Alphaearth Foundations helps us map our planets to unprecedented details

The new AI model integrates petabytes of Earth Observation Data to generate a unified representation of data that revolutionizes global mapping and monitoring

Every day, satellites capture information-rich images and measurements, providing scientists and experts with near-real-time views of our planet. This data is extremely impactful, but its complexity, multimodality and refresh rate create new challenges in connecting different data sets and using them all effectively.

Today we are introducing Alphaearth Foundations, an artificial intelligence (AI) model that functions like a virtual satellite. By integrating vast amounts of Earth observation data into unified digital representations that computer systems can easily handle, or “embedded,” it accurately and efficiently characterizes the planet’s entire land and entire coastal waters. This allows the model to provide scientists with a more complete and consistent image of our planet’s evolution, helping them make more informed decisions on important issues such as food security, deforestation, urban expansion, and water resources.

To accelerate research and unlock use cases, we are releasing an annual embedding collection from the Alphaearth Foundations as Google Earth Engine satellite embedding datasets. Over the past year, we have worked with over 50 organizations to test this dataset in real-world applications.

Our partners have already seen great benefits in using data to better classify unmapped ecosystems, understand agricultural and environmental changes, and significantly improve the accuracy and speed of mapping tasks. In this blog, we are excited to highlight some of their feedback and showcase the specific impact of this new technology.

How Alphaearth Foundations Work

Alphaearth Foundations offers a powerful new lens for understanding our planet by solving two key challenges: data overload and inconsistent information.

First, it combines information from many different public sources, including optical satellite images, radar, 3D laser mapping, and climate simulations. Weaving all this information together, analyzing the world’s lands and coastal waters in sharp 10x10m squares, allowing you to track changes over time with incredible accuracy.

Second, it becomes practical to use this data. A major innovation in the system is the ability to create very compact summaries for each square. These summaries tested the cost of planetary-scale analysis and require 16 times more storage space than the capacity generated by other AI systems that dramatically reduced the cost.

This breakthrough allows scientists to do things that are impossible to do before. Create detailed and consistent maps, on-demand, of our world. No longer will you need to rely on a single satellite passing over your head, such as monitoring crop health, tracking deforestation, or observing new construction. Now they have a new kind of foundation for geospatial data.

We have tested its performance rigorously to ensure that Alphaearth Foundations is ready for real use. When compared to both traditional methods and other AI mapping systems, Alphaearth Foundations was consistently the most accurate. They excel at a wide range of tasks over a variety of periods, including identifying land use and estimating surface properties. Importantly, if there is a lack of label data, we achieved this in the scenario. On average, Alphaearth Foundations exhibited an error rate of 24% lower than the models tested, showing superior learning efficiency. Find out more in our paper.

Generate custom maps with satellite embedded datasets

The satellite embedded dataset from Google Earth Engine with Alphaearth Foundations is one of the largest types with an embedded footprint of over 1.4 trillion per year. This annual embedded collection is already in use by organizations around the world, including the United Nations Food and Agriculture Organization, Harvard Forest, Earth Observation Group, MAPBIOMAS, Oregon State University, Spatial Informatics Group, and Stanford University.

For example, Global Ecosystems Atlas, an initiative aimed at creating the first comprehensive resource for mapping and monitoring global ecosystems, uses this dataset to help classify unmapped ecosystems into categories such as coastal shrubl and extreme deserts. This first role in the first resource will play a key role in helping countries prioritize conservation areas, optimize reconstruction efforts, and combat biodiversity losses.

In Brazil, Mapbiomas tests its dataset to better understand agriculture and environmental changes across the country. This type of map informs conservation strategies and sustainable development initiatives in important ecosystems like Amazon Rainforest.

As Tasso Azevedo, founder of Mapbiomas, said, “Satellite embedded datasets can change how your team works. Now there’s a new option to create more accurate, accurate and faster maps.

For more information about satellite embedded datasets, see the Google Earth Engine blog tutorial.

Empower others with AI

The Alphaearth Foundations represent important advances in understanding the changing planetary states and dynamics. Currently, we believe that using Alphaearth Foundations to generate annual embeddings and combine them with popular inference LLM agents like Gemini will be even more useful in the future. As part of Google Earth AI, we continue to apply the best way to apply the time-based capabilities of our models, a collection of Earth-space models and datasets, to help us address the most important needs of our planets.