Wu et al. (2025) Graph Fourier Deep Learning for Spatiotemporal and Hydrogeological Interpretation of Groundwater Levels in the Yellow River Basin

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water Resources Research
• Year: 2025
• Date: 2025-11-01
• Authors: Zhenjiang Wu, Yingying Yao, Shuitao Guo, Shuai Yang, Xin He, Michele Lancia, Chunmiao Zheng
• DOI: 10.1029/2025wr041215

Research Groups

Information not available from the provided text.

Short Summary

This study proposes a novel Graph Fourier Network (GFN) model that integrates hydrogeological prior information for regional groundwater level prediction. The GFN model significantly outperforms baseline models, achieving high accuracy and enhanced extrapolation capability for lead times up to 25 days, effectively capturing the complex dynamics of groundwater systems.

Objective

• To propose and evaluate a novel Graph Fourier Network (GFN) model that integrates Fourier domain modeling, dynamic graph structure generation, multi-head graph attention, and hydrogeological prior information to improve the accuracy, interpretability, and extrapolation capability of regional groundwater level predictions.

Study Configuration

• Spatial Scale: Yellow River Basin (YRB), China, utilizing data from 473 groundwater monitoring wells.
• Temporal Scale: Groundwater level predictions for extended lead times up to 25 days.

Methodology and Data

• Models used: Graph Fourier Network (GFN) (proposed). Baseline models: Long Short-Term Memory (LSTM), Extreme Gradient Boosting (XGBoost), Spatio-Temporal LSTM (ST-LSTM).
• Data sources: Groundwater monitoring data from 473 wells across the Yellow River Basin. Hydrogeological units embedded as prior information. Climatic and ecological drivers are used as inputs.

Main Results

• The proposed GFN model significantly outperforms baseline models (LSTM, XGBoost, ST-LSTM), achieving an R² of up to 0.90.
• Incorporating hydrogeological prior knowledge enhances the extrapolation capability of groundwater level predictions, particularly for extended lead times up to 25 days.
• Predicted groundwater level dynamics exhibit lagged and nonlinear responses to climatic and ecological drivers.
• The spatial heterogeneity of these response patterns is shaped by hydrogeological conditions, including prolonged influence of precipitation and spatially varying effects of evapotranspiration and vegetation across different hydrogeological units.

Contributions

• Proposes a novel Graph Fourier Network (GFN) model that integrates Fourier domain modeling, dynamic graph structure generation, and multi-head graph attention for regional groundwater level prediction.
• Demonstrates the effectiveness of incorporating hydrogeological prior knowledge into deep learning models to enhance extrapolation capability and capture complex system characteristics.
• Provides a fast and reliable approach for short- and long-term regional groundwater prediction under varying hydrogeological conditions.

Funding

Information not available from the provided text.

Citation

@article{Wu2025Graph,
  author = {Wu, Zhenjiang and Yao, Yingying and Guo, Shuitao and Yang, Shuai and He, Xin and Lancia, Michele and Zheng, Chunmiao},
  title = {Graph Fourier Deep Learning for Spatiotemporal and Hydrogeological Interpretation of Groundwater Levels in the Yellow River Basin},
  journal = {Water Resources Research},
  year = {2025},
  doi = {10.1029/2025wr041215},
  url = {https://doi.org/10.1029/2025wr041215}
}