1. Introduction
Central to understanding and detecting climate change is a knowledge of the natural variability of the oceans and their interaction with the atmosphere. Despite reports of teleconnections between the El Niño–Southern Oscillation (ENSO) and the atmosphere in the midlatitudes, previous work has only shown weak associations between ENSO and the subsurface ocean in the main gyre of the South Pacific. Wyrtki (1984) and Meyers and Donguy (1984) have shown that an intensification of the South Equatorial Current during the 1982–83 ENSO event was associated with a southward migration of the subtropical gyre. Harris et al. (1988) suggested that during ENSO years the subtropical convergence in the southern Tasman Sea appears to move northward, causing a cooling of about 3°C over 3° of latitude. Hsieh and Hamon (1991) suggested that there is a weak ENSO signal in the East Australian Current at about 34°S off the east Australian coast.
More recent work on variability and long-term changes in the southwest Pacific Ocean stem from the examination of the seasonal and interannual variability of the subtropical gyre by repeated expendable bathythermograph surveys, conducted between Auckland, New Zealand, and Suva, Fiji, over the period of 4 yr (Roemmich and Cornuelle 1990), and by the repeat of the SCORPIO hydrographic sections in the Tasman Sea 22 yr later (Bindoff and Church 1992). Despite the contributions made in this region by these studies, work in the southwest Pacific region has been limited by the relatively short investigation period, the spatial area of investigation, or the deficiencies associated with the evaluation of two “snapshots” separated by decades. Typically, past observational studies have been unable to provide important spatial and temporal information in the southwest Pacific region as a whole for more than a few years at a time, raising the possibility that aliasing could mask the underlying oceanography.
This paper addresses the limitations of previous studies by examining the most complete set of upper-ocean temperature measurements from bathythermograph casts obtained over the entire southwest Pacific region from 0° to 50°S and 140° to 180°E. The paper examines the large-scale interannual and decadal changes occurring within the region during the period 1955–88. From these data a new upper-ocean temperature “atlas” has been generated, focusing on spatial scales of at least 2° of latitude and longitude and temporal scales of 3 months and longer. This choice of scales is consistent with the sampling of the ocean data over this period.
Three separate analyses (summarized in Table 1) using a rotated principal components analysis (RPCA) are presented in order to describe features of the spatial and temporal variability in the mapped ocean temperature field. The first analysis (analysis 1) examines intraocean correlations of the upper-ocean temperature field from the surface to a depth of 100 m for the 34-yr period from 1955 to 1988. This analysis permits not only the examination of El Niño variations, but also provides a valuable base for the examination of decadal-scale variability in the southwest Pacific.
The second analysis (analysis 2) examines the shorter 16-yr period of temperature observations between 1973 and 1988, where the profiles typically extend to depths of at least 450 m and are more evenly distributed. The third analysis (analysis 3) examines the cross correlation between ocean temperatures at 250-m depth and the Comprehensive Ocean–Atmosphere Data Set (COADS) sea level pressure and winds over the observation period between 1973 and 1988.
2. Data and analysis
The ocean temperature data used in this study are from the archives of the National Oceanographic Data Center (National Oceanographic Data Center 1991) and contain the most comprehensive compilation of oceanographic measurements available in this region. A total of almost 40 000 mechanical and expendable bathythermograph (MBT and XBT) casts are included in this analysis of the temperature data for the southwest Pacific region between 0° and 50°S and 140° and 180°E, and during the period from 1955 to 1988. The distribution of these data is shown in Figs. 1 and 2, and a description of these data and the quality checks performed is presented in Holbrook (1994).
Wind and sea level pressure data from the COADS dataset (Woodruff et al. 1987) are also used in this study. These data provide monthly averaged zonal and meridional components of the winds and sea level pressures on a 2° × 2° grid between 1947 and 1987. In this paper we only use those data from 1973, in order to examine cross correlations between the atmosphere and ocean fields, the ocean temperatures are from the XBTs. The wind and sea level pressure data coverage is also better from the early 1970s. Monthly mean sea level data used in this study are from the Permanent Service for Mean Sea Level (PSMSL) (Woodworth 1991).
The upper-ocean temperature measurements are four-dimensional. However, the profiles are irregularly distributed in space and in time. In contrast, the COADS data are simpler, consisting of regularly gridded two-dimensional spatial fields averaged to monthly values. To homogenize these two data sources, and after first interpolating the profiles vertically at 5-m depths, the ocean temperature data were interpolated both in horizontal space and time. The spatial grid is the same as the COADS spatial grid but restricted to the months of January, April, July, and October of each year. The mapping of the ocean temperature data in time and space was performed using a combination of empirical orthogonal functions (EOFs) and objective mapping. This approach follows the methods used by Fukumori and Wunsch (1991) and Bindoff and Wunsch (1992) in their analyses of hydrographic data. The COADS data were also objectively mapped in order to improve the spatial and temporal consistency and reduce the effects of gaps in the data. As a result, both the atmospheric and ocean data have been filtered with similar space and timescales.
The data matrix used in the EOF procedure consisted of normalized differences between the temperature data and the monthly means at the data points. Thus, the seasonal cycle has been removed from each of the analyses so that the objective mapping focuses on the interannual and decadal changes that have occurred in the region. The normalization factor is the estimate of the a priori noise as described in Holbrook (1994). The a priori noise represents the noise associated with the unresolved physical processes such as internal waves, mesoscale eddies, and other sources of noise that occur with instrument type and reporting technique. The objective mapping was based on spatial scales of 4° longitude and 2° latitude, and a timescale of 90 days. These scales are consistent with the data distribution in time and space over the period for which observations are available. Because the ocean temperature field is quasi-zonal in this region, a longer zonal length scale was used.
