Glonce

Google DeepMind's ForTy maps world forest types at 10-meter resolution

researchThe Glonce editorsThe Decoder🔥64Glonce rating64Impact55Novelty70Relevance80Surprise35Players80Credibility85
Google DeepMind's ForTy maps world forest types at 10-meter resolution

Google DeepMind and the World Resources Institute have developed ForTy, short for Forest Typology, an AI-built map that tells apart primary forests, timber plantations and tree crop farms worldwide. Ordinary satellite maps show whether trees grow somewhere but not what kind of forest or crop it is. ForTy assigns the world's land area in 2020 to six classes at 10-meter resolution: primary forest, naturally regenerating forest, planted forest, plantation forest, tree crops such as oil palm, rubber or cocoa, and other land. The authors say it is the first global map of its kind; earlier datasets, including one from the European Commission, cover only parts of the world.

The main driver is the EU Regulation on Deforestation-free Products (EUDR), which prohibits goods linked to forests cleared or degraded after December 31, 2020, including goods produced after primary forest was converted to a plantation. Oil palm and cocoa farms count as agricultural land, so companies assessing supply chain risk need to know how a site was used in 2020. ForTy is meant to give them that reference point, and it could also feed other initiatives that rely on the same kind of data, including the Tropical Forest Forever Facility (TFFF) and the UN goal of protecting 30 percent of the world's land by 2030. The difficulty, DeepMind says, is that a forest's type depends on its history, not just how it looks: an old planted mixed forest can look natural from space, and a recently cleared plantation still counts as forest by definition even though satellite images show bare ground.

ForTy runs on DeepMind's AlphaEarth Foundations model, which compresses data from multiple satellites into a compact description of each pixel. Several image recognition models build on those descriptions, starting with one trained on verified examples. Its predictions, combined with outside data on tree heights, buildings and forest loss, become expanded training data for a second model. A final step analyzes several years of data together to smooth outliers. Google's Gemini language model helped adapt older training data to the EU's definitions. In correction rounds, experts added about 43,000 examples to fix errors such as vineyards in Europe classified as tree crops, natural pine forests in Central America labeled as planted forests, and oil palms missed in Africa and Indonesia. Satellites still cannot see how land is managed, so the researchers use a 40-year rotation period to separate plantations from other planted forests, and that cutoff does not fit every region.

To validate the map, 2 experts reviewed each of 8,190 randomly selected sites; the map was correct in 90.2 percent of cases. The authors warn that this overall score gets a large boost from the easy-to-identify "other land" class. Results vary sharply by type. The map correctly identified 89 percent of the primary forests that experts confirmed, but found only about half of tree crops and planted forests and 60 percent of plantations. Seen from the other side, roughly a quarter to a third of areas mapped as plantations or tree crops turned out to be misclassified, and the map was wrong for a third of areas labeled as planted forest. These gaps hit the boundaries that matter most for the EUDR. Uncertainty in the evaluation may also flatter the results slightly: experts were most often unsure about managed areas, and the team excluded those cases. On the same test data, ForTy still beats European Commission maps made specifically for the EUDR: 95.5 percent against 93.6 percent for forest versus nonforest, and 90.9 percent against 87.8 percent for forest types. The model also inherited a weakness from its training data, whose annotators rarely labeled forests near roads or settlements as primary forest, even when they looked untouched.

The researchers estimate that the world has about 4.1 billion hectares of forest, including nearly 3.9 billion hectares of natural forest, of which about 2 billion hectares are primary forest. Planted forests cover about 250 million hectares and tree crops nearly 200 million hectares, mostly in Asia. Regional errors are large: the map overestimates primary forest in Asia by about 200 million hectares, largely in Russia's conifer forests, and misses about 94 million hectares of forest in Africa, likely because of open dry forests and savannas. In Sweden, about two-thirds of forest area is planted and natural forest covers only around one-fifth of the country. The authors say overall tree cover there may be adequate, but the share of natural forest, which matters for biodiversity, could be critically low. The Democratic Republic of the Congo, Indonesia and Brazil have the highest shares of natural forest.

