Figure 5

Comparison of the d¹⁸O measured in a Tarawa Atoll coral with large-scale indices of Pacific climate from Wright (1989). Shaded bars denote El Niño events identified by Quinn and Neal (1992) from historical information; strongest events are shaded most darkly (from Cole, 1992). Strong and moderate El Niño events in 1907, 1917, 1932, and 1943 have no counterparts in the instrumental or coral records, indicating anomalies localized to coastal South America in those years. The coral record indicates a shift to wetter conditions at Tarawa between 1976-1989 that is unprecedented in the past century.

histories, especially with regard to anomaly intensity. However, their compilation of El Niño events remains widely (if perhaps inappropriately) used as a history of large-scale ENSO variability. Both climatological and coral data from across the Pacific indicate that the Quinn index of El Niño events in coastal South America does not consistently track the state of the Pacific-wide ENSO system during the past century (Rasmusson et al., 1995, in this volume; Cole et al., 1993).

Frequency-domain analysis of the Tarawa d¹⁸O record suggests that interannual variance across periods of 1.9-2.5, 3.0, 3.6, and 5.6 years are highly coherent with large-scale ENSO indices; at these periods 80 to 90 percent of the variance in the Tarawa d¹⁸O record is linearly related to variance in the instrumental ENSO indices (Cole, 1992; Cole et al., 1993). These results, shown in Figure 6, fit into the general framework of studies that indicate biennial and low-frequency concentrations of variance in ENSO-related climatological data. With 96 years of coral d¹⁸O data, we can address the issue of whether the spectral signature of ENSO has changed over the period of our record.

We examine the distribution of variance among dominant periods over the length of the record by performing spectral analysis on a series of 30-year windows of d¹⁸O data, each shifted by 2 years from the previous. This evolutionary spectral analysis suggests that the variance spectrum of ENSO may have changed during the past century (Figure 7; Cole et al., 1993). Our 30-year intervals are too short to identify the specific frequencies noted above with statistical confidence, but this analysis suggests intriguing broad-scale changes in the variance spectrum of rainfall at Tarawa. Figure 7 maps the changing concentrations of variance at periods between 1 and 10 years over the past 96 years. In

Figure 6

Cross-spectral comparison between SST index from Wright (1989) and Tarawa d¹⁸O record (Cole, 1992). Upper panel shows the variance spectra of the Tarawa d¹⁸O record (solid fine line) and the Wright SST index (plus symbols), plotted against frequency. The coherency between these records, which represents the degree of correlation between them as a function of frequency, is shown by the heavy line; the squared coherency gives the percent variance in common between the records. The lower panel shows the phasing of the records as a function of frequency. Shaded bars indicate frequency bands over which variance peaks are aligned; coherency is non-zero at more than 80-95 percent significance. At interannual periods characteristic of ENSO (here centered at 5.6, 3.5, 3.0, and 2.1 years), these records share at least 80 percent of their variance and are in phase. These results are consistent with cross-spectral analyses between the Tarawa record and other instrumental indices of ENSO (Cole, 1992).