Mesoscale circulations induced by differential boundary layer heating due to surface inhomogeneities on scales of 5 km and more can significantly change the average properties and the structure of the convective boundary layer. This numerical case study reveals for the first time that the onset of a thermally induced mesoscale circulation (TMC) can induce a temporal oscillation of the boundary layer flow, which significantly changes both average statistical properties and mean profiles. TMCs exhibit a considerably larger average kinetic energy than convection under homogeneous conditions.

The case study uses the Hannover Large-Eddy Simulation model PALM with prescribed 1D sinusoidal surface heat flux variations on wavelengths from 2.5 to 40 km. The resulting TMCs are analyzed by means of domain-averaged cross-sections, time averaged and normalized with the boundary layer height, as well as domain-averaged time series.

The simulated TMCs were periodic. Vertical profiles and time series demonstrate that the TMC onset triggers off a temporal boundary layer oscillation whose period and amplitude depend on the surface heat flux perturbation wavelength and amplitude and on the background wind component perpendicular to the orientation of the surface inhomogeneity.

The TMC itself accounts responsible for the oscillation mechanism, for it effectively reduces the horizontal temperature and pressure gradients that were responsible for its onset. Its forcing being reduced, the TMC itself then also decreases in strength. Meanwhile, however, the differential boundary layer heating continues, and it restores the TMC.

The results of this study might call into question those turbulence parameterizations employed by general circulation and other large-scale models that rely entirely on homogeneous control runs of high-resolution models. The domain averaged value of the sinusoidal surface heat flux used in this study exactly equals that of a completely homogeneous model run – but the kinetic energy, for instance, on which many sub-gridscale parameterizations are based, deviates considerably.