2D vs. 3D Storm Surge Sensitivity in ADCIRC: Case Study of Hurricane Isabel

Differences in ADCIRC 2D depth-integrated and ADCIRC 3D results are explored with particular attention to strong, wind forced applications. The 3D model results are sensitive to the turbulent vertical eddy viscosity (EV) formulation and bottom stress formulation. At high values of EV, the 3D velocity profile becomes uniform with depth and the simulation results match the 2D results. EV values typically range from 0.05 to 1.0 m²/s, with higher values in regions of wave breaking. Sensitivity to EV is tested using steady state, spatially uniform forcing and a domain of uniform along shelf depth. Maximum water elevations from the 3D model can differ from those predicted by the 2D model in a range of 0 percent for high values of constant EV to 9 percent for EV computed using Mellor-Yamada 2.5. We expect that the Mellor-Yamada parameterization may underestimate the vertical mixing during strong wind events such as tropical cyclones due to the effects of breaking wave induced turbulence, and therefore, that the actual differences will be substantially less than 9% during these events. The bottom roughness parameter, z₀, governs the 3D bottom drag coefficient which determines the bottom stress response, and can range from 0.10 m to 0.0001 m. We present results from a series of 3D simulations of Hurricane Isabel using the Mellor-Yamada 2.5 turbulence closure model and constant bottom roughness for a range of roughness values. These results are compared against 2D simulations of Hurricane Isabel using a Manning's n type bottom friction. Our results show that the differences in maximum elevations between the 3D and 2D results range from ±5% over most of the domain to ±20% in localized regions of inundation. There is no indication that, with the current state of the science as applied to storm surge modeling, the 2D and 3D results differ beyond the levels of intrinsic error in specifying the vertical mixing or bottom friction parameterization that currently exists for each method. We show that the differences in maximum elevation of storm surge between a 2D and 3D model is probably on the order of ±5%.