CHARLES D. KEELING AND TIMOTHY P. WHORF1
Since 1958, global air temperature has been found to correlate with the concentration of atmospheric CO2 on interannual time scales, from biennial to decadal. The decadal variations also correlate approximately with the 11-year sunspot cycle. Although CO2 data are lacking before 1958, global temperature data are available back to the 1850s. No consistent solar correlation is evident in this longer record. Instead, the record shows a decadal pattern that can be expressed by two oscillations having periods of approximately 9 and 10 years. These oscillations show reinforcement in the 1880s and 1970s and interference in the 1920s. Thus, decadal oscillations appear to have been relatively prominent in the latter half of the nineteenth and twentieth centuries, while being weak or absent from about 1905 to 1940. In the absence of an identified mechanism that would explain this pattern, the results must be treated as tentative.
Several years ago we detected a possibly cyclic variation in the concentration of atmospheric CO2 (Keeling et al., 1989). The period of variation, of the order of 10 years, appeared to be synchronous with a decadal oscillation in air temperature when both records were band-passed to remove high- and low-frequency fluctuations and the influence of fossil-fuel combustion on the CO2 concentration. As shown in the upper panel of Figure 1, the adjusted 32-year CO2 record shows maximum concentrations near 1961, 1971, and 1981, the same years in which decadal maxima in air temperature occur. The maxima in CO2, especially the latter two, nearly coincide with maxima in the number of sunspots (lower panel) as though both CO2 and temperature were influenced by changes in solar irradiance. The investigation was not extended back to early times, because of a lack of CO2 data.
Globally averaged surface air temperature anomalies have been determined back to 1854, however (Jones and Wigley, 1991). To extend our analysis, we have accordingly computed the difference between two smoothing splines fit to this temperature record, where the differing stiffnesses of the splines result in band-passed data similar to those shown in Figure 1. The results, plotted in Figure 2 together with the record of sunspots, suggest that the positive correlation of global temperature with sunspot numbers is transi-