The validity of convective parameterization is dependent on the existence of a scale-separation between the sub-grid and resolved systems. At mesoscale resolution, however, the scale of individual convective clouds is close to the grid-scale. Most clouds are sub-grid but errors incurred through a convective parameterization scheme may result in a more useful forecast being one without a sub-grid parameterization. Up-scale growth of convection across the grid-scale associated with mesoscale organisation increases uncertainty for the decision to parameterize and how the scheme should be formulated. It is possible that these mesoscale circulations will be represented explicitly to some extent by a mesoscale model.

Through simulations of two Mesoscale Convective Systems (MCSs) over the UK this presentation addresses the successes and limitations of approaches to convective parameterization at mesoscale resolution. The two MCSs are found to differ in their predictability of triggering through an ensemble of simulations by perturbing the levels of convective initiation. Predictability refers here to the ability of the model to capture the trigger mechanism. Simulations are presented of an MCS with predictable initiation and an MCS with unpredictable initiation. Contrasting cases allow deeper analysis of approaches to parameterization.

Performance of the Gregory-Rowntree scheme is compared with fully explicit convection using the Met Office hydrostatic mesoscale model at 12km resolution. Size, intensity and propagation mechanisms of model solutions to the large-scale forcings differ significantly between the experiments and from observations. Realistic effects of convection on the large-scale dynamics can be essential for an accurate forecast of downstream development. Upper-level jet maxima in the fully explicit simulation are found to be double that in the Gregory-Rowntree simulation. Results suggest representations of convection other than the Gregory-Rowntree scheme provide more useful information with increased benefit to the forecaster. Results motivate the use of the Kain-Fritsch scheme which is currently being implemented into the model.