Rapid simulation for real-time flood depth prediction using support vector machine

Area of Science:

• Environmental science
• Hydrology
• Urban planning

Background:

• Climate change intensifies Local Intensive Precipitation (LIP), leading to severe urban flooding.
• Traditional hydrodynamic models (SWMM, FLO-2D) are accurate but computationally intensive, limiting real-time flood prediction.
• Urban areas like Gangnam, Seoul, face significant flood risks.

Purpose of the Study:

• To develop a rapid flood depth prediction model for urban areas.
• To integrate machine learning with physical simulations for enhanced flood forecasting.
• To support timely disaster response in flood-prone urban environments.

Main Methods:

• A Support Vector Machine (SVM) model was developed for rapid flood depth prediction.
• The SVM model was trained using data generated from a 1D-2D coupled hydrodynamic simulation (SWMM-FLO-2D).
• Input variables included cumulative rainfall and manhole overflow data over 1 to 5-hour scenarios.

Main Results:

• The integrated SVM model demonstrated high performance with R² = 0.988, NSE = 0.987, % difference = 1.080, and RMSE = 0.098 m.
• The 1D-2D hydrodynamic model (SWMM-FLO-2D) was validated against observed flood records with a 64% match.
• The SVM model accurately predicted flood depths when compared to FLO-2D simulation results.

Conclusions:

• Integrating machine learning with physical simulations offers a fast and reliable approach to flood prediction.
• The developed SVM model can significantly aid in real-time urban flood risk management.
• This approach enhances the efficiency of disaster response systems in urban areas facing climate change impacts.