New Google AI model maps world in 10-meter squares for machines to read
(2025/07/31)
- Reference: 1753995907
- News link: https://www.theregister.co.uk/2025/07/31/google_ai_maps_world/
- Source link:
Google has released a new AI model that maps the world in 10-meter squares for machines to read.
The company's AlphaEarth Foundations model has been trained on vast amounts of Earth observation data from satellites and other sources to produce embeddings for computer programs, rather than visual imagery or some other form of output intended for human interpretation.
AlphaEarth Foundations, [1]according to Google DeepMind researchers, "accurately and efficiently characterizes the planet’s entire terrestrial land and coastal waters by integrating huge amounts of Earth observation data into a unified digital representation, or ' [2]embedding, ' that computer systems can easily process."
[3]
The model thus is able to provide scientists with a comprehensive view of Earth over time, to help with research into environmental and agricultural concerns like deforestation and [4]water resources drained by datacenters running AI models.
[5]
[6]
Embeddings are vectors – numerical representations of data points – created by machine learning models that encode data about some object. They define how different objects within a model are related to each other.
Vectors in math are modelled in dimensional space. A two-dimensional vector, for example, could be represented by two numbers, X and Y.
[7]
AlphaEarth's embeddings have 64 dimensions, each representing a 10-meter pixel that encodes data from multiple sources about terrestrial surface conditions around that plot over the course of a year.
AlphaEarth makes Earth observation information available as a data set of embeddings so that it can be processed efficiently by deep learning applications like Google Earth Engine. According to its creators, the model's succinct data summaries for its 10-meter squares allow it to operate with 16x less storage space and lower cost than other AI systems. The model is also said to be more accurate, with an error rate 24 percent lower than other models.
Google DeepMind researchers say AlphaEarth improves upon prior geospatial foundation models like SatMAE (Cong et al., 2022) and SatCLIP (Klemmer et al., 2025) by averaging the embeddings across multiple sources, by including time into the modeling framework, and by offering spatial resolution at high precision.
[8]Bitter fight over 2020 Microsoft quantum paper both resolved and unresolved
[9]The TSA likes facial recognition at airports. Passengers and politicians, not so much
[10]Lethal Cambodia-Thailand border clash linked to cyber-scam slave camps
[11]How Google profits even as its AI summaries reduce website ad link clicks
With the announcement of AlphaEarth, Google is releasing its [12]Satellite Embedding dataset , produced by the model, for use in applications like Earth Engine.
Google research scientist Valerie Pasquarella and product manager Emily Schechter [13]suggest the dataset can be used for conducting a similarity search (identifying areas similar to other areas based on specified criteria), detecting geographical change, discovering hidden patterns, and creating maps without manual labeling.
[14]
Christopher Seeger, professor and extension specialist of landscape architecture and geospatial technology at Iowa State University, told The Register in a phone interview that he expects AlphaEarth Foundations will be quite beneficial.
"Dealing with discrete data sets alone can be challenging, but when you bring in multiple disparate data sets and are trying to do the analysis it can be an intensive use of the machine," he said. "And even at best, you can usually only do small areas of what you're studying."
Leaning on AI to look at features across multiple data sets, he said, looks like an interesting use of 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 said the application of AI to this sort of data makes a lot of sense.
"What is interesting is that they're able to get down to 10 by 10 meter squares, which is phenomenal," he said. "It's going to be great for decision makers." ®
Get our [15]Tech Resources
[1] https://deepmind.google/discover/blog/alphaearth-foundations-helps-map-our-planet-in-unprecedented-detail/
[2] https://developers.google.com/machine-learning/crash-course/embeddings/embedding-space
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aIvnd9JAbqbT_UXxyh4sdAAAAIg&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://www.nytimes.com/2025/07/14/technology/meta-data-center-water.html
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aIvnd9JAbqbT_UXxyh4sdAAAAIg&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aIvnd9JAbqbT_UXxyh4sdAAAAIg&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aIvnd9JAbqbT_UXxyh4sdAAAAIg&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://www.theregister.com/2025/07/31/microsoft_quantum_paper_science/
[9] https://www.theregister.com/2025/07/31/tsa_facial_recognition/
[10] https://www.theregister.com/2025/07/31/thai_cambodia_war_cyberscam_links/
[11] https://www.theregister.com/2025/07/29/opinion_column_google_ai_ads/
[12] https://developers.google.com/earth-engine/datasets/catalog/GOOGLE_SATELLITE_EMBEDDING_V1_ANNUAL#description
[13] https://medium.com/google-earth/ai-powered-pixels-introducing-googles-satellite-embedding-dataset-31744c1f4650
[14] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aIvnd9JAbqbT_UXxyh4sdAAAAIg&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[15] https://whitepapers.theregister.com/
The company's AlphaEarth Foundations model has been trained on vast amounts of Earth observation data from satellites and other sources to produce embeddings for computer programs, rather than visual imagery or some other form of output intended for human interpretation.
