Satellites orbiting Earth to capture images and measurements of land, forests, cities and coastal waters have long helped scientists better understand our planet. But because this data comes from disparate sources, it is often difficult to bring together into a single, clear picture.

Google’s artificial intelligence team, DeepMind, recently announced a model called AlphaEarth Foundations that can create extremely detailed maps of the world in near real time, Euronews reported.

“AlphaEarth Foundations acts as a virtual satellite, mapping the Earth anywhere, anytime,” Christopher Brown, a research engineer at Google DeepMind, said during a press conference in July.

“Whether they’re monitoring crop yields, tracking deforestation, or monitoring new construction, researchers no longer have to rely on a single satellite pass. They’re getting a new foundation for geospatial data,” Google DeepMind said in a statement last month.

The system combines trillions of images from dozens of public sources—including satellite imagery, radar scans, 3D laser mapping, and climate simulations—to map the entire land and coastal waters of the planet.

According to Google, the model can provide accurate enough information about ecosystems down to an area of 10 square meters. What’s more, AlphaEarth Foundations’ data takes up much less space than similar AI systems, making large-scale analysis much more practical.

In tests with data from 2017–2024, the model outperformed similar AI solutions in recognizing land use and assessing surface features, with an average error of 24% lower, according to a scientific paper published by DeepMind.

Google hopes the technology will help scientists study global changes related to food security, deforestation, urban expansion and water resource management.

Artificial intelligence – the new wave in environmental science

AlphaEarth Foundations is part of a growing trend in environmental research, in which AI turns a continuous stream of satellite observations into practical tools for studying the Earth.

The high resolution and regular updates of the data allow researchers to precisely measure environmental changes and understand their causes. They can be used to track the effects of climate change, plan conservation activities and manage resources such as water and agricultural land.

For example, in 2020, scientists from NASA and the University of Copenhagen mapped 1.8 billion individual tree crowns in the Sahel and Sahara regions of West Africa using AI trained to recognize trees in satellite images. Without AI, this would take years of work by millions of people, the study authors note.

Starting in 2022, NASA’s Surface Water and Ocean Topography (SWOT) satellite will take high-resolution measurements of oceans, lakes, reservoirs, and rivers, covering 90% of the planet’s surface. NASA’s Jet Propulsion Laboratory (JPL) says the technology provides a resolution ten times better than previous methods.

Meanwhile, in 2024, the European Space Agency (ESA) will launch the EarthCARE satellite, which will study how clouds and airborne particles affect Earth’s temperature balance.

Much of the data used by DeepMind also comes from long-standing NASA and ESA missions—such as the Landsat satellites, the Moderate Resolution Imaging Spectroradiometer (MODIS) and the Sentinel fleet—that monitor vegetation, coastlines, bodies of water, snow and ice.

AlphaEarth Foundations has already been tested by more than 50 organizations around the world for ecosystem monitoring and urban planning.

For example, the Brazilian initiative MapBiomas uses data from the model to better understand changes in agriculture and the environment, including in the Amazon rainforest.

The annual datasets have given the team “new opportunities to create more accurate, precise and rapidly produced maps – something that was previously impossible,” MapBiomas founder Tasso Azevedo said in a statement from Google.

The company announced that it would make the dataset available through Google Earth Engine, Google’s platform for environmental data, in order to encourage further research. | BGNES