Zhang et al. (2026) Deployment-oriented evaluation of stage-specific information for rice yield prediction across years, sites and site-year environments in southern China

Identification

• Journal: Frontiers in Plant Science
• Year: 2026
• Date: 2026-09-08
• Authors: Jinshuai Zhang, Yongquan Xiong, Shuyao Zhou, Qing Ouyang, Ming Huang, Tao Guo, Chun Chen, Yongzhu Liu
• DOI: 10.3389/fpls.2026.1915306

Research Groups

• National Engineering Research Center of Plant Space Breeding, College of Agriculture, South China Agricultural University, Guangzhou, China
• School of Accounting, Anyang Institute of Technology, Anyang, China
• College of Artificial Intelligence and Low Altitude Technology, South China Agricultural University, Guangzhou, China

Short Summary

This study presents a deployment-oriented framework for rice yield prediction using multi-year, multi-site field-trial data from southern China. The framework evaluates the performance of different information sources at various decision points and identifies limitations in model generalization.

Objective

• Evaluate the effectiveness of machine-learning models with inputs that match the intended decision point and evaluation schemes that reflect the target deployment setting.
• Identify which information sources are useful at different decision points and where model generalization remains limited.

Study Configuration

• Spatial Scale: 10 test sites in southern China, including Guangdong Province, Guangxi Zhuang Autonomous Region, Hainan Province, and Fujian Province.
• Temporal Scale: Seven growing seasons from 2016 to 2022.

Methodology and Data

• Models used: Linear regression, ridge regression, ElasticNet, random forest, XGBoost, LightGBM, CatBoost.
• Data sources: Official national variety registration trial reports for the Southern Rice Region of China, daily meteorological data from the China Meteorological Data Service Centre.

Main Results

• Random five-fold cross-validation produced high apparent accuracy, with the best field-survey-enhanced model reaching R^2 = 0.844 and RMSE = 340.59 kg ha^-1.
• Deployment-oriented performance was lower, with F4 ElasticNet achieving a macro mean R^2 of 0.359 in forward-year prediction, 0.290 in leave-one-site-out validation, and 0.131 in leave-one-site-year-out validation.

Contributions

• This study supports a deployment-oriented framework for rice yield prediction, rather than a general claim of universal out-of-sample improvement.
• The main value is to show how prediction skill, feature utility, and model interpretation change when validation is aligned with realistic temporal, spatial, and site-year deployment settings.

Funding

• National Engineering Research Center of Plant Space Breeding, College of Agriculture, South China Agricultural University, Guangzhou, China.

Citation

@article{Zhang2026Deploymentoriented,
  author = {Zhang, Jinshuai and Xiong, Yongquan and Zhou, Shuyao and Ouyang, Qing and Huang, Ming and Guo, Tao and Chen, Chun and Liu, Yongzhu},
  title = {Deployment-oriented evaluation of stage-specific information for rice yield prediction across years, sites and site-year environments in southern China},
  journal = {Frontiers in Plant Science},
  year = {2026},
  doi = {10.3389/fpls.2026.1915306},
  url = {https://doi.org/10.3389/fpls.2026.1915306}
}