DeepMind and WRI release ForTy, a 10-meter global forest type map with 90.2% overall accuracy

Google DeepMind and the World Resources Institute have released ForTy (Forest Typology), a global map that assigns the world's 2020 land area to six classes at 10-meter resolution. Expert review found it correct in 90.2% of cases, but accuracy drops sharply for plantations and tree crops.

The six classes are primary forest, naturally regenerating forest, planted forest, plantation forest, tree crops (such as oil palm, rubber, or cocoa), and other land. The authors say ForTy is the first global map of its kind. They say earlier datasets, including one from the European Commission, cover only parts of the world.

Built for EU deforestation rules

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. That includes 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. DeepMind says the map could also feed into the Tropical Forest Forever Facility (TFFF) and the UN goal of protecting 30 percent of the world's land by 2030.

How it works

ForTy runs on AlphaEarth Foundations, which compresses data from multiple satellites into a compact description of each pixel. Three model stages build on those descriptions:

• A first model learns from 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.

Gemini helped adapt older training data to the EU's definitions. Over multiple active learning rounds, experts added about 43,000 examples to fix errors such as European vineyards classified as tree crops, natural pine forests in Central America labeled as planted forests, and oil palms missed in Africa and Indonesia. To separate plantations from other planted forests, the researchers use a 40-year rotation period, which does not fit every region.

Accuracy

Two experts reviewed each of 8,190 randomly selected sites. The authors warn that the 90.2% overall score is boosted by the easy-to-identify "other land" class.

• Primary forest: 89% of expert-confirmed sites correctly identified.
• Tree crops and planted forests: only about half identified.
• Plantations: 60% identified.
• Precision: 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.

On the same test data, ForTy scored 95.5% against 93.6% for the European Commission's EUDR maps on forest versus nonforest classification. On forest types it scored 90.9% against 87.8%. The team excluded cases where experts were unsure, mostly managed areas, which may make results look slightly better than they are.

Estimates and known gaps

The researchers estimate about 4.1 billion hectares of forest worldwide, 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 significant. 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 due to open dry forests and savannas. Annotators rarely classified forests near roads or settlements as primary forest, a bias the model inherited. Shaded cocoa and coffee under natural canopy are still sometimes labeled forest. ForTy has no separate open woodland class, which the EUDR treats separately, though the researchers say future versions should add one.

About two-thirds of Sweden's forest area is planted, and natural forest covers around one-fifth of the country. The Democratic Republic of the Congo, Indonesia, and Brazil have the highest shares of natural forest.

Availability

The map, classification layer, and test data are free under a CC BY 4.0 license, including in Google Earth Engine. Per-class confidence scores let users set their own thresholds. Code is in the GeeFlow and JEO repositories. DeepMind introduced AlphaEarth Foundations in July 2025. It combines optical imagery, radar, laser scans, and climate simulations, with annual embeddings since 2017 available in Google Earth Engine.

What this means

ForTy gives EUDR compliance work a global, openly licensed 2020 baseline that, on the shared test data, beats the European Commission's EUDR maps. But the weakest classes, plantations, tree crops, and planted forests, are the boundaries the regulation cares about most. The authors themselves note these gaps, so the map is better suited to screening and risk assessment than to definitive site-level rulings. The released confidence scores let users tune thresholds to their own tolerance for error.