Kayhomayoon et al. (2025) Improving the performance of daily pan evaporation (Evp) prediction using the ensemble empirical mode decomposition combined with deep learning models

Identification

• Journal: Scientific Reports
• Year: 2025
• Date: 2025-12-04
• Authors: Zahra Kayhomayoon, Naser Arya Azar, Sami Ghordoyee Milan, Ronny Berndtsson, Peiman Kianmehr
• DOI: 10.1038/s41598-025-27255-8

Research Groups

• Department of Geology, Payame Noor University, Tehran, Iran
• Water Engineering Department, Tabriz University, Tabriz, Iran
• Water Engineering Department, Faculty of Agricultural Technology, University College of Agriculture & Natural Resources, University of Tehran, Tehran, Iran
• Centre for Advanced Middle Eastern Studies & Division of Water Resources Engineering, Lund University, Lund, Sweden
• Civil Engineering Department, American University in Dubai, Dubai, UAE

Short Summary

This study presents a novel hybrid approach for daily pan evaporation (Evp) prediction, combining gamma test and genetic algorithm (GTGA) for optimal input selection, ensemble empirical mode decomposition (EEMD) for noise reduction, and deep learning models (LSTM and CNN). The EEMD-CNN hybrid model demonstrated superior performance, significantly enhancing prediction accuracy for water resource management in arid and semi-arid regions.

Objective

• To develop and evaluate a novel hybrid methodology for daily pan evaporation (Evp) prediction by integrating gamma test and genetic algorithm (GTGA) for optimal input selection, ensemble empirical mode decomposition (EEMD) for noise reduction, and deep learning models (CNN and LSTM).

Study Configuration

• Spatial Scale: Kardeh Dam catchment area, northeastern Iran (north of Mashhad City), covering approximately 540 km². Elevations range from 1291 m to 2948 m.
• Temporal Scale: Daily data collected between 2015 and 2017.

Methodology and Data

• Models used:
  • Gamma Test (GT) and Genetic Algorithm (GA) for input selection (GTGA)
  • Ensemble Empirical Mode Decomposition (EEMD) for data preprocessing
  • Long Short-Term Memory (LSTM) neural network
  • Convolutional Neural Network (CNN)
  • Hybrid models: EEMD-LSTM, EEMD-CNN
• Data sources:
  • Daily meteorological observations from Kardeh Dam catchment area, Iran (2015-2017).
  • Input variables included maximum temperature, minimum temperature, precipitation, and pan evaporation (Evp) from previous days (Evp(n-1), Evp(n-2), Evp(n-3)).

Main Results

• The GTGA identified maximum temperature, minimum temperature, precipitation, Evp(n-1), and Evp(n-2) as the optimal input variables for Evp prediction.
• EEMD successfully decomposed each input variable into approximately nine Intrinsic Mode Functions (IMFs), effectively reducing noise and simplifying complex patterns.
• Hybrid models (EEMD-CNN and EEMD-LSTM) significantly outperformed standalone deep learning models (CNN and LSTM) in Evp prediction.
• The EEMD-CNN hybrid model exhibited the best performance on test data with an RMSE of 0.87 mm, MAE of 0.62 mm, SI of 0.155, and NSE of 0.973.
• Taylor's diagram further confirmed EEMD-CNN as the best model, showing a correlation coefficient of 0.997, RMSE of 0.81 mm, and standard deviation of 5.2 mm.
• Uncertainty analysis revealed larger uncertainty bands for Evp values between 0 and 5 mm/day, while values exceeding 10 mm consistently had smaller uncertainty.

Contributions

• Introduction of a novel hybrid methodology integrating GTGA for optimal feature selection, EEMD for noise reduction, and deep learning models (CNN and LSTM) for enhanced daily pan evaporation prediction.
• Demonstrated the significant improvement in deep learning model performance (CNN and LSTM) when coupled with EEMD for preprocessing hydrological time series data.
• Identified a parsimonious yet effective set of input variables (maximum temperature, minimum temperature, precipitation, and previous two days' Evp) for accurate Evp forecasting in semi-arid regions.
• Provided a robust and accurate predictive tool for water resource managers, aiding in better understanding future evaporation trends and optimizing water resource management, particularly in water-scarce regions.

Funding

• Middle East in the Contemporary World (MECW) project at the Centre for Advanced Middle Eastern Studies, Lund University.

Citation

@article{Kayhomayoon2025Improving,
  author = {Kayhomayoon, Zahra and Azar, Naser Arya and Milan, Sami Ghordoyee and Berndtsson, Ronny and Kianmehr, Peiman},
  title = {Improving the performance of daily pan evaporation (Evp) prediction using the ensemble empirical mode decomposition combined with deep learning models},
  journal = {Scientific Reports},
  year = {2025},
  doi = {10.1038/s41598-025-27255-8},
  url = {https://doi.org/10.1038/s41598-025-27255-8}
}