Google ForTy Maps Global Forest Types at 10 Meter Resolution
Google's ForTy maps six forest and land-use types at 10-meter resolution, with 2017 to 2025 change data for EUDR compliance and conservation work.
Summary
On October 22, 2026, Google DeepMind and Google Research, with the World Resources Institute and International Institute for Applied Systems Analysis, released Forest Typology 2020, or ForTy. The open global map classifies every 10-by-10-meter land patch as primary forest, naturally regenerating forest, planted forest, plantation forest, tree crops and agroforestry, or other land. An early annual time series covering 2017 to 2025 identifies losses, regrowth and conversions, including natural forest replaced by timber plantations or farms. It extends 2025’s Natural Forests of the World 2020 map, which only separated natural forest from other tree cover.
ForTy was trained on about 1.7 million labeled locations from expert surveys, public datasets and targeted reviews. Its custom remote-sensing architecture uses AlphaEarth Foundations embeddings, compact summaries of years of satellite observations, to detect planting patterns, growth cycles and harvest rhythms. Validation by two experts at more than 9,000 randomly selected sites found 90% forest-type accuracy and 95% accuracy distinguishing forest from non-forest, outperforming other global maps. Class probability scores let users adjust precision thresholds, including companies managing EU Deforestation Regulation risks involving coffee, cocoa, palm oil, furniture and other wood products. Described in Scientific Data, the map and time series are freely available under an open license through Google Earth Engine. Governments, companies, scientists and conservation groups can use them for supply-chain checks, forest finance and conservation, while the partners refine the early time series over the coming months.
Positives
- 10-meter global coverage distinguishes six forest and land-use classes instead of treating all tree cover alike.
- 2017 to 2025 annual data exposes natural forests replaced by plantations or farms, even when tree cover quickly returns.
- 90% forest-type accuracy and 95% forest-versus-non-forest accuracy exceeded other global forest maps across more than 9,000 sites.
- 1.7 million labeled locations and AlphaEarth Foundations embeddings help detect subtle planting, growth and harvesting patterns.
- Open licensing through Google Earth Engine gives governments, companies, conservation groups and scientists free access.
- Probability scores allow EUDR users and researchers to choose thresholds suited to regulatory risk or forest-extent analysis.
Risks & concerns
- 90% forest-type accuracy leaves some sites incorrectly classified, requiring cautious use for regulatory decisions.
- The 2017 to 2025 time series remains an early release and will be refined over the coming months.
- Land-use categories can appear nearly identical from space because plantations and farms may quickly restore green cover.
- EUDR compliance requires distinctions that ordinary tree-cover maps cannot provide, including conversions between primary, regenerating, planted and plantation forests.