The Large Eddy Simulation (LES) regime of boundary-layer turbulence modelling is expected to be reasonably independent of sub-grid model choices. In contrast, numerical weather predictions at 1 km grid length are in the grey zone of the convective boundary layer where the sub-grid model and other sources of diffusion play a much larger role. Also, the expectation at the limit of the grid length ( \(\varDelta x\) ) being the same order as the boundary-layer depth (h) is that turbulence should collapse due to being no longer resolved. The rate of transition from resolved to unresolved flow with increasing grid length is a key question in grey-zone research. In this paper we investigate the role of sub-grid diffusion magnitude on the simulation of turbulence at grid lengths ranging from LES to the grey zone, focusing on three key features. For the LES, and at high wavenumbers, we look at the fall-off of the spectrum with respect to the inertial sub-range. In the grey zone, we examine the collapse of turbulence at \(\varDelta x \sim h\) and the ability to spin up turbulence. Our methodology is to compare a numerical weather prediction model (the Met Office Unified Model) with an established LES model with a completely different dynamical core (MONC) but both using a Smagorinsky sub-grid model. We use an established convective boundary layer case. Both models require an enhancement of diffusion to give a collapse of turbulence at \(\varDelta x \sim h\) . In contrast, the spin up of turbulence within the grey zone ideally requires a reduced diffusion. This indicates that grey-zone diffusion levels are regime dependent. We propose a classification of the useable grey zone in terms of the spectrum. When there is a discernable peak and inertial subrange, we argue there is useful information in the simulation and the turbulence should not be suppressed.
