Considerations for Stochastic Convective Parameterization

Abstract

Convective parameterizations in general circulation models (GCMs) generally only aim to simulate the mean or first-order moment of convection; higher moments associated with subgrid variability are not explicitly considered. In this study, an empirically based stochastic convective parameterization is developed that uses an assumed mixed lognormal distribution of rainfall, tuned with parameter values derived from observations, to control selected nonmean statistical properties of convection. Testing of this stochastic convective parameterization reveals that large-scale model dynamics interacts heavily with the convective parameterization, in ways such that the resulting output is fundamentally different from the input. This suggests stochastic parameterizations cannot be calibrated outside of a model's dynamical framework. Implications are discussed for the relative merits of the empirical approach versus another approach that introduces the stochastic process within the framework of the convective parameterization. Inclusion of the variance arising from unresolved scales by stochastic parameterization of convection is found to have a substantial impact upon atmospheric variability in the Tropics, including intraseasonal and longer timescales.