Throughout this section, the discussion will be motivated by the difference in MJO activity between the 19 and 30 level GCM experiments shown in fig. 2. We will attempt to explain this difference with reference to the behaviour of the convection scheme in the aqua-planet integrations.
As shown in section 3.4, the convective clouds at L19 and L30 over the warmest SST behave quite differently. In L19, convective clouds are almost always deep cumulonimbi, extending into the upper troposphere, whereas in L30 there are periods when the convective clouds only extend up to the mid-troposphere, with tops around the more stable layer that periodically develops at the melting level. The convection scheme assumes that solid precipitation melts as it falls from a model layer where the temperature is below 0 
C to a layer where the temperature is above 0 C. This melting is achieved at the expense of removal of heat from the atmosphere in that model level, thus cooling the layer. This results in an increase in stability at that level in the atmosphere. In L19, the layer which is cooled is 100 hPa thick so the reduction in temperature and resulting change in stability are quite small. In L30, the layer which is cooled is only 50 hPa thick so the cooling and increase in stability are proportionally larger. This could explain the periods of greater stability in L30 shown in fig. 12. The increase in stability in L30 then results in more convective plumes terminating in mid-troposphere, giving periods when the convective clouds are mainly cumulus congestus clouds as indicated in fig. 11 and by the peak at 570 hPa in fig. 15. These clouds predominantly moisten the mid-troposphere thus increasing the moist static energy and pre-conditioning the atmosphere for the next burst of deep convection.
Mechanisms for the development of stable layers around the melting level are discussed in the observational study of [Johnson et al.(1996)]. Soundings from tropical islands in the west Pacific warm pool region frequently exhibit stable layers near the 0 C level which are believed to be a consequence of melting precipitation in a layer of about 500m thick just below the melting level. At the altitude of the 0 C level, 500m is equivalent to 35-40 hPa, so by increasing the vertical resolution of the GCM from 100 hPa to 50 hPa in this region, we are starting to resolve the impact of melting precipitation on the stability of the atmosphere.
We propose that this mechanism is at least partly responsible for the difference in MJO activity between the 19 and 30 level full GCM integrations. The development of a stable layer around the tropical melting level acts to reinforce the transition from the enhanced convective phase to the suppressed phase of the MJO, and then moistening of the mid-troposphere during the suppressed phase acts to reinforce the transition back to the active phase. This proposed mechanism is supported by the observations of [Johnson et al.(1999)]. They show that, during a suppressed phase of the MJO (late November to early December 1992), convection in the west Pacific was dominated by cumulus congestus clouds. The associated detrainment deepened the moist layer, pre-conditioning the atmosphere to the next burst of deep convection which occurred in mid-December.
[Blad'e and Hartmann(1993)] propose a ``recharge-discharge'' theory for the MJO. In this theory they suggest that the MJO timescale may be set by the time it takes for the moist static energy in the Indian Ocean to build up following the decay of the previous convective event, together with the duration of the convective episode. This means that the propagation of the eastward moving MJO right around the globe and back into the warm SST region is not necessary to trigger a subsequent MJO event. Even if the large scale eastward moving wave breaks down, the convection in the Indian Ocean region may still re-activate once the necessary instability has been realised locally. It may be that the recharging of the moist static energy is achieved in part by the injection of moisture into the mid-troposphere by the cumulus congestus clouds that dominate during the suppressed phase of convection.
There may be other reasons for the change in behaviour between the L19 and L30 versions of the model. In particular, changes to the mean state of the Tropics which may allow more wave energy to propagate in from the extra-tropics cannot be completely ruled out, as indicated by fig. 8(b) which shows that the L30 version has more westerly flow in the Tropics at around 100 hPa than the L19 version. However, this mechanism would not in itself explain the very different nature of the convective moistening at L30 which must be due to some different behaviour of the convection scheme itself.
Some of the moistening by convection in the L30 integration is due to the rather noisy forced detrainment at this vertical resolution described in section 3.4(ii). However, comparison with observations such as those presented by [Lin and Johnson(1996)] show that moistening of the mid-troposphere is a critical element in the organization of deep convective events. It appears that the most significant, extended moistening events in the model are indeed preceded by peaks in the cooling of the 0 C layer, as indicated by the composite event shown in fig. 14.