Google DeepMind Releases Global 10-Meter Forest Typology Map

Mapping Global Forests at 10-Meter Resolution

Google DeepMind, Google Research, the World Resources Institute (WRI), and the International Institute for Applied Systems Analysis (IIASA) have released Forest Typology (ForTy) 2020. Released on October 22, 2026, the global map categorizes every 10-by-10-meter patch of land on Earth into six distinct forest and land-use types to help track environmental changes and assist with compliance like the European Union’s deforestation regulation (EUDR).

Alongside the 2020 baseline map, the organizations launched an early time series tracking yearly changes from 2017 to 2025. This allows users to observe not just where tree cover was lost, but also what replaced it, such as timber plantations or tree-crop farms.

How the ForTy Model Classifies Land Use

Standard tree-cover maps often fail to differentiate between an old-growth rainforest and a single-species timber plantation because both appear green from space. ForTy addresses this limitation by sorting land into six specific categories based on UN Food and Agriculture Organization (FAO) definitions:

• Primary forest: Native forest with no visible human disturbance.
• Naturally regenerating forest: Forest that has grown back independently following events like fires or logging.
• Planted forest: Trees intentionally planted or seeded and managed over long cycles.
• Plantation forest: Intensively managed, fast-growing tree farms typically featuring a single species.
• Tree crops and agroforestry: Commercial farms growing non-wood products such as oil palm, rubber, cocoa, or fruit.
• Other land: Everything else, including grasslands, croplands, urban areas, and water bodies.

Training, Validation, and Practical Application

The underlying machine learning model utilizes AlphaEarth Foundations embeddings—compact summaries of multi-year satellite data for each pixel—to detect subtle planting patterns, growth cycles, and harvest rhythms. Training relied on roughly 1.7 million labeled locations compiled from expert surveys, regional datasets, and targeted reviews.

During validation against more than 9,000 randomly chosen global sites assessed by trained experts, ForTy correctly identified forest types 90% of the time and accurately separated forest from non-forest land in 95% of cases. Rather than offering rigid binary labels, the system provides probability scores for each class, giving practitioners flexibility to adjust certainty thresholds depending on specific regulatory or research requirements.

Availability and Open Access

The Forest Typology 2020 map and the accompanying time series are available freely on Google Earth Engine under an open license. The 2020 map data has also been described in Scientific Data. Google intends to refine the time series data continuously based on user feedback and partner collaboration.

Source

Official announcement or documentation