The global dataset classifies land at 10-metre resolution into six categories: primary forest, naturally regenerating forest, planted forest, plantation forest, tree crops and agroforestry, and other land.
The release also includes an early annual time series for 2017–2025. Google says it can distinguish changes such as natural forest being replaced by a timber plantation, rather than recording only changes in tree cover. The model was trained with about 1.7 million labelled locations and uses AlphaEarth Foundations embeddings derived from years of satellite observations.
The partners report validation against more than 9,000 randomly selected sites, with 90% forest-type accuracy and 95% accuracy for separating forest from non-forest land. The dataset provides class probabilities rather than only fixed labels. ForTy 2020 and the time series are freely available under an open licence in Google Earth Engine.
The source does not establish that the map alone verifies compliance with the EU Deforestation Regulation, or that its reported global accuracy applies equally to every region, class, or individual land parcel.