Google recently released a new artificial intelligence model called AlphaEarth Foundations, which is humorously referred to as a "virtual satellite." Its goal is to completely revolutionize Earth observation, allowing scientists to track the dynamic changes of the Earth in ways never before possible.
According to the introduction, AlphaEarth Foundations uses a technology called "embedding" to collect massive public data every day from multiple sources such as optical satellites, radar, and climate simulations, and integrates and processes it. The model divides land and coastal areas into 10-meter × 10-meter grid cells, and tracks the changes in these micro-regions over a long period of time. Each grid is assigned a specific color code based on its attributes (such as vegetation type, surface material, etc.), forming a clear and intuitive data map.
Compared with traditional AI systems, AlphaEarth's "highly compact summary" mechanism can compress data storage requirements to 1/16, significantly reducing the cost of Earth observation, and providing feasibility for more extensive scientific research and commercial applications.
In terms of application, Google stated that scientists can use this model to generate high-resolution maps, widely used in crop health monitoring, deforestation tracking, urban expansion analysis, and other fields. In tests at multiple time scales, AlphaEarth Foundations has shown excellent performance, becoming an important tool for studying the state and dynamic changes of the Earth.
In the past year, more than 50 organizations have been the first to use the model's "Satellite Embedding" dataset to verify its value in real-world scenarios. Currently, Google has officially launched this dataset on the Google Earth Engine platform, making it available to global researchers, helping the scientific community to better understand the process of Earth's changes.