Shen et al. (2026) SMFS-RF: a knowledge-guided machine-learning method for crop phenology extraction from fine-resolution vegetation index data

Identification

• Journal: Remote Sensing of Environment
• Year: 2026
• Date: 2026-09-04
• Authors: Ruoque Shen, Qiongyan Peng, Xiangqian Li, Jianxi Huang, Jin Chen, Jie Dong, Xiuzhi Chen, Wenping Yuan
• DOI: 10.1016/j.rse.2026.115632

Research Groups

• State Key Laboratory of Climate System Prediction and Risk Management, Nanjing Normal University
• School of Atmospheric Sciences, Sun Yat-sen University
• College of Science, Shihezi University
• Faculty of Geosciences and Engineering, Southwest Jiaotong University
• State Key Laboratory of Earth Surface Processes and Resource Ecology, Institute of Remote Sensing Science and Engineering, Faculty of Geographical Science, Beijing Normal University
• College of Geomatics & Municipal Engineering, Zhejiang University of Water Resources and Electric Power
• Institute of Carbon Neutrality, Sino-French Institute for Earth System Science, College of Urban and Environmental Sciences, Peking University

Short Summary

The study introduces SMFS-RF, a knowledge-guided machine learning framework that combines shape model fitting with random forest regression to accurately extract eight key rice phenological stages from fine-resolution NDVI data.

Objective

• To develop a method that overcomes the limited range of extractable phenological stages and the low accuracy/poor transferability of existing remote-sensing approaches when applied to heterogeneous cropping regions.

Study Configuration

• Spatial Scale: National scale (China), validated using 226 agricultural meteorological stations.
• Temporal Scale: Seasonal (crop growing cycle).

Methodology and Data

• Models used: SMFS-RF (integrates Shape Model Fitting by Separate Phenological Stage [SMF-S] and Random Forest [RF] regression).
• Data sources: Fine-resolution Normalized Difference Vegetation Index (NDVI) time series and ground-truth phenological records from agricultural meteorological stations.

Main Results

• Successfully estimated eight rice phenological stages: transplanting, regreening, tillering, stem elongation, booting, heading, milk ripening, and maturity.
• Achieved mean spatial Root Mean Square Errors (RMSE) ranging from 8.69 to 10.41 days.
• Achieved mean temporal RMSEs ranging from 5.52 to 6.36 days.

Contributions

• Expands the capacity of remote sensing to monitor a broader range of phenological stages beyond simple start/end dates.
• Improves the robustness and accuracy of phenology extraction across spatially heterogeneous agricultural landscapes by integrating physical knowledge (shape models) with machine learning (RF).

Funding

Not provided in the text.

Citation

@article{Shen2026SMFSRF,
  author = {Shen, Ruoque and Peng, Qiongyan and Li, Xiangqian and Huang, Jianxi and Chen, Jin and Dong, Jie and Chen, Xiuzhi and Yuan, Wenping},
  title = {SMFS-RF: a knowledge-guided machine-learning method for crop phenology extraction from fine-resolution vegetation index data},
  journal = {Remote Sensing of Environment},
  year = {2026},
  doi = {10.1016/j.rse.2026.115632},
  url = {https://doi.org/10.1016/j.rse.2026.115632}
}