Hwang et al. (2025) Explainable deep learning-based simulation for evaluating climate-driven future groundwater level changes in South Korea

Identification

• Journal: Groundwater for Sustainable Development
• Year: 2025
• Date: 2025-10-28
• Authors: Bing‐Fang Hwang, Kang‐Kun Lee
• DOI: 10.1016/j.gsd.2025.101541

Research Groups

• Korea Institute of Geoscience and Mineral Resources
• School of Earth and Environmental Sciences, Seoul National University

Short Summary

This study developed explainable deep learning models to simulate future groundwater level changes in South Korea under Shared Socioeconomic Pathways, revealing that high-emission scenarios lead to more unstable and significantly declining groundwater levels, particularly in alluvial aquifers.

Objective

• To develop and apply explainable deep learning models to simulate long-term groundwater level variations through the end of the 21st century under Shared Socioeconomic Pathways (SSP1-2.6 and SSP5-8.5) in South Korea, and to evaluate the direct impacts of climate change on groundwater systems while ensuring model plausibility.

Study Configuration

• Spatial Scale: National groundwater monitoring wells installed in alluvial and bedrock aquifers across South Korea.
• Temporal Scale: Long-term projections through the end of the 21st century.

Methodology and Data

• Models used: Explainable deep learning models; AI interpretation algorithms were applied to verify hydrogeologically plausible relationships before long-term simulations.
• Data sources: Meteorological inputs (for climate change scenarios), data from national groundwater monitoring wells.

Main Results

• Explainable deep learning models were successfully developed for long-term groundwater level simulation, allowing for verification of hydrogeologically plausible relationships.
• Groundwater level patterns were projected to be more unstable under the high-emission scenario (SSP5-8.5) than under the low-emission scenario (SSP1-2.6), exhibiting prolonged low-groundwater level periods and a long-term decline.
• Relative reductions in period-mean groundwater levels from the reference to future periods reached -1.41 % under SSP5-8.5, compared to less than -0.56 % under SSP1-2.6.
• Statistically significant declining trends appeared mainly under SSP5-8.5 and were more pronounced in alluvial aquifers than in bedrock aquifers.

Contributions

• Developed and applied explainable deep learning models for long-term groundwater level simulation, addressing the critical need for assessing model plausibility and explainability in AI applications for hydrological projections.
• Provided a comprehensive evaluation of the direct impacts of climate change on groundwater levels in South Korea under different Shared Socioeconomic Pathways.
• Offered valuable insights for developing adaptive groundwater management strategies tailored to specific aquifer types under increasing climate stress.

Funding

• [No funding information explicitly provided in the given text.]

Citation

@article{Hwang2025Explainable,
  author = {Hwang, Bing‐Fang and Lee, Kang‐Kun},
  title = {Explainable deep learning-based simulation for evaluating climate-driven future groundwater level changes in South Korea},
  journal = {Groundwater for Sustainable Development},
  year = {2025},
  doi = {10.1016/j.gsd.2025.101541},
  url = {https://doi.org/10.1016/j.gsd.2025.101541}
}