Zhang et al. (2025) Optimization of Sensor Combinations for Simplified Estimation of Reference Crop Evapotranspiration Using Machine Learning and SHAP Interpretation

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

Identification

• Journal: Agriculture
• Year: 2025
• Date: 2025-12-31
• Authors: Qiong Zhang, Yang Xu, Cheng Ding, Weining Xiu, Chang Liu, Shufen Dai
• DOI: 10.3390/agriculture16010093

Research Groups

Not explicitly stated in the provided text, but likely institutions in China given the study location.

Short Summary

This study systematically evaluates machine-learning (ML) models for estimating reference crop evapotranspiration (ET0) in data-sparse regions of China, finding that the Random Forest model performs best and can maintain accuracy even with reduced sensor inputs, while SHAP analysis reveals key regional and national drivers.

Objective

• To systematically evaluate several machine-learning models as data-efficient alternatives to the traditional Penman–Monteith method for estimating daily reference crop evapotranspiration (ET0) in data-sparse regions.

Study Configuration

• Spatial Scale: 698 meteorological stations across China.
• Temporal Scale: Daily observations.

Methodology and Data

• Models used: Random Forest (RF), multiple linear regression, Penman–Monteith (PM) (as a benchmark). SHapley Additive exPlanation (SHAP) for model interpretability.
• Data sources: Daily meteorological observations from 698 meteorological stations across China.

Main Results

• The Random Forest (RF) model demonstrated the best performance (coefficient of determination, R² = 0.957; mean absolute percentage error, MAPE = 9.214%).
• RF significantly outperformed multiple linear regression and approached the accuracy of the traditional Penman–Monteith method.
• SHAP analysis identified maximum temperature, sunshine duration, and month as the most influential factors nationwide for ET0 estimation.
• Geographic variables contributed less overall but became important in specific regions, such as Southwest China.
• The study revealed pronounced spatial heterogeneity in the drivers of ET0, emphasizing the need for regionalized interpretations.
• Sensor-reduction experiments showed that reasonable estimation accuracy can be maintained even without radiation or wind-speed observations.

Contributions

• Provides transparent model comparisons for machine-learning-based ET0 estimation.
• Uncovers regional differences in the controlling factors of ET0.
• Offers practical insights for designing meteorological monitoring strategies, particularly in data-limited environments, by demonstrating the feasibility of sensor reduction.

Funding

Not explicitly stated in the provided text.

Citation

@article{Zhang2025Optimization,
  author = {Zhang, Qiong and Xu, Yang and Ding, Cheng and Xiu, Weining and Liu, Chang and Dai, Shufen},
  title = {Optimization of Sensor Combinations for Simplified Estimation of Reference Crop Evapotranspiration Using Machine Learning and SHAP Interpretation},
  journal = {Agriculture},
  year = {2025},
  doi = {10.3390/agriculture16010093},
  url = {https://doi.org/10.3390/agriculture16010093}
}