Warm-season quantitative precipitation forecasting is well known for its vexingly low skill. The root cause is, at least partly, deficiencies in convective parameterization. Improved parameterization has high priority but has proven equally vexing. An obvious answer is the explicit approach: resolve the convection. But it will be a long time before cloud-resolving models can truly resolve all kinds of convection. At that juncture, the parameterization problem will have moved on to longer range prediction.

Here we examine the mesoscale organization of deep convection, a neglected process identified with a formidable gap in our knowledge of interaction between organized convection and its environment. Extensively studied as a process, mesoscale organization has received little attention in the parameterization community. A paradigm shift in thinking from ordinary convection needed to represent organized convection. By `ordinary convection’ we mean cumulus that interact weakly among themselves and with their environment. By 'organized convection' we mean the combined effect of fields of cumulus and convectively generated mesoscale circulations associated with sheared environments.

We simulate organized convection using non-hydrostatic numerical models having computational domains that span most of continental U.S. A representative active period (July 3 through July 9 2003) is chosen for study. The observed distribution of precipitation is from NEXRAD data collated by Carbone et al. Two sets of simulations are performed. One set applies the Betts-Miller convective parameterization at 10-km grid resolution. The other explicitly represents the mesoscale organization at 3-km grid-resolution, the highest resolution practicable in the large computational domain. At 10-km grid-resolution, Betts-Miller generates sequences of precipitation that stem partly from the parameterized convection and partly from grid-scale circulations. A method for parameterizing mesoscale organization, based on the dynamics of organized convection, is formulated and tested. Finally, the implications for representing mesoscale organization in next-generation prediction models are discussed.