Benchmarking Geospatial Foundation Models for Agriculture Applications

2026-06-28

AI Summary

This study systematically evaluates the cross-regional generalization capabilities of geospatial foundation models in agricultural tasks. To address the significant performance drop observed in unseen regions, the authors establish a controlled benchmark by testing Prithvi, SpectralGPT, and SatMAE on multi-temporal crop semantic segmentation and change detection across four U.S. states, with strictly disjoint training, validation, and test regions to isolate geographic transfer performance. The work reveals, for the first time, that current models suffer severe degradation due to distributional shifts across regions and struggle to recognize rare crops. It also demonstrates that standardized input formats affect model performance heterogeneously. These findings highlight critical limitations of existing foundation models in agricultural applications and underscore the need for region-aware evaluation protocols.

Abstract

Geospatial foundation models pretrained on satellite imagery promise broad generalization across remote sensing tasks and regions, but their geographic transferability has not been systematically tested, especially in agriculture applications. This paper presents a controlled benchmark that evaluates three models, Prithvi, SpectralGPT, and SatMAE, on multi-temporal crop segmentation and change detection across four U.S. states, Iowa, North Carolina, California, and Minnesota. By assigning each train, validation, and test split to a separate region, we measure how well each model transfers to land it has not seen. All three degrade sharply under regional distribution shift, predicting only the most common crops while missing rare ones. We further find that fitting these models to a shared input format affects each one differently, which complicates direct architectural comparison. These results expose key limitations of current geospatial foundation models for agriculture and point to region aware evaluation as a necessary standard.

Problem

• geospatial foundation models
• agriculture applications
• geographic transferability
• crop segmentation
• distribution shift

Innovation

• geospatial foundation models
• cross-regional transfer
• agricultural remote sensing
• controlled benchmarking
• distribution shift