Matrix scenario-based urban flooding damage prediction via convolutional neural network.

Area of Science:

• Environmental Science
• Urban Planning
• Artificial Intelligence

Background:

• Traditional flood assessment methods struggle with urban heterogeneity and complex climate scenarios.
• Urban pluvial flooding poses significant risks to infrastructure and economies.
• Accurate hazard assessment is crucial for effective urban planning and climate resilience.

Purpose of the Study:

• To introduce a high-resolution tool for urban pluvial flood hazard assessment using convolutional neural networks.
• To account for urban spatial heterogeneity and complex climate change impacts.
• To improve the accuracy and interpretability of flood risk prediction.

Main Methods:

• Development of a convolutional neural network (CNN) model for flood hazard assessment.
• Case study application in Shenzhen, China.
• Performance analysis comparing predicted and actual flood depths, with focus on extreme events.

Main Results:

• The CNN model demonstrated high accuracy in predicting most urban flood depths, with limitations in extreme flood prediction (>35 cm).
• Western and central Shenzhen, areas with extensive construction, were identified as highly vulnerable.
• Estimated economic losses under the severe Scenario 25 reached approximately $25,484 million, with commercial and residential areas most affected.

Conclusions:

• Findings highlight escalating flood impacts under climate change scenarios, necessitating reinforced mitigation strategies.
• Flood control should prioritize reducing inundation areas and integrating land-use characteristics.
• The tool aids in precise flood-prone area identification, enhancing evacuation plans, urban sustainability, and climate resilience.