STAC has become the standard for discovering and accessing geospatial data, providing robust search capabilities based on spatial, temporal, and metadata filtering. However, as GeoAI workflows mature, there is growing demand for new forms of search that go beyond structured queries, such as similarity-based and feature-level search.
This talk explores how vector embeddings can extend, rather than replace, STAC-based search. In emerging architectures, STAC continues to serve as the foundation for data discovery and access, while embedding-based systems enable semantic search over image content, derived features, and learned representations.
We’ll examine patterns for combining these approaches, including workflows that use STAC to identify candidate data and vector indexes to support similarity search and analysis. We’ll also introduce the emerging STAC Embeddings extension, which provides a standardized way to describe and reference embedding data within the STAC ecosystem.
Finally, we’ll discuss tradeoffs and design considerations: where STAC’s model is sufficient, where embedding-based approaches add value, and how to build systems that integrate both effectively.
This session provides a practical view of how structured and semantic search can be combined to support modern GeoAI applications.