Asadollah et al. (2025) Climate-responsive crop forecasting: an EEMD-LSTM fusion approach for improved strategic crop yield simulation

Identification

• Journal: Acta Geophysica
• Year: 2025
• Date: 2025-12-26
• Authors: Seyed Babak Haji Seyed Asadollah, Yusef Kheyruri, Ahmad Sharafati, Asaad Shakir Hameed
• DOI: 10.1007/s11600-025-01764-6

Research Groups

• Department of Civil Engineering, University of Alicante, Spain
• Department of Civil Engineering, SR.C, Islamic Azad University, Tehran, Iran
• New Era and Development in Civil Engineering Research Group, Scientific Research Center, Al-Ayen University, Thi-Qar, Iraq
• Department of Mathematics, General Directorate of Thi-Qar Education, Ministry of Education, Nasiriyah, Thi-Qar, Iraq
• Petroleum Engineering College, Al-Ayen University, Nasiriah, Thi-Qar, Iraq

Short Summary

This study developed an EEMD-LSTM fusion model to improve strategic crop yield forecasting for barley, lentils, pea, and wheat across all provinces of Iran. The research demonstrates that integrating Ensemble Empirical Mode Decomposition (EEMD) as a signal denoising technique generally enhances the predictive accuracy of the Long Short-Term Memory (LSTM) model by reducing noise in climate input data.

Objective

• To integrate Ensemble Empirical Mode Decomposition (EEMD) with a Long Short-Term Memory (LSTM) model to enhance crop-specific yield estimation for strategic crops (barley, lentils, pea, and wheat) in Iran.
• To compare the predictive outcomes of the EEMD-LSTM hybrid model against a traditional standalone LSTM model to ascertain the extent to which EEMD affects prediction accuracy.

Study Configuration

• Spatial Scale: National scale, covering all 31 provinces of Iran. Climate data has a spatial resolution of approximately 0.5 degrees (50 kilometers).
• Temporal Scale: Annual crop yield data from 2005 to 2020 (15 years), disaggregated to monthly values. Monthly climate parameters are used as input.

Methodology and Data

• Models used:
  • Ensemble Empirical Mode Decomposition (EEMD): Used as a signal denoising and modification technique to decompose nonstationary time series into intrinsic mode functions (IMFs).
  • Long Short-Term Memory (LSTM): A deep learning recurrent neural network (RNN) variant for time series prediction.
  • Kriging model: Used for spatial extrapolation of prediction errors across the study area.
• Data sources:
  • Annual crop yield data (barley, lentils, pea, wheat): Iran’s Ministry of Agriculture (2005–2020).
  • Climate parameters (minimum temperature, maximum temperature, precipitation, Standardized Precipitation–Evapotranspiration Index (SPEI)): NASA POWER database, derived from the MERRA-2 reanalysis dataset.

Main Results

• The application of EEMD generally enhanced the predictive performance of the LSTM model for most agricultural products, particularly barley, lentils, and peas.
• For lentils, the EEMD-LSTM method yielded notable improvements, with reductions of 15.7% in Mean Absolute Error (MAE), 13.8% in Root Mean Squared Error (RMSE), and 8% in Pearson Correlation Coefficient (PCC) compared to the standalone LSTM.
• For wheat, while the impact was not statistically significant, EEMD still reduced RMSE by approximately 42%.
• Spatial assessment revealed clear geographic differences in EEMD's effectiveness:
  • Substantial reduction in prediction errors for barley in the northwestern region.
  • A similar, though smaller, improvement was observed for lentils in the northwestern region.
  • Conversely, EEMD increased prediction errors for pea production in both the northwest and eastern regions.
• Temporal analysis showed that both LSTM and EEMD-LSTM models generally aligned well with observed data trends, with most predicted values falling within the 80% confidence interval.

Contributions

• Presents the first application of an EEMD-LSTM hybrid model for crop yield estimation in agricultural studies using time series data.
• Demonstrates the effectiveness of EEMD as a noise removal tool to significantly enhance the prediction accuracy of LSTM for strategic crops, particularly barley and lentils.
• Offers a climate-responsive crop forecasting approach that can be utilized for developing food security policies and improving agricultural product performance and productivity.
• Highlights the spatial variability of model performance, providing insights into regional applicability and areas requiring further investigation (e.g., warmer and drier climates).

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Asadollah2025Climateresponsive,
  author = {Asadollah, Seyed Babak Haji Seyed and Kheyruri, Yusef and Sharafati, Ahmad and Hameed, Asaad Shakir},
  title = {Climate-responsive crop forecasting: an EEMD-LSTM fusion approach for improved strategic crop yield simulation},
  journal = {Acta Geophysica},
  year = {2025},
  doi = {10.1007/s11600-025-01764-6},
  url = {https://doi.org/10.1007/s11600-025-01764-6}
}