The temporal and spatial characteristics of deep tropical convection is currently a topic of great interest. There is a general feeling (Ricciardouli and Garcia, 2000; Amodei et al. 2001) that current parameterizations of deep cumulus convection in GCMs produce too little temporal variability. In particular, cloud-based mass flux (CBMF) schemes, which tend to produce very reasonable seasonal mean distributions of tropical precipitation, are associated with the least amount of variability. This is an extremely important modelling issue since the variability of latent heating associated with deep convection in the tropics has a hand in shaping the intraseasonal oscillation, and forcing tropical waves which play a role in the quasi-biennial, and semi-annual, oscillations.

In this study, the properties of tropical precipitation in the Canadian Centre for Climate Modelling and Analaysis (CCCma) third generation GCM (GCM3) are investigated. GCM3 employs the CBMF scheme of Zhang and McFarlane (1995; ZM) for the parameterization of deep cumulus convection. Here it is found that the temporal variability of the precipitation produced by the ZM scheme depends strongly on several of its internal parameters and on the type of closure condition employed. The analysis indicates that the CBMF scheme of ZM can match the level of "observed" variability (Horinouchi, 2002). An unexpected result of the present study is that the parameterized deep convection and resolved stratiform precipitation in the tropics experience a strong reciprocal interaction. That is, a change that increases the variability (and mean value) of one, decreases the variability (and mean value) of the other. This turns out to have a strong influence on the model biase of mean precipitation in the tropics.