Abdi et al. (2026) Groundwater level and drought prediction with hybrid artificial intelligence and deep learning models and data preprocessing techniques

Identification

• Journal: Acta Geophysica
• Year: 2026
• Date: 2026-01-05
• Authors: Somaye Abdi, Hossein Fathian, Mehdi Asadi Lour, Aslan Igdernejad, Ali Asareh
• DOI: 10.1007/s11600-025-01768-2

Research Groups

• Department of Irrigation and Drainage, Ahv.C., Islamic Azad University, Ahvaz, Iran
• Department of Water Resources Engineering, Ahv.C., Islamic Azad University, Ahvaz, Iran

Short Summary

This study developed a hybrid modeling framework combining signal preprocessing (Wavelet Transform, Complete Ensemble Empirical Mode Decomposition) with artificial intelligence and deep learning models (Long Short-Term Memory, Group Method of Data Handling) to predict groundwater levels and drought in the Nahavand aquifer, western Iran. The W-GMDH model achieved the highest accuracy (coefficient of determination = 0.954, root mean square error = 0.027 m) and predicted continued severe groundwater drought for the region.

Objective

• To investigate and compare the performance of standalone (LSTM, GMDH) and hybrid (W-LSTM, W-GMDH, CEEMD-LSTM, CEEMD-GMDH) models in forecasting groundwater levels (GWL) in the Nahavand aquifer, western Iran.
• To assess groundwater drought (GWD) conditions using the Groundwater Resource Index (GRI) based on the predicted GWL.

Study Configuration

• Spatial Scale: Nahavand aquifer, western Iran, covering an alluvial area of approximately 502.2 square kilometers within a sub-basin of 1,779.2 square kilometers.
• Temporal Scale: A 28-year statistical period (1996–2024) for data analysis and model training, with monthly scale predictions for GWL and GWD, including a forecast for the first six months of the 2024–2025 water year (April–September 2025).

Methodology and Data

• Models used:
  • Artificial Intelligence/Deep Learning: Long Short-Term Memory (LSTM), Group Method of Data Handling (GMDH).
  • Signal Decomposition (Preprocessing): Wavelet Transform (WT), Complete Ensemble Empirical Mode Decomposition (CEEMD).
  • Hybrid Models: W-LSTM, W-GMDH, CEEMD-LSTM, CEEMD-GMDH.
  • Drought Index: Groundwater Resource Index (GRI).
  • Software/Techniques: MATLAB (for model implementation), HEC-4 (for unit hydrograph and data reconstruction), Geographic Information System (GIS) with Kriging interpolation (for spatial mapping).
• Data sources:
  • Precipitation, temperature, and evaporation records from meteorological stations.
  • Groundwater level (GWL) data from 37 selected observation wells in the Nahavand aquifer, provided by the Regional Water Company of Hamedan Province.

Main Results

• The hybrid models significantly outperformed standalone models; WT improved the coefficient of determination (R²) by 9.1% for LSTM and 7.9% for GMDH, while CEEMD improved R² by 9.5% for LSTM and 6.8% for GMDH.
• The W-GMDH hybrid model demonstrated the best performance for GWL prediction, achieving a coefficient of determination (R²) of 0.954 and a root mean square error (RMSE) of 0.027 m. It also had the lowest Akaike information criterion (AIC) of 157.15.
• Analysis of the Groundwater Resource Index (GRI) revealed prolonged and severe groundwater drought (GWD) conditions in the Nahavand aquifer, particularly in the water years 2022–2023 and 2023–2024.
• Forecasts for the first six months of the 2024–2025 water year indicate a continuation of severe GWD in the region.

Contributions

• Addresses a critical research gap by providing the first comprehensive investigation into water resources and drought dynamics in the Nahavand aquifer, a vital agricultural and domestic water source in western Iran.
• Develops and evaluates an integrated hybrid modeling framework combining advanced signal decomposition techniques (WT, CEEMD) with AI/DL models (LSTM, GMDH) for improved GWL forecasting and GWD assessment.
• Demonstrates that preprocessing techniques significantly enhance the accuracy and reliability of AI/DL models in capturing complex, nonlinear groundwater dynamics, particularly in semi-arid regions.
• Provides both historical and future projections of GWL and drought conditions, offering a robust and accurate decision-support tool for proactive groundwater resource management and policy development under climate change and increasing water demand.

Funding

Not explicitly stated in the paper. The Hamadan Regional Water Company is acknowledged for providing essential data.

Citation

@article{Abdi2026Groundwater,
  author = {Abdi, Somaye and Fathian, Hossein and Lour, Mehdi Asadi and Igdernejad, Aslan and Asareh, Ali},
  title = {Groundwater level and drought prediction with hybrid artificial intelligence and deep learning models and data preprocessing techniques},
  journal = {Acta Geophysica},
  year = {2026},
  doi = {10.1007/s11600-025-01768-2},
  url = {https://doi.org/10.1007/s11600-025-01768-2}
}