A new AI model uses petabytes of Earth observation data to map it.
Google has introduced AlphaEarth Foundations, an AI model that acts as a virtual satellite. The system maps changes on Earth in detail and does so more efficiently than existing models.
Compact Summaries
AlphaEarth operates with a technique called “embedding”. It collects large amounts of public information daily, such as satellite images, radar data, and climate simulations. This data is combined and divided into blocks of 10 by 10 meters. Each block is assigned a color code based on characteristics like vegetation or material type.
AlphaEarth makes its data user-friendly by generating a compact summary for each block. According to Google, this requires 16 times less storage space than comparable models. This makes large-scale Earth observation cheaper and more accessible.
Mapping the Planet
Scientists can use the model to monitor crop health or track deforestation, for example. Google states that AlphaEarth performed strongly in tests across various domains and time scales. More than 50 organizations have already gained access to the underlying dataset over the past year.
This dataset is now part of the Google Earth Engine, allowing researchers worldwide to work with it.
read also