Zeng et al. (2026) TandemNet: A Multi-Scale Multiple-Instance Learning Framework for Early-Season Rice Yield Prediction

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-09-05
• Authors: Meiqi Zeng, Wenxi Wu, Ran Yang, Wanxin Zhang, Siya Du, Xingzhi Huang, Luo Liu
• DOI: 10.3390/rs18173040

Research Groups

• Department of Agricultural Engineering, University of Northeast China
• Key Laboratory for Agro-ecological Processes in Middle and Lower Reaches of Yangtze River
• National Engineering Research Center for Information Technology in Agriculture

Short Summary

This study proposes TandemNet, a multi-scale multiple-instance learning framework for early-season prediction of japonica rice yield in Northeast China. The model outperforms several baselines by incorporating cross-scale interactions between pixel-level growth trajectories and county-level statistical responses.

Objective

• Investigate the feasibility of using remote-sensing data to predict early-season crop yields at a county level, considering spatial heterogeneity and partial-season observations.

Study Configuration

• Spatial Scale: County-level yield prediction in Northeast China.
• Temporal Scale: Early-season prediction (approximately 2–3 months before harvest).

Methodology and Data

• Models used: TandemNet, Random Forest, XGBoost, LSTM, Transformer
• Data sources: Sentinel-1, Sentinel-2, MODIS satellite data from 2018 to 2023.

Main Results

• TandemNet achieves an R2 of 0.69 and an RMSE of 655.98 kg/ha at the tillering stage.
• Ablation and attention analyses indicate a stage-dependent shift from local heterogeneity to county-level consistency.

Contributions

• This study demonstrates that modeling cross-scale interactions improves the timeliness, accuracy, and interpretability of rice yield prediction.
• The proposed TandemNet framework can be applied to other crop types and regions with similar spatial and temporal scales.

Funding

• National Key Research and Development Program (2018YFD0200300)
• National Natural Science Foundation of China (61903244)

Citation

@article{Zeng2026TandemNet,
  author = {Zeng, Meiqi and Wu, Wenxi and Yang, Ran and Zhang, Wanxin and Du, Siya and Huang, Xingzhi and Liu, Luo},
  title = {TandemNet: A Multi-Scale Multiple-Instance Learning Framework for Early-Season Rice Yield Prediction},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18173040},
  url = {https://doi.org/10.3390/rs18173040}
}