Google DeepMind has introduced the AI system AlphaEarth Foundations , which allows for rapid analysis of satellite images and the creation of digital maps of the Earth’s surface with an accuracy of up to 10 meters. The system integrates data from various sources, including optical satellite images, radar, 3D scanning, and climate models, and compresses this information into compact digital “embeddings.” This reduces the volume of stored data by 16 times compared to other AI systems, and the average error rate in analysis decreases by nearly 24 percent.
AlphaEarth Foundations enables organizations to create detailed maps for monitoring agriculture, forest cover changes, water resources status, and urban development. The system works with historical data from 2017 to 2024 and supports change analysis over any time period. It also addresses the typical problem of satellite images—cloud cover and data gaps—by combining information from different sensors and filling in observation gaps.
More than 50 organizations are already testing the system in real projects. For example, the Global Ecosystems Atlas initiative uses AlphaEarth Foundations to classify unexplored ecosystems, while MapBiomas in Brazil analyzes changes in agriculture and the state of the Amazon. Researchers note that the new “Satellite Embedding” dataset helps create more accurate and faster maps that were previously unavailable.
AlphaEarth Foundations data is available in Google Earth Engine as the “Satellite Embedding” dataset, containing over 1.4 trillion digital imprints per year. Users can use this data to search for similar ecological conditions, detect changes, automatically group areas, and classify without large training datasets. For scientists, Google also offers research grants using this dataset.
