Google DeepMind has introduced AlphaEarth Foundations, a new AI model designed to map Earth's terrestrial land and coastal waters with high precision. By processing petabytes of Earth observation data, the system offers scientists a detailed and real-time view of the planet, supporting efforts to monitor climate change, deforestation, urban expansion, and other environmental issues.

AlphaEarth Foundations integrates data from multiple public sources, including optical satellite images, radar, 3D laser mapping, and climate simulations. It creates a unified digital representation, or "embedding," for each 10x10 meter square of land and coastal waters across the globe. These embeddings provide a compact summary of data, requiring 16 times less storage space than comparable AI systems, which reduces the cost of large-scale analysis.

The system acts like a virtual satellite, enabling researchers to track changes over time with greater consistency and accuracy. Unlike traditional methods that rely on single satellite passes, AlphaEarth Foundations combines diverse datasets to produce a comprehensive picture of Earth's surface.

Performance and Applications

In testing, AlphaEarth Foundations outperformed traditional methods and other AI systems, achieving an average error rate 24 percent lower than competing models. It excelled in tasks such as identifying land use and estimating surface properties, even when labeled data was limited, demonstrating its efficiency in learning from sparse datasets.

The model supports a range of applications, including monitoring crop health, tracking deforestation, and observing urban development. Organizations like the United Nations' Food and Agriculture Organization, Harvard Forest, Group on Earth Observations, MapBiomas, Oregon State University, the Spatial Informatics Group, and Stanford University have already used AlphaEarth Foundations' dataset to create custom maps for their research.

Open Access to Data

DeepMind has made a collection of annual embeddings from AlphaEarth Foundations available through Google Earth Engine as the Satellite Embedding dataset. This dataset, one of the largest of its kind, contains over 1.4 trillion embedding footprints per year. Over the past year, more than 50 organizations have tested this dataset for real-world applications, contributing to insights that inform environmental decision-making and policy.

Supporting Global Research

AlphaEarth Foundations is designed to adapt to the evolving landscape of Earth observation, where new measurement platforms are launched, and others are retired. By providing a flexible and efficient tool for geospatial analysis, the system supports scientists and policymakers working on critical issues like food security and water resource management.