The objective mapping procedure gives an objective estimate of the random error at each grid point. This error estimate takes into account variations in data density caused by, for example, changes in ship routing. The atlas provides temperature information for January, April, July, and October, covering the periods 1955–88 (to 100-m depth) and 1973–88 (to 450-m depth). Confining the analysis of the longer time series to only the upper 100 m was motivated by the shallower depth limitations of most of the MBTs and the possible quality differences between MBT and XBT data below 125 m (Ridgway 1995). Due to computational constraints, the data were subdivided in time to include between 1100 and 1500 data points. Each subdivison was mapped separately. Furthermore, data within each time subdivision overlapped adjacent periods by 6 months in order to minimize the errors at the temporal boundaries. By dividing the data into these smaller subsets and using the EOF procedure to describe the vertical temperature structure, significant reductions were made in the computational costs associated with the matrix inversion and the mapping of these data was made tractable.
Statistical checks of the objective mapping were also performed to ensure that the residuals were unbiased and consistent with the a priori noise (Holbrook 1994). It was found that the vertical structure of the temperature data could be described by five EOFs, which explained more than 86% of the total variance in the data. The small number of modes required shows that the vertical temperature profiles contain strong vertical correlations.
3. Rotated principal components analysis
In this study the objectively mapped ocean temperature anomalies and COADS anomalies (each with the seasonal mean removed) are analyzed using the RPCA. The RPCA is particularly good at examining all timescales, thus avoiding the aliasing of different physical phenomena that occurs in simple subsampling and filtering techniques of data commonly used to examine climate change.
An S-mode principal components analysis (PCA; see, e.g., Richman 1986) is used in this study in order to examine the spatial distribution of various fields and how these patterns evolve over time. Here, the spatial information comes from the principal component (PC) loadings, while the time amplitudes are the PC scores. The PCA itself is performed using the singular value decomposition of the correlation matrix. Each term in this matrix comprises the cross correlation between normalized temperature (and/or atmospheric field) anomalies, the normalization factor being pointwise estimates of the data error estimated from the objective mapping procedure. Thus, for the temperature data, regions of low data density will have lower weight, reflecting the smaller number of measurements and hence the greater uncertainty. One such area is subregion 2, prior to 1973 (Fig. 1). As a result of this normalization scheme, individual modes describing this area and period will tend to have PC scores with small magnitudes.
A total of 413 spatial points out of the maximum of 500 (25 × 20) were utilized from each variable (field) in the PCA. The 87 spatial points that were not used represent either land or data-sparse areas. In each analysis, cross correlations were taken over the four fields, including ocean temperatures at various vertical levels and/or the sea level pressures and surface winds.
Rotation of the PCs was performed using the orthogonal rotation known as varimax (Kaiser 1958). The varimax rotation of these data only slightly changed the spatial patterns in the PCs. Overall, there were only very minor improvements to the modes, with slight increases in the pattern correlations and congruences as defined by Richman and Lamb (1985). These small improvements (discussed further in Holbrook 1994) indicate that the unrotated PCs are robust due to the relatively good representation of the data by the unrotated modes and suggest that the ocean temperature data used in this study, at the scales of interest, are well sampled. This is also borne out by the very close similarity of the spatial modes and scores between different analyses. For example, in analysis 1 the rotated principal components (RPCs), (and PCs) have very similar spatial patterns and scores to the RPCs (and PCs) found in analysis 2. This is despite the two analyses differing in their respective periods covered (34 vs 16 yr), in instrument type (MBT and XBT compared with just XBT), and in ship route and data density.
4. Results
A simple measure of El Niño–La Niña variability frequently used in climate studies is the Southern Oscillation index (SOI). In this study the SOI (slightly smoothed using a five-point filter) is correlated with the RPC scores and tested for significance (where the degrees of freedom are determined using the method of Davis 1976) in order to identify associations between these orthogonal modes and El Niño. The form of the SOI used is that that is defined by the standardized difference in the sea level pressure anomalies between Darwin and Tahiti (Troup 1965). As will be shown, many of these modes that define regions outside of the Tropics are significantly correlated at various lags and leads with this large-scale pressure index.
a. Intraocean correlation analyses
1) Analysis 1
Analysis 1 was designed to examine the interannual and decadal temperature variability within the mixed layer and upper seasonal thermocline by utilizing temperature data at the sea surface and 30-, 60-, and 100-m depth between 1955 and 1988. Figure 3 shows a plot of the first 30 eigenvalues from the nonseasonal (seasonal cycle removed) PCA of the data correlation matrix. Also shown on this figure are the sampling errors on the eigenvalues, as described by North et al. (1982). The error for each eigenvalue is defined by λ(2/N)1/2, where λ is the eigenvalue and N is the number of time increments (in this case 136). The 95% confidence levels, indicated by the dashed line in Fig. 3, were calculated from a Monte Carlo simulation comprising 100 eigenvalue analyses of individual correlation matrices containing randomly generated Gaussian variables (Overland and Preisendorfer 1982).
Interpretation of Fig. 3 suggests that 8 modes are significant, with an initial break in slope at about 4 modes and then again at about 7 to 8 modes. The “scree” (i.e., the noise component in the eigenvalue sequence) is represented by those modes that approach a straight line after 8 modes. This is clearer in the natural log eigenvalue (LEV) plot (Craddock and Flood 1969) shown in the lower graph of Fig. 3 and is consistent with the higher modes reflecting the background noise level. Interestingly, the Monte Carlo simulation at the 95% confidence level suggests that there could be more than 30 significant modes. However, the higher modes are physically inconsistent with the information in the data, and thus the more conservative choice is made of retaining only 8 modes. We recognize that there is degeneracy in the eigenvalue sequence according to North et al. (1982). However, we believe that the first 8 modes (34% of the total variance) are robust because, as discussed at the end of section 3, in different analyses over different time periods and with different instruments, the spatial patterns of the retained PCs were very similar. Furthermore, there were only minor changes to the modes through rotation. Thus, these spatial patterns reflect the true physical structures that occur in this region and are not merely associated with the choice of modes or the temporal and spatial variations in the data distribution.
