1. Introduction
Recent observational studies have shown significant and consistent changes in Antarctic Intermediate Water (AAIW) and Sub-Antarctic Mode Water (SAMW) properties along sections in the Pacific and Indian Oceans over the 20–30 yr between the 1960s and about 1990. Bindoff and Church (1992) and Johnson and Orsi (1997) found statistically significant cooling and freshening of these water masses on neutral density surfaces on several pairs of repeat sections in the southwest Pacific Ocean over this period. For a number of other important sections in the Pacific and Indian Oceans, section data from the earlier decade do not exist or were not considered suitable; however, by objectively mapping distributed historical data onto modern synoptic World Ocean Circulation Experiment (WOCE) sections, Wong et al. (1999, 2001) found similar changes on five sections in the Pacific Ocean over the same period, as did Bindoff and McDougall (2000, hereafter BM) on the 32°S section in the Indian Ocean.
Interior properties can be regarded as the integrated and smoothed result of rapidly changing and inadequately observed surface processes. Understanding how changes in these properties come about may be important in the context of climate change. From an analysis of the Third Hadley Centre Coupled Ocean–Atmosphere GCM simulations (Gordon et al. 2000), Banks et al. (2000) concluded that the observed patterns of cooling, and hence freshening, on isopycnals were more similar to those diagnosed from an integration with anthropogenically caused increasing levels of greenhouse gases than from one in which the atmospheric components remained constant. They found SAMW in the Indian Ocean to be a particularly sensitive indicator of possible anthropogenic change because of its homogeneity in potential density and high volume production [estimated to be 19.8 Sv (1 Sv ≡ 106 m3 s−1) for the range σθ = 26.52–26.80 kg m–3 by Marsh et al. (2000)].
SAMW forms in a band of deep mixed layers that develop as a result of wintertime convection in the Sub-Antarctic Zone (SAZ) on the northern side of the Sub-Antarctic Front (SAF) in the South Indian Ocean (McCartney 1977, 1982). Maximum densities are attained at the eastern end of the band, where convection reaches a depth of 300–500 m. McCartney (1982) recognized the mode as a pycnostad or potential vorticity (PV) minimum in zonal hydrographic sections taken in the interior of the subtropical gyre using data from the Atlantis II cruises of 1965 at 32°S (Wyrtki 1971) and of 1976 at 18°S (Warren 1981). The potential density of the PV minimum at 32°S, which lies several degrees north and downstream of the SAMW subduction zone, varied between σθ = 26.7 kg m–3 in the west of the section and 26.85 kg m–3 in the east, implying a similar gradient of winter surface densities with longitude in the SAZ.
The entire 32°S section has been traversed four times, by the RRS Discovery in April–May 1936, the RV Atlantis II in June–July 1965, and the RRS Charles Darwin in November–December 1987 and March–April 2002; in addition, the western and eastern ends of the section were surveyed in 1995 by the RV Knorr. The first two full-section cruises closely followed the 32°S parallel and took bottle casts at 5° longitude intervals. The more recent surveys sampled at higher density in longitude and depth using CTDs and followed tracks running essentially east–west but diverging from the 32°S parallel and from each other by up to 3° of latitude.
BM analyzed sectional changes between the 1960s and the WOCE cruise of 1987. Values for the earlier period were projected onto the WOCE I5 section and assigned a mean date of 1962 from casts taken between 24° and 40°S, mostly in the 1950s and 1960s. The analyzed data included an important contribution from casts taken at 32°S on the 1965 Atlantis II cruise. BM found an almost monotonic cooling and freshening of the SAMW (0.54°C, 0.13 psu at 26.8 kg m–3) and AAIW (0.33°C, 0.06 psu at 27.45 kg m–3) on neutral surfaces over the 1962–87 period and showed that the freshening on isopycnals above the intermediate salinity minimum was most likely the result of warmed surface waters.
The analysis of temporal changes has been repeated and extended by Bryden et al. (2003) and McDonagh et al. (2005, hereafter M05), but using only the data from the five synoptic sections. M05 found that there had been increases in salinity of up to 0.093 and 0.069 psu on isopycnals in the western and eastern parts of the section, respectively, in waters above the 10°C isotherm (corresponding to σθ = 26.9 kg m–3 and a depth of 700 m) but decreases (–0.031 and –0.035 psu) below this level. They noted that the recent changes had reversed those of the previous period in the upper thermocline (essentially, the SAMW) but had extended them in the lower thermocline (the AAIW above the salinity minimum).
