Sadeghzadeh et al. (2026) A Paradigm Shift to Automated Machine Learning for Local and External Reference Evapotranspiration Estimation with Uncertainty Implication

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

Identification

• Journal: Water
• Year: 2026
• Date: 2026-04-13
• Authors: Mostafa Sadeghzadeh, Sepideh Karimi, Amir Hossein Nazemi, Pau Martí, Jalal Shiri
• DOI: 10.3390/w18080927

Research Groups

Not specified in the provided text.

Short Summary

This study evaluates various automated machine learning (AutoML) algorithms coupled with base models for estimating daily reference evapotranspiration (ET0) across three diverse climatic regions. The research demonstrates that hybrid AutoML models significantly improve ET0 estimation accuracy and generalizability compared to standalone models, with optimal performance being dependent on the specific climatic conditions.

Objective

• To evaluate the performance and generalizability of different automated machine learning (AutoML) algorithms (Neural Architecture Search, Optuna, Enhanced Grey Wolf Optimization, Quantum Whale Optimization) coupled with base machine learning models (Random Forest, Neural Networks, Light Gradient Boosting) for estimating daily reference evapotranspiration (ET0) across diverse climatic regions (Cairo, Singapore, London).

Study Configuration

• Spatial Scale: Three distinct climatic regions: Cairo (arid), Singapore (tropical), and London (temperate).
• Temporal Scale: Daily estimation of reference evapotranspiration (ET0).

Methodology and Data

• Models used: Hybrid models combining automated machine learning (AutoML) algorithms (Neural Architecture Search (NAS), Optuna, Enhanced Grey Wolf Optimization (EGWO), Quantum Whale Optimization (QWOA)) with base machine learning models (Random Forest, Neural Networks (NN), Light Gradient Boosting).
• Data sources: Meteorological input data for ET0 estimation. Specific sources (e.g., satellite, observation, reanalysis) are not detailed in the provided text.

Main Results

• In local validation, the NN-NAS model achieved the highest accuracy in Cairo (R² = 0.969, RMSE = 0.432 mm/day) and Singapore (R² = 0.657, RMSE = 0.596 mm/day).
• For London, the NN-Optuna model provided the highest performance accuracy (R² = 0.941, RMSE = 0.370 mm/day).
• Hybrid AutoML models improved the coefficient of determination (R²) by 5–15% and reduced the root mean square error (RMSE) by 10–20% compared to standalone models.
• In external validation, NN-NAS and NN-Optuna demonstrated superior generalizability, with R² values reaching up to 0.899 in London and 0.680 in Cairo, respectively.
• The performance of hybrid models was climate-dependent: NN-NAS was identified as the best model for arid sites, while NN-Optuna showed the highest accuracy in temperate climates.
• Shapley Additive Explanations (SHAP) analysis indicated that solar radiation had the highest impact on ET0 estimation across all three studied climatic contexts, although its degree of importance varied with climate.

Contributions

• Provides a comprehensive evaluation of various automated machine learning (AutoML) algorithms for reference evapotranspiration (ET0) estimation across diverse climatic conditions.
• Demonstrates significant improvements in accuracy and generalizability of hybrid AutoML models over traditional standalone machine learning models for ET0 prediction.
• Identifies climate-dependent optimal AutoML strategies, suggesting that the best model architecture and tuning approach can vary with the specific climatic context.
• Utilizes SHAP analysis to provide insights into the relative importance of meteorological variables in ET0 estimation within an AutoML framework.

Funding

Not specified in the provided text.

Citation

@article{Sadeghzadeh2026Paradigm,
  author = {Sadeghzadeh, Mostafa and Karimi, Sepideh and Nazemi, Amir Hossein and Martí, Pau and Shiri, Jalal},
  title = {A Paradigm Shift to Automated Machine Learning for Local and External Reference Evapotranspiration Estimation with Uncertainty Implication},
  journal = {Water},
  year = {2026},
  doi = {10.3390/w18080927},
  url = {https://doi.org/10.3390/w18080927}
}