DeepMind opens the ForTy forest map
On 22 October 2026 Google DeepMind and Google Research, with the World Resources Institute (WRI) and IIASA, released Forest Typology (ForTy) 2020, a global map that sorts every 10-by-10-metre patch of land into six forest and land-use types, and an early annual time series for 2017–2025. Both are free on Google Earth Engine under an open licence. The model takes AlphaEarth Foundations satellite embeddings as input.
Why it matters
Tree-cover maps say where trees stand; this one tries to say what kind of forest it is and what replaced it, a distinction the page says companies and regulators need under the EU's deforestation rules. The accuracy is the authors' own check: on more than 9,000 sites reviewed by two experts each, the forest type was right 90 per cent of the time and forest versus non-forest 95 per cent.
What the post says. ForTy sorts land into six classes based on FAO definitions: primary forest, naturally regenerating forest, planted forest, plantation forest, tree crops and agroforestry, other land. The training set is about 1.7 million labelled locations (expert surveys, national and regional datasets, targeted reviews); the key input is AlphaEarth Foundations embeddings. Validation: more than 9,000 randomly chosen sites, each studied by two experts with high-resolution imagery and years of satellite records; the forest type was right 90% of the time, forest against non-forest 95%, "outperforming other global forest maps". A probability is given for every class, so users set their own threshold. The 2017–2025 time series is an "early-release"; the authors say they will keep improving it "over the coming months". The map is described in Scientific Data (the paper was not read). The datasets are on Google Earth Engine "under an open license". The page's title reads "Not All Forests Are Equal"; its address still carries the name "Beyond tree cover". What the record does not claim. The accuracy and the advantage over other maps are the authors' statements; no independent check was found. The map describes 2020 and the series is early, and the page does not assess its quality. The data were not downloaded.