AlphaEarth Foundations, [1]according to Google DeepMind researchers, "accurately and efficiently characterizes the planet’s entire terrestrial land and coastal waters by integrating huge amounts of Earth observation data into a unified digital representation, or ' [2]embedding, ' that computer systems can easily process."
[3]
The model thus is able to provide scientists with a comprehensive view of Earth over time, to help with research into environmental and agricultural concerns like deforestation and [4]water resources drained by datacenters running AI models.
[5]
[6]
Embeddings are vectors – numerical representations of data points – created by machine learning models that encode data about some object. They define how different objects within a model are related to each other.
Vectors in math are modelled in dimensional space. A two-dimensional vector, for example, could be represented by two numbers, X and Y.
[7]
AlphaEarth's embeddings have 64 dimensions, each representing a 10-meter pixel that encodes data from multiple sources about terrestrial surface conditions around that plot over the course of a year.
AlphaEarth makes Earth observation information available as a data set of embeddings so that it can be processed efficiently by deep learning applications like Google Earth Engine. According to its creators, the model's succinct data summaries for its 10-meter squares allow it to operate with 16x less storage space and lower cost than other AI systems. The model is also said to be more accurate, with an error rate 24 percent lower than other models.
Google DeepMind researchers say AlphaEarth improves upon prior geospatial foundation models like SatMAE (Cong et al., 2022) and SatCLIP (Klemmer et al., 2025) by averaging the embeddings across multiple sources, by including time into the modeling framework, and by offering spatial resolution at high precision.
[8]Bitter fight over 2020 Microsoft quantum paper both resolved and unresolved
[9]The TSA likes facial recognition at airports. Passengers and politicians, not so much
[10]Lethal Cambodia-Thailand border clash linked to cyber-scam slave camps
[11]How Google profits even as its AI summaries reduce website ad link clicks
With the announcement of AlphaEarth, Google is releasing its [12]Satellite Embedding dataset , produced by the model, for use in applications like Earth Engine.
Google research scientist Valerie Pasquarella and product manager Emily Schechter [13]suggest the dataset can be used for conducting a similarity search (identifying areas similar to other areas based on specified criteria), detecting geographical change, discovering hidden patterns, and creating maps without manual labeling.
[14]
Christopher Seeger, professor and extension specialist of landscape architecture and geospatial technology at Iowa State University, told The Register in a phone interview that he expects AlphaEarth Foundations will be quite beneficial.
"Dealing with discrete data sets alone can be challenging, but when you bring in multiple disparate data sets and are trying to do the analysis it can be an intensive use of the machine," he said. "And even at best, you can usually only do small areas of what you're studying."
Leaning on AI to look at features across multiple data sets, he said, looks like an interesting use of 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 said the application of AI to this sort of data makes a lot of sense.
"What is interesting is that they're able to get down to 10 by 10 meter squares, which is phenomenal," he said. "It's going to be great for decision makers." ®
Get our [15]Tech Resources
[1] https://deepmind.google/discover/blog/alphaearth-foundations-helps-map-our-planet-in-unprecedented-detail/
[2] https://developers.google.com/machine-learning/crash-course/embeddings/embedding-space
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aIvnd9JAbqbT_UXxyh4sdAAAAIg&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://www.nytimes.com/2025/07/14/technology/meta-data-center-water.html
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aIvnd9JAbqbT_UXxyh4sdAAAAIg&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aIvnd9JAbqbT_UXxyh4sdAAAAIg&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aIvnd9JAbqbT_UXxyh4sdAAAAIg&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://www.theregister.com/2025/07/31/microsoft_quantum_paper_science/
[9] https://www.theregister.com/2025/07/31/tsa_facial_recognition/
[10] https://www.theregister.com/2025/07/31/thai_cambodia_war_cyberscam_links/
[11] https://www.theregister.com/2025/07/29/opinion_column_google_ai_ads/
[12] https://developers.google.com/earth-engine/datasets/catalog/GOOGLE_SATELLITE_EMBEDDING_V1_ANNUAL#description
[13] https://medium.com/google-earth/ai-powered-pixels-introducing-googles-satellite-embedding-dataset-31744c1f4650
[14] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aIvnd9JAbqbT_UXxyh4sdAAAAIg&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[15] https://whitepapers.theregister.com/