Mohammadnezhad et al. (2025) A novel hybrid model for actual evapotranspiration estimation in data-scarce arid regions: Integrating modified Budyko and machine learning models using deep learning

Identification

• Journal: The Science of The Total Environment
• Year: 2025
• Date: 2025-09-17
• Authors: Mahdi Mohammadnezhad, Kamran Davary, Pooya Shirazi, Mohammad Javad Rezvanpour, Seyed Majid Hasheminia
• DOI: 10.1016/j.scitotenv.2025.180438

Research Groups

Water Science and Engineering Department, College of Agriculture, Ferdowsi University of Mashhad, Iran.

Short Summary

This study developed a novel hybrid model integrating a modified Budyko framework with machine learning (XGBoost) using deep learning to accurately estimate monthly actual evapotranspiration (ETa) in data-scarce arid regions, demonstrating superior performance over standalone models.

Objective

• To develop and evaluate a novel hybrid model for accurate monthly actual evapotranspiration (ETa) estimation in data-scarce arid regions by integrating a modified Budyko framework with an optimized machine learning model using deep learning.

Study Configuration

• Spatial Scale: A watershed in California’s Central Valley.
• Temporal Scale: Monthly.

Methodology and Data

• Models used: Modified Budyko framework (specifically Zhang equation), Machine Learning models (XGBoost, among thirteen others), Deep Learning for optimization, SHAP for feature importance analysis.
• Data sources: ERA5 (remote sensing), TerraClimate (remote sensing), Eddy Covariance Towers (observed data).

Main Results

• The novel hybrid Budyko-ML model, optimized with deep learning, consistently and significantly outperformed both standalone modified Budyko (e.g., Zhang equation) and pure machine learning models in estimating monthly ETa.
• The hybrid approach demonstrated superior predictive capability, mitigating weaknesses of both conceptual/physical and pure data-driven models.
• Methodological innovation by optimizing the temporal scale of Budyko’s parameter improved performance by accounting for non-steady-state hydrological conditions.
• Feature importance analysis using SHAP values highlighted climate, soil, and vegetation indices as primary drivers.

Contributions

• Development of novel hybrid Budyko-ML models using deep learning for accurate actual evapotranspiration (ETa) estimation in data-scarce arid regions.
• Significant improvement in predictive accuracy compared to both standalone modified Budyko and pure machine learning models.
• Introduction of a methodological innovation by optimizing the temporal scale of Budyko’s parameter for non-steady-state hydrological conditions.
• Provides a scalable and cost-effective solution for water management utilizing globally available remote sensing data.

Funding

Not specified in the provided text.

Citation

@article{Mohammadnezhad2025novel,
  author = {Mohammadnezhad, Mahdi and Davary, Kamran and Shirazi, Pooya and Rezvanpour, Mohammad Javad and Hasheminia, Seyed Majid},
  title = {A novel hybrid model for actual evapotranspiration estimation in data-scarce arid regions: Integrating modified Budyko and machine learning models using deep learning},
  journal = {The Science of The Total Environment},
  year = {2025},
  doi = {10.1016/j.scitotenv.2025.180438},
  url = {https://doi.org/10.1016/j.scitotenv.2025.180438}
}