A new global map aims to show more than where trees grow. ForTy, developed by Google DeepMind and the World Resources Institute, classifies land by forest type and use, offering a 2020 reference point for organizations assessing deforestation and biodiversity.
Why forest type matters
Many satellite maps identify tree cover without distinguishing a natural forest from a plantation or a crop farm. ForTy divides the world's land area in 2020 into six classes at 10-meter resolution: primary forest, naturally regenerating forest, planted forest, plantation forest, tree crops, and other land. Tree crops include oil palm, rubber, and cocoa.
The distinction is relevant to the EU Regulation on Deforestation-free Products (EUDR). The regulation prohibits goods linked to forests cleared or degraded after December 31, 2020, including goods produced after primary forest is converted to a plantation. Since oil palm and cocoa farms count as agricultural land, companies need to know how a site was used in 2020 when assessing supply chain risks.
The map could also support other efforts that rely on land-use information, including the Tropical Forest Forever Facility (TFFF) and the UN goal of protecting 30 percent of the world's land by 2030. The researchers describe ForTy as the first global map of its kind; earlier datasets, including one from the European Commission, cover only parts of the world.
How ForTy classifies land
A forest's history can matter as much as its visible appearance. An old planted mixed forest may look natural from space, while a recently cleared plantation may appear bare even though it still counts as forest by definition. Satellite images alone rarely reveal how land is managed.
ForTy starts with AlphaEarth Foundations, a Google DeepMind model that combines data from multiple satellites into a compact description of each pixel. Image-recognition models then use these descriptions to classify land. The first learns from verified examples; its predictions, along with outside data on tree heights, buildings, and forest loss, help train a second model. A final stage compares several years of data to smooth outliers.
Google's Gemini language model also helped adapt older training data to the EU's definitions. During correction rounds, experts added about 43,000 examples to address mistakes such as vineyards in Europe labeled as tree crops, natural pine forests in Central America labeled as planted forests, and oil palms missed in Africa and Indonesia.
Some boundaries remain difficult to infer. Because satellite images rarely reveal management practices, the researchers use a 40-year rotation period to distinguish plantations from other planted forests. They note that this cutoff does not fit every region.
Accuracy varies by class
Two experts reviewed each of 8,190 randomly selected sites. The map was correct in 90.2 percent of cases overall, though the authors say the score benefits from the easy-to-identify “other land” category.
Performance was uneven across forest types. ForTy correctly identified 89 percent of primary forests confirmed by experts, but found only about half of tree crops and planted forests, and 60 percent of plantations. Roughly a quarter to a third of areas labeled as plantations or tree crops were misclassified; the map was wrong for a third of areas labeled as planted forest.
The map scored higher than European Commission maps made specifically for the EUDR on the same test data: 95.5 percent compared with 93.6 percent for forest versus nonforest classification, and 90.9 percent compared with 87.8 percent for forest types. Still, the researchers caution that uncertainty could make the results look slightly better. Experts were most often unsure about managed areas, and uncertain cases were excluded from the assessment.
Training data also carried a bias: forests near roads or settlements were rarely labeled as primary forest, even when they appeared untouched. The map does not yet resolve every distinction that users may need.
What the map suggests—and misses
The researchers estimate that the world has about 4.1 billion hectares of forest, including nearly 3.9 billion hectares of natural forest and about 2 billion hectares of primary forest. Planted forests cover about 250 million hectares, while tree crops cover nearly 200 million hectares, mostly in Asia.
Regional errors complicate the global picture. ForTy overestimates primary forest in Asia by about 200 million hectares, largely in Russia's conifer forests. In Africa, it misses about 94 million hectares of forest, likely because open dry forests and savannas are difficult to classify.
In Sweden, about two-thirds of forest area is planted, while natural forest covers only around one-fifth of the country. The authors say overall tree cover may be adequate there, 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.
Shaded cocoa or coffee grown under a natural canopy can still be labeled as forest, and selectively logged or burned forests also pose challenges. ForTy has no separate class for open woodlands, which the EUDR treats separately, though the researchers say future versions should add one. Users can apply their own thresholds using the dataset's confidence scores.
The map, classification layer, and test data are freely available under a CC BY 4.0 license, including in Google Earth Engine. Google DeepMind has also released code in the GeeFlow and JEO repositories.