1.3

This publication is a mirror of Global drivers of forest loss at 1 km resolution (Sims et al. 2025), version 1.3, coverage 2001 to 2025. We added a virtual icechunk store that mirrors the .tif file.

Producer and source

Producer: World Resources Institute (WRI) and Google DeepMind.
Source: Zenodo record 19485190 (v1.3), file drivers_forest_loss_1km_2001_2025_v1_3.tif. Version DOI https://doi.org/10.5281/zenodo.19485190. Concept DOI https://doi.org/10.5281/zenodo.14162799.
Reference: Sims, M., et al. (2025). Global drivers of forest loss at 1 km resolution. Environmental Research Letters. https://doi.org/10.1088/1748-9326/add606

Licence

CC-BY 4.0. Attribution is required. Cite the paper and the Zenodo record. The mirror does not imply endorsement by WRI or Google DeepMind.

Changes to Original Publication

Nothing. The original file is a COG (Cloud-optimized GeoTIFF). We mirror it unchanged.

Coverage

Property	Value
Extent	Global, longitude -180 to 180, latitude -56 to 84
Resolution	0.01° (~1 km), grid 36,000 × 14,000
CRS	EPSG:4326
Years	2001 to 2025, one dominant driver per cell
Bands	8 × uint8, NoData 255

Bands

The data has 8 bands. Band 1 gives the classification. Bands 2 to 8 give the quantized pseudo-probabilities.

Band	Name	Meaning
1	`classification`	Dominant driver, values 1 to 7
2 to 8	`probability_1` to `probability_7`	Class probability, 0 to 250. Multiply by 1/250 for [0, 1]

Classes:

1 permanent agriculture
2 hard commodities
3 shifting cultivation
4 logging
5 wildfire
6 settlements and infrastructure
7 other natural disturbances

Files

How to read

Example of an anonymous read of the COG through the source.coop proxy:

Example of a GDAL-free read with icechunk and xarray. Import virtual_tiff first. This import registers the codec that the chunks use:

Known limitations

The product does not separate natural forest loss from the clearing of planted trees.
The product gives only one dominant driver per 1 km cell for the whole period. It does not record co-occurring drivers or their sequence.
Loss detection inherits the accuracy of Global Forest Change v1.13.
This mirror does not include the training and validation data from the original study.

1v1_3/drivers_forest_loss_1km_2001_2025_v1_3.tif
2icechunk/v1_3/                                   virtual icechunk store

1v1_3/drivers_forest_loss_1km_2001_2025_v1_3.tif
2icechunk/v1_3/                                   virtual icechunk store

1# Requires: aiohttp, fsspec, rioxarray
2import rioxarray
3da = rioxarray.open_rasterio("https://data.source.coop/epoch/global-forest-loss-drivers/v1_3/drivers_forest_loss_1km_2001_2025_v1_3.tif")

1# Requires: aiohttp, fsspec, rioxarray
2import rioxarray
3da = rioxarray.open_rasterio("https://data.source.coop/epoch/global-forest-loss-drivers/v1_3/drivers_forest_loss_1km_2001_2025_v1_3.tif")

1# Requires: aiohttp, fsspec, icechunk, virtual_tiff, xarray
2# Note zarr may need to be downgraded to 3.2.1 due to virtual_tiff incompatibility issues
3import icechunk as ic, xarray as xr, virtual_tiff
4bucket = "us-west-2.opendata.source.coop"
5repo = ic.Repository.open(
6    storage=ic.s3_storage(bucket=bucket, prefix="epoch/global-forest-loss-drivers/icechunk/v1_3", region="us-west-2", anonymous=True),
7    authorize_virtual_chunk_access={f"s3://{bucket}/": ic.Credentials.S3(ic.S3Credentials.Anonymous())},
8)
9ds = xr.open_zarr(repo.readonly_session("main").store, consolidated=False)
10da = ds["forest_loss_drivers"]        # dims (band, latitude, longitude)
11print(da.isel(latitude=1000, longitude=1000).values)

1# Requires: aiohttp, fsspec, icechunk, virtual_tiff, xarray
2# Note zarr may need to be downgraded to 3.2.1 due to virtual_tiff incompatibility issues
3import icechunk as ic, xarray as xr, virtual_tiff
4bucket = "us-west-2.opendata.source.coop"
5repo = ic.Repository.open(
6    storage=ic.s3_storage(bucket=bucket, prefix="epoch/global-forest-loss-drivers/icechunk/v1_3", region="us-west-2", anonymous=True),
7    authorize_virtual_chunk_access={f"s3://{bucket}/": ic.Credentials.S3(ic.S3Credentials.Anonymous())},
8)
9ds = xr.open_zarr(repo.readonly_session("main").store, consolidated=False)
10da = ds["forest_loss_drivers"]        # dims (band, latitude, longitude)
11print(da.isel(latitude=1000, longitude=1000).values)