Effects of sub-filter scale turbulence models on vertical velocity structure in neutral boundary-layer flow

This work reports on variations in the flow-field fluctuations observed in large-eddy simulations (LES) with different subfilter-scale turbulence models. The Advanced Regional Prediction System (ARPS) mesoscale model was used in an LES configuration to simulate neutrally-stratified boundary-layer flow over a flat, rough surface using six different sub-filter scale models (Smagorinsky, dynamic Wong-Lilly, and a dynamic reconstruction model with various types and levels of reconstruction). Previous work indicated that the dynamic reconstruction model (DRM) gave significant improvements over traditional eddy-viscosity closure models (such as the Smagorinsky model) in comparisons of mean velocity profiles to the logarithmic profile predicted by similarity theory. Comparisons of vertical shear profiles as well as turbulent stress profiles also indicated significant improvement using the DRM over traditional models (Chow et al. 2005).

While previous results focused on profiles of mean quantities, this work presents statistical distributions and snapshots of horizontal variations in flow-field fluctuations to give a detailed description of flow structures and how they are affected by different turbulence models. Specifically, snapshots of the distributions of wind speed, vertical motion, and spanwise vorticity are analyzed at 2500 s intervals from 280,000 to 300,000 s (simulated time). Horizontal layer means of speed and spanwise vorticity differ among the models by small, but usually statistically significant amounts. The differences in the layer standard deviations for the three variables differ from model to model, by amounts that are almost always statistically significant. When horizontal distributions of spanwise vorticity and vertical velocity are plotted, there are pronounced differences in the near surface (below about 15 m) patterns. The Smagorinsky patterns are characterized by extended structures (about 5 times as long as wide) that, in the case of vertical motion, could be interpreted as vortices aligned at about 10° clockwise from the westerly forcing wind direction. Most of the other models produced patterns that displayed a mix of extended and more compact and symmetric features. The differences among models in vorticity and vertical velocity structures diminish with height above the ground. However even near the top of the model domain, differences in layer means and standard deviations between the different turbulence models are statistically significant in the majority of cases. This is contrary to the popular belief that turbulence models affect only the near-surface flow; rather, the large differences near the surface are felt throughout the flow domain. Our findings support the use of the turbulence model that compares best with theoretical profiles and higher resolution results, because the turbulence model can have a significant impact on transport of momentum and atmospheric scalars throughout the boundary layer.