He et al. (2026) Explaining urban flood susceptibility under rainfall uncertainty through probabilistic modeling and interpretable machine learning

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-03-31
• Authors: Sijing He, Jeffrey Neal, Chengguang Lai, Zhaoli Wang
• DOI: 10.1016/j.jhydrol.2026.135430

Research Groups

• School of Civil Engineering and Transportation, State Key Laboratory of Subtropical Building and Urban Science, South China University of Technology, Guangzhou, China
• School of Geographical Sciences, University of Bristol, Bristol, UK

Short Summary

This study proposes a comprehensive framework integrating probabilistic hydrodynamic simulations with interpretable machine learning to systematically analyze urban and environmental factors influencing urban flood susceptibility at the grid scale, revealing that terrain features, drainage capacity, and urban form are key predictors with depth- and scale-dependent patterns.

Objective

• To systematically analyze urban and environmental factors influencing urban flood susceptibility at the grid scale under rainfall uncertainty.

Study Configuration

• Spatial Scale: Grid scale, local scales, building-centered 2D and 3D morphological features within an urban environment.
• Temporal Scale: Focus on flood event susceptibility under rainfall uncertainty, considering future climate uncertainty and long-term urban resilience.

Methodology and Data

• Models used: Probabilistic hydrodynamic simulations, interpretable machine learning (XGBoost, SHAP).
• Data sources: Stochastic rainfall inputs for hydrodynamic simulations; seventeen identified predictors including terrain features (elevation minimum, digital elevation model height), drainage capacity, and urban form factors (e.g., openness, morphological spatial pattern analysis metrics).

Main Results

• Average conditional inundation probability decreases at higher water depth thresholds, while the spatial pattern of high-risk areas remains consistent across different thresholds.
• Seventeen predictors of urban flood susceptibility were identified.
• Terrain features, such as elevation minimum and digital elevation model height, demonstrated the strongest predictive contributions.
• Drainage capacity and urban form factors also showed notable contributions to flood susceptibility.
• The influence of urban form was found to be nonlinear and threshold-dependent; higher openness was generally associated with lower predicted probabilities, though this association varied spatially.

Contributions

• Development of a comprehensive framework integrating probabilistic hydrodynamic simulations with interpretable machine learning (XGBoost and SHAP) for urban flood susceptibility assessment.
• Systematic, grid-scale analysis of urban and environmental factors influencing flood susceptibility under rainfall uncertainty.
• Identification of key predictors, including terrain features, drainage capacity, and urban form, and characterization of their depth- and scale-dependent influences.
• Highlighting the necessity of multi-scale, probabilistic, and interpretable approaches for effective urban flood susceptibility assessment and mitigation planning.

Funding

• Not specified in the provided text.

Citation

@article{He2026Explaining,
  author = {He, Sijing and Neal, Jeffrey and Lai, Chengguang and Wang, Zhaoli},
  title = {Explaining urban flood susceptibility under rainfall uncertainty through probabilistic modeling and interpretable machine learning},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.135430},
  url = {https://doi.org/10.1016/j.jhydrol.2026.135430}
}