thereby reducing the poleward heat transport. It has also been suggested that increasing concentrations of sulfate aerosols over the Northern Hemisphere might have been responsible for the cooling during that period (Charlson et al., 1991, 1992). Hence, there is no shortage of mechanisms that may have contributed to the observed interdecadal variability in the climate record. The problem is to quantify the arguments, through the use of climate models, so that the dominant mechanisms can be identified (Battisti, 1995; Rind and Overpeck, 1995).
The results presented in the foregoing sections serve to illustrate the need for a flexible, pragmatic approach toward the smoothing of monthly mean data, and stratifying of data by season and by latitude belt. For example, in displaying tropospheric data, a three-way partitioning into means for the tropics and Northern and Southern Hemisphere extratropics appears to be more informative than the more commonly used two-way partitioning by hemisphere, because it isolates the tropical ENSO signal. For stratospheric data it may be best not to do any partitioning at all, since global averaging tends to filter out spurious, dynamically induced temperature fluctuations.
I would like to thank Yuan Zhang and Elena Yulaeva for providing material from their graduate research projects, and Clifford Watson for assistance with the preparation of the figures. Clifford Mass offered a number of helpful comments. The work was supported by the National Science Foundation under Grant 9215512.
HENRY F. DIAZ
NOAA/Environmental Research Laboratories
Before we begin a discussion, I just want to encapsulate some of the primary conclusions or at least principal points Dr. Wallace discussed. First, I think that the ENSO is manifested primarily in tropical temperatures. When you look at global indices or hemispheric indices, what you are seeing is essentially the folding of the tropical signal into the overall record. A very important point that we should consider in the context of the previous talks and the slides that Professor Lindzen showed is that the volcanic temperature signal on the hemispheric temperature record is relatively short-lived. It seems to me that we are talking about thermal relaxation on the order of 3 to 5 years from the time when a major volcanic eruption takes place, through the drop in the temperatures, to the recovery. In that sense, I think that we are not talking about decades but less than a decade.
The volcanic effect also says something about the sensitivity of climate to greenhouse perturbations. There is something interesting about the difference between the Northern and Southern Hemispheres' temperature responses. Even in the 1979-to-present record, the atmospheric response is quite rapid, and there are clear differences between the two hemispheres. We should try to discover whether they are due primarily to the fact that the Southern Hemisphere reflects the ocean damping effects, and whether the Antarctic ice cap has something to do with the difference. Also, the lower stratosphere has a much more spatially homogenous response than the troposphere, and perhaps long-term monitoring of lower stratospheric temperatures would turn out to be a useful tool. Large temperature changes in the lower stratosphere have been projected, and I think we should consider how the observations and the predictions are coming together.
Dr. Wallace mentioned that interdecadal temperature signals are different for surface and upper air indices. This is suggested by data for the last decade, although the record is not really long enough that we can say so with certainty. The correlation between, say, Jim Angell's 63-station radiosonde network and temperatures at the surface is sufficiently high to explain about 50 percent of the variance. For comparison, the correlation between that radiosonde network and the MSU data is about 0.8 or higher.