The purpose of this paper is to review the sectional changes that have taken place during the periods between the cruises of 1965, 1987, and 2002 in the context of simulations performed using an ocean general circulation model. We have attempted to reproduce the subsurface changes by modeling the ocean response to the application of gridded surface forcings from 1948 to the present and have investigated how these changes are related to variations in conditions at the surface. To limit the scope of this study somewhat, we have chosen not to concern ourselves with the processes, such as local heat and/or freshwater fluxes, cross-frontal Ekman transport, and advection by geostrophic currents, which contribute to the water properties found in the formation regions, but rather to investigate the following questions (section 3): given a credible series of sea surface temperature and salinity (SST and SSS) fields in the formation region, can the model reproduce the subsurface changes reported by BM and M05, and what patterns of temporal variation do these form a part? Because the sea surface properties, in particular salinity, are not known with great certainty, there is an issue of providing suitable forcings, and this is discussed in sections 2b and 2d.
2. Method
a. The model
To obtain adequate resolution in the region of interest while minimizing computational expense and allowing some experimentation with forcings, a curvilinear grid with a resolution of 0.6° × 0.8° in the Indian Ocean sector and 4° × 6° on the opposite side of the globe was used. The grid is similar to that employed by Murray and Reason (2001b) but with a further compression of grid contours in the South Indian Ocean. The vertical grid is composed of 21 levels ranging in thickness from 25 m at the surface to 450 m at depth.
The model is an orthogonal curvilinear version of the Geophysical Fluid Dynamics Laboratory (GFDL) model (Murray and Reason 2001a). It includes a Kraus–Turner mixed layer scheme derived from that used by Chen and Rothstein (1994) and isopycnal mixing and eddy-induced transport (implemented as a skew diffusion; Griffies 1998), both of the latter using Cox (1987) numerics. To keep the scale of resolved features approximately in proportion to the grid spacing, the lateral mixing coefficients were made to vary according to grid size, but with minimum values of horizontal viscosity AM = 0.3 × 109 cm2 s–1 and isopycnal and thickness diffusivity AI = AE = 0.5 × 107 cm2 s–1, which applied in most of the Indian Ocean sector. Within the mixed layer, vertical mixing coefficients were set to 250 cm2 s–1; beneath it, a background vertical viscosity (KM) of 1 cm2 s–1 was used, and the vertical diffusivity (KH) followed a profile based on Kraus (1990), with values KH = 0.2 cm2 s–1 in the upper ocean, increasing to 1.4 cm2 s–1 below 4000 m. Convective adjustment of tracers between unstable layers was simulated with diffusivities of KH = 106 cm2 s–1.
The model was spun up to a cyclic equilibrium using National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) Reanalysis 1 (NCEP 1) climatological monthly heat and freshwater fluxes and wind stresses (averaged for each month of the year over the period 1948–2000 inclusive) and a 30-day relaxation to climatological monthly SSTs and SSSs. World Ocean Atlas 2001 (Conkright et al. 2002) climatological monthly 10-m fields were used for the restoration of salinity but not for temperature, which, for reasons of consistency explained in section 2b, was restored to 1948–2002 climatologically averaged SSTs from the Hadley Ice and SST (HadISST) dataset (Rayner et al. 2003). From the equilibrium state, a number of transient simulations were run on using NCEP monthly mean forcings for the period 1948–2002, as discussed in section 2b.
Accelerated convergence with a tracer time step of 1 day was used during the earlier part of the equilibrium integration (Bryan 1984). Time step inequalities were then progressively graded to equal time steps of 1.5 hours at 600 yr, after which the model was run for a further 40 yr. The solution at 640 yr was taken as the initial condition as at 1 January 1948 for the 55-yr transient synchronous simulation. During the final 40 yr preceding the transient simulation, temperature tendencies averaged over zonal regions for each level and for a particular month of the year were mostly about 5 × 10−5 °C yr−1, or 0.003°C over 55 yr. This amount of drift is negligible in comparison with the interannual changes of 0.2°–1°C, which occurred in waters down to 1500 m during the transient simulations.
