Simulation of subtropical marine stratocumulus clouds in convection-resolving models

Even though the complexity and resolution of global climate models (GCMs) has increased over the last decades, the inter-model spread of equilibrium climate sensitivity has not narrowed. The representation of subtropical low-level clouds and their associated radiative feedbacks in climate models still poses a major challenge. A fundamental problem underlying the simulation of such clouds is their multiscale nature. On the one hand, current GCMs allow to capture the large-scale processes but are too coarse to represent the mesoscale and microscale dynamical processes governing their formation and dissipation. On the other hand, large eddy simulations (LES) resolving the micro scale are bound to small domains and thus lack a robust representation of the large-scale flow and the mesoscale organisation of the clouds. Convection-resolving models (CRMs) are an attractive compromise between the former two since they allow for simulations at much higher resolution than in conventional GCMs and on larger domains than in LES.

Here we analyse how CRMs simulate stratocumulus decks and investigate causes for inter-model differences. We consider a set of ten CRMs (nine GCMs that are run at convection-resolving resolution during a short time period as part of the pioneering DYAMOND initiative, and the limited area model COSMO run by ourselves) used to simulate stratocumulus clouds over the South-East Atlantic during a 40 day period. The simulations cover a range of horizontal grid spacings between 5 and 1 km.

We find pronounced differences in the mean cloud cover among the analysed CRMs. In comparison to observed radiation (CERES), most of them underestimate cloud cover, in particular the low-lying stratus decks close to the African coast. Nevertheless, the simulated mesoscale cloud organisation is realistic and similar in the set of CRMs, with few exceptions showing organisation on larger scales than in the other models. In general, the simulated cloud field appears to be more sensitive to the model choice than to the horizontal resolution.

Despite the differences in the cloud cover, most models capture the subtropical inversion and its spatial structure relatively well. Therefore, differences in the inversion strengths do not suffice to explain variability in the simulated cloud cover fraction between models. However, we find a relation between the mean height of the stratocumulus layer (or inversion layer) and its cloud cover fraction: Models with higher inversions tend to simulate a higher cloud cover fraction, bringing them closer to observations. Similarly, stronger vertical mixing within the boundary layer and enhanced surface latent heat fluxes appear to be related to higher cloud cover. Such relations may help to determine the physical processes responsible for the differences among CRMs in the simulated stratocumulus field.