Earth Embedding Products for Geospatial Analysis: Foundations, Applications, and Open Challenges

Yongchuan Cui , Keli Shi, Shunlin Liang

Earth observation satellites generate petabytes of imagery each year, but extracting useful information remains constrained by limited labels, heterogeneous sensors, and the cost of processing large archives. Foundation models reduce part of this burden by learning transferable representations from multi-source observations. More recently, these representations have been precomputed over continental and global extents and released as embedding layers or datasets. Existing reviews primarily examine model architectures, pretraining, and transfer learning; released embedding products also require attention to spatial and temporal support, storage, access, versioning, and evaluation as fixed geospatial features. This survey reviews the development of Earth embeddings from foundation model pretraining and reusable encoders to global and near-global embedding products. We organize the literature around data and representation learning, model and product design, distribution and access, evaluation benchmarks, and downstream applications. Experimentally, we compare representative embeddings through their geometry, downstream performance, and sensitivity to dimension across discrete and continuous land-surface tasks. Finally, we summarize the technical and scientific challenges surrounding global embedding products and discuss priorities for their evaluation, maintenance, and future development.