Google DeepMind launches AI model that works like ‘virtual satellite’

Google DeepMind

, Google’s AI unit, has introduced an AI model,

AlphaEarth Foundations

, that integrates a massive cache of

Earth observation data

and generates a data representation that can help scientists and researchers understand and monitor our planet. The company says that the newest

AI model

functions like a “virtual satellite,” to provide insights into global changes.According to Google DeepMind, the model accurately and efficiently characterises the planet’s entire terrestrial land and coastal waters by integrating huge amounts of Earth observation data into a unified digital representation, or "embedding," that computer systems can easily process. This allows the model to provide scientists with a more complete and consistent picture of our planet's evolution.

How AlphaEarth Foundations AI model works

According to the company, the data has been taken by satellites that capture information-rich images and measurements, providing scientists and experts with a nearly real-time view of our planet. The data is impactful, however, its complexity, multimodality and refresh rate creates a new challenge of connecting different datasets and making use of them all effectively.

The AI model visualises the rich details of the world by assigning the colours red, green and blue to three of the 64 dimensions. For instance, in Ecuador, AlphaEarth Foundations can “see through persistent cloud cover” to detail agricultural plots. In Antarctica, it maps complex surfaces in clear detail, an area notoriously difficult for irregular satellite imaging. It can even reveal variations in Canadian agricultural land use that are invisible to the naked eye.Google says it tested AlphaEarth Foundations, consistently finding it to be the most accurate when compared against traditional methods and other AI mapping systems. To accelerate research and unlock use cases, Google is releasing a collection of AlphaEarth Foundations’ annual embeddings as the Satellite Embedding dataset in

Google Earth Engine

. The company is currently using the model to generate annual embeddings and believes its utility could be further amplified by combining it with general reasoning LLM agents like

Gemini

in the future.