A primer on embeddings

Development Seed has published an approachable primer on Earth data #embeddings, available both as a poster and as a zine. It explains how these representations learned by geospatial #FoundationModels are generated and why they are useful for search, retrieval, and natural language interfaces.
Author

Ralph Straumann

Published

October 9, 2026

A while back, I saw their call for interesting resources on the topic. Now, Development Seed offers their primer on Earth data embeddings1. From their blog post:

Embeddings are one way we make use of the rich representations learned by geospatial foundation models2, supporting everything from search and retrieval to natural language interfaces. Yet for many practitioners, the concept remains opaque. This primer explores how embeddings are generated, the kinds of questions they help answer, and why they’re useful across the geospatial ecosystem.

The Development Seed team has made available both a poster and a zine3. The aforementioned blog post shows animated pieces and graphics from the latter.

A high-level workflow diagram of building and using Earth data embeddings (source: Development Seed)

This looks very promising. Last year, Development Seed published a zine on cloud-native4 the quality of which I appreciated.

Footnotes

  1. A geospatial embedding is a compact, high-dimensional vector representation of a geospatial location, scene, or feature, learned so that similar places or patterns end up close together in vector space.↩︎

  2. Geospatial foundation models are large AI models pre-trained on vast amounts of Earth observation and other geospatial data, which can then be adapted to many downstream tasks.↩︎

  3. “Zine” is short for “magazine”.↩︎

  4. I.e., taking full advantage of the cloud computing model; in the data world typically through enabling partial downloads of file contents through so-called HTTP range requests.↩︎