Google DeepMind and the World Resources Institute have created ForTy, a global satellite map that classifies the world's land cover for 2020 at a 10-meter resolution. The system categorizes land into six distinct classes, including primary forest, naturally regenerating forest, planted forest, plantation forest, tree crops like oil palm and rubber, and other land. Using Google DeepMind's AlphaEarth Foundations model alongside image recognition stages and Google's Gemini language model, the tool achieved an overall accuracy of 90.2 percent on test data.
The project was primarily driven by the need to support the EU Regulation on Deforestation-free Products, which prohibits goods linked to forests cleared or degraded after December 31, 2020. Because a forest's classification often depends on its history and management rather than just visual appearance, the system analyzes several years of satellite data, growth and harvest cycles, and outside datasets. Experts also incorporated thousands of corrective examples during development to refine the classifications and align them with strict definitions.
Despite its high overall accuracy, the map's performance varies significantly across different categories. While it successfully identifies primary forests, it proved less reliable for planted forests, plantations, and tree crops, identifying only about half of tree crops and planted forests and sixty percent of plantations. Why it matters: This AI-driven mapping tool provides companies and regulators with a crucial 2020 reference point to enforce deforestation-free supply chain rules and track global conservation goals.