Before drawing conclusions about the effectiveness of a convective parametrization in a GCM, it should be borne in mind that any scheme is designed primarily to represent the impacts of convection on the large-scale distribution of heat and moisture within the atmopshere. Producing the correct distribution of cloud and precipitation is also important in determining the radiation and water budgets of the earth-atmosphere system but it would be too harsh to judge the performance of a convective parametrization on how well the simulated day-to-day variations in cloudiness and precipitation compare to observations. This study shows that small scale physical processes occurring within convective clouds can have an impact on how the distributions of temperature, water vapour and cloud evolve, which can in turn have wider impacts on the global circulation patterns. However, we would need a detailed cloud resolving model or a much more complex parametrization of convection to fully capture the impact of these effects.
[Johnson et al.(1999)] point out that many conceptual models of tropical convection are based on a bimodal cloud distribution, emphasizing shallow ``trade-wind'' or boundary layer cumuli and deep cumulonimbi. Consequently, many GCM cumulus parametrizations are also designed to emphasize these cloud types. Although increasing the vertical resolution of the model in this study did produce a more trimodal distribution of tropical cloud, the model is still strongly biased towards shallow cumulus and deep cumulonimbus clouds as shown by fig 15. Analysis of observations from TOGA-COARE, together with re-analysis of earlier tropical observing programmes has shown the ubiquity of the 0 
C stable layer in regions of active convection and has demonstrated the importance of this layer in organizing convective clouds. It may well be that for a GCM to produce a more realistic cloud distribution with all its associated consequences, the whole design of the cumulus parametrization needs to be re-considered in the light of these studies.
One particular question raised by this study is how the cooling of the melting level during periods of heavy precipitation should be modelled. By applying this cooling over a layer of the atmosphere 100 hPa thick, the L19 version of the model is unable to develop the thin stable layer seen in observational studies, and it may be better to impose a stability condition on the atmosphere during periods of heavy precipitation rather than explicitly trying to model the development of the stable layer within a framework which is unable to resolve it.