Difference maps like Figure 1 are useful to understand gross features of the response, but they may fail to give a clear picture of the coherency of variations over large scales. EmpiricaI orthogonal function (EOF) analysis is useful for just such a purpose. In order to better isolate stationary features from propagatings ones, we use an extended EOF (EEOF) analysis of 400m temperature whereby time-lagged maps of the variables are included in the input vector. Also, we use combined EEOF analysis (2 input variables) to identify relations between different variables. Seasonal anomalies (with monthly mean climatological values removed) defined as DJF, etc., are analyzed with lags out to 12 seasons (three years) in constructing the covariance matrix. We consider only the region of the North Pacific east of 155E and north of 20N in order to avoid including the poorly resolved (in both the model and XBT dataset) Kuroshio region and the ENSO-dominated low-latitudes.
3. Results
In all of our results two dominant time scales emerge from the EEOF analyses. The first is a decadal change as depicted in Figure 1. The second is a pair of propagating EEOF modes which hear ENSO time scales. Typically the decadal mode explains 25% of the field variance while the two ENSO modes together explain 10% of the total variance. Both signals are equivalent barotropic in the sense that similar changes at 250-300m are coherent with the 400m response but have larger amplitude at shallower depths.
a. Decadal thermocline change.
Figure 2 shows a synopsis of the decadal temperature change in observations and in the model. Since this mode is nearly stationary, we plot the average of the 13 lags together rather than showing the pahs lags. The top two maps show the first combined EEOF of observations and model 400m temperature with the time series (scaled amplitude) shown as the thin line in the bottom plot. (The two temperature fields are normalized by 0.25C and 0.13C, respectively, because model variability is weaker than observed). The observed pattern is essentially the same as that of the first EEOF of observations alone (not shown) and the combined EEOF time series has very similar time variation as the EEOF of observations alone (bottom, thick line). Both the model and observations reveal a cooling of the basin-scale thermcline from the early 1970s to the early 1980s (as seen in the time coefficient), which is western intensified as expected from inspection of Figure 1. Since this signal explains 29% of the combined variance it represents a significant deviation of the basin-scale thermocline structure and can be expected to be associated with gyre-scale changes in upper-ocean circulation. Indeed, the combined EEOF of model 400m temperature and model 400m velocity (second panel of Figure 2 and dotted time series at bottom) reveals that the decadal signal is nearly geostrophically balanced over this 10-year transition time scale. The flow field reveals a 10% increase in the strength of the Kuroshio extension and the subpolar gyre return flow (cf. Sekine, 1991; Trenberth, 1991; Trenberth and Hurrell, 1994; Akitomo et al., 1996). A stronger than normal northward flow into the central Gulf of Alaska (cf. Tabata, 1991; Lagerloef, 1995) is also seen during the early 1980s in the model diagnosis. It is interesting to note that little change in the California Current System is associated with this decadal signal even though the thermocline (Figures 1,2) did deepen off the West Coast of America.
Since wind stress curl is the dominant forcing function for the gyre-scale circulation, we anticipate that it is associated with the observed and modeled changes seen in Figures 1 and 2. Indeed combined extended EEOFs of wind stress curl and north-south transport (integrated from 0m to 1500m) yields an EEOF mode (third panel of Figure 2 and dashed time series at bottom) which corresponds to the decadal change in temperature and velocity (panels 1 and 2). The pattern of basin-scale wind stress curl seen in Figure 2 is the same pattern one get obtains when differencing the early 1980s and early 1970s as done in Figure 1. Figure 2 indicates that in the eastern basin the flow field is nearly in Sverdrup balance (to within a factor or two) and supports the notion that the wind stress is the forcing function of the decadal thermocline change. Further details od this analysis are presented by Miller et al., 1996a.
1. ENSO-sca1e thermocline variations
The second most important signal in the EEOF analysis of the 400m temperature observations alone
