Google DeepMind maps global forest types at 10-meter resolution

Google DeepMind released a global forest mapping AI that categorizes land into 12 distinct forest types at a 10-meter resolution. The model moves beyond simple tree-cover detection to identify specific ecosystems such as mangroves, tropical moist forests, and needleleaf forests. This granular data allows for more accurate tracking of biodiversity and carbon sequestration across the planet.

What's new

The Google DeepMind forest mapping model uses satellite imagery from the Sentinel-2 mission to provide a detailed taxonomy of Earth's forests. As of late 2024, the model classifies every 10-meter patch of land, offering a significant resolution improvement over previous global datasets that often relied on 30-meter or 100-meter scales. The system identifies 12 specific classes, including flooded forests, deciduous broadleaf forests, and evergreen needleleaf forests. By analyzing seasonal patterns and spectral data, the AI can distinguish between forest types that look similar in static imagery but function differently within their local ecosystems. This structural classification is critical for understanding the health of a forest, as it differentiates between primary growth and secondary plantations.

How it fits your workflow

For environmental researchers, documentarians, and spatial analysts, the Google DeepMind forest mapping AI serves as a critical tool for habitat analysis and location scouting. It replaces or augments general land-cover datasets like the ESA WorldCover or the Global Forest Watch tree-cover layers. While those tools are useful for detecting general deforestation, they often lack the taxonomic detail required to understand which specific species or ecosystems are at risk.

Filmmakers focusing on conservation can use this data to find specific biomes, such as rare mangrove stands or primary tropical forests, with higher accuracy than previous satellite-based tools. The 10-meter resolution ensures that even small fragments of forest are captured, making it a valuable asset for VFX artists or digital twin creators who need accurate ecological data for environmental modeling. In terms of spatial intelligence, this model competes with high-resolution land-cover products from ESRI, but with a specialized focus on forest taxonomy that is essential for calculating carbon credits or assessing wildfire vulnerability in specific biomes. This level of detail helps editors and producers verify the ecological context of footage or plan complex shoots in specific forest types.

What it costs / how to try it

The Google DeepMind forest mapping data is available as an open-access dataset. Researchers and developers can access the map layers through Google Earth Engine for use in environmental monitoring and conservation projects as of late 2024.