🔍 Read the full analysis: OlmoEarth V1.3: AI Embeddings On A Global Scale 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 has released v1.3 of its satellite embedding model, annual global data at 10-meter resolution and a browser-based explorer. The team says its 128-dimensional version 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, alongside annual global embeddings at 10-meter resolution and an online map explorer, as described in the original analysis. The developers say the model ranked highest in 20 of 27 evaluations, but those results are the team’s own and have not been independently verified in the source material.
OlmoEarth v1.3 processes yearly time series from Sentinel-1 radar and multispectral imagery from Sentinel-2 and Landsat, with custom embedding exports available for downstream analysis. It produces a 128-number representation, or embedding, for each 10-meter pixel; users can also use the first 64 dimensions as a smaller representation. The team says the precomputed global data is available through Source Cooperative.
The accompanying OlmoEarth Embeddings Explorer lets users view the data as false-color maps, switch between years and compare the results with Sentinel-2 imagery. According to the release description, users can select a location to find visually similar places or add labeled points to train a small classifier. The tool is presented as usable without signing in or advanced technical knowledge.
In comparisons with AlphaEarth Foundations and Tessera v2, the team reports that the 128-dimensional embeddings ranked highest in 20 of 27 evaluations and the 64-dimensional version in 18. The evaluations covered six continents and multiple tasks. The team also introduced PLANTEUR, a crop-type benchmark focused on tropical farming, and says OlmoEarth generalized best to locations the classifier had not seen during training. Full methods and results are in the team’s technical report.
Reusing Satellite Data for Mapping
Precomputed embeddings could reduce the work involved in using satellite imagery for land-cover mapping and change detection. Rather than downloading and processing large image archives for each project, a user can work with the model’s compact representations and train a smaller classifier for a specific task. That may help researchers and organizations without substantial computing resources explore satellite-based analysis.
The release does not remove the need to validate results. A model’s aggregate evaluation ranking does not establish how well it will perform for every region, label set or operational decision. Users still need to check maps against underlying imagery and other evidence, particularly when the results may inform environmental monitoring or public policy.
The team says it has published the model weights, pretraining data and code. It estimates that reproducing the final pretraining run for OlmoEarth v1.3 Base would cost about $2,720 in rented H100 computing time. That is a developer estimate, not a guaranteed cost; prices can vary by provider, availability and setup.
satellite imagery analysis software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How OlmoEarth Reached Version 1.3
Earlier OlmoEarth workflows centered on fine-tuning a model for a particular mapping task through OlmoEarth Studio, then running it over satellite imagery. The team says that approach requires GPU access and the ability to download and store imagery. Embeddings offer another workflow: process imagery into reusable vectors that can support multiple mapping tasks.
The developers say v1.3 used 53% of the computing time of v1.2 and 18% of the computing time of v1 to train. They also report emissions equal to 50% of v1.2 and 22% of v1, and say v1.3 matches v1.2 in fine-tuning performance across research and partner labels. These are team-reported comparisons; the supplied material does not provide details of the emissions accounting.
The release illustrates the embeddings with maps of mining-related landscape change in Peru’s Madre de Dios region, including differences around La Pampa across 2017–2020 and 2020–2025. The material cites Amazon Conservation’s MAAP monitoring program and reporting on mining activity. The maps indicate detected change, but embeddings alone do not 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.
Validation and Coverage Questions
The reported evaluation results come from the developers. Although the team’s technical report contains the methods, the supplied material does not include an independent replication of the 27 evaluations. How task selection, regional variation and specific labels affect results will matter when users assess the model for a particular application.
The release description also does not specify how often the annual embeddings will be updated, what data gaps or quality limits apply in every location, or what access and computing requirements users may face for large-scale work. The source material does not state whether the browser explorer supports every use case or what operational limits apply.
In the Peru example, differences between annual embeddings indicate landscape change, but they do not by themselves prove that mining caused it or identify the exact timing or driver. Those conclusions require checks against satellite imagery, monitoring records or field evidence. The supplied source also does not provide a publication date for the release.
As an affiliate, we earn on qualifying purchases.
Testing the Release in Practice
Users can examine the precomputed data through Source Cooperative and explore the maps in the browser-based tool, according to the release description. Researchers can consult the team’s technical report and available code, weights and pretraining data to review the methods and test the embeddings against local labels and imagery.
The next useful evidence will be independent evaluations across additional regions and tasks, alongside reporting on data coverage, update schedules and practical access requirements. Until then, the team’s scores and resource estimates should be treated as developer-reported results rather than independently established performance.
land cover classification software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is OlmoEarth v1.3?
It is the OlmoEarth team’s satellite embedding model. The developers say it processes yearly Sentinel-1, Sentinel-2 and Landsat imagery into compact numerical representations for 10-meter pixels.
What does the release include?
The release includes annual global embeddings at 10-meter resolution, a browser-based explorer and, according to the team, model weights, pretraining data and code.
How did it perform in the team’s evaluations?
The team reports that the 128-dimensional version ranked highest in 20 of 27 evaluations, while the 64-dimensional version led in 18. These results have not been independently verified in the supplied material.
Can the embeddings identify why a landscape changed?
No. A difference between yearly embeddings can signal detected change, but it does not establish the cause. Users need to check results against imagery and other monitoring or field evidence.
Where can users explore the data?
The team says the precomputed data is available through Source Cooperative and that the OlmoEarth Embeddings Explorer offers browser-based maps and comparisons with Sentinel-2 imagery.
Primary source: Hugging Face · via ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
