Evaluation of multiple hydraulic models in generating design/near-real time flood inundation extents under various geophysical settings

Application of land surface/hydrologic models within an operational flood forecasting system can provide probable time of occurrence and magnitude of streamflow at specific locations along a stream. Creating time-varying spatial extent of flood inundation and depth requires the use of a hydraulic or hydrodynamic model. Models differ in representing river geometry and surface roughness which can lead to different output depending on the particular model being used. The result from a single hydraulic model provides just one possible realization of the flood extent without capturing the uncertainty associated with the input or the model parameters. The objective of this study is to compare multiple hydraulic models toward generating ensemble flood inundation extents. Specifically, relative performances of four hydraulic models, including AutoRoute, HEC-RAS, HEC-RAS 2D, and LISFLOOD are evaluated under different geophysical conditions in several locations across the United States. By using streamflow output from the same hydrologic model (SWAT in this case), hydraulic simulations are conducted for three configurations: (i) hindcasting mode by using past observed weather data at daily time scale in which models are being calibrated against USGS streamflow observations, (ii) validation mode using near real-time weather data at sub-daily time scale, and (iii) design mode with extreme streamflow data having specific return periods. Model generated inundation maps for observed flood events both from hindcasting and validation modes are compared with remotely sensed images, whereas the design mode outcomes are compared with corresponding FEMA generated flood hazard maps. The comparisons presented here will give insights on probable model-specific nature of biases and their relative advantages/disadvantages as components of an operational flood forecasting system.