Numerical Modeling of Fluvial Urban Floods: Implications for Flood Mitigation Strategies

Abstract

Fluvial urban floods, also known as riverine urban floods, occur when water levels in urban streams or rivers rise rapidly due to heavy rainfall, snowmelt, or dam releases. Climate change significantly affects the severity, frequency, and predictability of floods, presenting new challenges to the relevant studies. In this thesis, numerical modeling is used to investigate the morphodynamic and inundation processes of fluvial urban floods and to explore engineering solutions to the associated problems. The first study employed Delft-3D to develop a numerical model for examining the morphodynamic processes of the 2013 Bow River flood in Calgary, Canada. The model was calibrated using velocimetry data and validated with post-flood bathymetry data. The temporal and spatial distributions of the modeled flow and morphodynamic data were presented and analyzed. Results indicate that the timing of morphological changes during the flood varies among different morphodynamic units (MUs) but remains consistent within similar MUs. It was demonstrated that, given the same flood peak and duration, a regulated flood event with a brief rising period, as opposed to a prolonged rising period, might result in reduced bank erosion and bar growth. Additionally, bedload transport rates were found to be more sensitive to flow velocities than to bed sediment sizes in the Bow River case, due to the greater spatial and temporal variation of velocities during the flood. Another issue arising from the 2013 Bow River flood was the flood-induced bar growth, which constricted the river channel and increased future flood risk. The second study focused on exploring the optimal bar management solution for the Bow River. Using the developed morphodynamic model, we compared the effectiveness of a traditional bar removal plan with a novel bar realignment plan. Results indicate that while appropriate bar realignment can protect aquatic habitats and provide river recreation opportunities, bar removal is more effective in reducing future flood peak levels. The findings also suggest that manipulating instream bars has minimal morphological impact on downstream reaches. This study also highlights that creating a less obstructed channel is a fundamental strategy for flood mitigation. The third study emerged from surface image velocimetry analysis of the flow field in a large-scale physical model of flood mitigation strategies for the Bow River. We developed a new post-processing algorithm called Time Frequency Analysis (TiFA) to address challenges in Large Scale Particle Image Velocimetry (LSPIV) under unfavorable tracer conditions. TiFA involves three steps: (1) plotting the temporal frequency distribution of PIV-recognized velocities at a specific location; (2) fitting a bimodal Gaussian distribution model to the plot to identify the “most likely” velocity at that location; and (3) repeating these steps at all locations to generate a spatially distributed velocity map. We evaluated the performance of TiFA, the traditional temporal-averaging method, and the ensemble correlation method using the scaled physical model surface imagery. Results showed that TiFA produced lower errors compared to the temporal-averaging method and was at least 40% faster than the ensemble correlation method, demonstrating the great potential of TiFA in LSPIV post-processing. While the flow characteristics of open channel confluences have been extensively studied, the inundation dynamics of urban confluence floods remain unexplored. The fourth study aims to fill this research gap by investigating a 100-year flood event at the Ottawa-Gatineau (OG) confluence in Canada, utilizing in-situ measurements, remote sensing, and a two-dimensional (2D) hydrodynamic model. A flow rating curve was first developed at the confluence outlet, showing that the total discharge and water level follow a power-law function, with minimal influence from confluence discharge ratio. The developed rating curve also showed a distinct segmentation behavior, where water level increases faster with total flow discharge in the overbank flow stage than that in the in-channel flow stage, probably due to the combined effects of the change of roughness and friction slope. Then, we employed MIKE+ to develop a two dimensional (2D) unsteady hydrodynamic model to study the confluence flood dynamics. The developed 2D model reproduced well the measured inland floodwater velocity and urban flood inundation extent, demonstrating its reliability in simulating large-scale urban confluence floods. Model results show that confluence flood inundation extent is mainly a function of total flow discharge, with minimal impact from discharge ratio or flow unsteadiness. Therefore, using steady-state models may be appropriate in future modeling of urban confluence floods.