What's in an Earth Embedding? An Explainability Analysis of Location Encoders

2026-06-23

AI Summary

This work addresses the limited interpretability of positional embeddings generated by geographic Implicit Neural Representations (INRs), which obscure the underlying geospatial and semantic information they encode. The study presents the first systematic dissection of such embeddings by integrating sparse autoencoders, SpLiCE for language-based concept extraction, and CLIP Surgery saliency maps to decompose them into sparse, human-interpretable concept units. This approach achieves high-fidelity reconstruction while effectively uncovering complementary geographic semantic features embedded within the representations—such as biomes (e.g., forests, deserts, urban areas), infrastructure (e.g., roads), landmarks, and signals related to climate and land cover. By rendering these latent structures explicit and comprehensible, the method substantially enhances the interpretability of geographic INRs.

Abstract

Geographic implicit neural representations (INRs) learn to map any coordinate on Earth to a location embedding, implicitly encoding geospatial data into the weights of a neural network. Location embeddings are widely used off the shelf as general-purpose geospatial representations, yet users lack principled tools to audit what geographic or semantic information these embeddings capture. In this work, we analyze the information content of geographic INRs through their location embeddings. We decompose these embeddings into human-interpretable features$\unicode{x2014}$namely, (i) sparse latent concepts, (ii) natural language concepts, and (iii) visual features. The latent concept embeddings are learned using sparse autoencoders. To recover natural language concepts, we apply sparse linear concept embeddings (SpLiCE) over a predefined geospatial dictionary. Finally, visual features are extracted using saliency maps derived from CLIP Surgery. We show that location embeddings can be decomposed into human-interpretable representations while retaining high reconstruction capability, revealing interpretable geographic structures such as forests, deserts, and urban features. Across methods, sparse decompositions expose systematic differences in encoded information, ranging from urban structures to broader biome and climate signals, and pretraining-space saliency maps further highlight complementary features such as roads and landmarks. We hope this work provides a first step toward interpretable geospatial representations.

Problem

• location embeddings
• geographic implicit neural representations
• explainability
• geospatial representations
• interpretability

Innovation

• location embeddings
• geographic implicit neural representations
• sparse autoencoders
• SpLiCE
• CLIP Surgery