Figure 4 shows the spatial patterns and corresponding time series for modes 1, 2, 3, 5, 7, and 8 of the 8 varimax rotated modes. Modes 4 and 6 are not shown since they are less interesting for this paper. These 2 modes together describe the variability along the east Australian coast associated with the bifurcation of the westward flowing South Equatorial Current, which is continuous with the southward East Australian Current and the northward component toward the Torres Strait. Details regarding these modes are given in Holbrook (1994). The modes presented in this paper reveal important new information about the upper-ocean climate of the southwest Pacific Ocean.
Mode 1 (Fig. 4a, 5.9% of the variance) of analysis 1 describes the dominant spatial pattern and temporal signature associated with El Niño–La Niña in the upper southwest Pacific temperature field. Immediately apparent are the strong peak amplitudes in 1983 and 1987 during the 1982–83 and 1986–87 El Niño events. In both cases the peak amplitudes occur in July of those years. The small amplitudes of the mode 1 scores prior to 1980 are due mainly to the data sparsity in subregion 2 in these earlier years. The mapping errors are large for the gridded data during this period and, consequently, the normalized temperature anomalies are given low weight in the RPCA.
The largest loadings associated with this mode are in a large-scale zonally oriented pattern extending eastward from Papua New Guinea, with a strong center over the Solomon Islands. A significant feature in the spatial patterns is that the largest-magnitude RPC loadings occur at 100-m depth. From analysis 2 it was found that the loadings for the corresponding mode covering the deeper depths were even larger at 250-m depth. The correlation coefficients between the mode 1 scores and the SOI at zero-, 3-, 6-, and 9-month lags are respectively −0.27, −0.50, −0.52, and −0.39. The correlation coefficients with 3- and 6-month leads are respectively −0.08 and 0.04. Here, the maximum correlation coefficients occur at lags of 3 and 6 months and are significant at the 99% level. These results indicate that, on interannual timescales, the dominant mode of ocean temperature variability associated with ENSO is due to vertical movements of the subsurface tropical ocean thermocline at a lag of between 3 and 6 months. These results also reinforce that heat budget studies cannot be properly undertaken without adequate knowledge about subsurface temperature variability, and for the purposes of predicting the strength of El Niño events, subsurface temperature anomalies are the most significant.
It should be noted that although the spatial pattern of this mode is qualitatively consistent with individual ENSO events, it actually represents an averaged ENSO. Associated with the cooling in this region during ENSO, lowered sea levels due to thermal contraction are implied and have been observed in sea level and steric height studies in the area by Wyrtki (1985) and Ridgway et al. (1993). Wyrtki (1985) indicated that, for example, in May 1983 sea levels were depressed over a wide area from the Solomon Islands to Tahiti, with the lowest values of −41 cm observed at Funafuti. On the other hand, during the reportedly weaker ENSO event of 1986–87, Ridgway et al. (1993) reported the largest depression between December 1986 and April 1987 at Honiara, where sea level fell by approximately 25 cm. However, they further suggested that steric sea level appeared to vary by up to 45 cm south of New Ireland.
The mode 2 (Fig. 4b, 5.5% variance) loadings are shown to extend southeastward in a broad pattern from Papua New Guinea across the Coral Sea. This predominantly surface pattern looks very similar to and is approximately aligned with the mean position of the South Pacific convergence zone. This mode is almost identical to the mode 5 pattern at the surface for analysis 2. In both of these modes the loadings at 100-m depth are much smaller. For the deeper analysis the loadings are negligible at 250-m and 450-m depths. This suggests that, on interannual timescales, the ocean below the mixed layer does not significantly interact with the atmosphere in this region. In both mode 2 and mode 5, the scores over the same time periods are also similar, though slightly out of phase (the phase difference being explained by the significant RPC loadings describing the mixed layer at 30-m and 60-m depth in mode 2, presented here).
The maximum correlation coefficient between the mode 2 scores in analysis 1 and the SOI is 0.47 at zero lag (significant at the 99% level), while the surface mode 5 scores in analysis 2 lead the SOI by 3 months (with a correlation coefficient of 0.61 significant at the 99% level). It appears, therefore, that the sea surface temperature anomalies in this Papua New Guinea–Coral Sea region may be a useful indicator for the onset of El Niño, with a 3-month lead. This will be discussed further in the following section.
The spatial pattern for mode 3 is centered in the subtropical gyre, with loadings in excess of 0.4. All of the depth levels show a very similar spatial pattern with similar loadings. This subtropical gyre mode has a very similar structure to mode 3 for analysis 2, which is dominant at 250-m and 450-m depth, thus confirming that the spatial pattern associated with this mode extends into the main thermocline to at least 450 m.
A significant proportion of the interannual variability in mode 3 (Fig. 4c, 5.3% variance) can also be attributed to the El Niño phenomenon. The largest correlation coefficients are, respectively, 0.34 and 0.41 at zero lag and 3-month leads (ocean scores leading the SOI). These correlations are significant at the 98% and 99% levels. In addition to the El Niño component, the scores show significant decadal variability not present in the lower-order modes. The scores for this mode show that the subtropical gyre is at its warmest in the mid-1970s and has been cooling since. The cooling between the mid-1970s and 1988 implies that the whole of the subtropical gyre cooled during this time. This cooling would also have been accompanied by a decrease in the steric sea level in the main gyre and a weakening of the meridional gradients, leading to a decrease of the East Australian Current. Sea level measurements at Noumea confirm this trend toward reduced sea level in this region (Fig. 5).
