On October 22, 2026, Google DeepMind published “Beyond tree cover: mapping the world’s forest types” by Maxim Neumann and Charlotte Stanton, introducing Forest Typology (ForTy) 2020. The map classifies every 10-by-10-meter patch of land worldwide into six categories that follow UN FAO definitions: primary forest, naturally regenerating forest, planted forest, plantation forest, tree crops and agroforestry, and other land. The work was done with Google Research, the World Resources Institute and the International Institute for Applied Systems Analysis.
The model was trained on about 1.7 million labeled locations drawn from expert surveys, national and regional datasets and targeted reviews of hard areas, and uses AlphaEarth Foundations embeddings, which summarize years of satellite observations, as a key input. Validation against more than 9,000 randomly chosen sites, each assessed by two trained experts, found it identified the correct forest type 90% of the time and separated forest from non-forest correctly 95% of the time; DeepMind says this beats other global forest maps.
Instead of a single label, ForTy outputs a probability per class so users can choose stricter or looser thresholds. An early-release annual series for 2017 to 2025 shows where forest was lost or regrew and what replaced it, such as plantation replacing natural forest. The 2020 map and time series are free on Google Earth Engine under an open license.
Why it matters: tree-cover maps say whether trees are present, not what kind of forest they are. Distinguishing natural forest from plantations and tree crops matters for carbon accounting, biodiversity and deforestation rules such as the EU’s EUDR.
What it does not show: the time series is explicitly early-release, some classes depend on land use rather than appearance, and the authors note changes such as logging followed by replanting can be hard to detect from space. The accuracy figures come from the developers’ own validation.