Jia et al. (2025) Evapotranspiration estimation at different land surface scales in semi-arid areas using gene expression programming and the FAO56 Penman-Monteith model

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2025
• Date: 2025-11-20
• Authors: Tianyu Jia, Asaad Y. Shamseldin, Tingxi Liu, Yongzhi Bao, Yiran Zhang, Limin Duan, Xin Tong, Mingyang Li, Zixu Qiao, Zhiting Wang
• DOI: 10.1016/j.ejrh.2025.102952

Research Groups

• State Key Laboratory of Water Engineering Ecology and Environment in Arid Area, Inner Mongolia Agricultural University, China
• Institute of Water Sciences, Zhejiang University of Water Resources and Electric Power, Hangzhou, China
• Inner Mongolia Key Laboratory of Ecohydrology and High-Efficient Utilization of Water Resources, Hohhot, China
• Autonomous Region Collaborative Innovation Center for Integrated Management of Water Resources and Water Environment in the Inner Mongolia Reaches of the Yellow River, Hohhot, China
• Department of Civil and Environmental Engineering, The University of Auckland, Auckland, New Zealand
• Water Resources Research Institute of Shandong Province, Shandong Provincial Key Laboratory of Water Resources and Environment, Jinan, China

Short Summary

This study developed a GEP-PM model by integrating Gene Expression Programming (GEP) with the FAO56 Penman-Monteith (PM) model to improve actual evapotranspiration (ETa) estimation in semi-arid regions. The model accurately predicts ETa across diverse land surfaces (sand dunes and meadows) at a 30-minute temporal scale, demonstrating superior performance compared to traditional methods.

Objective

• To optimize the Penman-Monteith model by recombining its radiation and aerodynamic terms using Gene Expression Programming (GEP) and comprehensive meteorological, environmental, and soil data from 2017 to 2021, thereby enhancing its applicability in semi-arid regions.
• To analyze evapotranspiration data at a 30-minute time scale, enabling more accurate capture of dynamic ET changes than traditional daily-scale models.
• To compare evapotranspiration dynamics across two representative surfaces, sand dunes and meadows, to highlight their impact on ET and improve the model’s adaptability to different environments.

Study Configuration

• Spatial Scale: Agula Eco-hydrological Experimental Station, located in the southeastern part of the Horqin Sandy Land, Tongliao City, Inner Mongolia, North China (43°18′48″–43°21′2″N, 122°32′30″–122°41′00″E), covering approximately 55 square kilometers. The study focused on nine test sites representing sand dunes, meadows, and farmlands.
• Temporal Scale: Data collected from 2017 to 2021, with analyses performed at a 30-minute time step.

Methodology and Data

• Models used:
  • FAO56 Penman-Monteith (P-M) model (standard reference).
  • Gene Expression Programming (GEP).
  • GEP-PM model (hybrid model developed in this study).
  • Priestley-Taylor model (for comparison).
• Data sources:
  • Observation: Meteorological parameters (wind speed, wind direction, air temperature, relative humidity, net radiation, soil temperature, soil moisture content, rainfall) from Campbell Scientific Inc. weather stations.
  • Observation: Flux data (H₂O and CO₂ fluxes, sensible heat flux, soil heat flux) from open-circuit eddy covariance (EC) systems (CSAT-3, LI-7500A).
  • Study Area: Agula Eco-hydrological Experimental Station, Inner Mongolia, China.
  • Data Processing: Eddy Pro software for raw flux data, ReddyProc Web tool for gap filling.
  • Model Development: GeneXproTools 4.0 software.
  • Input Variables for GEP-PM: Soil water content (SWC) at 10 cm depth, air temperature (Ta), relative humidity (RH), wind speed at 2 m (u2), saturated vapor pressure deficit (e₀-eₐ), and user-defined function symbols representing the radiation and aerodynamic terms from the FAO56 P-M model.

