These appear as irregularly shaped patches of load across the map that can be seen to move in a weather-like fashion as the simulation progresses. The imbalance contributed from within the CONVAD subtree is mostly of this nature (Figure 3). In the CONVAD subtree we notice that the pattern for CONVAD is largely similar to that of CMFMCA (mass flux convective parameterization). The mean computation time in this routine is 2.3 milliseconds, and its contribution to total load imbalance is 2.6 mill-seconds. Thus we note that most of the imbalance in this subtree is caused by mass flux convective parameterization processes. This is understandable because the routine CMFMCA is invoked only when the atmosphere is unstable to moist convection and not all regions have this instability. Load tends to be higher closer to the equator, in the inter-tropical convergence zone, where there is more moist convective activity.
The mass flux convective parameterization is called for every time step in the model, and the characteristics of load do not vary in this routine over the three types of CCM2 time step.