The model ocean was global except for the Arctic seas, which were considered unimportant for this study and were blocked at the Greenland–Iceland–Scotland Ridge. No sea ice model was used, as the water properties in the (mainly Southern Hemisphere) sea ice regions were forced by prescribed and restoration fluxes in the same way as in open ocean regions. To keep deep-ocean water mass properties approximately correct, tracer values at bottom points below 500-m depth in the path of the Denmark Strait and Weddell Sea outflows were restored to Levitus (1982) climatological late winter values. Wall restorations were not applied at these high-latitude boundaries but were applied between 100 and 1000 m at the locations of the unresolved straits of Gibraltar (March), Bab el Mandeb, and Hormuz (September).
b. Transient simulations
In the seasonal equilibrium simulation, ocean surface tracer values were forced by a combination of prescribed fluxes and sea surface values, the fluxes forcing the annual cycle without lag and the relaxation keeping the surface values close to the climatology. There is a problem in applying this prescription to transient simulations. We have NCEP reanalysis flux data going back to 1948, but not a full interannual time series of surface restoration data. Two historical SST datasets cover the period of interest: the Hadley dataset, previously mentioned, and the Reynolds, Stokes, and Smith Extended Reconstructed SSTs (RSSER) dataset (Smith and Reynolds 2003, 2004); however, there is no comparable monthly mean SSS dataset to use in conjunction with the temperature time series to give the correct surface buoyancy restoration. As a first attempt at solving this problem, we applied monthly mean fluxes in association with a 30-day relaxation to climatological restoration fields for both temperature and salinity (experiment 1). This procedure was expected to capture seasonal variations fairly well but strongly damp interannual and especially interdecadal variations due to ocean advection and storage.
In an attempt to obviate the damping effect, we diagnosed a seasonal climatology of restorative fluxes from the equilibrium simulation and reran the transient simulation using the combined NCEP and diagnosed restorative fluxes. Using only the adjusted fluxes without any restoration based on instantaneous conditions resulted in wide excursions of surface values and some drift over the integration period. As a compromise, a weak 180-day relaxation to surface climatology was also applied. This relaxation has a “restoring power” one-sixth that of the standard 30-day relaxation; so, to keep the seasonal cycle of the restoring fluxes comparable to that of the equilibrium simulation, the adjustment to the prescribed fluxes was made to five-sixths of the diagnosed restorative fluxes (experiment 2).
Another method that was employed was to use reconstructed monthly mean SSTs for the temperature restoration together with a repeating cycle of climatological monthly SSSs for the salinity restoration. The Hadley SSTs were used in this study; however, a test simulation run with the RSSER SSTs showed similar responses. It may be objected that a restoration to interannual SST and climatological SSS fields produces an inconsistent interannual component of the buoyancy restoration. The provisional answer to this is that, at midlatitudes, temperature variations usually have a greater impact on density than do salinity variations; however, this is a question that will need to be critically examined later. Two experiments were performed using the hybrid restoration, both with a 30-day temperature relaxation, one (experiment 3) using a 180-day restoration to SSS and a five-sixths weighting of diagnosed restoration fluxes and the other (experiment 4) using a 30-day relaxation and no flux adjustment.
Two further simulations (experiments 5 and 6) were conducted using low- and high-frequency components of the forcing fields of experiment 4. The purpose of these experiments was to test whether subduction occurs preferentially during years of extreme forcing, in which case epoch-averaged or time-smoothed boundary conditions would not be effective in producing interior water masses of the right properties. The low-pass-filtered forcings were obtained by time-smoothing the interannual series of each forcing field using a diffusive filter with a 4-yr time scale. This was done for each month separately so that the seasonal cycle itself should not be smoothed but should be preserved with full amplitude (as averaged over several years by the smoothing). The high-pass fields for each month of the year were computed from the residual (full minus smoothed) series, to which was added the climatological monthly mean in order to restore the seasonal cycle to the full monthly series.
