Three very different views of the mean structure and variability of deep convection over the tropical east Indian and west Pacific Oceans, provided by three different reanalysis datasets for 1980–93, are highlighted. The datasets were generated at the National Centers for Environmental Prediction, the National Aeronautics and Space Administration's Goddard Laboratory for Atmospheres, and the European Centre for Medium-Range Weather Forecasts (ECMWF). Precipitation, outgoing longwave radiation (OLR), and 200-mb wind divergence fields from the three datasets are compared with one another and with satellite observations. Climatological means as well as interannual and intraseasonal (30–70 day) variability are discussed. For brevity the focus is restricted to northern winter (DJF).
The internal consistency of the datasets is high, in the sense that the geographical extremes of rainfall, OLR, and divergence in each dataset correspond closely to one another. On the other hand, the external consistency, that is, the agreement between the datasets, is so low as to defy a simple summary. Indeed, the differences are such as to raise fundamental questions concerning 1) whether there is a single or a split ITCZ over the west Pacific Ocean with a strong northern branch, 2) whether there is more convection to the west or the east of Sumatra over the equatorial Indian Ocean, and 3) whether there is a relative minimum of convection near New Guinea. Geographical maps of interannual and intraseasonal variances also show similar order 1 uncertainties, as do regressions against the principal component time series of the Madden–Julian oscillation. The annual cycle of convection is also different in each reanalysis. Overall, the ECMWF reanalysis compares best with observations in this region, but it too has important errors.
Finally, it is noted that although 200-mb divergence fields in the three datasets are highly inconsistent with one another, the 200-mb vorticity fields are highly consistent. This reaffirms the relevance of diagnosing divergence from knowledge of the vorticity using the method described in Sardeshmukh (1993). This would yield divergence fields from the three datasets that are not only more consistent with each other, but also more consistent with the 200-mb vorticity balance. Further, as proxies of deep convection, they would help resolve many of the issues raised above.