Xie et al. (2026) Quantifying multidimensional drought risks under climate change: A deep learning-based downscaling and DDF curves approach

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-10-29
• Authors: Xuan Xie, Guohe Huang, Yuke Li, Shuguang Wang, Yurui Fan
• DOI: 10.1016/j.jhydrol.2026.136717

Research Groups

• School of Water Conservancy and Environment, University of Jinan, Jinan, China
• School of Environmental Science and Engineering, Shandong University, Qingdao, China
• Institute for Energy, Environment and Sustainable Communities, University of Regina, Regina, Saskatchewan, Canada
• Section of Sanitary Engineering, Department of Water Management, Faculty of Civil Engineering and Geosciences, Delft University of Technology, Delft, Netherlands
• Shandong Key Laboratory of Water Pollution Control and Resource Reuse, School of Environmental Science and Engineering, Shandong University, Qingdao, China
• Department of Civil and Environmental Engineering, Brunel University of London, Middlesex, United Kingdom

Short Summary

This study develops an integrated framework combining multi-criteria GCM selection, deep learning downscaling, and Duration-Deficit-Frequency (DDF) analysis to quantify multidimensional drought risks under climate change. It reveals that future extreme droughts in the Upper Yangtze River Basin will generally exceed historical levels, with increased cumulative precipitation deficits and significant uncertainties from GCMs and downscaling methods.

Objective

• To develop an integrated analytical framework for reliably estimating future drought characteristics and associated risks, addressing limitations in spatial resolution of traditional datasets, direct representation of precipitation deficits, and the interdependence of drought characteristics.

Study Configuration

• Spatial Scale: Analyzed stations in the Upper Yangtze River Basin.
• Temporal Scale: Future climate scenarios (under four Shared Socioeconomic Pathway (SSP) scenarios) compared to historical levels.

Methodology and Data

• Models used: Multi-criteria General Circulation Model (GCM) selection, deep learning-based ensemble downscaling (ECA-CNN mentioned in keywords), Duration-Deficit-Frequency (DDF) curves constructed using best-fit probability distributions, Bootstrap analysis for uncertainty, Multilevel Factorial Analysis (MFA) for uncertainty source identification.
• Data sources: Large-scale precipitation datasets (GCM projections), downscaled precipitation series.

Main Results

• The occurrence of extreme droughts generally exceeds historical levels under future climate scenarios, with the magnitude of change varying substantially among stations and SSP scenarios.
• DDF analysis indicates increased cumulative precipitation deficits, particularly for longer-duration drought events.
• Uncertainty in the DDF curves increases with drought duration and return period.
• GCM ensemble configuration and downscaling methodology are identified as the dominant sources of uncertainty.

Contributions

• Proposes an integrated analytical framework that combines multi-criteria GCM selection, deep learning-based ensemble downscaling, and Duration-Deficit-Frequency (DDF) analysis for quantifying multidimensional drought risks.
• Provides a physically interpretable and uncertainty-aware approach for assessing station-specific drought characteristics and associated risks under alternative SSP scenarios.
• Links drought-index-based event identification with physical precipitation-deficit quantification.

Funding

• Not explicitly mentioned in the provided text.

Citation

@article{Xie2026Quantifying,
  author = {Xie, Xuan and Huang, Guohe and Li, Yuke and Wang, Shuguang and Fan, Yurui},
  title = {Quantifying multidimensional drought risks under climate change: A deep learning-based downscaling and DDF curves approach},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136717},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136717}
}