The reasons for these changes in the gyre heat content and circulation are unclear. The SOI also exhibits a decreasing trend due to the increased tendency toward the occurrence of El Niños during this time. This decadal cooling could possibly be related to the increased incidence or persistence of El Niño during the last two decades (e.g., Salinger et al. 1996). However, the decreasing trend may also be related to interdecadal variability such as that described by Wang (1995) for the Pacific, unrelated to the El Niño. Numerical experiments using coupled ocean and atmosphere models and that examine multidecadal variations caused by unstable air–sea interactions between the subtropics and high latitudes in the North Pacific have shown similar patterns of temperature variability in that gyre (Latif and Barnett 1994).
The mode 5 spatial pattern (Fig. 4d, 3.7% variance) highlights the region between the subtropical convergence and the Tasman front. This mode is coherent through the mixed layer to 100-m depth. The subtropical convergence lies just to the south of Tasmania and extends eastward in a meandering pattern across the Tasman Sea to the southern end of New Zealand at approximately 45°S. The RPC scores for this mode indicate a long-period variability of about 15 yr.
The spatial pattern for mode 7 (Fig. 4e, 3.2% variance) occupies an area between the north island of New Zealand and southeastern Australia. The pattern of spatial variation is east and south of the Tasman front (e.g., Stanton 1976; Andrews et al. 1980) and north of the subtropical convergence. It is a surface intensified mode, with the significant loadings describing the upper 30 m. The scores show strong decadal variations over the entire period. This region of the Tasman Sea is a circulation “cul-de-sac,” with the fast-flowing East Australian Current to the north and the eastward-moving Sub-Antarctic mode water to the south. The shallowness of this mode suggests that most of the variability here is due to air–sea interactions rather than ocean advection.
Finally, mode 8 (Fig. 4f, 3% variance) scores show a long-term warming trend in the waters to the east and south of Tasmania and in Bass Strait. Superimposed on this trend are time variations consistent with ENSO. The long-term warming in the southwest Tasman Sea is discussed further in section 5.
2) Analysis 2
Analysis 2 examines correlations in upper-ocean temperatures between the sea surface and 100-, 250-, and 450-m depth during the period 1973–88 using the RPCA. This analysis focuses on the XBT data after 1972. Seven significant modes were retained and rotated based on the scree and LEV plots. These 7 modes explain approximately 36% of the total variance. Two of these modes (modes 3 and 5), presented here, extend the findings in analysis 1. All of the modes for this analysis have comparable features to those presented in analysis 1, although not necessarily in the same order.
The spatial pattern of mode 3 (Fig. 6a, 5.8% variance) shows that there was an overall cooling in the center of the subtropical gyre between 1973 and 1988. In particular, thermocline depths greater than 100 m all cooled during this period. These results have dynamic implications for the ocean circulation because they imply significant changes in the steric sea level across the subtropical gyre. Reiterating the results in the previous subsection, the time amplitudes for mode 5 (Fig. 6b, 4% variance) from analysis 2 complement the mode 2 results from analysis 1 (Fig. 4b). However, mode 5 of this analysis correlates strongly with the SOI, but with a 3-month lead rather than at zero lag. This result is discussed further in section 5.
b. Atmosphere–ocean correlation analysis
1) Analysis 3
Analysis 3 was designed to examine atmosphere–ocean correlations between nonseasonal anomalies of sea level pressure, the zonal and meridional components of the wind stresses, and subsurface temperatures at 250-m depth. The depth of 250 m was selected as a central indicator of movements of the thermocline in association with interannual surface wind and pressure changes observed between 1973 and 1988. Seven modes were again retained and rotated based on the scree and LEV plots.
Mode 1 (Fig. 7, 8.6% variance) is the only mode arising from this analysis that showed a significant correlation between the ocean temperature anomalies at 250-m depth and the atmospheric variables. The higher modes showed significant correlations between the atmospheric variables themselves but not with the 250-m ocean temperature anomalies. A number of attempts were made to reweight the variables jointly in order to ensure that each variable was given equal importance. This did not improve the correlation between ocean and atmosphere, and led us to the conclusion that for the higher modes the ocean temperature field is not correlated with local atmospheric forcing, but instead is likely to be responding to forcing outside the data domain. In particular, the curl of the wind stress is strongest in the eastern Pacific. Thus, only the first mode is presented here. For details of other modes see Holbrook (1994).
This first mode shows the dominant ENSO ocean–atmosphere pattern for the southwest Pacific. The spatial pattern shown in the first panel, associated with ocean temperatures at 250-m depth, is characteristic of the dominant ENSO modes in each of the two intraocean analyses (analyses 1 and 2). From the second panel it is apparent that a large pressure anomaly pattern extends southeastward from the northwest of the region. The scores highlight the higher-than-“normal” pressures that persisted in this region during the 1982–83 and 1986–87 ENSO events. This sea level pressure pattern is characteristic of the western arm of the planetary-scale Southern Oscillation, corresponding to the out-of-phase pressure relationship between the Indonesian–Australian region and the southeast Pacific.
