Charting the Planet with Precision Detail
Google DeepMind, in collaboration with Google Earth Engine, has made a significant stride in environmental research with the release of the AlphaEarth Foundations dataset. This extensive dataset, now publicly available through Google Earth Engine, offers a comprehensive collection of annual global embeddings from 2017 to 2024.
Each pixel in the dataset represents a 10x10 meter area of Earth’s surface, compactly summarising fused satellite imagery, radar, 3D mapping, and climate data into 64-dimensional vectors for efficient analysis. This innovative format reduces storage needs by a factor of 16 compared to conventional satellite data, enabling scalable planetary-scale environments monitoring and research.
Accessing and Utilising the AlphaEarth Satellite Embeddings Dataset
To access and use the dataset, simply sign up for or log into Google Earth Engine and search for or load the AlphaEarth Satellite Embeddings dataset (covering 2017-2024, 10m resolution). You can then use Earth Engine’s JavaScript or Python API to query embedding vectors for geographic areas and time periods of interest.
Tutorials and tools have been released to help visualize and work with these embeddings, including a 3D visualization tutorial and example notebooks implemented in Python with open-source libraries such as leafmap and geemap. These resources make it easier for researchers, policymakers, and businesses to analyse environmental changes, monitor ecosystems, and support sustainable development.
Key Steps to Get Started
- Sign up for or log into Google Earth Engine.
- Search for or load the AlphaEarth Satellite Embeddings dataset.
- Use Earth Engine’s JavaScript or Python API to query embedding vectors for geographic areas and time periods of interest.
- Apply ML or visualization techniques—examples include 3D plotting and temporal analysis using available notebooks and tutorials.
A Unique Resource for Environmental AI Tasks
The AlphaEarth geospatial embeddings dataset is one of the largest publicly available datasets of its kind, covering the entire planet with no gaps in coverage. These representations, called geospatial embeddings, provide consistent summaries for every 10x10 meter area of the planet's land and coastal waters.
The dataset is intended for use by researchers, organisations, and developers, allowing them to efficiently track changes in land use, vegetation, water, and development over time. With its high spatial and temporal precision, the dataset offers a unique and consistent way to compare and analyse different areas of the planet.
Best of all, the dataset has no associated costs for its use or download. Google DeepMind and Google Earth Engine have released this large dataset of highly detailed digital representations of Earth’s surface, totalling 1.8 billion geospatial embeddings, to foster a better understanding of our planet and support sustainable development.
For practical guidance, a tutorial video (dated August 6, 2025) and accompanying open notebooks are excellent resources to get started with accessing and applying this dataset. With these tools at your disposal, you can delve into the fascinating world of environmental AI tasks such as change detection and classification, making significant contributions to scientific research and sustainable development.
- The AlphaEarth Satellite Embeddings dataset, an extensive collection of annual global embeddings from 2017 to 2024, is a result of collaboration between Google DeepMind and Google Earth Engine, and it can be utilized for environmental research given its AI-powered format.
- This innovative dataset, now available through Google Earth Engine, offers a unique resource for researchers, policymakers, and businesses, as each pixel represents a 10x10 meter area of Earth’s surface and offers fused satellite imagery, radar, 3D mapping, and climate data in a 64-dimensional vector format.
- In addition to its significance for environmental science, the AlphaEarth dataset also promotes the development of artificial intelligence as it coincides with the growing interest in environmental AI tasks, such as change detection and classification, which could contribute to scientific research and sustainable development.
