AlphaEarth Foundations, developed by DeepMind, is a breakthrough AI model that integrates petabytes of multimodal Earth observation data to generate unified digital representations of terrestrial land and coastal waters. This “virtual satellite” enables consistent, high-resolution mapping and monitoring of the planet, overcoming longstanding challenges of fragmented datasets and limited satellite coverage. By compressing vast data into compact embeddings—requiring 16 times less storage than previous systems—AlphaEarth Foundations allows for rapid, large-scale analysis and detailed tracking of environmental changes. The model underpins the new Satellite Embedding dataset now available in Google Earth Engine, which is already being leveraged by over 50 organizations globally to classify ecosystems, monitor agricultural changes, and enhance conservation efforts. With superior accuracy (24% lower error rates than comparable models) and robust performance even with limited label data, AlphaEarth Foundations sets a new standard for geospatial intelligence. Its integration with broader AI agents, like Gemini, signals a future of even more powerful, automated environmental insights for research, policy, and industry.

Key Points

  • AlphaEarth Foundations unifies diverse Earth observation data into a single, highly efficient digital embedding for each 10x10m land or water square.
  • The model drastically reduces storage and computational costs, enabling planetary-scale mapping with 16x smaller data summaries than prior AI systems.
  • Outperforms traditional and AI-based mapping solutions, delivering 24% lower error rates and excelling even with sparse label data.
  • The Satellite Embedding dataset, powered by AlphaEarth, is driving real-world applications for ecosystem classification, agricultural monitoring, and conservation planning across global organizations.
  • Future integration with general reasoning LLMs like Gemini is expected to further enhance automated, actionable environmental insights.