Predominantly, the wind stress anomaly patterns (lower two panels in Fig. 7) simply describe changes in the geostrophic component of the winds associated with the regional-scale pressure changes. However, it can be seen that there is an eastward component of anomalous winds in the northeast of the region of opposite sign to the band extending southeastward from Papua New Guinea. These westerly (eastward) anomalies in the 1982–83 and 1986–87 ENSO periods represent the relaxation of the easterly winds along and near the equator and, more particularly, the presence of anomalous westerly winds, which were observed during the 1982–83 ENSO (e.g., Rasmusson and Wallace 1983). The maximum correlation coefficients at zero- and 3-month lags (RPC scores lagging the SOI) between the mode 1 scores and the five-point filtered SOI are both 0.63 and are significant at the 99% level.
5. Discussion
The main aim of this study was to identify and understand important climate signals in the upper South Pacific Ocean, west of 180°E. Rotated principal components analysis was the primary tool used to describe the interannual and decadal upper-ocean temperature variability in this region and the important processes that these modes represent. The spatial patterns of the significant RPCs associated with analyses 1 and 2 show striking similarities in the patterns of the major water masses and frontal systems in the Tasman Sea, thus connecting classical water mass analysis with statistical techniques.
The dominant El Niño pattern from the analyses is represented as a zonal line extending eastward from Papua New Guinea, with maximum RPC loadings in excess of 0.6 at the depth of the thermocline. The largest time amplitudes corresponding to this pattern are associated with the 1982–83 and 1986–87 El Niños, and describe the large vertical movements of the thermocline across the tropical southwest Pacific Ocean during these events. Furthermore, lagged correlations between the mode scores and the SOI (maximum correlations at 3- to 6-month lags) suggest that the largest and most persistent vertical excursions of the thermocline in the tropical southwest Pacific Ocean are a response to and not a cause of the large-scale pressure differences across the Pacific Ocean.
A coherent regionalization of positive RPC loadings extending southeastward from Papua New Guinea at the sea surface is observed in mode 2 from analysis 1 and in mode 5 from analysis 2 (Figs. 4b and 6b). Below 60-m depth the temperature loadings are small and localized. In fact, loadings for 100-m depth are poorly correlated with those above this level, indicating that these modes primarily reflect surface (or mixed layer) temperature perturbations independent of the large-scale circulation.
Maximum correlation coefficients of 0.61 and 0.48 were found at 3- and 6-month leads (significant at 99% level) between the mode 5 temperature scores from analysis 2 and the SOI. There is some consistency between these findings and the suggestions of Rasmusson and Carpenter (1982), who indicated that during a typical El Niño (from a composite of El Niño events between 1950 and 1976) sea surface temperature anomalies extending over much of the southwest Pacific Ocean are anomalously cool between August and October of the onset year.
Overall, mode 5 from analysis 2 represents a significant portion of the surface ENSO signal. The sea surface temperature pattern and amplitudes for this mode indicate a cooling of the sea surface during the 1976–77, 1982–83, and 1986–87 El Niño events, as well as during the 1978–80 period, when the SOI remained negative. Zhang and McPhaden (1995) demonstrated that in the equatorial Pacific, for timescales in excess of 90 days and for sea surface temperatures less than about 300.5 K (<27.5°C), latent heat flux increases with sea surface temperature. In contrast, for sea surface temperatures above 300.5 K (>27.5°C), latent heat flux decreases. These variations also reflect changes in the convection and wind strength. Although this region is outside the equatorial zone, it is likely that the sea surface temperature anomalies reported here in mode 5 represent large interannual changes in latent heat fluxes in the region, which are closely linked to the northeastward migration of the South Pacific convergence zone during El Niño (Vincent 1994). These time variations of the sea surface temperature anomalies in the Coral Sea, in advance of the large upward movements of the thermocline across the tropical ocean (nearer to the equator), are a valuable finding and may be a useful indicator for the onset of El Niño.
Once the El Niño has been excited, the dynamic response is to lower the western equatorial sea level due to a raising of the thermocline with a lag of 3–6 months, as observed in mode 1 from analysis 1. This separation between the phases of the surface lead and subsurface responses shown by two distinct RPCs indicates the sensitivity of the analysis to the ocean–atmosphere interactions and El Niño dynamics.
It was also shown that there has been a cooling trend in the subtropical gyre, with variations clearly suggesting that an ENSO-related phenomenon is affecting the gyre on interannual timescales. This ENSO signal is supported by significant correlations between the time amplitudes from mode 3 of analysis 1 (Fig. 4c) and the five-point filtered SOI (98% at zero lag and 99% at 3-month lead, with the ocean leading the pressure index). The warming during 1983 is probably related to the southward migration of the gyre in response to El Niño, as suggested by Wyrtki (1984). On the other hand, the combined pattern loadings and scores for those years prior to and during onset of the El Niños over the 34 yr represent sea surface temperature changes in the center of the gyre, as described by van Loon and Shea (1985), with sea surface temperatures being warmer in the winter and spring of the year preceding onset and cooler during the year of onset. Hence, these sea surface temperature anomalies may be indicative precursors to ENSO.
This point is seen most clearly in the mode 3 time amplitudes over the period 1981–82, when a strong upper-ocean cooling occurred prior to El Niño. Van Loon and Shea (1985) suggest that the sea surface temperature changes in this region and period resulted from atmospheric changes in the midlatitudes of the South Pacific. Other examples of this feature in the time amplitudes for mode 3 of analysis 1 are the similar warmer ocean temperatures during spring (October) of 1956, 1971, 1975, 1981, and 1985, prior to the onset years of the 1957, 1972, 1976, 1982, and 1986 El Niño events, where the subtropical gyre is again cooler. However, this trend is not apparent prior to the 1963 and 1969 events.
