To quantify the turbulent transport at gray zone length scales between 1 and 10 km, the Lagrangian evolution of the CONSTRAIN cold air outbreak case was simulated with seven large eddy models. The case is characterized by rather large latent and sensible heat fluxes and a rapid deepening rate of the boundary layer. In some models the entrainment velocity exceeds 4 cm/s. A significant fraction of this growth is attributed to a strong longwave radiative cooling of the inversion layer. The evolution and the timing of the breakup of the stratocumulus cloud deck differ significantly among the models. Sensitivity experiments demonstrate that a decrease in the prescribed cloud droplet number concentration and the inclusion of ice microphysics both act to speed up the thinning of the stratocumulus by enhancing the production of precipitation. In all models the formation of mesoscale fluctuations is clearly evident in the cloud fields and also in the horizontal wind velocity. Resolved vertical fluxes remain important for scales up to 10 km. The simulation results show that the resolved vertical velocity variance gradually diminishes with a coarsening of the horizontal mesh, but the total vertical fluxes of heat, moisture, and momentum are only weakly affected. This is a promising result as it demonstrates the potential use of a mesh size-dependent turbulent length scale for convective boundary layers at gray zone model resolutions.

Cold air outbreaks can bring snow to populated areas and can affect aviation safety. Shortcomings in the representation of these phenomena in global and regional models are thought to be associated with large systematic cloud-related radiative flux errors across many models. In this study, nine regional models have been used to simulate a cold air outbreak case at a range of grid spacings (1–16 km) with convection represented explicitly or by a parametrization. Overall, there is more spread between model results for the simulations in which convection is parametrized when compared to simulations in which convection is represented explicitly. The quality of the simulations of both the stratocumulus and the convective regions of the domain are assessed with observational comparisons 24 h into the simulation. The stratocumulus region is not well reproduced by the models, which tend to predict open cell convection with increasing resolution rather than stratocumulus. For the convective region the model spread reduces with increased resolution and there is some improvement in comparison to observations. Comparing models that have the same physical parametrizations or dynamical core suggest that both are important for accurately reproducing this case.