The requirements for a credible large-eddy simulation of neutral, turbulent
flow over hills with an aerodynamically rough surface are discussed, in or
der to select a suitable case for simulation. As well as providing adequate
resolution within the dynamically important inner region, obtaining a real
istic upstream or undisturbed mean velocity profile is also of critical imp
ortance. A distributed drag canopy formulation has been introduced to the m
odel to allow it to obtain such a profile while resolving very close to the
rough surface. Simulations have then been performed of flow over ridges of
varying heights. The results from the steepest case, which is just on the
verge of separation, are compared with wind-tunnel observations. It is show
n that the large-eddy simulation results are in much better agreement with
the experimental data than are the results from a simple first-order mixing
-length closure model. An encouraging lack of sensitivity of the simulation
results to numerical resolution is also demonstrated.