There are some similarities between the spatial pattern for mode 3 in analysis 2 at 250-m and 450-m depths, shown in Fig. 6a, and the first empirical orthogonal function pattern representing temperature changes at 450-m depth in a study by Ridgway and Godfrey (1996). In both of these studies, the maximum spatial pattern loadings are centered near New Caledonia. However, the higher time resolution of this study reveals a significant correlation with the SOI. The El Niño component of the RPC scores is as large as the observed decadal decrease over the last 16 yr of the record. We speculate that at the onset of El Niño, when the intensity of the South Equatorial Current reduces, the decreased circulation and relaxed Ekman convergence in the subtropical gyre are dynamically balanced by an upward movement of isotherms in the center of the gyre, and hence a cooling occurs at all depths.
An analogous dynamic mechanism for cooling the gyre, but on greenhouse timescales, was proposed by Church et al. (1991). In their study they investigated regional temperature and steric sea level changes using a simple model based on the concepts of subduction and Sverdrup balance. In their paper they discuss sea level rise using a model of the ocean that permits heating through subduction and horizontal advection. Results from their analyses showed that by preferentially heating the ocean surface at higher latitudes, the Sverdrup balance in the subtropical gyres would be affected, causing large-scale circulation changes. Overall, the circulation would relax due to a reduction in the horizontal density gradients that drive the flow. This relaxation would cause an upward movement of isotherms in the center of the gyre, resulting in a fall in steric sea level. This gyre-scale cooling (suggested by this simple model) is consistent with the subsurface cooling implied by the decreasing trend in the RPC scores in mode 3 of analyses 1 and 2 (Figs. 4c and 6a). However, the significant ENSO variations in the mode scores make it difficult to discern whether the cooling is due to such processes operating on these much longer timescales.
There is evidence that there has been a significant warming in the Southern Ocean. The southwest Tasman Sea region shown in mode 8 of analysis 1 identifies the main long-term warming trend within the southwest Pacific Ocean, north of 50°S. In fact, the spatial pattern identified in this mode exhibits strong similarities to the sea surface temperature changes in this region over the three decades between 1950 and 1989, as described by Parker et al. (1994).
At Maria Island (42°36′S, 148°16′E), off the east coast of Tasmania, a time history of sea surface temperature measurements has been recorded since 1945. In this region sea surface temperatures have been increasing since 1945 (Harris et al. 1988), with recent estimates indicating a rate of increase of about 0.025°C yr−1 (J. Church 1994, personal communication). To further investigate temperature changes in the Maria Island region, the gridded upper-ocean temperature data from the present study were examined at the grid point 43°S, 149°E to 100-m depth. Figure 8a shows the vertically averaged temperature changes for the upper 100 m of the water column at 43°S, 149°E between 1955 and 1988. A linear least squares fit to these data indicates that the upper-ocean temperatures at this location warmed by almost 0.5°C over the 34 yr. This represents a depth integrated warming rate of about 0.015°C yr−1 and a contribution to sea level rise, through thermal expansion, of about 0.3 mm yr−1 for the upper 100 m.
The temperature data were finally examined as a zonal band extending the width of the Tasman Sea meridionally between 39° and 49°S to 100-m depth. After removing the seasonal cycle, combined vertical and horizontal spatial averages were calculated every 3 months between 1955 and 1988. This time series is shown in Fig. 8b, with a least squares fitted line indicating a weak warming trend of about 0.13°C during the 34-yr period. A zero-lag correlation coefficient of 0.47 (significant at the 99% level) exists between these anomalies, detrended by the least squares fitted straight line, and the SOI, indicating that ENSO variations also extend into the high latitudes of the Tasman Sea. The weak warming trend apparent in these upper-ocean temperature data south of 39°S corroborates both the surface temperature changes in the region reported by the Intergovernmental Panel for Climate Change (Houghton et al. 1990) and a recent study by Bindoff and Church (1992).
6. Concluding remarks
Although it is clear from this paper that the dominant ocean temperature anomaly pattern that represents El Niño in the upper southwest Pacific Ocean extends eastward from Papua New Guinea in the tropical ocean, we have also shown that there are significant El Niño relationships over the entire southwest Pacific domain, particularly centered in the subtropical gyre. There are also significant decadal variations and trends in the modes, some of which are plausibly consistent with long-term climate change. However, modeling studies are required to determine whether these trends are related to a response to observed surface warming in the Southern Ocean or are a part of unrelated natural variability with a timescale of decades. It is clear that without the appropriate analysis, long-term trends in the data can easily be aliased with the observed decadal and El Niño variations.
Acknowledgments
Helpful comments were provided by Dr. Charles Macaskill from the School of Mathematics and Statistics, University of Sydney in New South Wales Australia; Dr. Richard Coleman from the Department of Surveying and Spatial Information Science, University of Tasmania in Hobart, Tasmania, Australia; Dr. John Church from the Commonwealth Scientific and Industrial Research Organisation Division of Oceanography in Hobart; Prof. Bill Budd from the Antarctic CRC, University of Tasmania in Hobart; and two anonymous reviewers. The authors are also grateful to Mrs. Judy Davis of the School of Earth Sciences, Macquarie University in North Ryde, New South Wales, Australia, for redrafting some of the computer-generated figures. This project was supported by a National Greenhouse Advisory Committee grant.
REFERENCES
Andrews, J. C., M. W. Lawrence, and C. S. Nilsson, 1980: Observations of the Tasman front. J. Phys. Oceanogr.,10, 1854–1869.
Bindoff, N. L., and J. A. Church, 1992: Warming of the water column in the southwest Pacific Ocean. Nature,357, 59–62.
——, and C. Wunsch, 1992: Comparison of synoptic and climatologically mapped sections in the South Pacific Ocean. J. Climate,5, 631–645.
