Direct Numerical Simulation of Entrainment in Dry Convective Boundary Layers

The widely employed 0.25/Ri growth rate law for the dry CBL is still not without controversy: results from atmospheric observations and laboratory experiments appear mutually inconsistent and display substantial scatter. Our goal was to end this controversy by conducting ground truth Direct Numerical Simulation (DNS) of convective boundary layers. DNS employs no empirical rules as it fully resolves the entire spatial spectrum of turbulence. Of course one cannot simulate the high Reynolds number of atmospheric turbulence, but present computer resources do allow one to faithfully simulate the classical laboratory experiments that gave rise to the existing growth rate laws for the ABL, and to even reach Reynolds numbers ten times higher than the classical experiments. By varying the Reynolds number over three orders of magnitude we aimed to get detailed insight into the impact of the Reynolds number on the ABL growth rate, so as to get insight in origin of the existing controversies.

The simulations were conducted within the DEISA (Distributed European Infrastructure for Supercomputing Applications) Extreme Computing Initiative framework with a resource allocation equivalent to 2.6 million CPU-h; this entailed simulations on five different european supercomputing platforms, including the Juelich Bluegene supercomputer. The largest simulation used 3072x3072x1536 gridpoints employing 32,768 cores in parallel. The simulations shed light on why different laboratory experiments, conducted in the past by various groups using different methods, gave different growth-laws. By mimicking these experimental conditions in our simulations, that is by accounting for the actual fluid properties that were used in the experiments, we could exactly simulate those historical experiments and get insight into how the fluid-properties (in particular its viscosity, conductivity/diffusivity) must have influenced previous findings on the boundary layer growth-rate. Finally, the results indicate which entrainment law is most appropriate for huge Reynolds numbers, i.e. for the case of atmospheric convection.