Google DeepMind has introduced AlphaEarth Foundations, an AI model built to turn vast and uneven Earth observation data into a single, usable digital layer. The system is meant to improve environmental analysis and support decisions around food security, deforestation, water resources, and ecosystem mapping.
DeepMind describes the model as a "virtual satellite" because it characterizes all land masses and coastal waters on Earth at a 10x10 meter resolution. Instead of relying on one kind of input, AlphaEarth Foundations brings together several sources and compresses them into data-efficient representations that can be used for downstream analysis.
How AlphaEarth Foundations Maps the Planet
The core idea behind AlphaEarth Foundations is to make complex Earth data easier to compare, search, and analyze. Satellite systems generate enormous amounts of information, but that information can be inconsistent, incomplete, or difficult to combine. DeepMind says the model addresses two core problems: data overload and inconsistent information.
AlphaEarth Foundations integrates multiple data sources, including optical satellite imagery, radar, 3D laser mapping, and climate simulations. It then compresses those inputs into 64-dimensional embeddings. These embeddings are compact numerical representations designed to preserve useful information while making the data easier to work with.
The training process used over three billion observations from more than five million locations worldwide. The system draws on satellite missions such as Sentinel-2 and Landsat, and it also incorporates text sources like Wikipedia articles and species observations.
That mix of inputs matters because environmental analysis often depends on seeing patterns across space and time. A single satellite image may show only part of the picture. AlphaEarth Foundations is designed to combine many signals into a more coherent representation of what is happening on land and along coastlines.
What The Model Can Detect
According to the source article, AlphaEarth Foundations can operate in difficult mapping conditions. It can see through persistent cloud cover, map intricate surfaces in Antarctica, and identify subtle variations in Canadian agriculture that are invisible to the human eye.
Those examples point to one of the model's main uses: extracting useful environmental information even when the raw data is hard to interpret. Clouds, irregular observations, and complex terrain can all make Earth monitoring less reliable. A system that can combine different measurements may help analysts work around those gaps.
In head-to-head tests against traditional approaches and other AI mapping systems, AlphaEarth Foundations achieved an average of 24 percent lower error rates, according to the research paper cited in the source. It also outperformed all other methods across 15 evaluation datasets.
Those datasets covered several task types:
- land use classification
- biophysical variable estimation
- change detection
The system is also described as useful in places where little processed data exists. Its continuous temporal analysis supports predictions across arbitrary time periods, including periods that do not fully align with the original input data.
Why Time Is Central To The System
AlphaEarth Foundations uses a "Space Time Precision" (STP) architecture. The source article describes this approach as treating satellite images from one location over time like frames in a video.
That structure lets the model learn relationships across space, time, and measurement types. Instead of looking only at isolated images, the system builds embeddings that capture local context and temporal trajectories. In plain terms, it is designed to represent both what a place looks like and how it changes.
Satellite images often arrive at irregular intervals. AlphaEarth Foundations arranges those images chronologically and transforms them into continuous embeddings. The result is a smoother view of sites over time, paired with measurement data.
For environmental monitoring, that time-aware design is important. Changes in agriculture, ecosystems, water resources, or forests are rarely captured by one clean snapshot. The value comes from tracking patterns and differences across years, seasons, and available measurements, while still working within the limits of the source data.
Real-World Testing And Available Data
More than 50 organizations are already testing AlphaEarth Foundations in real-world applications. The Global Ecosystems Atlas is using the data to classify previously unmapped ecosystems, including categories such as coastal shrublands and hyper-arid deserts.
Nick Murray, director of the James Cook University Global Ecology Lab, described the importance of the Satellite Embedding dataset for conservation work:
"The Satellite Embedding dataset is revolutionizing our work by helping countries map uncharted ecosystems - this is crucial for pinpointing where to focus their conservation efforts."
In Brazil, MapBiomas is using the data to gain deeper insight into agricultural and environmental changes, especially in critical ecosystems like the Amazon rainforest. The source article does not describe the specific outputs of that work, but it identifies the use case as environmental and agricultural change analysis.
Google is making the annual embeddings available as the Satellite Embedding Dataset in Google Earth Engine. The dataset generates over 1.4 trillion embedding footprints per year, making it one of the largest datasets of its kind. Annual embeddings from 2017 to 2024 are available under the Apache-2.0 license.
According to Google Earth Engine, the dataset supports several applications. These include similarity search for finding comparable environmental conditions worldwide, change detection by comparing embedding vectors across years, automatic clustering without predefined labels, and smarter classification using much less training data.
Research Support And Next Steps
Google is also offering research grants of up to $5,000 to scientists working on Satellite Embedding use cases. The source article says submissions will be accepted over the coming months, with the goal of accelerating research and publication focused on the dataset.
The developers frame AlphaEarth Foundations as a step toward "understanding the state and dynamics of our changing planet." They also believe that combining the system with general reasoning LLM agents like Gemini could enable more powerful applications.
For now, the immediate significance is practical. AlphaEarth Foundations packages many kinds of Earth observation data into a format that organizations can test, compare, and apply through Google Earth Engine. If the system performs as described, it could make planetary-scale environmental mapping more consistent, searchable, and useful for real-world decisions.