The choice of how to define the hemispheric averages in Figures 3 and 4 is, of course, rather arbitrary. In this particular case it does not make much difference whether one considers the entire hemisphere, the extratropics poleward of 20° latitude, or a smaller polar-cap region. As one moves the outer boundary poleward, the interdecadal trends become somewhat larger, but so does the sampling noise associated with regional and short-term variability, so one's overall impression of the "signal-to-noise ratio" does not change a great deal.
Walker and Bliss (1932) showed that their Southern Oscillation Index (SOI)4 modulates surface air temperature, averaged over the tropics, as well as over certain regions of higher latitudes, such as southwestern Canada. This result was confirmed by the later studies of Newell and Weare (1976), Angell and Korshover (1978), Newell (1979), and Horel and Wallace (1981), which showed that the ENSO signature is evident in time series of mean tropical tropospheric temperature, lagged by about a season or two relative to sea surface temperatures in the equatorial Pacific. This effect is clearly evident in time series of the radiance in Channel 2 of the MSU carried aboard the TIROS satellites (an indicator of the mean temperature of the 1,000 to 300 hPa layer) shown in Figure 5. The upper time series is unsmoothed monthly mean sea surface temperature (SST) averaged over the equatorial Pacific from 160°E to the South American coast and from 5°N to 5°S, which shows clearly the signature of the three most recent warm episodes (19821983, 1986-1988, and 1991-1992). The corresponding MSU-2 time series, averaged over the latitude belt from 20°S to 20°N shown just below it, exhibits the same features, some of them lagged by about a season, as noted in the studies cited above. A remarkably similar signature is evident in the mean surface air temperature of the tropics based on gridded data from land stations, shown in the lowest curve. The much larger amplitude of the SST time series and the phase lag between SST and the other time series are consistent with the notion that the ENSO signature in tropical tropospheric temperature is a forced response to local perturbations in the heat balance at the air-sea interface in the equatorial eastern Pacific. The fact that the ENSO signature is virtually identical in the MSU-2 and land-
surface temperature time series, but slightly larger in the former, suggests that it is transmitted from the equatorial Pacific to the remainder of the tropics via thermally direct circulations in the free atmosphere.
A strong correspondence between equatorial Pacific SST and surface air temperature averaged over the tropical belt is also evident in the longer time series shown in Figure 6. Because of the slight phase lag between the SST anomalies and the land-surface temperature anomalies, the correlation coefficient between the two time series is only about 0.5; nonetheless, the correspondence between warm and cold phases of the ENSO cycle and tropical tropospheric
The SOI is the difference between normalized sea level pressures at Tahiti (11°S, 150°W) and Darwin (12°S, 131°E). It is a measure of the strength of the zonal pressure gradient that drives the trade winds in the equatorial Pacific, which, in turn, govern the volume and coldness of the water brought to the surface by equatorial upwelling.