Continuous biome representations from Earth observation embeddings

2026-06-09

AI Summary

Traditional biome maps rely on discrete classifications, which struggle to capture the continuous gradients characteristic of ecological transition zones. This work proposes a continuous representation method leveraging embeddings from the Clay v1.5 foundation model trained on satellite imagery: a linear classifier is trained on its 1024-dimensional Earth observation embeddings to output softmax-normalized biome probability vectors, thereby preserving categorical semantics while explicitly modeling gradual transitions between biomes. Evaluated across six major Brazilian biomes, this continuous representation achieves a mean AUC of 0.618 in predicting species distributions—significantly outperforming discrete biome labels (AUC = 0.570)—and demonstrates consistently improved performance across varying distances from biome boundaries.

Abstract

Biotic communities vary continuously across space, yet biome maps impose categorical boundaries that compress this variation, particularly at ecotones where transitional communities are ecologically distinct. Could Earth observation (EO) foundation models, which encode spectral, spatial, and temporal information with dense embeddings, convert discrete biome maps into continuous representations that better capture ecological variation? Here, we fit a linear classifier on Clay v1.5 satellite image embeddings to predict biome labels from a categorical map. The softmax output yields a continuous probability vector whose dimensions correspond to named biome classes. We evaluate this approach using six Brazilian biomes, 1.3 million embeddings, and 10,015 withheld forest inventory plots spanning 4,672 plant species. The continuous biome representation outperforms discrete biome labels for predicting species occurrence (mean per-species AUC 0.618 vs. 0.570 across 10 spatial cross-validation folds). Decomposing this gain shows that continuity in the graded probability output, rather than label reassignment, accounts for the improvement; the pattern holds across all distances from biome boundaries. The raw 1024-dimensional embedding remains the strongest predictor we tested (mean AUC 0.646 vs. 0.618), but the continuous representation recovers most of the embedding's gain over discrete labels. This simple approach provides a probabilistic replacement for categorical map labels, preserving their meaning while encoding graded variation that discrete maps suppress.

Problem

• biome
• ecotone
• continuous representation
• Earth observation
• species occurrence

Innovation

• continuous biome representation
• Earth observation foundation models
• satellite image embeddings
• ecotone
• species occurrence prediction