🔍 Read the full analysis: Understanding Earth Through OlmoEarth V1.3's AI Embeddings on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
The OlmoEarth team released v1.3 of its satellite embedding model, annual global data at 10-meter resolution and a browser-based explorer. The team says the model ranked highest in 20 of 27 evaluations, but those results have not been independently verified in the supplied material.
The OlmoEarth team has released version 1.3 of its satellite embedding model, as described in the original analysis, together with annual global embeddings at 10-meter resolution and a browser-based map explorer. The team reports that the model ranked highest in 20 of 27 evaluations, but those results are its own and have not been independently verified in the supplied material.
OlmoEarth v1.3 turns yearly satellite-image time series into 128-number representations for each 10-meter pixel. It uses Sentinel-1 radar data and multispectral imagery from Sentinel-2 and Landsat. A smaller version uses the first 64 dimensions. The team says precomputed global data is available through Source Cooperative, allowing users to work with the representations rather than first processing large satellite-image archives.
The accompanying OlmoEarth Embeddings Explorer lets users view the data as false-color maps, switch between years and compare it with underlying Sentinel-2 imagery. The release description also says users can select a location to find visually similar places or add labeled points to train a small classifier. The browser tool is presented as requiring no sign-in or advanced technical knowledge; the source does not specify what computing or access requirements may apply to larger projects.
In comparisons with AlphaEarth Foundations and Tessera v2, the developers say the 128-dimensional embeddings led in 20 of 27 evaluations and the 64-dimensional version led in 18. The evaluations covered six continents and multiple tasks. The team also introduced PLANTEUR, a benchmark for crop types in tropical farming, and reports that OlmoEarth generalized best to locations not seen by a classifier during training. Its technical report contains the methods and full results, according to the release.
Reusable Data for Land Mapping
Precomputed embeddings could reduce the storage and computing work involved in satellite-based mapping. Rather than repeatedly processing extensive image archives for each research question, users can apply smaller models to reusable numerical representations or compare them across years. This may make land-cover classification and change analysis more accessible to teams without dedicated GPU resources.
That convenience does not make the resulting maps self-validating. The reported benchmark rankings are developer-reported results, and performance in a particular region or task can differ from an aggregate score. Researchers and operational users still need to check outputs against suitable imagery, labels and other evidence before relying on them.
The team says model weights, pretraining data and code are available, and estimates that reproducing the final pretraining run for OlmoEarth v1.3 Base would cost about $2,720 in rented H100 computing time. That is the developers’ estimate; actual costs depend on provider, availability and setup.
satellite imagery analysis software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Earlier OlmoEarth work focused on fine-tuning a model for a specific mapping task through OlmoEarth Studio. According to the team, that workflow requires GPU access and the ability to download and store imagery. Embeddings offer another approach: imagery is processed into reusable features that can support different classification or change-mapping tasks.
The team says v1.3 used 53% of v1.2’s computing time and 18% of the computing time used for v1. It also reports emissions equal to 50% of v1.2 and 22% of v1, and says v1.3 matched v1.2 in fine-tuning performance across research and partner labels. The release material does not provide details of the emissions accounting, so those figures should be understood as the developers’ comparisons.
The announcement includes an example from Peru’s Madre de Dios region, where maps for 2017–2020 and 2020–2025 show changes around La Pampa. The team relates the example to mining activity and cites Amazon Conservation’s MAAP monitoring program and the 2019 Operation Mercurio crackdown. A change in the embeddings indicates a difference in the mapped landscape, but does not by itself establish its cause.
“OlmoEarth’s 128-dimensional embeddings score highest on 20 of 27 evals and rank highest overall.”
— The OlmoEarth team
As an affiliate, we earn on qualifying purchases.
Independent Checks Still Needed
The supplied release material does not establish whether independent researchers have replicated the 27 evaluation results. The technical report is identified as the source for methods and full results, but the selection of tasks and variation across regions will matter when judging how well the model suits a particular use.
The release description also leaves open how often the annual global embeddings will be updated, what data gaps or quality limits apply in particular locations, and what access or computing requirements users may face at scale. In the Peru example, the maps show detected change; attributing that change to mining, recovery or another cause requires corroboration from imagery, monitoring records or field evidence.
As an affiliate, we earn on qualifying purchases.
Testing the Maps Beyond Benchmarks
Users can explore the precomputed data through Source Cooperative and the browser-based explorer, while researchers can consult the team’s technical report and available code, weights and pretraining data. The next useful test is whether independent groups can reproduce the reported comparisons and measure performance on new locations and tasks.
For practical use, researchers will also need to assess how the embeddings behave with local labels, changing land conditions and incomplete imagery. The announcement does not give a schedule for future data updates or specify a next evaluation milestone, so those details remain pending.
satellite image classification software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did OlmoEarth release?
The team released OlmoEarth v1.3, annual global satellite embeddings at 10-meter resolution and a browser-based explorer for viewing and comparing them.
What is a satellite embedding?
It is a compact numerical representation learned from satellite imagery. OlmoEarth v1.3 produces a 128-number vector for each 10-meter pixel, with a 64-dimensional version also available.
Has the model’s reported performance been independently confirmed?
Not in the supplied source material. The OlmoEarth team reports that its model ranked highest in 20 of 27 evaluations; independent replication is not provided there.
Can the maps identify why a landscape changed?
No. Embedding differences can indicate detected landscape change, but identifying a cause such as mining requires corroborating evidence.
Primary source: Hugging Face · via ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
