sustained oscillations in ocean-to-atmosphere heat flux with periods of 19, 38, and 110 years for the Atlantic Ocean and 110 years for the Pacific Ocean. Similar results have been obtained from a GFDL coupled ocean-atmosphere model (Delworth et al., 1993). Figure 9 shows the power spectrum of detrended mean annual surface air temperatures for the Southern Hemisphere generated by the GFDL model (Delworth et al., 1993). There is a statistically significant (a priori p < 0.05) oscillation in simulated temperatures with a mean period of about 110 years, along with minor secondary peaks at about 56 and 29 years. Thus, there appears to be a reasonable model-based argument for the Tasmanian temperature oscillations' having an internal origin in the Southern Hemisphere ocean-atmosphere system.
With regard to possible solar forcing, we compared our 79- and 204-year waveforms with those related to the 80-to-90-year Gleissberg sunspot cycle and an approximately 200-year oscillation found in high-precision radiocarbon measurements from tree rings (Sonett, 1984; Stuiver and Braziunas, 1989), which Sonett and Suess (1984) have suggested could influence temperatures. These comparisons produced suggestive but equivocal results.
Figure 10 shows the waveform of the 79-year temperature oscillation and the envelope of the annual Wolf sunspot numbers (i.e., the Gleissberg cycle) estimated by SSA for the common period 1700 to 1987. There is a surprising degree of agreement between the two, with temperatures lagging sunspots by about 25 years over the central portion of the curves. However, the phasing apparently breaks down after 1900, with temperatures lagging sunspots by only 10 years. The phase drift may simply mean that the good relationship in the central portion is spurious. However, if sunspot cycle length is used instead of sunspot number as an index of solar activity (Friis-Christensen and Lassen, 1991), the phasing between that index and Tasmanian temperatures actually improves. Peaks in solar cycle length
occurred around 1770, 1840, and 1940 (Friis-Christensen and Lassen, 1991). The corresponding peaks in the 79-year temperature waveform occurred around 1805, 1880, and 1970. In turn, solar-cycle-length minima occurred around 1805 and 1900, whereas temperature minima are found around 1840 and 1930. Thus, there appears to be a reasonably consistent lag relationship of 30 to 40 years between Tasmanian warm-season temperatures and solar cycle length since the early 1700s. In being nearly 180° out of phase, this putative relationship either implies that (1) the direct correlation between solar cycle length and Northern Hemisphere temperatures found by Friis-Christensen and Lassen (1991) is effectively opposite in the SH sector around Tasmania, (2) the thermal inertia of the SH oceans significantly delays the response of SH surface air temperatures to solar forcing, or (3) the results of Friis-Christensen and Lassen (1991) and those presented here are spurious. With only about three realizations of the solar-cycle-length envelope available from the sunspot record, it is difficult to make a strong case for either of the first two possibilities.
To compare our 204-year temperature oscillation with the approximately 200-year term in Δ14C, we obtained high-precision decadal radiocarbon measurements of dendro-chronologically dated wood from Stuiver and Becker (1986) covering the period 2500 B.C. to A.D. 1950. These measurements were detrended with a fifth-order polynomial to remove the geomagnetic dipole effect, and the data after 1885 were deleted to eliminate the industrial "Suess effect." Figure 11 shows the residuals from the polynomial curve since 300 B.C., which have also been smoothed to emphasize the century-scale fluctuations in Δ14C. Notable persistent departures are indicated in the A.D. 1100 to 1220 period (the Medieval Solar Maximum) and the A.D. 1450 to 1550 and 1650 to 1700 periods (the Spörer and Maunder Minima,