CivicEmbed: Feature-specific embeddings for efficient geographic reasoning and retrieval

Diverse geography, from natural landscapes to urban areas, poses challenges for terrain-sensitive development. Existing geo-embeddings are largely monolithic, entangling distinct geographic drivers and limiting factor-specific comparison. We present CivicEmbed, a lightweight approach that learns feature-specific embedding spaces for topography, water proximity, vegetation coverage, and road networks using self-supervised contrastive learning on thematic raster layers. These modular encoders enable structured analogical reasoning by retrieval: users can isolate a single driver (e.g., topography) or compose drivers with simple weights, then retrieve regions that match the selected criteria. Feature-specific encoders achieve $32\times$ compression ($128$-D vs $4,096$-D raw patches) while improving retrieval metrics on multiple features. We implemented a FAISS-backed retrieval system and interactive interface at the scale of Switzerland, providing a foundation for data-driven decisions in architecture, transit design, and land-use planning.