Create Custom AI Embeddings Effortlessly With OlmoEarth

📊 Full opportunity report: Create Custom AI Embeddings Effortlessly With OlmoEarth on ThorstenMeyerAI.com — validation score, market gap, and execution plan.

Prime Big Deal Days · Oct 6–7Offer from Amazon

Get hardware and tech essentials delivered free — and shop member deals

• Fast, free delivery on millions of items
• Access to Prime Big Deal Days deals on October 6–7
• Prime Video, Amazon Music and more included

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

OlmoEarth has announced a new feature in its Studio platform that allows users to generate custom satellite data embeddings on demand. This capability enables researchers and developers to create numerical representations of satellite imagery tailored to specific locations, time periods, and sensor sources, streamlining tasks such as similarity search and land-cover classification. Learn more in the original analysis.

The new feature supports defining an area of interest via drawing or uploading polygons, with options for selecting time spans from one to twelve months, spatial resolutions between 10 and 80 meters per pixel, and imagery sources including Sentinel-2 L2A and Sentinel-1 RTC. Users can choose among three encoder variants: Nano (128 dimensions), Tiny (192 dimensions), and Base (768 dimensions), with trade-offs between computational load and representation detail.

Results are delivered as Cloud-Optimized GeoTIFF files, with embedding vectors stored as signed 8-bit integers. For more on how these embeddings are generated, see the original analysis. Floating-point vectors can be reconstructed using a published dequantization method. Because each request is computed on demand, the output reflects the specific geography, dates, and satellite inputs selected by the user, rather than a fixed archive. The platform also offers the option to perform similarity searches, cluster analysis, and basic land-cover classification, although the performance and accuracy of these applications in real-world scenarios remain to be fully validated.

Implications for Earth Observation and AI Applications

This development lowers barriers for Earth observation analysis by providing a flexible, on-demand way to generate meaningful data representations without extensive model training. It enables faster experimentation with similarity search, clustering, and classification tasks, potentially accelerating research and operational workflows in environmental monitoring, land management, and climate studies. However, the actual performance of these embeddings in diverse conditions and applications is still being evaluated, and users should validate outputs for their specific use cases.

Background on OlmoEarth and Earth Observation Embeddings

OlmoEarth is an open-source project that offers foundation models for satellite imagery analysis. Its platform allows users to compute and export embeddings, which are compressed representations of satellite data that can be used for various downstream tasks. Prior to this update, generating such embeddings typically required training custom models, a process that can be resource-intensive and complex. The new on-demand export feature simplifies this process, making advanced analysis more accessible to a broader range of users.

Existing applications of embeddings include similarity search, land-cover segmentation, and clustering, with some reports indicating promising results in case studies like mangrove mapping in Vietnam. The platform supports multiple satellite sources and resolutions, providing flexibility for different environmental and research needs.

“OlmoEarth Studio now lets you compute and export embedding vectors.”

— Thorsten Meyer, OlmoEarth team

Performance and Accessibility Still Under Evaluation

It is not yet clear how well the embeddings perform across different climates, sensors, and real-world applications. The platform’s access terms, processing times, and geographic restrictions remain unspecified, and validation of the embeddings’ accuracy and robustness is ongoing. Users are advised to conduct their own testing before deploying in operational settings.

Next Steps for Users and Developers

Interested users can request access to OlmoEarth Studio to try the new features. Future updates may include performance benchmarks, expanded geographic coverage, and additional application examples. The platform’s open-source nature allows researchers to compute embeddings independently and tailor models further for specific tasks.

Key Questions

Can I use OlmoEarth’s embeddings for operational land management?

While the platform supports generating embeddings for various applications, the performance in operational settings has not been fully validated. Users should conduct their own testing and validation before relying on the outputs for critical decisions.

Is the OlmoEarth platform free to use?

The announcement indicates that users can request access, but details about pricing, availability, and restrictions are not yet specified. The source code and models are publicly available for independent use.

What types of satellite imagery are supported?

The platform supports Sentinel-2 L2A and Sentinel-1 RTC imagery, with options to combine sources for richer data inputs.

Can I generate embeddings for any location worldwide?

In principle, yes, but access and performance may vary depending on the availability of imagery and processing capacity. Validation for specific regions is recommended.

Are there any limitations on the resolution or time span?

Users can select resolutions between 10 and 80 meters per pixel and time spans from one to twelve months, but the accuracy and usefulness depend on the specific application and data quality.

Source: ThorstenMeyerAI.com