Thick sea ice on the polar oceans acts as a good insulator between the cold atmosphere and the relatively warm water, especially during late winter and early spring, when air temperatures are low. A turbulent exchange of energy at the atmosphere-ocean interface is then almost limited to regions with polynyas or leads. The latter represent elongated channels in sea ice, either ice-free or covered by thin, new ice. Large temperature differences between the surface of leads and the near-surface atmosphere cause strong turbulent convection (convective plumes) and enhanced transports of heat, moisture, and momentum. These rather small-scale processes can have a profound impact on the characteristics of the lower atmosphere on different spatial scales. In regional climate or numerical weather prediction models, leads with potentially different characteristics are not resolved with the typical grid sizes of such models. Open water is accounted for only by a continuous open-water fraction rather than by separated leads as in nature. Thus, neither leads, nor related convective plumes, nor a changing lead configuration (pattern) from one grid cell to the next one can be resolved. We examined potential implications for such models by not resolving these subgrid-scale patterns, which in nature are inhomogeneously distributed over an area of a typical grid cell. To study such drawbacks, we used a model with a grid size to appropriately resolve convection over individual leads but not the turbulent eddies as in large eddy simulation (LES) models. Nevertheless, its turbulence parametrization was adjusted to results of LES applied to scenarios of convection over leads. We considered simulations of the flow over the same domain but with six different lead patterns. In five domains, we prescribed leads of different widths separated by 100% sea ice on a microscale grid. The sixth domain was considered as representative for a few grid cells of a regional climate model with much coarser grid spacing. Its grid cells consisted of a fractional sea ice cover of 91% to account for the leads indirectly. However, averaged over each domain, the mean sea ice concentration was the same in all simulations. We found considerable differences in domain-averaged vertical profiles of wind, temperature, and turbulent fluxes. This points to a strong impact of both the leads and their geometry on atmospheric patterns and to a relevance of the treatment of leads in regional climate modelling.