Comparison of numerical model and laboratory simulator tornado wind fields with radar observations of the Spencer, South Dakota tornado

Radar observations taken around the time an F4 tornado passed through Spencer, South Dakota on May 30 1998 are compared with numerical simulations performed with the Fluent CFD model and measurements taken within a laboratory simulator that allows a large translating vortex. The numerical simulations use the radar data outside the tornado as boundary condition input. The radar observations, consisting of the output of an axisymmetric model driven by raw data collected from the Doppler-on-Wheels radar, are used as a benchmark for comparison with the flow characteristics within the laboratory simulator. Sensitivity tests are performed with the numerical model to better understand the impact of some assumed parameters in the design of the model domain, and also the impact of some inflow characteristics on the vortex structure.

Radar observations show peak flow occurring at the lowest sampled level, around 20 m above ground with a relatively narrow core radius of around 100 m in the lowest 65 m and a sudden broadening of the vortex to roughly 200 m at higher elevations. Observations in the laboratory simulator were taken using a spherical 18-hole pressure probe. The lab simulator vortex also has stronger velocities closer to the ground with a narrower core radius there, similar to the Doppler data but with less dramatic changes. Differences between the lab data and radar data could reflectdifferences in surface roughness and Swirl ratio. Numerical simulations also show strongest flow at levels very near the ground (20 m) and a gradual decrease in core radius toward the surface. Data normalized by length and velocity scales depicts greatest tangential velocities in some of the numerical simulations and weakest in the laboratory simulations. Profiles of wind components suggest better agreement between the lab results and the radar observations than between the numerical results and radar observations. The numerical results are found to be somewhat sensitive to a variety of assumed parameters such as the radius of the simulation domain, depth of inflow layer, mesh size, and height at which the upper-level outflow occurs.