Wang et al. (2026) Predicting emitter irrigation duration for optimized irrigation system operation: An interpretable multi-model ML approach with SHAP

Identification

• Journal: Agricultural Water Management
• Year: 2026
• Date: 2026-07-27
• Authors: Hui Wang, Pengliang Yang, Xiaotao Hu, Qianqian Zhu, Wenè Wang, Xiaopeng Ma
• DOI: 10.1016/j.agwat.2026.110652

Research Groups

Institute of Agricultural Resources and Environment, Xinjiang Academy of Agricultural Sciences, Urumqi 830091, China
College of Civil Engineering, Hunan University, Changsha 410082, China
College of Water Resources and Architectural Engineering, Northwest Agriculture and Forestry University, Yangling 712100, China

Short Summary

This study develops an interpretable multi-model machine learning framework to predict emitter irrigation duration for optimized irrigation system operation. The proposed framework reveals the underlying physical mechanisms, particularly the interaction between water quality and emitter structure.

Objective

• Develop a multi-model machine learning framework for accurately predicting emitter irrigation duration across different water sources and emitter designs.
• Systematically evaluate and compare the performance and robustness of eight machine learning algorithms under complex, cross-scenario conditions.
• Employ SHAP interpretability techniques to identify key influencing factors and reveal the underlying physical mechanisms.

Study Configuration

• Spatial Scale: Laboratory-scale drip irrigation system
• Temporal Scale: Real-time prediction

Methodology and Data

• Models used: Support Vector Regression (SVR), K-nearest neighbor (KNN), Decision Tree (DT), Random Forest (RF), Adaptive boosting (AdaBoost), Extreme gradient boosting (XGBoost), LightGBM, Categorical boosting (CatBoost)
• Data sources: Experimental database of 450 samples, covering three water types and three emitter structures

Main Results

• The proposed framework demonstrates superior performance in predicting emitter irrigation duration compared to existing models.
• XGBoost emerges as the most outstanding model for cross-condition generalization due to its reduced CoV and ideal Mean.
• SHAP analysis reveals sediment concentration as the dominant factor negatively influencing emitter irrigation duration.
• A critical nonlinear interaction is identified: emitter channel depth (D) exhibits a distinct regulatory threshold at approximately 0.5 mm.

Contributions

• This study provides a reliable predictive tool for optimized irrigation system operation, advancing data-driven approaches from pure prediction toward informed design optimization and precision management of drip irrigation systems in complex global environments.
• The proposed framework reveals the underlying physical mechanisms, particularly the interaction between water quality and emitter structure, providing quantitative support for optimized emitter design and precision field management.

Funding

• This research was funded by the National Natural Science Foundation of China (Grant No. 51909254)

Citation

@article{Wang2026Predicting,
  author = {Wang, Hui and Yang, Pengliang and Hu, Xiaotao and Zhu, Qianqian and Wang, Wenè and Ma, Xiaopeng},
  title = {Predicting emitter irrigation duration for optimized irrigation system operation: An interpretable multi-model ML approach with SHAP},
  journal = {Agricultural Water Management},
  year = {2026},
  doi = {10.1016/j.agwat.2026.110652},
  url = {https://doi.org/10.1016/j.agwat.2026.110652}
}