Large-Eddy Simulation: How Large is Large Enough?

Abstract

The length scale evolution of various quantities in a clear convective boundary layer (CBL), a stratocumulus-topped boundary layer, and three radiatively cooled (?smoke cloud?) convective boundary layers are studied by means of large-eddy simulations on a large horizontal domain (25.6 ? 25.6 km2). In the CBL the virtual potential temperature and the vertical velocity fields are dominated by horizontal scales on the order of the boundary layer depth. In contrast, the potential temperature and the specific humidity fields become gradually dominated by mesoscale fluctuations. However, at the mesoscales their effects on the virtual potential temperature fluctuations nearly compensate. It is found that mesoscale fluctuations are negligibly small only for conserved variables that have an entrainment to surface flux ratio close to ?0.25, which is about the flux ratio for the buoyancy. In the CBL the moisture and potential temperature flux ratios can have values that significantly deviate from this number. The geometry of the buoyancy flux was manipulated by cooling the clear convective boundary layer from the top, in addition to a positive buoyancy flux at the surface. For these radiatively cooled cases it is found that both the vertical velocity as well as the virtual potential temperature spectra tend to broaden. The role of the buoyancy flux in their respective prognostic variance equations is discussed. It is argued that in the upper part of the clear CBL, where the mean vertical stratification is stable, vertical velocity variance and virtual potential temperature variance cannot be produced simultaneously. For the stratocumulus case, in which latent heat release effects in the cloud layer play an important role in its dynamics, the field of any quantity, except for the vertical velocity, becomes dominated by mesoscale fluctuations. In general, the location of the spectral peak of any quantity becoming constrained by the domain size should be avoided. The answer to the question of how large the LES horizontal domain size should be in order to include mesoscale fluctuations will, on the one hand, depend on the type of convection to be simulated and the kind of physical question one aims to address, and, on the other hand, the time duration of the simulation. Only if one aims to study the dynamics of a dry CBL that excludes moisture, a rather small domain size suffices. In case one aims to examine either the spatial evolution of the fields of any arbitrary conserved scalar in the CBL, or any quantity in stratocumulus clouds except for the vertical velocity, a larger domain size that allows the development of mesoscale fluctuations will be necessary.