Church, J. A., J. S. Godfrey, D. R. Jackett, and T. J. McDougall, 1991: A model of sea level rise caused by ocean thermal expansion. J. Climate,4, 438–456.
Craddock, J. M., and C. R. Flood, 1969: Eigenvectors for representing the 500 mb geopotential surface over the Northern Hemisphere. Quart. J. Roy. Meteor. Soc.,95, 576–593.
Davis, R. E., 1976: Predictability of sea surface temperature and sea level pressure anomalies over the North Pacific Ocean. J. Phys. Oceanogr.,6, 249–266.
Fukumori, I., and C. Wunsch, 1991: Efficient representation of the North Atlantic hydrographic and chemical distributions. Progress in Oceanography, Vol. 27, Pergamon Press, 111–195.
Harris, G. P., P. Davies, M. Nunez, and G. Meyers, 1988: Interannual variability in climate and fisheries in Tasmania. Nature,333, 754–757.
Holbrook, N. J., 1994: Temperature variability in the southwest Pacific Ocean between 1955 and 1988. Ph.D. dissertation, University of Sydney, 229 pp.
Houghton, J. T., G. J. Jenkins, and J. J. Ephraums, Eds., 1990: Climate Change. Cambridge University Press, 365 pp.
Hsieh, W., and B. V. Hamon, 1991: The El Niño–Southern Oscillation in southeastern Australian waters. Aust. J. Mar. Freshwater Res.,42, 263–275.
Kaiser, H. F., 1958: The varimax criterion for analytic rotation in factor analysis. Psychometrika,23, 187–200.
Latif, M., and T. P. Barnett, 1994: Causes of decadal climate variability over the North Pacific and North America. Science,266, 634–637.
Meyers, G., and J. R. Donguy, 1984: South Equatorial Current during the 1982–83 El Niño. Tropical Ocean–Atmos. Newsl.,27, 10–11.
National Oceanographic Data Center, 1991: CD-ROMs NODC-02 and NODC-03: Global ocean temperature and salinity profiles. National Oceanographic Data Center Informal Rep. 11, 14 pp. [Available from National Oceanographic Data Center, User Services Branch, NOAA/NESDIS E/OC21, 1825 Connecticut Ave. NW, Washington, DC 20235.].
North, G. R., T. L. Bell, R. F. Cahalan, and F. J. Moeng, 1982: Sampling errors in the estimation of empirical orthogonal functions. Mon. Wea. Rev.,110, 699–706.
Overland, J. E., and R. W. Preisendorfer, 1982: A significance test for principal components applied to cyclone climatology. Mon. Wea. Rev.,110, 1–4.
Parker, D. E., P. D. Jones, C. K. Folland, and A. Bevan, 1994: Interdecadal changes of surface temperature since the late nineteenth century. J. Geophys. Res.,99, 14 373–14 399.
Rasmusson, E. M., and T. H. Carpenter, 1982: Variations in tropical sea surface temperature and surface wind fields associated with the Southern Oscillation/El Niño. Mon. Wea. Rev.,110, 354–384.
——, and J. M. Wallace, 1983: Meteorological aspects of the El Niño/Southern Oscillation. Science,222, 1195–1202.
Richman, M. B., 1986: Rotation of principal components. J. Climatol.,6, 293–335.
——, and P. J. Lamb, 1985: Climatic pattern analysis of 3- and 7-day summer rainfall in the central United States: Some methodological considerations and a regionalization. J. Climate Appl. Meteor.,24, 1325–1343.
Ridgway, K. R., 1995: An application of a new depth correction formula to archived XBT data. Deep-Sea Res.,42, 1513–1519.
——, and J. S. Godfrey, 1995: Long term temperature and circulation changes off eastern Australia. J. Geophys. Res.,101, 3615–3627.
——, ——, G. Meyers, and R. Bailey, 1993: Sea level response to the 1986–1987 El Niño–Southern Oscillation event in the western Pacific in the vicinity of Papua New Guinea. J. Geophys. Res.,98, 16 387–16 395.
Roemmich, D., and B. Cornuelle, 1990: Observing the fluctuations of gyre-scale ocean circulation: A study of the subtropical South Pacific. J. Phys. Oceanogr.,20, 1919–1934.
Salinger, M. J., and Coauthors, 1996: Observed variability and change in climate and sea-level in Australia, New Zealand and the South Pacific. Greenhouse 94: Coping with Climate Change, W. J. Bouma, G. I. Pearman and M. R. Manning, Eds., CSIRO Publishing, 100–126.
Stanton, B. R., 1976: An oceanic frontal jet near the Norfolk Ridge northwest of New Zealand. Deep-Sea Res.,23, 821–829.
Troup, A. J., 1965: The Southern Oscillation. Quart. J. Roy. Meteor. Soc.,91, 490–506.
van Loon, H., and D. J. Shea, 1985: The Southern Oscillation. Part IV: The precursors south of 15°S to the extremes of the oscillation. Mon. Wea. Rev.,113, 2063–2074.
Vincent, D. G., 1994: The South Pacific convergence zone (SPCZ): A review. Mon. Wea. Rev.,122, 1949–1970.
Wang, B., 1995: Interdecadal changes in El Niño onset in the last four decades. J. Climate,8, 267–285.
Woodruff, S. D., R. J. Slutz, R. L. Jenne, and P. M. Steurer, 1987: A Comprehensive Ocean–Atmosphere Data Set. Bull. Amer. Meteor. Soc.,68, 1239–1250.
Woodworth, P. L., 1991: The permanent service for mean sea level and the global sea level observing system. J. Coastal Res.,7, 699–710.
