The principal roadblock to dynamic hydraulic modeling over regional-to-continental scale river networks is in obtaining adequate in-channel bathymetric data. Existing data are typically: (1) incomplete in spatial coverage, (2) stored in a variety of formats across different organizations, and (3) surveyed with different methods and precision. Conducting new comprehensive surveys to fill these gaps would be costly and, more importantly, might be unnecessary if more limited data is sufficient to approximate the bathymetry for hydraulic modeling. This study examines entropy maximization techniques for synthesizing submerged cross-section bathymetry based on visible bathymetry between bankfull and the water surface; i.e., data that can be obtained by remote sensing techniques. Previously surveyed cross-sections for river networks in four counties of Alabama (USA) are used to generate synthetic bathymetry. The relationships between entropy-produced synthetic cross-sections for flow depth, area, and hydraulic radius (key parameters for dynamic modeling) are compared to values for the true surveyed data (including submerged sections) as well as to synthetic bathymetry using trapezoidal, power law, and triangular cross-sections developed using established methods. Results show that all four methods introduce systematic biases in flood depth and conveyance predictions. Triangular and entropy methods underpredict channel conveyance by roughly 24% and 12% respectively, leading to conservative flood estimates, while power law and trapezoidal methods overpredict conveyance by roughly 14% and 30% respectively, which leads to underestimation of flood risk. The entropy method maintains the closes bankfull capacity to surveyed values and its bias remains stable regardless of channel visibility.
Cesar Davila-Hernandez, Ben R. Hodges, Ahmad A. Tavakoly