Experiment 4 has been used as the standard simulation, and results refer to it unless otherwise mentioned. The other five experiments were included for the purpose of checking the appropriateness of the experiment-4 forcing and the reliability of the conclusions. The surface boundary conditions used in experiments 1–6 are summarized in Table 1. All of the forcing combinations notionally include a full weighting of the seasonal cycle of prescribed forcings and 30-day restoration fluxes used in the equilibrium integration when considered over the 55 yr of the experiments. This aids comparison between experiments and justifies the use of a single spinup solution as the initial condition for the six experiments.
3. Results
a. Surface properties
In this study, we were interested in the subduction and dispersal of SAMW associated with the deep mixed layers that form in winter in the SAZ in the South Indian Ocean, and hence we desired a plausible distribution of mixed layer depths (MLDs) in this region. This seems to have been achieved in the 1 September solution from the equilibrium simulation (Fig. 1). A zonally oriented axis of deep mixed layers, increasing in depth eastward to greater than 350 m, extends across the Indian Ocean at latitudes of 40°–45°S as far as the South Australian basin. The partly disconnected region farther north at about 38°S between 105° and 120°E is associated with a loop in the upper-ocean currents extending northwestward from near Cape Leeuwin in southwestern Australia. MLDs > 350 m are also found farther to the east in the ocean south of Tasmania and the Tasman Sea.
Some appraisal of the model’s realism in this region can be obtained by comparing the 1 January 1995 solution (experiment 4) to observations taken on the December 1994 WOCE cruise along the I8S section, which runs north along the 95°E meridian between 45° and 30°N. The sections plotted in Fig. 2 show contours of potential temperature and potential density anomaly (σθ), hereafter referred to as just temperature and density. Although the model contains biases in the position and sharpness of the SAF, both model and observations represent very clearly the large pool of weakly stratified water in the density range 26.5–26.9 kg m–3, which lies north of the SAF and reaches a depth of 700 m. This water mass is the SAMW. During winter, it is entrained into the mixed layer and is exposed to convective renewal, but, during summer, it is capped at about 80 m by a seasonal thermocline, which isolates it from a shallow wind-generated mixed layer of 20–60-m depth. North of 35°S, the SAMW is cut off from the mixed layer above by the permanent thermocline all year: it is therefore not formed locally but has been advected from a source region to the southeast of this part of the section. One feature at odds with observations is the reversal of the horizontal density gradient between 42° and 34°S, which implies an exaggerated westward component of the flow in upper levels.
To identify an area representative of the formation of SAMW, we looked at the principal modes of variability in the model winter SSTs. Figure 3 shows the first empirical orthogonal function (EOF) of the 1 September SST variation and its time series computed over the region 55°–20°S, 20°–145°E from experiment 4. The pattern indicates a large area of coherent temperature changes in the midlatitudes of the southern Indian Ocean, oscillating over a 50–60-yr period, with low values in the 1960s and high values in the 1980s. The east–west trough of low but positive function values (<0.5°C) lying between 35° and 45°S can be attributed to the moderating effect of surface contact with deep mixed layers; this feature is absent in summer. Two areas (marked “W” and “E” in the figure), encompassing the western and eastern parts of the trough in the EOF pattern, were chosen as representative averaging regions for surface conditions in the zone of ventilation.
Figures 4a,b show, respectively, the time series of reconstructed SSTs averaged over region E from the HadISST and RSSER datasets for August and September and the time series of net prescribed heat and evaporative fluxes averaged over the region and the cooling season (March–August) of each year. The 50–60-yr oscillation discovered in the EOF analysis is seen in the heat fluxes and the SSTs, which reached minimum values in 1965 and maximum values in 1987 in this region. The SST graphs for the two datasets agree fairly well in respect of the long-term oscillation and the timing of many individual maxima and minima but show considerable differences in the patterns for particular groups of years and in the amplitudes of year-to-year temperature variations. In both SST analyses, EOFs were used to fill gaps in the data, but with different algorithms. The differences between the curves reflect this and give some indication of the errors that may have been introduced as a result of truncating the EOF series or of extrapolating modern modes of variability in areas that have been less well sampled in the past. There are also differences between the August and September curves for each dataset. These differences are rather more pronounced in the RSSER set, which suggests that they are not necessarily indicative of real changes over the 1-month interval. For comparison with model output, the Hadley SSTs were used and averaged over the two adjacent months.
