Detecting and Validating a Global Levee Dataset From High-Resolution Dem for Large-Scale Flood Modeling

Key Points

• The aim is to improve flood modeling accuracy by extracting a global levee dataset using high-resolution DEM.
• Collected LiDAR DEM data from 33 countries.
• Extracted approximately 100,000 km of levees based on geometric indicators.
• Performed accuracy evaluation by comparing with measured data.
• Incorporated extracted levee data into flood models analyzing 30 past floods.
• Achieved an F1 score greater than 0.5 for well-maintained levees.
• Lower accuracy for narrow levees and meandering rivers.
• Demonstrated a 37% reduction in flooding during the 2003 Rhine flood due to levee incorporation.
• Established a positive correlation between extraction accuracy and flood area improvement rate.