Sun et al. (2026) Integrating deep learning and groundwater dynamics for drought vulnerability assessment under climate scenarios

Identification

• Journal: Groundwater for Sustainable Development
• Year: 2026
• Date: 2026-02-05
• Authors: Wei Sun, Li‐Chiu Chang, Jun Jie Lin, Fi‐John Chang
• DOI: 10.1016/j.gsd.2026.101591

Research Groups

• Department of Bioenvironmental Systems Engineering, National Taiwan University, Taipei, Taiwan
• Department of Environmental Science, Policy and Management, University of California, Berkeley, CA, United States of America
• Department of Water Resources and Environmental Engineering, Tamkang University, New Taipei City, Taiwan

Short Summary

This study develops an AI-driven framework (DCDVI) integrating deep learning with climate, groundwater, and socio-environmental factors to assess future drought vulnerability, demonstrating improved groundwater prediction and highlighting increased drought severity under the SSP5-8.5 climate scenario.

Objective

• To develop an AI-driven framework to assess future drought risk from climate, groundwater, and socio-environmental drivers, particularly in groundwater-dependent regions.

Study Configuration

• Spatial Scale: Taiwan's Zhuoshui River alluvial fan; basin-wide.
• Temporal Scale: Calibration period: 22 years (historical data). Projection period: 2021–2100.

Methodology and Data

• Models used: Hybrid Convolutional Neural Network–Backpropagation model (CNN-BP), Backpropagation Neural Network (BPNN) for benchmark comparison, Deep Learning-based Comprehensive Drought Vulnerability Indicator (DCDVI) framework.
• Data sources: Gridded precipitation, temperature, Standardized Precipitation Index (SPI), groundwater levels from 18 monitoring wells, physiographic data (soil, land use, elevation, slope, distance to river), socio-economic data (population).

Main Results

• The CNN-BP model significantly improved groundwater prediction accuracy compared to BPNN, increasing the correlation coefficient by 35.85% and reducing Mean Absolute Error (MAE) by 19.51%.
• Integration of multi-factor drivers (climatic, groundwater, physiographic, socio-economic) improved the overall drought risk evaluation.
• Projections under the SSP5-8.5 climate scenario indicate greater drought severity and deeper groundwater decline.
• A projected population decline may slightly mitigate future drought-related water demand.

Contributions

• Development of a novel deep learning-based comprehensive drought vulnerability indicator (DCDVI) framework that explicitly integrates groundwater dynamics with climatic and socio-environmental factors.
• Introduction of a robust hybrid CNN-BP model for accurate groundwater level prediction under future climate scenarios, outperforming traditional neural network approaches.
• Provides an integrated and advanced approach for drought vulnerability assessment, addressing limitations of existing drought indices that implicitly treat groundwater and AI studies that focus solely on prediction without integrated vulnerability metrics.

Funding

• Not specified in the provided text.

Citation

@article{Sun2026Integrating,
  author = {Sun, Wei and Chang, Li‐Chiu and Lin, Jun Jie and Chang, Fi‐John},
  title = {Integrating deep learning and groundwater dynamics for drought vulnerability assessment under climate scenarios},
  journal = {Groundwater for Sustainable Development},
  year = {2026},
  doi = {10.1016/j.gsd.2026.101591},
  url = {https://doi.org/10.1016/j.gsd.2026.101591}
}