Announcing Sprints to Define Standards for Earth Observation Vector Embeddings
CNG, Planet, and Clark University are convening a two-day in-person sprint on March 10-11, 2026 at Clark University in Worcester, Massachusetts focused on defining standards for Earth observation (EO) vector embeddings. The goal of this sprint is to collaboratively draft and align on a practical specification that can be tested, extended, and adopted by the broader EO community.
The potential of EO vector embeddings
EO vector embeddings – compressed representations generated by AI models trained on satellite imagery – enable new approaches to understanding our planet. These embeddings can make AI-driven geospatial analysis accessible on consumer-grade hardware, making it possible for many more people around the world to access applications like global content-based search, change detection, and pattern recognition. Moreover, they have the potential to replace raw satellite observations as input to many downstream applications and empower new localized models.
Adoption of EO vector embeddings has been limited by the lack of standardized formats and metadata specification. While organizations like Google, Earth Genome, ESA, and Clay have published open EO vector embeddings, there are no established best practices for cataloging or discovering embedding datasets. This fragmented approach limits community adoption, research potential, and application development.
Goal for the Sprints
With funding support from Navigation Fund, we are organizing a pair of sprints this year to facilitate adoption of EO vector embeddings through development of a standard and publication of best practices.
This first meeting will convene data practitioners from throughout the geospatial community to draft a shared structure for storing and cataloging vector embeddings. Participants will review published embeddings and evaluate the approaches that have been taken so far. The deliverable will be a draft specification that teams can test in the following months.
The sprint will focus on hands-on demonstrations and capability mapping, and identifying gaps. Participants will document embedding capabilities, assess adoption barriers, and develop a vision for improved access. Our goal is to publish a draft working standard for EO vector embedding catalogs by the end of the meeting so various teams and users can test it before the second meeting later in the year.
A second workshop will be held at IBM’s office in Zürich in the summer. Between the two meetings, participants, and members of the community, will be encouraged to test the draft specification created at Clark University and provide real-world use cases and feedback on implementation challenges. The goal of the second workshop will be to establish performance benchmarks and develop a global adoption and support roadmap.
Who Should Attend
This sprint is for organizations building or using vector embeddings in Earth observation applications. We are keen in bringing together individuals who are hands-on with either the generation of EO vector embeddings, or are actively using them to develop downstream applications. You should commit to attend the 2-day sprint in-person in Worcester, MA and continue engaging with the group after the meeting.
Registration
The meeting is limited to ~15 participants. Please fill out this form to register your interest, and we will contact you to confirm your participation.
Travel support is available for participants based on need. The workshop will be followed by online coordination through the Cloud-Native Geospatial Forum Slack workspace and GitHub.
Deadline to submit your interest: Feb 6, 2026
Our blog is open source. You can suggest edits on GitHub.