The map runs on DeepMind's AlphaEarth Foundations, a model that condenses imagery from several satellites into a short summary for every pixel. Image recognition models then work from those summaries, with outside data on tree heights, buildings and forest loss folded into training, and Google's Gemini helping adapt older training data to the EU's definitions. Experts added roughly 43,000 examples during correction rounds, fixing errors such as European vineyards labelled as tree crops.

The main driver is the EU Regulation on Deforestation-free Products, which bans goods linked to forests cleared or degraded after 31 December 2020. Because a forest's type depends on its history rather than its appearance, companies need to know how a site was used in 2020, and ForTy provides that reference point. It could also feed into initiatives such as the Tropical Forest Forever Facility and the UN target of safeguarding 30 percent of global land by 2030.

Accuracy varies sharply by type. Two experts reviewing 8,190 randomly selected sites found the map correct in 90.2 percent of cases overall, but that score is boosted by the easy "other land" class. It identified 89 percent of confirmed primary forests, yet got tree crops and planted woods right only around half the time and plantations 60 percent of the time, while roughly a quarter to a third of the ground it marked as plantations or crop trees turned out to be mislabelled. Those gaps fall exactly on the boundaries that matter most for the EUDR. Even so, ForTy outperformed European Commission maps built for the same rules on the same test data.

The researchers estimate the world holds about 4.1 billion hectares of forest, including roughly 2 billion hectares of primary forest, though the map overestimates primary forest in Asia by around 200 million hectares, largely in Russia's conifer forests, and misses about 94 million hectares in Africa, likely open dry forests and savannas.