Main Results

• The GEP-PM model demonstrated high accuracy in estimating actual evapotranspiration (ETa) across both sand dunes and meadows.
  • For meadow areas, the model achieved R² values ranging from 0.77 to 0.83, Root Mean Square Error (RMSE) between 0.04 and 0.06 mm/h, Mean Absolute Error (MAE) from 0.02 to 0.03 mm/h, and Nash-Sutcliffe Efficiency (NSE) from 0.78 to 0.86 during calibration.
  • For dune areas, the model showed R² values from 0.71 to 0.77, RMSE between 0.05 and 0.07 mm/h, MAE from 0.03 to 0.05 mm/h, and NSE from 0.72 to 0.83 during calibration.
  • Overall validation performance for GEP-PM ranged from R² = 0.70 to 0.82, RMSE = 0.049 to 0.09 mm/h, MAE = 0.031 to 0.089 mm/h, and NSE = 0.70 to 0.83.
• Evapotranspiration rates were observed between 0 and 0.8 mm/h on meadows and 0 and 0.6 mm/h on sand dunes at a half-hourly scale.
• The GEP-PM model effectively captured seasonal (peak ET from June to August) and intraday (symmetrical patterns) dynamics of ET.
• Comparative analysis showed GEP-PM outperformed FAO56 Penman-Monteith and Priestley-Taylor models:
  • In dune ecosystems (daily scale): GEP-PM (R²=0.84, RMSE=0.85 mm/d, MAE=0.17 mm/d) was superior to FAO56 P-M (R²=0.72, RMSE=1.18 mm/d, MAE=0.24 mm/d) and Priestley-Taylor (R²=0.39, RMSE=4.47 mm/d, MAE=0.91 mm/d).
  • In meadow ecosystems (daily scale): GEP-PM (R²=0.92, RMSE=0.21 mm/d, MAE=0.04 mm/d) was superior to FAO56 P-M (R²=0.55, RMSE=2.09 mm/d, MAE=0.42 mm/d) and Priestley-Taylor (R²=0.53, RMSE=1.70 mm/d, MAE=0.34 mm/d).
• ET on meadow surfaces was primarily influenced by meteorological factors, while dune surfaces were more sensitive to changes in soil moisture content.
• Agricultural practices (rice and corn cultivation) significantly influenced ET rates at specific farmland sites.
• An upward trend in annual average evapotranspiration rates was observed across different stations from 2018 to 2021.

Contributions

• Developed a novel hybrid GEP-PM model that enhances the physical interpretability of machine learning approaches by integrating the radiation and aerodynamic terms of the FAO56 Penman-Monteith model with Gene Expression Programming.
• Demonstrated the model's superior accuracy and adaptability for actual evapotranspiration (ETa) estimation across contrasting land surface types (sand dunes and meadows) in semi-arid regions.
• Provided a robust methodology for optimizing model parameters for specific land surface conditions, improving ET prediction accuracy at a fine 30-minute temporal scale, which is crucial for dynamic water resource management.
• Highlighted the differential controls on ET in semi-arid environments, showing that meadows are primarily driven by meteorological factors, while sand dunes are more sensitive to soil moisture content, offering valuable insights for regional water balance dynamics.
• Offered a practical and reliable tool for sustainable water resource management, precision irrigation planning, and hydrological modeling in dryland ecosystems, addressing limitations of traditional models and site-specific machine learning approaches.

Funding

• National Natural Science Foundation of China (Grant 52439004, U2243234, 52169002, 51809141 and 52309021)
• Science and Technology Plan Project of Inner Mongolia Autonomous Region (Grant 2025KYPT0099)
• Inner Mongolia Agricultural University Basic Research Project (Grant BR251403)
• Higher Education Reform and Development Project- Postgraduate Research Innovation Funding Project (Grant no. B20231085Z)
• First-class Academic Subjects Special Research Project of the Education Department of Inner Mongolia Autonomous Region (Grant YLXKZX-NND-010 and YLXKZX-NND-028)
• Inner Mongolia Autonomous Region Science and Technology Leading Talent Team (2022LJRC0007)

Citation

@article{Jia2025Evapotranspiration,
  author = {Jia, Tianyu and Shamseldin, Asaad Y. and Liu, Tingxi and Bao, Yongzhi and Zhang, Yiran and Duan, Limin and Tong, Xin and Li, Mingyang and Qiao, Zixu and Wang, Zhiting},
  title = {Evapotranspiration estimation at different land surface scales in semi-arid areas using gene expression programming and the FAO56 Penman-Monteith model},
  journal = {Journal of Hydrology Regional Studies},
  year = {2025},
  doi = {10.1016/j.ejrh.2025.102952},
  url = {https://doi.org/10.1016/j.ejrh.2025.102952}
}