However, the modal decomposition analyses in the previous empirical studies have given a paradox to our community. Several scientists showed that SST anomalies consisted of an equatorial anti-symmetric dipole and a symmetric monopole (Nobre and Shukla 1996, Chang et al. 1997). But others argued that SST anomalies in the Northern tropics were independent of those in Southern tropics: Northern Hemisphere and Southern Hemisphere modes existed (Houghton and Tourre 1992, Mehta 1998, Enfield and Mestas-Nunez 1999).
From SST datasets over recent decades, Iarge amounts of variance are found on each of these time scales (Servain 1991, Huang and Shukla 1997) and the clear spectral gap is present between time scales (Mehta and Delworth 1995, Mehta 1998). The purpose of the present study provides dominant modes, which are physically reasonable, from statistical analyses by taking advantage of the spectral gap between decadal and interannual SST variability in the tropical Atlantic. Details of observational datasets and analysis procedures should be referred to Tanimoto and Xie (1998). The exact physical basis is presented with the ocean-atmosphere coupled model by Xie et al. (1998, also in this volume).
2. Dominant modes in the Pan-Atlantic basin
Figure 2 shows two leading empirical orthogonal functions (EOFs) and accompanying normalized principle components (PCs) of SST variability in the two subdomains of Atlantic. A monthly 2 degree latitude-longitude SST dataset is constructed from Global sea-Ice SST dataset (GISST; Folland and Parker 1995) for the North and South Atlantic (70｡?-50｡?) from 1946 through 1996. Before we compute those EOFs, we divide the Atlantic basin into two independent parts north and south of the equator. Then SST anomalies are averaged for the boreal winter (November through April) and filtered through the decadal band (8-16 year). A Butterworth type of a recursive filter is used to isolate decadal and interannual variability. These EOFs explain 44.2% and 43.2% of the band-passed variance on each of hemispheres, respectively.
Two EOFs (spatial patterns) are organized into meridionally lined up five centers of action. Maxima appear east of Newfoundland, in the Northern Tropics and in mid-latitude in the Southern Hemisphere, while minima appear off east coast of the United States and in Southern Tropics. The normalized PCs (time series) of their leading EOFs are highly correlated (correlation 0.64). Although the maxima and minima do not always coincide with one another, the sub-periods of upward and downward shifts are consistent. In this analysis, the two subdomains are statistically independent and there is no a priori reason for them to be correlated. This indicates that tropical Atlantic SST anomalies are not always independent in either side of equator, at least on decadal variations, further and that the tropical dipolar oscillation is associated with the Pan-Atlantic decadal variability.
In fact, the tropical dipolar oscillations show a good agreement with two PCs after late 1960s. The lowermost curve of three time series represents a dipole index, calculated from difference of zonal mean SST anomalies between 10-20 degree latitudes in either side of equator. Note that no filter is applied to this index. Figure 3 shows the simultaneous regressions (annual mean) of Atlantic SST anomalies onto the dipole index.
