A cross-city transferable convolutional neural network framework for assessing street-scale flood risks in urban...

Area of Science:

• Environmental Science
• Urban Planning
• Artificial Intelligence

Background:

• Urban flooding poses significant risks to infrastructure and public safety in rapidly urbanizing areas.
• Climate change exacerbates flood risks, necessitating advanced assessment tools.

Purpose of the Study:

• To develop and validate an AI-driven framework for street-scale urban flood risk assessment.
• To integrate hydrometeorological, topographic, and urban morphological data for enhanced flood prediction.
• To evaluate the spatial transferability of the AI model across different urban environments.

Main Methods:

• A convolutional neural network (CNN)-based deep learning framework was employed.
• The model was trained using data from Shenzhen and applied to Hong Kong.
• Flood inundation was predicted under various rainfall scenarios, including a 100-year recurrence interval event.

Main Results:

• Under extreme rainfall, 64.1 km² of Hong Kong's urban area was predicted to be flood-prone, with average depths of 15.6 cm.
• A road-level assessment identified 501 high-risk road segments (18.9% of the network).
• The CNN model demonstrated strong spatial transferability, indicating its potential for broader application.

Conclusions:

• The AI framework provides a robust method for detailed urban flood risk assessment.
• Findings highlight the need for region-specific flood mitigation strategies.
• The model's cross-regional transferability offers an innovative approach for flood risk management in similar urban settings.