Figure 2 shows a timeseries of an index of MJO activity for the L19 and L30 GCM integrations and ECMWF re-analysis (ERA) data for the period 1979-1993. This index is based on the variance of the zonal mean 200 hPa zonal wind component averaged between 10 
N and 10 S. The resulting timeseries is filtered using a 20-100 day band-pass filter and finally the timeseries is smoothed using a 100 day running mean. This index provides a useful ``quick-look'' at the amount of variability in the upper level flow associated with the MJO. The definition of this index and its representation of the response of the planetary scale circulation to MJO activity is discussed in detail in [Slingo et al.(1999)].
The time-series of the ERA index shows that activity peaks in the northern hemisphere winter and spring apart from 1981 when the strongest activity is in the northern hemisphere summer. There is considerable interannual variability in the index, with years such as 1984 and 1987 showing very little activity and years such as 1990 showing very strong activity. The L19 GCM integration has a strong peak in 1986 but otherwise the MJO activity is rather weak. Also, many of the peaks in the L19 time-series occur during northern hemisphere summer. The equivalent time-series from the other 5 members of the L19 ensemble (not shown here) indicate similar behaviour, with rather weak MJO activity and isolated peaks in activity which occur in different years in each member of the ensemble. This is consistent with the results of [Slingo et al.(1999)] who showed that, in a 4 member ensemble of a previous version of this GCM forced with observed SSTs, there is no reproducibility in MJO activity from year-to-year.
The L30 GCM integration shows much more MJO activity, with peaks in most years of comparable strength to those seen in the re-analysis data, although not necessarily matching the strength of the re-analysis signal in each individual year. The seasonality of the L30 index also seems to be improved over the L19 run. L30 reproduces the peak in the northern hemisphere summer of 1981, although there are still some other years (e.g. 1980,1991) when the L30 also produces a peak in MJO activity during northern summer which is not seen in the ERA timeseries.
One way to look at the MJO in more detail is by plotting Hovmöller (time-longitude) diagrams of a field such as the velocity potential (VP) at a level corresponding to the outflow from deep convective clouds. Hovmöller diagrams of 200 hPa VP for the period from July 1989 to July 1990 are shown in fig. 3 for the ECMWF re-analysis and the L19 and L30 AMIPII GCM runs. Evidence of an eastward propagating signal with an intraseasonal period is apparent in the re-analysis data from September onwards, with a large amplitude signal making several circuits of the equator between January and May. The L30 GCM also shows eastward propagating events of similar magnitude to the re-analysis data, but with a rather faster period and somewhat more noise. The strongest signal also occurs between January and May. The L19 plot shows very little coherent eastward propagation apart from the period between September and November when a rather weak signal is evident. This year is chosen to be fairly typical of the MJO in all 3 sets of data and it confirms that the MJO in the L30 GCM is stronger and more coherent than in the L19 configuration.
The eastward propagating signal of the MJO in the velocity potential fields shown in fig. 3 is accompanied in reality by an eastward propagating envelope of enhanced convective activity. This develops over the Indian Ocean, moves east over Indonesia and then propagates out to about the dateline, enhancing convection in the ITCZ and/or the SPCZ. Finally the convective activity dies out over the cooler waters of the east Pacific. This aspect of the coupling of convective activity to the large scale dynamical signal of the MJO is often not well captured by GCMs. [Sperber et al.(1997)] showed that, even in 2 GCMs with a strong signal of the MJO in the upper level winds, the convective anomalies did not propagate particularly realistically. Rather they showed at best a standing oscillation pattern in which convection develops and decays out of phase over the Indian Ocean and west Pacific at intraseasonal timescales with little or no propagation between the two centres.
This propagation of convective precipitation can be examined by performing a lag-correlation analysis of the precipitation and velocity potential from the model integrations. Timeseries of band-pass filtered daily values of convective precipitation for each model longitude, averaged between 10 N and 10 S are correlated with the velocity potential timeseries at 123 E, also band-pass filtered and averaged between 10 N and 10 S. 15 years of October-May data are used from both the L19 and L30 AMIPII integrations. Fig. 4(a) shows the resulting pattern from the AMIPII L19 integration. The pattern is dominated by a standing oscillation in the west Pacific with a period of about 30 days. The area of positive correlations centred at 90 E at lag 0 indicates that when convective precipitation is enhanced over the west Pacific in association with a minimum in VP, it is suppressed over the Indian Ocean. The orientation of the positive and negative correlations in the Indian Ocean indicate, if anything, westward propagation of precipitation anomalies. Figure 4(b) shows the same lag correlation analysis for the AMIPII L30 integration. Although the pattern is still dominated by the standing pattern in the west Pacific, there are now significant negative correlations in the Indian Ocean appearing 15-20 days before the VP minimum at 123 E. This region of negative correlation (i.e. enhanced precipitation) appears to propagate eastwards across the Indian Ocean and Indonesia, although the correlation coefficients become insignificant at the 95% level between 90 E and 100 E. There is also a region of significant negative correlations extending out to the dateline at a lag of +5 days indicating that convective precipitation continues to propagate eastwards in association with the VP minimum. This pattern is not seen in the L19 plot. The patterns seen in this figure are robust in that they are also seen if the data are split into shorter segments and the lag-correlations are computed for these segments. Neither L19 or L30 show a smooth propagation of precipitation anomalies across the Indian Ocean/west Pacific, but L30 shows more organized convection on the scale of the MJO and a more coherent, out-of-phase relationship bewteen intraseasonal convective anomalies in the Indian Ocean and west Pacific.
We need to understand why the change in vertical resolution has such an impact on the simulation of the MJO when the physics of the model has not changed. [Slingo et al.(1996)] make several suggestions as to why some GCMs produce strong MJO signals whereas others are very weak. One suggestion is that GCMs with a convective parametrization with a buoyancy closure simulate a more realistic MJO than GCMs in which the convective parametrization is closed on moisture convergence. This cannot explain the differences seen in the current study since all the physical parametrizations are the same in the L30 and L19 versions. Another suggestion from [Slingo et al.(1996)] is that the basic climate of the GCM is important in determining the strength of the MJO signal. GCMs with a tropical precipitation distribution which is closely correlated with warm SST produced more realistic MJOs in their study. GCMs with regions of strong upper level (200 hPa) westerly flow in the equatorial belt also produced stronger MJO activity. This may be due to the propagation of wave energy from the extra-tropics into the Tropics through regions of westerly winds, thus strengthening the MJO as proposed by [Hsu et al.(1990)].