into seasonal anomalies (November to March) to form winter-mean departures from normal. The winter-mean value was considered missing if two or more months in a winter were missing. Missing data have not been replaced.
Empirical orthogonal function (EOF) analysis was used to identify objectively the dominant modes of variability in the North Atlantic climate system during the period 1900 to 1989. EOF analysis was performed separately on the SST, air temperature, sea level pressure, and zonal wind fields. EOFs were calculated using only those grid squares with at least 60 years of winter-averaged data. A minimum of 50 grid squares containing data in each winter was required for reconstructing the principal components (time series of the EOFs).
Figure 1 shows the first two EOFs of SST over the North Atlantic, superimposed on the long-term mean SST distribution. The EOFs are based on un-normalized winter-mean (November to March) anomalies during the period 1900 to 1989. The first EOF (Figure 1a; hereafter referred to as E1 SST), which accounts for 45 percent of the variance, has uniform polarity over the entire basin. The largest loadings occur along the Gulf Stream (as indicated by the tightly packed isotherms in the climatology). The time series of E1 SST (Figure 1b) exhibits a sudden transition from below-normal values to above-normal values around World War II. It is well known that the technique for measuring SSTs aboard ships changed during the early 1940s, and that the bucket temperatures used before World War II were about 0.3° lower than the engine intake samples used later (cf. Wright, 1986; Folland et al., 1984). Over the Gulf Stream this difference can reach approximately 0.8°, due to the rapid evaporative cooling from the bucket samples (Bottomley et al., 1990). Thus, one may be tempted to disregard El SST; however, air temperatures exhibit a similar EOF pattern with a more realistic time series (see below).
The second EOF (Figure 1c; hereafter E2 SST), which