The Allen Institute for Artificial Intelligence (Ai2) has released OlmoEarth v1.3, a model that turns a year of satellite observations into numerical summaries of places on Earth. Ai2 also released precomputed summaries for the whole planet and a browser-based map tool, OlmoEarth Embeddings Explorer, which requires no sign-in.
The summaries, called embeddings, act like fingerprints for patches of land. Other software can use them to look for patterns such as forest, crops or mines. The smallest grid covers 10-by-10-meter patches; the source also lists 64-square-meter and 128-meter options. The model draws on radar and visible and infrared imagery from Sentinel-1, Sentinel-2 and Landsat.
Test land-change ideas without a GPU
Previously, using OlmoEarth generally meant fine-tuning the model for a particular task and running it over satellite imagery. That can require substantial computing power and downloading and storing the images. Ai2 says the Explorer lets users make an initial map or compare places in seconds, without training a model or using a GPU. More involved work can still call for fine-tuning.
The article describes comparing the Madre de Dios region of Peru across 2016–2020 and 2020–2024. The first comparison showed 34 square kilometers of significant change in a mining belt; the second showed 9 square kilometers. The article presents this as a way to see changes around illegal mining, but does not report an independent evaluation of those measurements.
Lower training costs, according to Ai2
Ai2 says v1.3 took 53% of the computing time required by v1.2 and 18% of the time required by the original v1. The company also reports that its carbon emissions were 50% and 22% of those earlier models, respectively. The source does not give a benchmark score or a detailed comparison of model accuracy, and says v1.2 remains useful for research and partner labs.
Why it matters
For environmental researchers and people exploring land use, the browser tool may make an initial comparison possible without access to a GPU or satellite-image storage. Ai2 says these first passes take seconds, but the source does not establish how accurate the maps are or whether the tool can replace expert review for decisions about mining, forests or agriculture.
