Large advances in computer resources have allowed researchers to simulate thunderstorms with very high resolution. For example, it is now possible to simulate deep moist convection (including an entire mesoscale convective system) with grid spacing on the order of 100 m. It is natural to ask whether such high resolution is necessary, and at what point simulations of convective processes are "well resolved".

Examination of the subgrid-scale turbulence parameterizations used in cloud models provides some guidance in the debate over adequate resolution. Although not widely acknowledged, the subgrid turbulence closures in most cloud models are taken directly from large eddy simulation (LES) models. Two assumptions inherent in LES are relevant to the issue of adequate resolution: 1) the grid spacing must be within the inertial subrange; and, 2) the scale of the phenomenon to be simulated must be several orders of magnitude larger than the grid spacing. Based on these assumptions, it is concluded that grid spacing on the order of 100 m is required for turbulence schemes used in cloud models to behave properly. In particular, this high resolution is necessary for the simulated flow to become turbulent.

Based on these arguments, we have conducted simulations of squall lines with grid spacings from 1 km to 125 m. In these simulations, specific details of the squall line are significantly modified as the resolution is increased. In particular, precipitation distribution and amount, phase speed, cloud depth, mesoscale flow patterns, and stability structure all change significantly. It is concluded that the differences in the solutions do not (necessarily) arise simply because thunderstorms are resolved better. Rather, as grid spacings approach 100 m, it becomes possible for the flow to become turbulent - and the development of the turbulence changes the physics of the convective process. For example, system-averaged vertical heat and moisture fluxes in the 125 m simulation are markedly lower than those in the 1 km simulation.

An analysis of velocity spectra from the model output confirms that the assumptions of LES are only satisfied in the highest resolution simulations. Therefore, these results infer that it may not be possible to accurately reproduce moist convective processes unless very high resolution is applied or unless new turbulence parameterizations are developed specifically for grid spacings of order 1 km.