From Pixels to Patches: Pooling Strategies for Earth Embeddings

TL;DR

This paper evaluates pooling strategies for geospatial pixel embeddings, demonstrating that richer statistical pooling methods significantly enhance spatial generalization over mean pooling.

Contribution

It introduces EuroSAT-Embed, benchmarks multiple pooling methods, and recommends distributional statistics pooling for better geographic generalization.

Findings

Richer pooling reduces geographic generalization gap by over 50%.

Distributional statistics pooling improves accuracy by up to 6%.

Simple statistics outperform mean pooling across multiple models.

Abstract

Geospatial foundation models increasingly expose pixel-level embedding products that can be downloaded and reused without access to the underlying encoder. In this setting, downstream tasks with patch- or region-level labels require a post-hoc aggregation step that maps dense pixel embeddings to a single representation. The default choice, mean pooling, discards within-patch variability and can underperform under spatial distribution shift. To study this setting, we introduce EuroSAT-Embed: 81,000 embedding GeoTIFFs derived from three foundation models: AlphaEarth, OlmoEarth, and Tessera. Using these fixed embedding products, we benchmark 11 training-free pooling methods and 2 train-set-fitted baselines under both random and geographically disjoint test splits. Richer pooling schemes reduce the geographic generalization gap by over 50% relative to mean pooling and improve accuracy by up to 6% on spatial splits. We recommend a three-tier strategy: (1) mean as a baseline, (2) stats pooling (min/max/mean/std) as the default at 4x the embedding dimension, and (3) covariance pooling for peak accuracy. Across all three embedding products, simple distributional statistics improve spatial-split performance over mean pooling.