Research & Search
AlphaEarth Foundations is Google DeepMind's foundation model for geospatial data, functioning like a virtual satellite. It integrates optical imagery, radar, 3D laser mapping, and climate simulations into a single compact "embedding" for every 10x10 meter square of the planet's land and coastal waters, tracking change over time far more efficiently than prior AI mapping systems (16x less storage, 24% lower error rate in DeepMind's benchmarks). Released as the Satellite Embedding dataset in Google Earth Engine, it's already used by the UN FAO, Harvard Forest, MapBiomas, and Stanford to classify unmapped ecosystems, monitor deforestation and agriculture, and track urban expansion and water resources — free and open for researchers to build custom maps on Earth Engine.