Google DeepMind, Google Research, WRI and IIASA release ForTy 2020, a global 10-meter forest-type map, plus an early 2017-2025 time series, on Earth Engine

Google DeepMind, Google Research, the World Resources Institute and IIASA released ForTy 2020, a global map that sorts every 10 by 10 metre patch of land into six classes: primary forest, naturally regenerating forest, planted forest, plantation forest, tree crops or agroforestry, and other land. It is free on Google Earth Engine under an open licence. An early annual time series for 2017 to 2025 is included, and it is meant to show conversions such as natural forest turning into plantation or farmland. The model uses AlphaEarth Foundations embeddings and about 1.7 million labelled locations, and it outputs a probability for each class. The authors report 90% forest-type accuracy and 95% forest versus non-forest accuracy, checked on over 9,000 random sites by two experts each. A paper is described in Scientific Data.

Ordinary tree-cover maps count a palm plantation and an old-growth forest as the same thing. A map that separates them is useful to researchers, commodity buyers and regulators who need to show that supply chains are deforestation-free under the EU Deforestation Regulation. The open licence and Earth Engine access lower the barrier for developers and NGOs. The impact is still conditional: it gives a baseline and a screening tool, not legal proof of compliance. The time series is labelled early, so conversion tracking is less mature than the static 2020 map.

The headline accuracy figures and the claim that it beats other global forest maps are self-reported, and the summary does not say which maps were compared or how much better it is. The compliance framing is the authors' own: nothing here says regulators or auditors accept ForTy as evidence. It also builds on last year's Natural Forests map, so it is a step forward rather than a new capability. A 90% overall figure can hide weaker results on hard classes such as primary versus regenerating forest, and in particular regions.

1. 01Independent validation or critiques of the per-class accuracy, especially primary versus naturally regenerating forest and plantation versus tree crops, in tropical regions, from academics or groups outside the authors' partners.
2. 02Evidence of real uptake: named companies, certification schemes or EU member state regulators saying they use ForTy for EUDR due diligence, and Earth Engine usage or citations in the months after release.
3. 03How the 2017 to 2025 time series matures: whether it moves from early release to a validated product, and whether the map is updated beyond the 2020 baseline.