Some cases remain hard. In some regions the map still labels shaded cocoa or coffee grown under a natural canopy as forest, and selectively logged or burned forests also cause trouble. ForTy has no separate class for open woodlands, which the EUDR treats separately, though the researchers say future versions should add one. Users can make their own classification choices with the dataset's per-class confidence scores and set their own thresholds. The map, classification layer and test data are freely available under a CC BY 4.0 license, including in Google Earth Engine, and DeepMind has released the code in the GeeFlow and JEO repositories. DeepMind introduced AlphaEarth Foundations in July 2025 as a kind of virtual satellite, combining optical images, radar, laser scans and climate simulations into a description of each 10-by-10-meter area, with annual embeddings since 2017 available in Earth Engine. ForTy is one of the first large mapping projects built on those embeddings and, Google says, is part of Earth AI.

Key facts

  • ForTy, built by Google DeepMind and the World Resources Institute, assigns the world's 2020 land area to six classes at 10-meter resolution, including primary forest, plantation forest and tree crops; the authors call it the first global map of its kind.
  • Overall accuracy is 90.2 percent on 8,190 randomly selected sites reviewed by 2 experts, but the authors warn the score is boosted by the easy 'other land' class.
  • The map found 89 percent of confirmed primary forests but only about half of tree crops and planted forests and 60 percent of plantations.
  • On the same test data it beats European Commission EUDR maps: 95.5 vs 93.6 percent for forest versus nonforest, and 90.9 vs 87.8 percent for forest types.
  • The map, classification layer and test data are free under CC BY 4.0, including in Google Earth Engine; code is in the GeeFlow and JEO repositories.

Why it matters

The EU deforestation regulation (EUDR) bans goods linked to forests cleared or degraded after December 31, 2020, and counts conversion of primary forest to a plantation. Oil palm and cocoa farms count as agricultural land, so companies need to know how a site was used in 2020. ForTy aims to supply that 2020 reference point, which the authors say no earlier global dataset provided; existing ones, including the European Commission's, cover only parts of the world. It could also feed initiatives such as the Tropical Forest Forever Facility and the UN goal of protecting 30 percent of land by 2030. The technical novelty is separating forest types by history and growth and harvest cycles, not appearance alone.

Who it affects

Companies assessing supply chain risk under the EUDR are the main target audience named in the article. Organizations working on forest protection initiatives such as TFFF and the 30-by-2030 land goal could use the same kind of data. Researchers can build on the open dataset and code. Countries come out with distinct pictures: the map suggests Sweden has plenty of trees but a possibly critically low share of natural forest, while the DRC, Indonesia and Brazil have the highest natural forest shares.

How to use it

The map, classification layer and test data are freely available under a CC BY 4.0 license, including in Google Earth Engine. DeepMind has released the code in the GeeFlow and JEO repositories. The dataset includes confidence scores for each class, so users can set their own thresholds and make their own classification choices, for example by being stricter where plantations and tree crops matter.

How solid is it

The accuracy figures come from expert review of 8,190 randomly selected sites, each reviewed by 2 experts. The 90.2 percent overall score is across all six classes and, the authors warn, is lifted by the easy 'other land' class. Per-class results are weaker: 89 percent of confirmed primary forests found, but only about half of tree crops and planted forests and 60 percent of plantations. The evaluation itself may flatter the map slightly, since experts were most often unsure about managed areas and those cases were excluded. The comparison with European Commission EUDR maps on the same test data favors ForTy by a modest margin. The first-of-its-kind claim is the authors' own.

Risks and caveats

Errors fall on the boundaries that matter most for the EUDR: roughly a quarter to a third of areas mapped as plantations or tree crops were misclassified, and the map was wrong for a third of areas labeled as planted forest. Regional errors are large: primary forest in Asia is overestimated by about 200 million hectares, largely in Russia's conifer forests, while about 94 million hectares of forest are missed in Africa. The model inherited a bias from training data, where forests near roads or settlements were rarely called primary forest. Shaded cocoa or coffee under natural canopy can still be labeled as forest, and selectively logged or burned forests are hard. The 40-year rotation cutoff does not fit every region. There is no open woodland class yet, though the EUDR treats those separately; researchers say future versions should add one. The article does not say any company or regulator has adopted ForTy for EUDR compliance.