Recoverable Geographic Location Information in Earth-Observation Embeddings

Peiwen Zhang; Kristie Hu; Jovana Knezevic; Shunde Yin; Kyle Gao

cs.CVSep 24, 2026

Recoverable Geographic Location Information in Earth-Observation Embeddings

Authors: Peiwen Zhang, Kristie Hu, Jovana Knezevic, Shunde Yin, Kyle Gao

Organizations: University of Waterloo · University of Cambridge · Aalto University

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.

Figures & tables

Figure 1: Sentinel-2 imagery and PCA-coloured embeddings for six solar farms; colours are comparable only within each rows.

Input	Dim.	Spearman $\rho$ [95% CI]	Mean error, km [95% CI]	Median error, km [95% CI]	Equal-country mean, km [95% CI]
AlphaEarth	64	0.751 [0.664, 0.808]	179.7 [153.9, 231.0]	144.5 [127.6, 188.2]	333.3 [201.3, 486.0]
Tessera v1.1	128	0.424 [0.344, 0.514]	235.9 [192.3, 331.3]	174.9 [150.3, 235.0]	613.9 [366.6, 923.7]
Tessera v1	128	0.386 [0.308, 0.492]	256.7 [207.0, 361.9]	178.0 [152.5, 250.3]	676.1 [432.1, 940.8]
Sentinel-2 center-pixel	13	0.294[0.179,0.443]	391.9[310.2, 518.9]	304.6 [236.5, 407.5]	815.2 [490.4, 1169.0]
Sentinel-2 $3\times 3$ patch	117	0.288 [0.178, 0.396]	380.8 [303.9, 502.3]	289.8 [245.1, 390.4]	807.4 [466.3, 1178.1]

Table 1: Primary embedding distance-geographic distance correlation and coordinate-decoding results. Brackets give 95% spatial-block-bootstrap confidence intervals.

Figure 2: Farm location prediction under spatial cross-validation. (a–c) Mean geodesic errors for three embedding products and six predictors. (d) Paired error differences relative to Sentinel-2 controls using the same predictor.

Figure 3: Sensitivity of farm location prediction to spatial separation. Mean geodesic error with equal farm weighting as the training exclusion distance increases from 25 to 100 km, using ordinary linear regression, ridge regression, and an 8-layer MLP. Test farms and folds remain fixed across distances.

Figure 4: Leave-one-country-out farm location prediction. Mean geodesic errors for three predictors and five input representations, with each country excluded in turn from predictor training.

Explore similar work

Earth Embeddings
Aug 4, 2026Adam J. Stewart, Heng Fang, Isaac A. Corley +1Earth Observation Land Cover
Chair of Data Science in Earth Observation, Technical University of Munich, Munich, Germany · Munich Center for Machine Learning, Munich, Germany · KTH Royal Institute of Technology, Stockholm, Sweden +1
Better Together: Evaluating the Complementarity of Earth Embedding Models
May 18, 2026Thijs L van der Plas, Jacob JW Bakermans, Vishal Nedungadi +3Earth Observation Land Cover
Wageningen University, NL · University College London, UK · University of Bonn, Germany
Characterizing AlphaEarth Embedding Geometry for Agentic Environmental Reasoning
Apr 20, 2026Mashrekur Rahman, Samuel J. Barrett, Christina LastGeospatial Reasoning Behavioral Embeddings
Dartmouth Libraries, Dartmouth College, Hanover, 03755, NH, USA. · LGND AI, Canarias, Spain. · TipplyAI, London, UK.

Aug 4, 2026, cs.CVJ/K move · Enter open · S save

Earth Embeddings

Adam J. Stewart, Heng Fang, Isaac A. Corley, Xiao Xiang Zhu

Chair of Data Science in Earth Observation, Technical University of Munich, Munich, Germany · Munich Center for Machine Learning, Munich, Germany · KTH Royal Institute of Technology, Stockholm, Sweden+1

Earth Observation Land Cover Satellite Imagery

Open paper AskPDF arXiv

Earth observation is moving from foundation models that users must run themselves toward embedding products that package model feature outputs as reusable data without needing to download and process the imagery used to generate them. Earth embeddings are vectors that summarize locations, image patches, or pixels, letting users analyze compact features instead of repeatedly training or running large models on raw satellite imagery. This chapter explains the main types of Earth embeddings, from implicit location encoders to explicit patch and pixel products, and compares their coverage, resolution, dimensionality, storage cost, licenses, and reproducibility. We review their use in land cover and crop mapping, ecological and hazard modeling, socioeconomic prediction, and semantic search, with evidence on when embeddings improve on conventional features and when pooling, fusion, or spatial transfer limit performance. Two case studies show practical workflows for similarity search and land cover mapping. We close with guidance for choosing, evaluating, storing, compressing, and publishing embeddings, and with open problems in oceanic and atmospheric coverage, uncertainty, and benchmarking.

Abstract

Explore similar work

Earth Embeddings

Better Together: Evaluating the Complementarity of Earth Embedding Models

Characterizing AlphaEarth Embedding Geometry for Agentic Environmental Reasoning