The time series of the 1 September region E SSTs for experiments 1–4 are compared with the August–September average Hadley SSTs in Fig. 5a. The high-frequency variations are strongly attenuated in the model. This may be due to disagreement between the flux and SST data or to an overly convective response during the cooling season; it is also possible that some of the short-period variability in the observed SSTs is an artifact of the data reconstruction.
The amplitudes of both short- and long-period SST oscillations in the model differ considerably between experiments. However, the phases of the oscillations and the times of the overall minimum and maximum SSTs agree fairly well. Experiment 1 experienced the greatest damping and underestimated the long-period variation in the September regional averages; these effects are the consequence of restoring to a constant annual cycle. In experiment 2, the variation is overestimated, indicating that the weak restoration was insufficient to correct model biases. Experiments 3 and 4, which both used the Hadley SST forcing fields and have very similar curves, provide the best match to the observed SST variation. The shorter-term variations in these experiments tended to follow the prescribed heat flux variations better than the SST variations, however (see Fig. 4b).
The SSS curves for the four model experiments (Fig. 5b) show a general downward trend, which is different from the SST variation. Here the greatest similarity is between experiments 1 and 4, both of which used a 30-day climatological SSS restoration. The short-term variations in experiments 2 and 3 show the same features as experiments 1 and 4, but these are greater in amplitude owing to the weaker relaxation; they also differ from each other and do not follow closely the time series for net evaporation Fig. 4b. This suggests that the model SSS variations may be responding more to advective changes controlled by temperature than to the freshwater forcing. We do not have SSS data with which to compare the curves in this figure. Fortunately, however, the effect of SSS variations on surface density is much smaller than that of SST variations, as can be seen from the similarity of the density curves for experiments 3 and 4. This and the fact that the amplitudes of the short-term SSS variations are not known from data were reasons for choosing experiment 4 as the standard experiment in this study.
Winter SSTs and MLDs from experiment 4 are shown for regions W and E in Fig. 6. In each region, MLDs are well correlated with SST, the lowest SSTs being associated with the deepest mixed layers. The sensitivity of MLD to SST is greater at high than at low frequencies. The MLDs in region E are not only deep but vary widely, between 250 and 450 m. The weaker-than-observed stratification down to 500 m in the model (Fig. 2) suggests that it may have overpredicted the depth and homogeneity of the winter mixed layer in the SAZ. This would partly explain the damping of short-period winter SST variations seen in Fig. 5a. During the colder period preceding 1975, minimum SSTs and maximum MLDs in the east occurred later than in the west, but, in the warmer period following, the variations were more in phase.
b. Sectional comparison
In Fig. 7, the model solution for 1 January 1988 interpolated to the 15 section (Fig. 7b) is compared to the (December) 1987 WOCE cruise data (Fig. 7a). The model reproduces most of the observed large-scale features, although gradients associated with the Agulhas and Leeuwin Currents are rather too broad. In both model and observed sections, a shallow near-surface stratification overlies a more uniform subsurface layer containing the SAMW. In consequence of the gentle upward slope of the density contours to the east across the gyre, the stratified surface layer is thicker (>200 m) in the eastern half of the section than in the western half. The model shows the shallower stratification in the western half of the section to be unrealistically weak and to be broken down in the winter months.
The SAMW occupies the layer of slack density gradients below 200-m depth. Its core is distinguished by the shading for static stabilities N 2 < 7.5 × 10−6 s–2. In the cruise section, the stability minimum lies at a greater density and depth on the eastern side of the section (26.8 kg m–3, 500 m) than on the western side (26.6 kg m–3, 400 m). The density variation is consistent with that shown by McCartney (1977) based on the 1965 data. In the model, the mode water core shows qualitatively similar characteristics, but the density of the stability minimum varies more strongly from east to west. In the eastern part of the section it is 26.7 kg m–3 and not greatly different from that observed, but west of 80°E it decreases to less than 26.0 kg m–3, causing the isopycnals (and isotherms) to bend sharply upward to the east across the pycnostad layer. In both model and observations, the isotherms and isopycnals are fairly closely aligned, implying that temperature and salinity vary only weakly on isopycnals and are essentially functions of density; this characteristic is what Bryden et al. (2003) referred to as a “tight” temperature–salinity (T–S) relationship.
