In this section we will investigate how convection in the Tropics behaves and organises on a day-to-day basis within the two aqua-planet runs. In particular we will describe how the convective cloud amount and convective moistening varies through periods of 90 days in the L19 and L30 runs. The diagnostics we will present were not available from the full AMIPII GCM integrations.
In order to show the different behaviour of convective precipitation in the L19 and L30 GCM, the daily precipitation totals for the 2 aqua-planet integrations in all grid-boxes between 2.5 
N and 2.5 S have been binned into 7 different classes and are shown as a histogram in fig. 10. The L19 integration has fewer days than the L30 integration with rainfall totals of between 1 and 10 mm per day, but considerably more days with rainfall greater than 15 mm per day. This shows that the L19 version has deep, strongly precipitating convection occurring throughout the equatorial region more often than the L30 version which has more periods when convection is suppressed and precipitation is weaker.
In order to study the evolution of convection we will look at 90 day timeseries of diagnostic fields averaged every 6 hours in time and over 3x3 model grid boxes (7.5 latitude x11.25 longitude) centred on the equator in space. This area is similar in size to the intensive flux array (IFA) of TOGA-COARE and so qualitative comparisons can be made with fields from that programme averaged over the IFA region (e.g. [Lin and Johnson(1996), Johnson and Lin(1997)]). The fields presented will be time-height cross sections, showing how the vertical profile of certain diagnostic quantities varies with time. Due to the zonal symmetry of the aqua-planet system, any 3x3 grid-box averaging area can be plotted to represent the behaviour of the convection in any similar sized area around the equator.
Figure 11 shows the time-height evolution of convective cloud in (a) the L30 aqua-planet run and (b) the L19 run. In L19, although there is considerable variation in cloud fraction through the period, there is very little variation in the vertical structure of the clouds, with the tops almost always extending up to about 100 hPa. In contrast, the L30 run shows several periods when the tops of most of the clouds in the area only extend up to 600-500 hPa (e.g. around day 88).
Figure 12 shows the time-height sections of lapse rate ( 
p) for the same 90 day period and grid boxes as fig. 11. Both L19 and L30 have a more stable layer (larger negative p) between 600 and 400 hPa. However, there are several periods when the stability in this layer in L30 becomes quite strong, compared to L19 when the stability is generally rather weaker. Comparison with fig. 11 shows that these periods of enhanced stability correspond to periods when the convective cloud tops in L30 are around 600 hPa, as would be expected when the atmosphere is more stable to convection above this level. The presence of this more stable layer over the west Pacific warm pool region during the TOGA-COARE period and in other observational studies is discussed in detail in [Johnson et al.(1996)], and fig. 12 can be qualitatively compared to figure 1(b) in that paper. The development of such a layer in the Tropics appears to be rather common, particularly in association with precipitating convection, but also in regions remote from active convection.
The lapse rate in the L30 model clearly suffers from problems of unrealistic noise, as shown by the alternating more/less stable layers between 780 hPa and 400 hPa which develop at several points through the 90 day period (e.g. between day 84 and day 90). This is almost certainly due to the way that convective detrainment is parameterized in this model. A full description of the convection scheme can be found in [Gregory and Rowntree(1990)] but a brief description of the relevant aspects will be given here. The scheme is of the mass-flux type, with a single buoyant plume representing an ensemble of convective clouds in the grid-box. In order for the plume to continue ascending if it becomes negatively buoyant, mass is detrained in such a way as to maintain the potential temperature of the plume at a positively buoyant value, defined as 0.2K warmer than the grid-box value of potential temperature. This ``forced'' detrainment acts as a source of heat and moisture to the large-scale environment at the level at which it occurs. When the thickness of the model layers is halved, the same amount of heat is detrained into a thinner layer. This leads to greater heating and the rather noisy stability profile of fig. 12(a). These variations in the vertical profile may themselves have an impact on the subsequent development of convection and some of the differences in convective effects between the L19 and L30 versions of the model may be partly due to this noise. However, within the framework of the convective parametrization it is not possible to eliminate this effect completely.
The final time-height sections presented here are of specific humidity increment by the convection scheme, shown in fig. 13. Deep, precipitating convection would be expected to dry the troposphere and there are many observational studies which show that deep convection is indeed a moisture sink for the mid-troposphere [Yanai et al.(1973), Thompson et al.(1979), Lin and Johnson(1996)]. This is because, although some of the moisture carried aloft by convective plumes is detrained into the free troposphere, some also falls back to the surface as precipitation, whilst the descending air between the convective plumes, necessary to balance the ascending mass flux, also dries the atmosphere. Less vigorous, weakly precipitating convection acts to moisten the atmosphere since the compensating descent is weaker, there is less precipitation and more detrainment of moisture from the ascending plume. The L19 cross-section shows that, apart from a few periods of very slight moistening, convection is always acting to dry the atmosphere. This indicates that the convection is mostly deep, penetrating above the melting level in agreement with the cloud fraction plot in fig. 11(a). The L30 plot indicates several periods during which the effect of convection is to moisten the atmosphere significantly between 800 and 450 hPa. These periods correspond to the times when the clouds in the L30 run are shallower and the stability is greater.
