In our quest to understand decadal oscillations that appear in the record of atmospheric carbon dioxide measurements, we have been led to investigate seemingly similar oscillations in global air temperature. Employing a much longer temperature record than that available for atmospheric CO2, we have found that decadal variations in global air temperature are characteristic of the past 135 years, except for an interval near the middle of the record. This pattern can be reproduced by assuming the existence of two oscillations with slightly different frequencies that beat on the time scale of about 100 years.
Is it possible that these decadal oscillations are caused by a cyclic process? Perceptions of cyclic behavior in the climatic record have in the past involved so many exceptions and inconsistencies that the subject does not enjoy good repute. Compelling evidence that cyclic behavior is actually caused by an identifiable forcing agent such as solar irradiance is difficult to find.
Decadal oscillations have not been emphasized in previous discussions of global temperature, but the existence of bidecadal oscillations has been alleged by Newell et al. (1989) on the basis of marine air-temperature data. An inspection of their Figure 2 indicates that the signal that they found is similar in phase and amplitude to that shown in our Figure 4, panel 4. There is no indication in their analysis of a weakened signal near 1925, however, like that found by our analysis of global air temperature. We are not aware of other finds of either decadal or bidecadal oscillations.
Of possible importance in deciding the significance of the oscillations in temperature that we report here is the indication that the decadal signal appears to be replaced by shorter oscillations, with periods of close to 6 years, from about 1905 to 1940. Although the spectral evidence for this, as shown in the low-passed record of Figure 5, panel 5, is based only on spectral peaks with periods greater than 6.5 years, the strong 6.0-year peak in the MEM spectrum produces an oscillation (not shown) with almost exactly the same phase. Thus a 6.0-year oscillation appears to be a significant contributor to the overall pattern of variability in the global temperature record, although not as prominent as the decadal signal.
The 6.0-year oscillation, and two other adjacent oscillations with periods of 4.8 and 5.2 years (see Figure 6), could well be associated with the low-frequency component of the El Niño/Southern Oscillation (ENSO) phenomenon, as pointed out by Ghil and Vautard (1991) in their spectral analysis of temperature. When these three oscillations are included in a low-frequency band-pass of temperature, warm periods in this band-passed record include most of the strong and many of the moderate El Niño events.
The 4.8- and 6.0-year oscillations, when summed, show an interference pattern with a beat period of 22.4 years. The timing of these beats shows a possible relationship to the 21.7-year spectral oscillation, inasmuch as the warm phases of this bidecadal oscillation tend to occur near the times of reinforcement of the two higher-frequency oscillations. If these higher-frequency oscillations are related to global aspects of the El Niño phenomenon, is it possible that the bidecadal oscillations that we have been discussing are also related? In the absence of a mechanism to explain any of these oscillations, and not even clear proof that they exist, it is not possible to resolve this question, but the topic appears to us to deserve further investigation.
Does the record of atmospheric CO2 provide any insight into possible causes of the interannual variations in global temperature? At first glance one would not expect so, because interannual variations in CO2 are presumably a consequence of climatic factors that do not depend on the earth's carbon cycle. Nevertheless, two aspects of the CO2 record are worth noting.
First, atmospheric CO2 anomalies show a striking relation to global temperature anomalies with respect to El Niño events (Keeling et al., 1989). Second, on all interannual time scales that can be investigated in the 32-year CO2 record up to 1990, CO2 variations tend to lag temperature variations by approximately a half-year (Kuo et al., 1990). The existence of a lag of that length on the decadal time scale is evident in the plot of Figure 1.
This lag, as shown by Keeling et al. (1989), cannot easily be explained as an oceanic phenomenon, because temperature-induced exchange of CO2 gas at the air-sea interface on a decadal time scale should result in a lag in atmospheric CO2 variation of about 1.5 years, whereas a change in upwelling of cold water should produce atmospheric CO2 variations of the opposite phase with respect to temperature, i.e., a lag of about 5 years. On the time scale of El Niño events, Keeling et al. suggest that atmospheric CO2 responds to global temperature, because vegetation on land tends to release CO2 during anomalously warm periods. With respect to this terrestrial response, a short lag in CO2 is to be expected.
Perhaps variations caused by terrestrial vegetation and driven by climatic change also produce decadal variations in CO2. Indeed, a decadal signal in CO2 might be simply a modulation in the magnitude or frequency of variations on El Niño time scales. If this should be the case, a nearly constant phase lag in CO2 with respect to temperature would be expected over a broad range of interannual frequencies, as observed.
In support of such an hypothesis are several papers' assertions of possibly non-random, long-term behavior in El Niño events. Their focus has been mainly on the decadal time scale (see, for example, Hanson et al., 1989; Michaelsen, 1989; and Enfield and Cid, 1990) but there is a hint of a decadal connection in the study of Barnett (1989),