Zhu et al. (2026) Modeling urban flood susceptibility and identifying key flood-inducing factor chains using Bayesian network

Identification

• Journal: Natural Hazards
• Year: 2026
• Date: 2026-01-01
• Authors: W. J. Zhu, Deyun Wang, Ludan Zhang
• DOI: 10.1007/s11069-025-07773-4

Research Groups

• School of Economics and Management, China University of Geosciences, Wuhan, 430074, China
• The Laboratory of Natural Disaster Risk Prevention and Emergency Management, China University of Geosciences, Wuhan, 430074, China

Short Summary

This study develops a Bayesian Network (BN) model to assess urban flood susceptibility in Beijing, quantifying uncertainties and interdependencies among climatic, topographical, hydrological, and socio-economic factors. The model successfully maps high-risk areas in central urban zones and identifies three key flood-inducing factor chains, highlighting rainfall, land use, and socio-economic factors as dominant drivers.

Objective

• To develop a Bayesian Network model for assessing urban flood susceptibility, quantifying uncertainties, and capturing interdependencies among influencing factors.
• To identify key flood-inducing factor chains and their propagation pathways to inform targeted flood mitigation and sustainable urban planning.

Study Configuration

• Spatial Scale: Beijing, China, spanning approximately 16,410 square kilometers. Data processed at 30 meter resolution, flood susceptibility mapped at 300 meter grid cells.
• Temporal Scale: Focus on the "23·7" extraordinary rainstorm flood disaster (July-August 2023). Flood season rainfall data from June to September 2023, and monthly average rainfall over the past three years. Satellite imagery from June 1-30, 2023 (pre-flood) and July 15-August 15, 2023 (post-flood).

Methodology and Data

• Models used:
  • Bayesian Network (BN) for flood susceptibility assessment, structure learning (Peter-Clark algorithm combined with domain knowledge), and parameter learning (Maximum Likelihood Estimation).
  • Modified Normalized Difference Water Index (MNDWI) for water body extraction from satellite imagery.
  • OTSU algorithm for automatic threshold calculation for MNDWI.
  • K-Means algorithm and natural breaks method for data discretization.
  • Pearson correlation and Variance Inflation Factor (VIF) analyses for variable selection.
• Data sources:
  • Climatic: Flood season rainfall, storm frequency.
  • Topographical and Hydrological: Digital Elevation Model (DEM), slope, Normalized Difference Vegetation Index (NDVI), river network density, river proximity.
  • Socio-economic: Population density, GDP density, road network density (from OpenStreetMap), land use type.
  • Satellite: Sentinel-2 imagery (processed on Google Earth Engine - GEE) for flood inundation areas.
  • Observation: Historical flood inundation data (for sample points).

Main Results

• The BN model achieved an overall accuracy of 0.85, precision of 0.83, recall of 0.89, F1-score of 0.86, and an Area Under the Curve (AUC) of 0.93, demonstrating high reliability.
• The flood susceptibility map classified Beijing into: 5.41% very high, 8.77% high, 11.02% moderate, 22.34% low, and 52.46% very low susceptibility.
• High-susceptibility areas are predominantly located in central urban zones characterized by flat terrain, extensive impervious surface coverage, and dense population.
• An overlay analysis comparing predicted moderate-to-very high susceptibility areas with actual inundation points showed a 76% matching ratio.
• Three key flood-inducing factor chains were identified:
  1. Climatic: Flood Season Rainfall → Storm Frequency → Flood Susceptibility (increases flood probability from 49% to 57%).
  2. Topographical/Hydrological: DEM → Slope → Land Use → NDVI → Flood Susceptibility (increases flood probability from 49% to 66% under specific conditions).
  3. Socio-economic: Population Density → GDP Density → Road Network Density → Flood Susceptibility (increases flood probability from 49% to 64% under specific conditions).
• Sensitivity analysis revealed that flood season rainfall, population density, DEM, GDP density, and land use play dominant roles in influencing flood susceptibility.

Contributions

• Proposes a novel systematic approach for urban flood susceptibility assessment using a Bayesian Network that explicitly quantifies uncertainties and interdependencies among diverse influencing factors, overcoming limitations of traditional methods.
• Integrates machine learning algorithms (PC algorithm, MLE) with domain knowledge to construct a robust and interpretable BN model structure for flood susceptibility.
• Identifies and quantifies three critical flood-inducing factor chains across climatic, topographical/hydrological, and socio-economic dimensions, providing a deeper understanding of flood propagation pathways.
• Generates a validated probabilistic flood susceptibility map for Beijing, offering practical guidance for targeted flood mitigation and sustainable urban planning.
• Provides a framework for identifying high-priority intervention points within the disaster chain, supporting decision-making for urban risk management.

Funding

• National Natural Science Foundation of China (Grant No.72274186)

Citation

@article{Zhu2026Modeling,
  author = {Zhu, W. J. and Wang, Deyun and Zhang, Ludan},
  title = {Modeling urban flood susceptibility and identifying key flood-inducing factor chains using Bayesian network},
  journal = {Natural Hazards},
  year = {2026},
  doi = {10.1007/s11069-025-07773-4},
  url = {https://doi.org/10.1007/s11069-025-07773-4}
}