onef1shy/Crop-Yield-Prediction-MODIS

Crop Yield Prediction MODIS

This repository hosts the processed MODIS data used for crop-yield regression in the DFYP project. It was prepared from the MODIS branch of the DFYP project.

Repository: https://github.com/onef1shy/DFYP.

Paper: https://doi.org/10.1109/TGRS.2026.3684831

Contents

• datasets/modis/processed_data/<year>/*.npy: preprocessed yearly samples indexed by year, county, and sample id
• datasets/modis/processed_data/histogram_all_full.npz: histogram data used by the model pipeline

Example files:

datasets/modis/processed_data/2003/2003_17_1.npy
datasets/modis/processed_data/2016/2016_5_95.npy
datasets/modis/processed_data/histogram_all_full.npz

The release is split into year folders on Hugging Face because a single directory cannot contain too many files. For local DFYP usage, flatten those year folders back into datasets/modis/processed_data/ before running the code.

Flattened examples:

datasets/modis/processed_data/2003_17_1.npy
datasets/modis/processed_data/2016_5_95.npy
datasets/modis/processed_data/histogram_all_full.npz

Preprocessing

The MODIS branch uses surface reflectance, surface temperature, and land cover products from 2003 to 2015. Because the products have different temporal resolutions and the release must stay within Hugging Face directory limits, we process and upload them year by year:

• surface reflectance and surface temperature are grouped by year
• land cover is reclassified into a crop mask
• the crop mask is applied to filter non-target areas
• reflectance and temperature channels are stacked into a 9-channel yearly representation
• the final spectral images are converted into histograms to reduce noise and stabilize the input representation

Citation

If you use this dataset, please cite the DFYP paper:

@article{zhang2026dfyp,
  title={DFYP: A Dynamic Fusion Framework with Spectral Channel Attention and Adaptive Operator learning for Crop Yield Prediction},
  author={Zhang, Juli and Yan, Zeyu and Zhang, Jing and Miao, Qiguang and Wang, Quan},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  year={2026},
  publisher={IEEE}
}