Do Foundation Model Embeddings Improve Cross-Country Crop Yield Generalisation? A Leave-One-Country-Out Evaluation in Sub-Saharan Africa
For researchers and practitioners in agricultural remote sensing, this paper provides a reproducible negative benchmark highlighting the generalisability gap in cross-country crop yield prediction.
The paper evaluates whether geospatial foundation model embeddings improve cross-country maize yield prediction in sub-Saharan Africa, finding that all feature sets perform poorly under leave-one-country-out testing with universally negative R², and frozen Prithvi-EO embeddings offer no advantage over traditional spectral features.
Accurate predictions of smallholder maize yields across national boundaries are critical for food security planning in sub-Saharan Africa, yet most published benchmarks report within-country performance that overstates true generalisability. This paper evaluates whether geospatial foundation model embeddings, specifically Prithvi-EO-1.0-100M and ViT-Base, outperform traditional Sentinel-2 spectral features under a Leave-One-Country-Out cross-validation scheme on 6,404 maize field observations from five African countries. The results show a clear generalisability gap: within-country random cross-validation yields moderate R^2 values, but all feature sets perform poorly under cross-country testing, with universally negative R^2. Frozen Prithvi-EO embeddings provide no meaningful advantage over engineered spectral features for cross-country prediction in this setting. The paper argues that the main limitation is a shift in yield distribution between countries rather than representation quality and releases a reproducible negative benchmark for future work.