c. Sectional differences
Within the 55-yr period in which the model could be integrated using NCEP forcings, there have been three cruises that have traversed the Indian Ocean at 32°S, providing data for calculating differences over two time intervals. The methods used for analyzing the sectional changes in the two intervals varied in the two studies cited. BM showed changes on neutral density surfaces between a section analyzed from historical data centered on 1962 and the 1987 WOCE section, while M05 showed changes between the 1987 and 2002 sections on isotherms. To keep the comparison consistent, we have used only cruise data and have analyzed changes on σθ surfaces for the periods 1965–87 and 1987–2002 (i.e., 1987 minus 1965 and 2002 minus 1987 differences). The tracks for all three cruises diverged from one another over 3° of latitude. We have computed differences by longitude, ignoring the separation of latitudes. This neglect is justified by the tight T–S relationship mentioned above.
On the 1987 and 2002 cruises, data were sampled at ½°–1° intervals; this allows us to compute difference fields with good resolution after interpolation to a ½° grid. Figure 8 shows changes between these dates as differences in isopycnal depth and temperature. The depth differences contain many vertically coherent structures a few degrees in extent caused by the presence of eddies. The analysis on isopycnal surfaces largely removes this source of noise from the temperature plot below the 200-m depth contour (shown dashed). Figure 8b is essentially the same as Fig. 6 of M05, except for the change of vertical coordinate, and shows a basinwide warming of the mode water above the 26.9 kg m–3 isopycnal and a cooling below, corresponding to the qualitatively similar changes in salinity that they analyzed above and below the 10°C isotherm.
To facilitate a comparison of the 1987–2002 with the 1965–87 period, which included data from the Atlantis II cruise, on which stations were separated by 5° intervals, all three cruise sections (1965, 1988, and 2002) were interpolated to a 1° grid and horizontally smoothed before differencing on isopycnals. Figure 9 shows the observed sectional changes on σθ surfaces for the two periods (Figs. 9a–d) and the corresponding model changes (Figs. 9e–h, discussed later). The observed temperature difference section for 1965–87 (Fig. 9c) shows a conspicuous basinwide cooling on isopycnals between 26.5 and 27.0 kg m–3 and a weaker, less uniform cooling at deeper levels. The plot corresponds to Fig. 9 of BM, although the details are different because of the exclusion of nonsynoptic data, mainly from the 1960s, in our analysis. M05 showed that there had been a near reversal of these changes during the 1987–2002 period in waters above the 10°C isotherm but a persistence of the freshening (cooling) trend below; this can be seen in isopycnal coordinates in our analysis by comparing Figs. 9c and 9d. Interestingly, isopycnal depths too have undergone basinwide changes in the same density range and, once again, a near reversal, in which the whole pattern of change exhibited during the 1965–87 period has been preserved, but with opposite sign.
The largest changes are found in the density range 26.5 and 26.9 kg m–3. This range corresponds to that of the SAMW, the core of which is indicated by a dashed contour delimiting the mean N 2 < 10 × 10−6 s–2 pycnostad. It will be seen that the axes of maximum depth and temperature change are closely aligned with that of the pycnostad, which, as also shown by McCartney (1982), slopes upward slightly to the west. M05 remarked similarly that the maximum 1987–2002 property changes occur near the temperature of the resident mode water at each end of the section, but it will be seen in Figs. 9a–d that it applies to both periods and both variables.
The broad-scale changes are shown as zonally averaged difference profiles in Fig. 10. The observed profiles (solid curves) are all similar in form, each being characterized by large changes in the density range 26.4–26.9 kg m–3, smaller changes of opposite sign for σθ < 26.4 kg m–3, and small changes of varying sign for σθ > 26.9 kg m−3. Differences in the major features of each variable for 1987–2002 are of opposite sign to those for 1965–87 and of only ½–⅔ the amplitude. Maximum changes for each variable occurred at a density of about 26.7 kg m–3, the peaks for 1987–2002 being at slightly lower densities than for 1965–87 and those for isopycnal depth being at slightly lower densities than for water properties. Maximum temperature changes for 1965–87 and 1987–2002 were –0.7° and 0.4°C, corresponding to salinity changes of –0.15 and 0.08 psu, respectively. Maximum isopycnal depth changes for the two periods were 100 and –50 m. The 100-m 1965–87 maximum shown here is somewhat larger and more peaked than that of BM, which was 70 m and spread over the range γn = 26.7–27.0 kg m−3. The lower and more dispersed maximum of BM was probably a consequence of including data from (and effectively averaging across) a range of latitudes and times.
