On the method of superparameterization and the coupling between large- and small-scales

The method of superparameterization (SP) has emerged as a promising approach for representing the effects of unresolved convective-scale processes in coarse-resolution dynamical solvers. Despite this promise, however, little attention has been given to fundamental questions surrounding the SP approach, such as how the predictions of such models depends on the choice of the time interval for coupling between the large- and small-scale model components. Here it is shown that changes to this coupling time interval have a substantial impact on the simulated large-scale organization of convection, which in turn affects the strength of the simulated large-scale cloud forcing. The reason for this sensitivity is shown to stem from the neglect of the finite response time of convective cloud fields to changes in their large-scale environment (due to the well-known process of gravity wave adjustment). Based on comparisons against high-resolution benchmark calculations, it is shown that the optimal coupling interval is around 20 minutes, regardless of the choice of the large-scale model grid length.