Adaptive fusion of multi-modal remote sensing data for optimal sub-field crop yield prediction
Published in Remote Sensing of Environment, 2025
DOI: 10.1016/j.rse.2024.114547
Data: https://yieldsat.github.io/
The data we used in this manuscript is now released as YieldSAT (at CVPR).
This work introduces a Multi-modal Gated Fusion (MMGF) framework for sub-field crop yield prediction using heterogeneous remote sensing and environmental data. The approach combines Sentinel-2 optical imagery, weather time series, soil properties, and topographic information through dedicated modality encoders and an adaptive Gated Unit that learns sample-specific fusion weights. Experiments across soybean, wheat, and rapeseed in Argentina, Uruguay, and Germany demonstrate that MMGF outperforms conventional models, particularly when integrating all available data sources. The learned fusion weights also vary across countries and crop types, reflecting differences in the importance of each modality. These results highlight the potential of adaptive multi-modal fusion for high-resolution crop yield prediction and precision agriculture.
Recommended citation: Mena, Francisco, et al. "Adaptive fusion of multi-modal remote sensing data for optimal sub-field crop yield prediction." Remote Sensing of Environment 318 (2025): 114547.
Download Paper