tory. The two decadal-scale peaks immediately prior to 1961 (in 1943 and 1952) came well after the sunspot maxima of 1938 and 1948; a series of decadal oscillations appeared before 1905, but the peak temperatures occurred near times of minimum sunspots, rather than near maxima. From about 1905 to 1940, instead of decadal oscillations, the temperature record shows oscillations with periods of only about 6 years.
Although these 6-year and decadal variations may reflect internal oscillations of the coupled atmospheric-oceanic circulation system (James and James, 1989), or simply a noisy record that falsely appears to be periodic from time to time (Gribbin, 1979), the present evidence does not, in our view, conclusively rule out the possibility of a persistent decadal periodicity in temperature. It is our purpose here to describe this possible decadal signal to help decide whether its recent appearance, where it correlates with atmospheric CO2, is short-lived, or whether instead it might reflect an oscillatory feature of the long-term temperature record, present in spite of the lack of evidence for a simple mechanism (such as solar forcing) that could produce it.
To investigate decadal-scale variations in temperature we have made use of a compilation of monthly averaged Northern and Southern Hemisphere temperature anomalies both over land and in surface seawater, originally summarized by Jones et al. (1986c) but later supplemented by additional data for the years 1854-1860 and 1985-1989, inclusive. The globally averaged data that we used directly in our computations are listed in Appendix A below.
The large scatter in monthly averaged global temperature data (the dots plotted in Figure 3) is not a strong encouragement to look for cyclic phenomena. Nevertheless, if the data are fit to a flexible smoothing spline (Reinsch, 1967) to suppress the high-frequency scatter (the solid curve in Figure 3), persistent periods of warmer and cooler conditions show in the record. The average spacing of warming and cooling events is between 3 and 4 years, depending on how many of the minor fluctuations are considered significant. A large number of the warmer periods occurred close to the times of El Niño events. It seems likely that the latter reflect phenomena that originate mainly in the Indo-Pacific region of the tropics (Quinn et al., 1987; Quinn and Neal, 1992) as a result of interactions of the atmosphere and oceans that arise from imbalances in the coupled system unrelated to any possible external forcing. If only the stronger warm events are counted, the recurrence interval becomes about 9 years (Quinn and Neal, 1992), so that El Niño events cannot be ignored in considering the possible causes of decadal variations in temperature. However, if they are so related, one would not expect them to be closely periodic.
To obtain a closer look at possible decadal-scale fluctua-