Google released a new AI map that maps the globe in 10-meter squares. The model is designed for machines.
The AlphaEarth Foundations model was trained using vast amounts of Earth-observation data from satellites, and other sources. This is a model that produces embeddings rather than visual imagery for human interpretation. AlphaEarth Foundations is a model that, according to Google DeepMind researchers “accurately characterizes Earth’s entire land surface and coastal waters, by integrating enormous amounts of Earth observation data in a unified digital format, or “a target=”_blank” ” rel=”nofollow” href=”https://developers.google.com/machinelearning/crash course/embeddings/embedding space[19659003[19659003
Vectors – numerical representations for data points – are created by machine-learning models to encode data about an object. They describe how objects in a model are related. In math, vectors are modeled in dimensional space. Two numbers, X & Y, can be used to represent a two-dimensional vector.
AlphaEarth embeddings are 64-dimensional, with each pixel representing a 10-meter area. This data is derived from multiple sources and encodes the surface conditions of the earth around that plot for a period of one year.
AlphaEarth provides Earth observation data as a set of embeddings that can be processed by deep learning applications such as Google Earth Engine. According to its creators the model’s concise data summaries allow it to operate on 16x less storage and at a lower cost than other AI system. The model is said to be more accurate with a 24 percent lower error rate than other models.
Google DeepMind researches say AlphaEarth is an improvement over previous geospatial foundations models like SatMAE and SatCLIP. These models were developed by Cong et. al. in 2022 and Klemmer et. al. in 2025. They claim that AlphaEarth offers spatial resolution with high precision, by averaging embeddings from multiple sources and by including time within the modeling framework. The TSA is a fan of facial recognition in airports. Passengers and politicians are not as enthusiastic.
Google has released its Satellite Embedding dataset, produced by the model for use in applications such as Earth Engine. Google’s Valerie Pasquarella, a Google research scientist, and Emily Schechter, a product manager, suggest that the datasetcan be used to conduct a similarity (identifying areas that are similar to others based on specified criteria), detect geographical change, discover hidden patterns, and create maps without manual labels. In a phone conversation, Christopher Seeger, professor at Iowa State University and extension specialist in landscape architecture and geospatial technologies, said that he believes AlphaEarth Foundations to be very beneficial. He said. “And even at best, you can usually only do small areas of what you’re studying.”
He said that using AI to analyze features in multiple data sets is an interesting way to use the technology.
“I’m happy to see that they are doing some ground truthing with this to find out how reliable the models are, of course. And so I look forward to seeing what’s possible with this, not just at a global scale, but at a more regional scale.”
Seeger stated that the application of AI for this type of data makes sense. He said
“What is interesting is that they’re able to get down to 10 by 10 meter squares, which is phenomenal,” . “It’s going to be great for decision makers.” (r)