These plots can be qualitatively compared to the timeseries of apparent moisture sink, 
, calculated for the intensive observing period of TOGA COARE and presented in [Lin and Johnson(1996)] (their fig. 10). During periods of intense deep convection, such as mid-December 1992, the time-series indicate a sink of moisture throughout the depth of the troposphere. However, during suppressed phases of convection, such as mid to late November 1992, the profiles indicate a source of moisture, with maximum values between 800 and 700 hPa. The computed budgets in this study include the horizontal and vertical convergence of the eddy moisture transports, but it is supposed that during periods of negative the main moisture source is upward transport and detrainment by cumulus clouds.
There is an apparent contradiction between fig. 13 which shows moistening by convection at L30, and figs 7 and 9(c) which show a drier troposphere at L30 in the zonal and time mean. However, the humidity profile in the model is a result of subtle interactions between the physical parametrizations and the large scale dynamics. Studies of the moisture tendencies of the individual components of the L30 model show that the convective moistening is offset by increased layer cloud formation around the mid-troposphere, and with interactions between cloud, radiation and large scale dynamics, the zonal and time mean profile reached by the model in the Tropics is drier.
In order to investigate the convective moistening more closely, a composite moistening event has been formed. This has been done by searching a 90 day section of the L30 aqua-planet integration for periods when the convective increment to specific humidity averaged between 2.5 N and 2.5 S is positive for 5 days or more in at least 2 adjacent model levels below the 0 C level. To be included in the composite, the convective moistening rate must be in excess of 2g.kg 
.day at some point through the moistening episode. This eliminates many events of short duration or very weak moistening. Nine separate events are included in the composite. The impact on the temperature of the melting level due to melting precipitation has also been calculated, and day 0 has been defined as the point at which this cooling peaks prior to the onset of convective moistening. The composite event is shown in fig. 14.
Figure 14(a) shows the increment to specific humidity from the convection scheme, presented as a time-height cross-section, with negative values (drying) shaded. Note that, due to the compositing, the values of convective moistening are quite small (up to 1g.kg .day ). Apart from some weak moistening above the freezing level around day 0, the moistening event begins during day 2 and continues until day 10, with the largest positive values occurring between 840 hPa and 660 hPa. The composite timeseries of cooling of the 0 C layer due to melting precipitation is shown in fig. 14(b). This shows that the moistening episode is preceded by a strong peak in the cooling of the melting layer which will increase the stability at this level. The term then becomes small as convection is largely confined to below the 0 C level through the moistening episode. Figure 14(c) shows the cloud fraction throughout the composite event. Some clouds still extend up to 200 hPa, but cloud fractions in excess of 0.04 are largely confined to below 540 hPa indicating that cumulus congestus is the dominant cloud type from day 2 to day 10. Although there are periods in the L30 integration when similarly sized peaks in the cooling due to melting precipitation do not lead to extended moistening episodes, all the strong, extended moistening events examined were preceded by a peak in melting precipitation.
[Johnson et al.(1999)] show that there is a trimodal distribution of convective clouds within the Tropics, with shallow cumulus, cumulus congestus and cumulonimbus clouds all being prominent at different times. This conclusion is based on radar observations of clouds in the TOGA-COARE region. In their study, individual clouds could be identified from their radar echo and their tops recorded. In a GCM, a comparable diagnostic technique is not possible since the convective parametrization does not model individual clouds but instead tries to represent the effects of an ensemble of clouds within the area of each grid-box. The convective cloud top value at each grid-box indicates that some clouds within the 2.5 x 3.75 area may be expected to reach that level, but there will also be a spectrum of clouds with lower tops within the area. However, it is still informative to compare the distribution of cloud top heights within the L19 and L30 models with each other, bearing in mind that the convective parametrization is representing an ensemble of clouds.
The convective cloud top in each model grid-box between 10 N and 10 S for both L30 and L19 every 6 hours throughout the 12 month integration has been counted and is displayed in fig. 15 as graphs of cloud top pressure versus the percentage of grid-boxes with the cloud top at each level. It must be borne in mind when studying this graph that since the L30 run has twice as many levels between 800 hPa and 300 hPa as the L19 run, the percentage of grid-boxes with convective cloud tops at each individual level would be expected to be about half that of the L19 run. Both L19 and L30 graphs show the major peak at just above 200 hPa. This represents times when the convective parametrization predicts the presence of deep cumulonimbus clouds within the grid-box, and it is implicit within the scheme that there would also be an ensemble of clouds with tops at lower levels. The secondary peak on both graphs is around 800 hPa indicating significant periods when the deepest clouds in each grid-box would be shallow cumulus. These periods occur most frequently on the fringes of the deep convective region, near 10 N and 10 S.
The L30 graph shows a third minor peak at around 570 hPa. This peak corresponds to periods when the deepest clouds within the grid-box are cumulus congestus clouds with tops around the melting level. This peak is absent from the L19 graph. This plot indicates that, at L30, the convective parametrization starts to represent periods when deep cumulonimbus clouds are not present and the deepest clouds within a grid-box are cumulus congestus with tops around the 0 C level. The cloud resolving model study of convection over the tropical ocean by [Liu and Moncrieff(1998)] also indicates the presence of a significant third peak in the cloud spectrum with cloud tops around the melting level.
The significance of these results will be discussed in section 4. However, it is apparent that convection behaves rather differently in the L19 and L30 runs, and these differences may have important consequences for the organization of tropical convection and may affect the maintenance of the MJO in the full GCM integrations.