The atmospheric boundary layer flow downstream of a smooth-to-rough (S \(\rightarrow \) R) roughness transition is studied using large eddy simulations (LES). Simulating the flow over an S \(\rightarrow \) R transition requires overcoming two numerical challenges. First, the flow behind the transition needs space in the vertical direction for the outer boundary layer to develop without adversely affecting the flow near the surface. Vertical padding of the computational domain is employed for this purpose. Numerical experiments show that it is sufficient to set the padding height equal to the incoming boundary layer height. Second, an abrupt change in aerodynamic roughness length ( \(z_0\) ) leads to large oscillations in the flow statistics. To reduce these oscillations below an acceptable threshold, the S \(\rightarrow \) R transition needs to be smeared over a few grid points. Numerical experiments for a range of roughness ratios show that smearing the roughness transition such that \(\text {d}z_0/\text {d}x\) is 0.01 or less keeps the oscillations in the wall shear-stress under 5% of the overall jump across the S \(\rightarrow \) R transition. With these features incorporated, the LES are validated against two experimental datasets with roughness ratios 4.34 and 150. Wind engineering design studies require the use of computationally inexpensive analytical models. Seven internal boundary layer height analytical models are evaluated using the LES data and three of them are found to be reasonable over a range of roughness ratios. An analytical model for the turbulence intensity, previously developed for a rough-to-smooth transition, is found to predict the turbulence intensity reasonably for S \(\rightarrow \) R transitions as well.