Wyrtki, K., 1984: A southward displacement of the subtropical gyre in the South Pacific during the 1982–83 El Niño. Tropical Ocean–Atmos. Newsl.,23, 14–15.
——, 1985: Sea level fluctuations in the Pacific during the 1982–83 El Niño. Geophys. Res. Lett.,12, 125–128.
Zhang, G. J., and M. J. McPhaden, 1995: The relationship between sea surface temperature and latent heat flux in the equatorial Pacific. J. Climate,8, 589–605.
Spatial positions and frequency histogram of all MBT casts retained after quality control of the data. Individual subregions are numbered accordingly, while their boundaries are defined to overlap 1° beyond their respective marked bounding lines.
Citation: Journal of Climate 10, 5; 10.1175/1520-0442(1997)010<1035:IADTVI>2.0.CO;2
Spatial positions and frequency histogram of all XBT casts retained after quality control of the data. Individual subregions are numbered accordingly, while their boundaries are defined to overlap 1° beyond their respective marked bounding lines.
Citation: Journal of Climate 10, 5; 10.1175/1520-0442(1997)010<1035:IADTVI>2.0.CO;2
The upper graph shows a scree plot of eigenvalues against increasing eigenvector number for the first 30 eigenvalues from the nonseasonal PCA of the correlation matrix containing normalized temperature anomalies at the sea surface and 30-, 60-, and 100-m depth for analysis 1. Also shown are the 95% confidence levels (indicated by the dashed line) from a Monte Carlo simulation (Overland and Preisendorfer 1982) and sampling errors on the eigenvalues, according to North et al. (1982). Natural log eigenvalues replace the eigenvalues in the lower representation.
Citation: Journal of Climate 10, 5; 10.1175/1520-0442(1997)010<1035:IADTVI>2.0.CO;2
Modes (a) 1, (b) 2, (c) 3, (d) 5, (e) 7, and (f) 8 of the 8 retained modes from a varimax rotated principal components analysis of the data correlation matrix containing normalized temperature anomalies at the sea surface and 30-, 60-, and 100-m depth (analysis 1). The spatial patterns are the RPC loadings, while the time series represents the corresponding amplitudes. Solid contour lines represent positive RPC loadings, while the dashed contours represent negative loadings. The contour interval is 0.2.
Citation: Journal of Climate 10, 5; 10.1175/1520-0442(1997)010<1035:IADTVI>2.0.CO;2
(Continued )
Citation: Journal of Climate 10, 5; 10.1175/1520-0442(1997)010<1035:IADTVI>2.0.CO;2
(Continued )
Citation: Journal of Climate 10, 5; 10.1175/1520-0442(1997)010<1035:IADTVI>2.0.CO;2
Five-point filtered monthly sea level anomalies (mm) from the seasonal mean at Noumea (New Caledonia). The monthly mean sea level data are from the PSMSL holdings, relative to the Revised Local Reference.
Citation: Journal of Climate 10, 5; 10.1175/1520-0442(1997)010<1035:IADTVI>2.0.CO;2
Modes (a) 3 and (b) 5 of the 7 retained modes from a varimax rotated principal components analysis of the data correlation matrix containing normalized temperature anomalies at the sea surface and 100-, 250-, and 450-m depth (analysis 2). The spatial patterns are the RPC loadings, while the time series represents the corresponding amplitudes. Solid contour lines represent positive RPC loadings, while the dashed contours represent negative loadings. The contour interval is 0.2.
Citation: Journal of Climate 10, 5; 10.1175/1520-0442(1997)010<1035:IADTVI>2.0.CO;2
Mode 1 of the 7 retained modes from a varimax rotated principal components analysis of the data correlation matrix containing normalized anomalies of temperature at 250-m depth, sea level pressure, and the zonal and meridional components of the wind stresses, respectively (analysis 3). The spatial patterns are the RPC loadings, while the time series represents the corresponding amplitudes. Solid contour lines represent positive RPC loadings, while the dashed contours represent negative loadings. The contour interval is 0.2.
Citation: Journal of Climate 10, 5; 10.1175/1520-0442(1997)010<1035:IADTVI>2.0.CO;2
(a) Vertically averaged temperature changes (anomalies from the seasonal mean) for the upper 100 m of the water column at 43°S, 149°E. These results (solid line) are from the gridded upper-ocean temperature time series data between 1955 and 1988 generated in the present study. A linear least squares fit to the data is also shown (dashed line) and suggests that the upper 100 m of the water column has warmed by almost 0.5°C during this period. (b) Average temperature change in the upper 100 m of the water column for the zonal band extending between 141° and 179°E and meridionally between 39° and 49°S for the period 1955–88 (solid line). Individual values were calculated as the horizontal and vertical average of the differences between the temperature time series and the seasonal cycle. A first-order least squares fit to the data (dashed line) is also shown, indicating a weak warming trend of about 0.13°C during the 34-yr period. The zero-lag correlation coefficients were, respectively, 0.37 (raw data) and 0.47 (detrended data) between the temperature anomalies and the SOI, and were significant at the 99% level. (c) Troup’s Southern Oscillation index between January 1955 and December 1988. The gray bar graph representation is the monthly mean SOI. Superimposed is the five-point filtered SOI used in this study and indicated by the solid line.
Citation: Journal of Climate 10, 5; 10.1175/1520-0442(1997)010<1035:IADTVI>2.0.CO;2
Summary of the three experiments performed using the rotated principal components analysis. “Nonseasonal” indicates that the seasonal cycle has been removed from the data. Here, T250 represents temperatures at 250-m depth; τx and τy are, respectively, the zonal and meridional components of the wind stresses; and SLP is sea level pressure.
