Google DeepMind and WRI release ForTy 2020: the first global forest-typology map at 10-meter resolution

📅 31 October 2026 • Google DeepMind Blog

Google DeepMind and Google Research, together with the World Resources Institute and IIASA, released Forest Typology 2020 (ForTy) - the first global map that classifies land into six forest and land-use types at 10-meter resolution. The classes follow FAO definitions: primary forest, naturally regenerating forest, planted forest, plantation forest, tree crops and agroforestry, and other land. The dataset paper is out in Scientific Data.

The model release behind the map. The classifier takes as a key input AlphaEarth Foundations embeddings - data-compact summaries of years of satellite observations for each 10-meter pixel. That lets it pick up subtle clues like planting patterns, growth cycles and harvest rhythms that separate a timber plantation from a natural forest. Training used about 1.7 million labeled locations from expert surveys, national and regional datasets and targeted reviews, combined in a cascaded deep-learning pipeline that also outputs per-class confidence maps. Because the labels follow FAO definitions, the classes line up with how governments and companies already report forest data.

Measured accuracy. With WRI and IIASA, the team validated ForTy against 8190 expert-labeled random sites: 90.2 percent overall accuracy across the six classes, 94.8 percent on natural forest, 95.5 percent on forest versus non-forest, outperforming the European Commission's comparable map on the same test data. The release also includes an early time series tracking how forests changed yearly from 2017 to 2025, including what replaced lost natural forest. The map is free on Google Earth Engine under an open license, and was built as a reference layer for the EU Deforestation Regulation with its 31 December 2020 baseline cutoff.

IMPACT. For geospatial engineers, conservation researchers and supply-chain teams, a free 10-meter typology layer changes what can be built in-house: class-level forest monitoring without commissioning custom classification, plus per-class confidence scores for tuning thresholds by region. Teams can move from tree-cover-only inputs to typology-aware pipelines on Earth Engine right away - the data is already published, with no application or waiting list involved.

Sources: https://deepmind.google/blog/beyond-tree-cover-mapping-the-worlds-forest-types/ https://www.nature.com/articles/s41597-026-08299-5 https://the-decoder.com/google-deepmind-creates-a-global-map-of-forests-with-satellite-data-and-ai/

Continue the conversation: https://t.me/AIAutomationArchitecture · Bluesky