Experience with models and schemes for representing the effects of the unresolved motions in large-eddy simulations (LES) suggests that the near-wall and main flow domains require quite different treatments. Virtually all schemes make some special provisions for the near-wall representations. To facilitate our understanding and to improve turbulence models in the near wall region, it is useful to consider velocity partitioning schemes such as those of Carati, et al. (2001), Zhou et al. (2001), and Hughes et al. (2001). Following Carati et al., this velocity partitioning results in a decomposition for the LES stress tensor (tau): the subgrid-scale stress portion (tau_A) depends on scales beyond the resolution domain of the LES, while the filtered-scale stress portion (tau_B) depends on the differences between the exact and filtered velocity fields within the resolution domain. This partitioning requires that the filter width be larger than the grid spacing. Near the wall, the tau_A terms, which must be modeled, become increasingly important.

Our goal was to learn what logical steps and procedures are required for subfilter-scale modeling to bring the simulated flow fields into agreement with theoretical expectations in the near-wall region. We examined a specific test case: the neutral, rotation-influenced, rough-wall, field-scale boundary-layer flow considered by Andren, et al. (1994). A standard atmospheric mesoscale simulation code is used, namely, the Advanced Regional Prediction System (ARPS). Because this is a finite volume LES code for irregular terrain, spectral methods and sharp Fourier cutoffs in filters are not viable options. The only major modifications to the code are those associated with the new subfilter-scale models we have implemented.

For the stress term tau_B, which can be expressed in terms of the resolved velocity, we have implemented the series model of Katopodes, et al. (2000), a task for which it is ideally suited. For tau_A, we have examined multiscale Smagorinsky models [cf., the work of Hughes et al. (2001)] and the effect of the 4th-order numerical smoothing employed in ARPS to remove high-frequency noise and aliasing. To better represent the near-wall region, we have adopted a hybrid approach. The series and Smagorinksy models are used in conjunction with the near-wall canopy stress term of Brown, et al. (2001). Because the vertical grid-spacings are invariably smaller than the horizontal ones, 2*dx is the minimum vertical distance from the wall for eddies of the horizontal grid size to be formed. Thus, in this near-wall region, such augmentation of the stress models is appropriate.

When the series, canopy, and Smagorinsky models are used together, the overshoot in the velocity shear near walls observed when the Smagorinsky model is used alone (represented by the nondimensional shear parameter Phi being greater than unity) is compensated by the designed tendency for the canopy model to produce Phi values less than unity. Our conclusion is that the context provided by Carati, et al. (2001), is useful and leads to insights about model behavior. We are able to achieve improved Phi profiles by a systematic use of the models cited above.