Earth Embeddings as Products: Taxonomy, Ecosystem, and Standardized Access

TL;DR

This work tackles the fragmentation and reproducibility gaps in Earth embeddings by formalizing a three-layer taxonomy (Data, Tools, Value) and operationalizing a unified API within TorchGeo to load and query diverse embedding products. By treating embeddings as $D$-dimensional geospatial vectors over coordinates $(x, y)$ and time $t$, the authors enable direct cross-model comparisons and streamlined downstream analysis. Key contributions include a comprehensive survey of seven embedding products, standardized data loaders, and a concrete roadmap for interoperable, transparent Earth observation workflows. The approach significantly lowers engineering barriers, enhances benchmarking, and fosters broader adoption of frozen embedding products in geospatial analysis.

Abstract

Geospatial Foundation Models (GFMs) provide powerful representations, but high compute costs hinder their widespread use. Pre-computed embedding data products offer a practical "frozen" alternative, yet they currently exist in a fragmented ecosystem of incompatible formats and resolutions. This lack of standardization creates an engineering bottleneck that prevents meaningful model comparison and reproducibility. We formalize this landscape through a three-layer taxonomy: Data, Tools, and Value. We survey existing products to identify interoperability barriers. To bridge this gap, we extend TorchGeo with a unified API that standardizes the loading and querying of diverse embedding products. By treating embeddings as first-class geospatial datasets, we decouple downstream analysis from model-specific engineering, providing a roadmap for more transparent and accessible Earth observation workflows.

Figures (1)

• Figure 1: Taxonomy of the Earth Embedding Landscape. We organize the ecosystem into three functional layers: Data, Tools, and Value. Highlighted blocks indicate the specific scope of the assessment in Section \ref{['sec:ecosystem']}.