Earth observation models encode recoverable geographic location data
Recoverable Geographic Location Information in Earth-Observation Embeddings
Computer Vision and Pattern Recognition
Summary
Earth observation models create digital summaries of satellite images, but it's unclear if these summaries include location details. The authors tested three such models by seeing if they could predict the exact places of solar farms in Europe from the summaries. All three models contained enough clues to estimate locations better than random guessing, with one model performing best. This means these models store hidden geographic information that can be retrieved when needed.
What this means in practice
- •For earth observation platform developers: Design and audit models that require or should exclude geographic information depending on application needs.
- •For geospatial data analysts: Use embeddings from existing earth observation models to recover location information for spatial analyses without extra metadata.
Authors
Peiwen Zhang, Kristie Hu, Jovana Knezevic, Shunde Yin, Kyle Gao
Abstract
Earth-observation (EO) foundation models provide reusable embeddings, yet downstream task accuracy does not reveal whether these representations encode geographic information, which may be beneficial for location-aware applications but potentially detrimental when representations invariant to geographic location are desired. We therefore evaluate the geographic coordinate robustness of Tessera v1, Tessera v1.1, and AlphaEarth by testing whether coordinates can be predicted from the embedding representations using 284 quality-verified European solar farms from 2024. We assessed geographic information content information through the association between cosine and geodesic distances and through prediction of projected coordinates in EPSG:3035. Embeddings from all three EO foundation models contain recoverable geographic information. All prediction models significantly outperform training-range uniform random sampling baselines, with AlphaEarth exhibiting the strongest distance association and lowest mean geodesic error. Both Tessera variants also yielded higher geographic distance correlations than the Sentinel-2 controls. These findings motivate geographic information content as an additional criterion for auditing EO foundation models.