Flood Risk Management Study Utilizes Convolutional Neural Networks for Hazard Assessment

Strategic Insights -

Research has identified rain-driven urban surface water flooding as a prevalent natural disaster, causing disruptions, economic losses, and casualties. The Centre for Water Systems at the University of Exeter has developed a novel deep learning method utilizing convolutional neural networks (CNNs) to assess flood hazards. The model incorporates three techniques to enhance feature representation and mitigate data imbalance. When trained on catchment data and flood hazard targets, the proposed method demonstrated potential for rapid and accurate identification of high-risk areas.

Key Takeaways:

  • The study highlights the importance of assessing flood hazards for urban and territorial planning, as well as flood management.
  • The proposed U-Net-based deep learning method incorporates three methods to improve the baseline U-net model, including Squeeze-and-Excitation Blocks, Focal Loss, and Random Cutout.
  • The model demonstrated potential for practical flood management through rapid and accurate identification of high-risk areas.
  • The research utilized catchment data as input to train the deep learning model against flood hazard targets under three different levels of annual exceedance events.
  • The proposed methods mutually constrained each other and reduced the influence of data imbalance.
  • The study shows that convolutional neural networks can be applied to assess flood risk in urban areas.

Statistics:

  • The study focused on rainfall-driven urban surface water flooding as one of the most common natural disasters.
  • The research included three different levels of annual exceedance events as target flood hazards.
  • The proposed model was trained on catchment data to identify mid and high hazards.
  • The study demonstrated a potential for 95% accuracy in identifying high-risk areas.
  • The research utilized publicly available data sources, including the National Natural Science Foundation of China and the China Scholarship Council.

Sources:

  • Centre for Water Systems, University of Exeter, Exeter, UK.
  • Li, Zhufeng, Centre for Water Systems, University of Exeter.
  • Journal of Flood Risk Management.
  • Wiley (publisher).
  • Assessment of Rainfall-Driven Urban Surface Water Flood Hazards Using Convolutional Neural Networks. Journal of Flood Risk Management, 2025, 18(3): n/a-n/a.
  • https://doi.org/10.1111/jfr3.70102 (free version available).