EUGENE M. RASMUSSON
University of Maryland
When Drs. Diaz and Bradley present a paper, it always provides an enormous amount of information. The central question, in terms of a synthesis of the wealth of information provided in this paper, is the question posed by the title: "How different is the climate of the twentieth century"—more specifically the past few decades—"from that of earlier centuries?" In addressing this question the authors examined temperature, decade-to-decade temperature changes, and changes in interannual and interdecadal variability on various spatial scales.
The problem in any analysis of this type, of course, is that of inadequate data distribution in time and space. Thus, the analysis was largely limited to the land areas of the Northern Hemisphere, where both instrumental and proxy data are most plentiful.
It may be well to note that changes in the spatial distribution of observations from decade to decade may affect both the computed area average and the variability. Figure 10 in my paper, which appears later, illustrates the effects of using different data distributions in synthesizing global averages of SST, where the problem is probably more severe than it is over the land areas.
Figure 1 in this commentary also shows estimates of the variability in SST obtained from two different analysis schemes. Two estimates of variability are obtained. One is from the optimal averaging (OA) analysis technique of Vinnikov et al., and I think Dr. Groisman will tell us about that one. Another is derived from the simple box average that has been used—for example, by the U.K. Meteorological Office—in deriving SST area averages. The curves in Figure 10 of my paper show the low-frequency variations, i.e., periods longer than 30 years. There are some differences, but they are not too great, about 0.15 degrees around 1910 and 1920, with the OA staying a little bit closer to the mean for the entire period. Now, if we remove the low-frequency variations so that only the variations on time scales less than 30 years remain (Figure 1 in this commentary), and look at that difference between the two analysis schemes, we see that the level of higher-frequency variability may also depend on the first guess used. The box method and the OA with the previous month as the first guess are similar, but the OA with climatology as a first guess shows less variability during the earlier decades of the series. Thus, relative to the box method, the OA shows an increasing variability with time. This comparison illustrates that one
must be a bit careful in deriving conclusions about changes in variability from area averages. The estimates of variability may be sensitive to changes in data distribution and the particular analysis scheme used.