CFD Online Discussion Forums - atmospheric roughness height

I've copied the following from www.adapco-online.com:

------------------------------------------------------- "The models for the RANS equation typically break down near the walls, where there is a very thin boundary layer with large velocity gradients. At high Re, this layer is so thin that it is difficult to get a sufficient number of fluid cells into this zone. Even if this is done, the forms of the k-e equation suitable for the interior do not work very well at the wall, as the flow transitions to a laminar sublayer.

STAR-CD offers three alternatives to deal with this situation. The usual procedure is to use an algebraic "Law of the wall" model in the first cell at the boundary. The boundary layer is assumed to be contained within this cell, and the shear stress at the wall, and the boundary conditions for the K and e equations are computed from the log law. For this to be applicable, the cell next to the wall must be big enough; the value of Y+ at the cell centroid should be greatar than 30. This is rarely a problem, except in the neighbourhood of separation points and reattachment regions. It can be checked easily by plotting the Y+ at the boundary in PROSTAR, after the solution.

At the next level of complexity are the two layer models. In these models, a low Reynolds number representation of the turbulence equation is used in the viscosity influenced near wall region. Since transport equations must be solved in this region, there must be a sufficient grid density in the direction normal to the wall. Typically about 10 cells are placed in the region of Y+ < 30. This usually results in a large cell count, increasing the computational requirements. There is also a (one time) penality where STAR-CD has the determine the wall distances of the viscous layer cells.

In these two layer models, the flow in the wall layer is treated by one of two ways; a one equation model where near-wall flow is simulated by a low Reynolds number model consisting of a transport equation for k and an algebraic prescription for the turbulence length scale; or a mixing length model where an algebraic expression is used for the turbulent viscosity. In either case, the solution is matched to that of the standard k- e equations at the edge of the viscosity-influenced region.

The third alternative for dealing with wall effects is a Low Reynolds Number K-e model. The meshing requirements near the wall are similar to the two layer models. This model uses the standard transport equation for K, but augments the equation for e with an extra term which depends on the distance from the nearest wall. Thus in principle, this model is valid all the way down to the solid wall. The startup computational cost is higher, because the wall distances need to be computed for all the cells." -------------------------------------------------------

With the Hi-Re models, you'll need to adjust the near-wall cell layer (to be within the recommended 30-100 Y+ value) before you get a nice answer. Hi-Re turb models will usually give bad results when accurate calculation of the near-wall flow is required, eg. conjugate heat transfer, which needs an accurate velocity profile to in turn get a good thermal b/l. ICEM even has a commercial product called 'Y+' that is supposed to automatically adjust your near wall cells according to your previous sim results.

OR: you can just TREAT it as if you were calculating for a Low-Re flow with the Low-Re k-e model. Problem now is that you've got to mesh it really fine at the walls. Doesn't say anywhere that it is wrong to use it if your (mean) flow Reynolds number is in the turbulent regime (in fact, they are the same, except for the a slight mod in the e equation).

You could try the two-layer models too.