The sectional differences simulated by the model (experiment 4) corresponding to those observed are shown in Figs. 9e–h. Because there is very little seasonal variation below the thermocline, it was convenient to compute differences from fields for 1 January in the same summer cruise season. Fields were interpolated to the WOCE 15 line, not to the track followed by the survey in that season. The latter produced noisier differences, caused by unrealistic spatial variations being aliased as time variations (especially near the west coast of Australia), and was judged inappropriate for computing model time changes.
The model difference sections (Figs. 9e–h) reproduce the main qualitative features of the observed, as can also be seen by comparing the model profiles (dashed lines) with the observed profiles (solid lines) in Fig. 10. The major changes occurred in the density range 26.4–26.8 kg m–3 and were of the same signs as observed, namely, a cooling and deepening of isopycnals during the first period and a warming and shallowing during the second. Maximum changes were at slightly lower densities in the second period, also as observed. On the other hand, the model differences did show a couple of biases. The axes of maximum change lie in the correct density range in the eastern half of the section but slope upward to the west much more than in the observations. These axes are, however, closely aligned with the time-mean model pycnostad (indicated by the 10 × 10−6 s–2 static stability contour), which also shoals more than in reality toward the west side of the section: in this respect, their behavior is consistent with that observed. The slope in the axes causes the bulges in the model zonal average profiles to be broader and displaced upward compared to the data. At any one longitude, the maxima are more sharply peaked. The other bias is that changes in temperature in the model tend to be concentrated in the eastern part of the section, rather than being spread uniformly across it as in the observations.
The model differences alter signs below the levels of maximum change right across the section. In the observed differences, this is seen for the second period, where alterations of sign occurred at about 26.9 kg m–3, but not for the first period, where none occurred. The disagreement in the sign of the changes at densities >26.9 kg m–3 may be partly fortuitous, because these changes were small and variable in longitude. In any event, it is probably not of significance. The deeper intermediate water has a longer renewal time than the mode water, and there may not have been sufficient time for the advection of surface forcing anomalies to have reached these depths during the earlier part of the integration. Such changes as were simulated at these levels may have been due to latitudinal displacements of current systems and changes in subsurface mixing.
d. Temporal variation in the model
The value of using a model is to show how interior properties may have varied between occasional section occupations and how the variations may have arisen. Figure 11 shows the time series of the temperature and depth of the 26.7 kg m–3 isopycnal for experiments 1–6 and of the 27.0 kg m–3 isopycnal for experiment 4, in each case averaged over the eastern part of the 15 section (80°–105°E). In each of the experiments with unaltered forcing (experiments 1–4), the temperature variation in the eastern sector is made up of a low-frequency component with a 50–60-yr period and, superposed upon it, 5–10-yr oscillations with maxima occurring in 1952, 1959, 1968, 1975, 1982, and rather irregularly afterwards. The pattern of variation for each experiment closely matches that of the corresponding model winter SST curve for region E (bold curve), shown in Fig. 6a, except that the sign is reversed for both high- and low-frequency oscillations and the overall maximum and minimum temperatures occur slightly later (1968 and 1990) than the corresponding minimum and maximum SSTs.
A reversal of sign was shown by BM to be the expected result of analyzing temperature variations on isopycnals in the region of positive T–S gradient above the salinity minimum when surface density changes are driven by changes in temperature alone. The reason for this is briefly reviewed in Fig. 12a. If water at the points along the T–S curve CC′ are warmed as indicated by the vector OW for water at O having the density of the isopycnal II′, the curve will be displaced upward and leftward to AA′. Although the temperature of individual water parcels has increased, the temperature at which the isopycnal II′ intersects the T–S curve has decreased (from O to U). In practice, surface temperature and salinity changes occur together, and a warming could be associated with a salinity change of either sign. When advective processes are important, temperature and salinity variations will often be partially density compensated, that is, of the same sign, as in the case of vectors OX and OY. In this case, the sign of the isopycnal temperature change will depend on the degree of compensation of the temporal surface variations relative to that of the mean spatial, that is, depth, variations in the water mass. When the ratio of the temperature change to the salinity change is greater than the gradient of the T–S curve (CC′), as for the vector OX, or the changes are of opposite sign, as for the vector OV, the sign of the temperature change on isopycnals will be reversed as previously envisaged; when it is less, that is, the temporal variations are more compensated than the mean, as for the vector OY, the T–S curve will be displaced to the right (to BB′), and the sign of the isopycnal change will be the same as the surface change.
