Google's AlphaEarth AI Maps Any 10-Meter Area on Earth Using Satellite Data (blog.google) 7
Google today announced AlphaEarth Foundations, a new AI model that processes terabytes of daily satellite data to track environmental changes across the planet. The system, part of Google's broader Earth AI initiative, uses machine learning to compress satellite imagery into color-coded maps showing material properties, vegetation types, groundwater sources, and human constructions down to 10-meter resolution.
The model uses a technique called "embeddings" that reduces storage requirements by 16 times compared to other AI tools Google tested, while delivering 23.9% higher accuracy than similar systems. AlphaEarth has already mapped complex Antarctic terrain and identified variations in Canadian agricultural land use invisible to direct observation.
The technology currently powers flood and wildfire alerts in Google Search and Maps. Research organizations including Brazil's MayBiomas and the Global Ecosystems Atlas are using the system to analyze rainforests, deserts, and wetlands. The model integrates with Google Earth Engine, providing agencies like NASA and the Forest Service access to over one trillion annual data points for environmental monitoring and mapping applications.
The model uses a technique called "embeddings" that reduces storage requirements by 16 times compared to other AI tools Google tested, while delivering 23.9% higher accuracy than similar systems. AlphaEarth has already mapped complex Antarctic terrain and identified variations in Canadian agricultural land use invisible to direct observation.
The technology currently powers flood and wildfire alerts in Google Search and Maps. Research organizations including Brazil's MayBiomas and the Global Ecosystems Atlas are using the system to analyze rainforests, deserts, and wetlands. The model integrates with Google Earth Engine, providing agencies like NASA and the Forest Service access to over one trillion annual data points for environmental monitoring and mapping applications.
USA (Score:2, Informative)
For the Americans, 10m resolution is approximately 91.134 milli-football-fields.
What will this show? "Green"? (Re:USA) (Score:1)
For the Americans, 10m resolution is approximately 91.134 milli-football-fields.
For the suburban American that owns a 1/4 acre lot this is about nine "pixels" on their map.
If I'm doing my math right (and nobody should math in public as there's inevitably to be an error to be embarrassed about) then a common 1/4 acre lot is 100 feet by 100 feet, or 30 meters by 30 meters. If this satellite system maps the Earth in 10 meter "pixels" then your little piece of suburban America shows up as nine little dots on their map.
They want to track environmental changes on this map? Okay, map the ch
direct observation? (Score:2, Informative)
"AlphaEarth has already mapped complex Antarctic terrain and identified variations in Canadian agricultural land use invisible to direct observation."
How can AlphaEarth even be trained without data obtained by "direct observation"? That AlphaEarth can possibly do this is proof that claim is wrong.
But hey, if it justifies a $200M compensation package.
Re: (Score:1)
This is a problem common to all satellites. They can detect variations in many things with considerable precision but without a reference point on the ground this means nothing. Consider something so basic as detecting an aircraft over the ocean. The satellite can see the ocean, the aircraft, but without a reference on where the satellite is at the time there will be questions on the altitude, position, speed of that aircraft., and more. It's things like this that comes up all the time, including data f
Re: (Score:2)
I don't know what msmash meant by that line, but the only place it appears is in their summary.
There's a LOT of remote sensing + ground truth data, where "ground truth" actually means what it says, and various models have been trained on and tested against it for decades.
This is good AI (Score:2)
So that means Google is killing this off soon in favor of Gemini.