Introduction

Cumulus convection activity in the tropics is not distributed uniformly in the longitudinal direction; it is most intense over the Maritime continent, the region from Indonesia to the western Pacific. The location of intense convective activity is usually explained by the fact that the value of SST (Sea Surface Temperature) is the highest over the Maritime continent.

However, the distribution of convective activity in the tropics is not necessarily determined by the (sea) surface condition. For instance, the climatic value of precipitation of the western Pacific in March is quite different from that of the eastern Indian Ocean ( Wallace et al., 1995), although there is little difference between the values of SST of the western Pacific and the eastern Indian Ocean (Reynolds and Smith, 1995).

As an attempt to understand the nature of the distribution of tropical precipitation activity, there is an approach from the framework of the thermal response problem. By assuming that the cumulus convection activity at the convection center is a heat source in the atmosphere, the approach is to try to understand the large scale features of the tropical atmosphere as a response to the heat source. Hosaka et al. (1998) performed an aqua planet experiment by the use of a three dimensional primitive model. They placed a localized warm SST area at the equator and observed long term averages of the calculated fields. What they found is that the amount of precipitation increases over an extensive region to the east of the warm SST area, while drying occurs to the west.

Now, the question is why the response of precipitation is asymmetric between the eastern and the western regions. Their results show that, to the east of the warm SST area, there appears a extensive region of low pressure anomaly. Hosaka et al. (1998) argues that the low level frictional convergence associated with the low pressure anomaly reinforce the convergence of moisture flux and intensifies the convective activity there. To the west of the warm SST area, on the other hand, there appears a region of high pressure anomaly. It corresponds to the region where precipitation decreases. However, according to the thermal response theory by the use of shallow water system (Gill, 1980; Heckley and Gill, 1984), Kelvin waves are emitted to the east, while Rossby waves are emitted to the west. Both of the waves are associated with low pressure anomaly, and hence, increase of precipitation should also be expected to the west of the warm SST area.

In order to reveal the reason for this discrepancy, it is required to observe how the precipitation and pressure anomalies develop after the switch-on of the warm SST area. However, the initial development of the response is hard to observe, since the signal is buried in the noise caused by randomly emerging individual convective activity which has a small spatial scale and a short life time. Moreover, in the tropics, there exists intraseasonal variability, which is a propagating disturbance with a global extent and a time scale of several tens of days. (Madden and Julian, 1972; Hayashi and Sumi, 1986). There is also a possibility that the response to the introduction of the warm SST area is changed according to the phase of the intraseasonal variability at the time of the switch-on; for instance, the longitudinal location of the maximum of low level convergence associated with the intraseasonal variability may affect the development of the response (Hosaka et al. 1998).

For the purpose of extracting the initial development of the response to the introduction of the warm SST area, it is expected that ensemble mean operation is effective in reducing both the noise by randomly appearing convective activities and the effects of intraseasonal variabilities. The operation is to take an average of the results obtained by a number of runs starting from different initial conditions but with the same boundary condition. By this operation, the random noise caused by the convective activities will be smoothed out and the effects of intraseasonal variabilities will be removed by choosing the initial values such that the phase of intraseasonal variability is randomly placed.

In the followings, we will perform an ensemble experiment of the sample size of 200 with the same configuration used by Hosaka et al. and we will exemplify that ensemble mean operation is quite useful in extracting the temporal and spatial structure of the response.