DeepMind Unveils AlphaEarth AI, Promising Sharper Planetary Mapping

Google DeepMind has unveiled AlphaEarth Foundations, an artificial-intelligence model that integrates optical, radar, LiDAR and climate data to create a unified digital representation of the planet. The company says the system cuts error rates by roughly 23.9% compared with existing approaches while compressing data 16-fold, enabling faster and cheaper analysis of petabyte-scale satellite imagery. The model produces "embedding fields" at 10-meter resolution for every 10×10-metre square on Earth, allowing users to track deforestation, crop conditions and other environmental changes in near real time. Google is releasing a Satellite Embedding dataset containing about 1.4 trillion footprints per year from 2017 through 2024 via Google Earth Engine, positioning the tool as part of its broader Earth AI portfolio. More than 50 organizations—including Brazil’s MapBiomas and the Global Ecosystems Atlas initiative—have tested AlphaEarth Foundations to monitor the Amazon rainforest and classify previously unmapped ecosystems. Google stresses that the resolution is sufficient for environmental monitoring but not detailed enough to identify individuals, addressing privacy concerns.