CropViT
Climate-Aware Vision Transformer Model for Continental-Scale Crop Yield Prediction.
CropViT is a novel deep learning architecture that fuses multi-temporal Sentinel-2 satellite imagery with climate variables and extreme weather indices to deliver high-resolution agricultural yield forecasts.
Published in IEEE Xplore and presented as first author presenter at JPGU-AGU 2026.
Key Contributions
- Multimodal Spatiotemporal Fusion: Combines multispectral bands (NDVI, EVI, RedEdge) with localized meteorological time-series.
- Extreme Weather Sensitivity: Accurately models localized shock events (droughts, heatwaves, frost) on yield output.
- Continental-Scale Generalization: Evaluated across expansive agricultural zones with strong zero-shot transfer capabilities.