Wang et al. (2025) A cross-city transferable convolutional neural network framework for assessing street-scale flood risks in urban networks

Identification

• Journal: Journal of Environmental Management
• Year: 2025
• Date: 2025-11-19
• Authors: Mo Wang, Junling Zhuang, Jiayu Zhao, Chuanhao Sun, Jun Li, L. Zhou, Haowen Fan, Shiqi Zhou, Jinda Qi
• DOI: 10.1016/j.jenvman.2025.127977

Research Groups

• College of Architecture and Urban Planning, Guangzhou University
• Architectural Design and Research Institute of Guangzhou University
• Department of Architecture, National University of Singapore
• Shanghai Research Institute for Intelligent Autonomous Systems, Tongji University
• College of Environmental Science and Engineering, Tongji University
• College of Design and Innovation, Tongji University

Short Summary

This study develops an AI-driven convolutional neural network (CNN) framework for street-scale flood risk assessment by integrating hydrometeorological, topographic, and urban morphological data. The model, trained on Shenzhen data and applied to Hong Kong, demonstrates strong spatial transferability and identifies critical flood-prone areas and high-risk road segments under various rainfall scenarios.

Objective

• To develop and validate a cross-city transferable convolutional neural network (CNN) framework for assessing street-scale flood risks in urban networks.

Study Configuration

• Spatial Scale: Street-scale urban networks, specifically applied to Hong Kong after training on Shenzhen data. Identified flood-prone areas cover 64.1 square kilometers.
• Temporal Scale: Four design rainfall scenarios, including an extreme 100-year recurrence interval, 60-minute rainfall event.

Methodology and Data

• Models used: Convolutional Neural Network (CNN) framework.
• Data sources: Hydrometeorological, topographic, and urban morphological data.

Main Results

• Under the extreme 100-year recurrence interval, 60-minute rainfall scenario, the flood-prone area in Hong Kong expanded to 64.1 square kilometers (5.79 % of the urban area).
• The average inundation depth reached 0.156 meters, with a maximum depth of 0.686 meters.
• A road-level flood risk assessment identified 501 high-risk road segments (18.9 % of the total network), primarily concentrated in the southern Kowloon Peninsula and northern Hong Kong Island.

Contributions

• Introduces an innovative AI-driven CNN framework for street-scale urban flood risk assessment, addressing limitations of traditional computationally intensive and less adaptable methods.
• Demonstrates strong spatial transferability of the deep learning model across different urban contexts (Shenzhen to Hong Kong), enhancing its applicability to cities with similar conditions.
• Provides a detailed, high-granularity assessment of flood inundation depths and identifies specific high-risk road segments, crucial for targeted urban planning and flood mitigation strategies.
• Integrates diverse data types (hydrometeorological, topographic, urban morphological) into a unified framework for comprehensive flood risk analysis.

Funding

Not specified in the provided text.

Citation

@article{Wang2025crosscity,
  author = {Wang, Mo and Zhuang, Junling and Zhao, Jiayu and Sun, Chuanhao and Li, Jun and Zhou, L. and Fan, Haowen and Zhou, Shiqi and Qi, Jinda},
  title = {A cross-city transferable convolutional neural network framework for assessing street-scale flood risks in urban networks},
  journal = {Journal of Environmental Management},
  year = {2025},
  doi = {10.1016/j.jenvman.2025.127977},
  url = {https://doi.org/10.1016/j.jenvman.2025.127977}
}