The winter surface T–S variations in region E for Experiments 1–4 shown in Figs. 12b–e present in another way the information given in Fig. 5. All of the curves possess common features but are spread out differently: the temperature variance reflects the degree of restraint (experiment 1) or freedom (experiment 2) in the thermal boundary conditions; the salinity variance does likewise for the freshwater boundary conditions, being smallest in experiments 1 and 4, which used a 30-day climatological salinity relaxation. The main point of interest here is that the scatterplots, in which distances on the horizontal and vertical axes have been scaled equally for their effects on density, show a clear temperature dominance; specifically, the major axes of variation (indicated by the thin solid lines) all slope upward and to the left of the gray dashed curve, which represents the time-mean vertical T–S variation of the mode water averaged over the eastern part of the 15 section. This relationship is consistent with the reversal of temperature change on isopycnals, noted above. A similar relationship was found between the surface variations in region W and the mean T–S curve for 40°–70°E (not shown).
The low- and high-frequency oscillations implied in the curve for experiment 4 have been separately reproduced, with fair agreement in amplitude and phase, by experiments 5 and 6, which were forced by the corresponding components of the experiment-4 boundary conditions (Fig. 11; see also Fig. 6a). This suggests a linear response; however, it is not quite linear because the maxima, corresponding to years of cold surface conditions, are more peaked than the minima. While this has tended to amplify the interannual oscillations to some extent, it is the low-frequency oscillation that has the greater amplitude and the greater role in explaining the observed changes in the SAMW.
The variation of depth of the 26.7 kg m–3 isopycnal for each experiment in the eastern sector over the integration period (Fig. 11b) is very similar in form to that of temperature, except for the truncation of the minimum in the 1960s, which is due to the shoaling of the isopycnal being suppressed by the stratification at the top of the modal layer. The reason for the similarity of the graphs is made clear in Fig. 13, which shows the contours of temperature and density as functions of time and depth in the western and eastern parts of the section. In the eastern part of the section in experiment 4 (Fig. 13b), large changes occurred throughout the SAMW range, from the base of the surface stratification to σθ ≈ 26.9 kg m–3 at 750-m depth. The shallowing of isopycnal surfaces during the 1950s and 1960s, when the surface forcing was colder than normal, was due to an expansion of the denser mode water layers (σθ > 26.80 kg m–3) and a contraction of the lighter layers; in the following period (1970s to the present), the reverse occurred. The cycle is not sinusoidal but possesses an asymmetry in which each successively denser layer expanded at the expense of the one immediately above it until the densest layer itself decayed and the cycle began anew from the top in about 1975–80. The asymmetry is only moderate in experiment 4 (the standard experiment) but is very obvious in experiment 2 (Fig. 13c), in which the restoration was weaker and the variations in surface properties and mixed layer depths were greater.
In the western part of the section (Fig. 13a), the large thickness changes took place in a lower density range (26.1–26.4 kg m–3) and closer to the surface (down to 500 m) than in the eastern part, and the variations of temperature relative to isopycnals spanned a smaller range (½°C as compared to 1½°C). Many of the upward excursions of the 26.2 and 26.3 kg m–3 isopycnals (1951–52, 1956, 1961–62, 1965, 1974, 1982, and 1989) match concurrent SST minima in region W (Fig. 6a), indicating that waters lighter than 26.4 kg m–3 are the result of convection close to the section in that half of the ocean. In the longer-period part of the variation, minimum isopycnal depths and maximum isopycnal temperatures occurred earlier in the western part of the section than in the eastern part. Maximum isopycnal depths also occurred earlier in the west than in the east, but by a smaller lead (in 1986 as compared to 1990). The comparative timing is in agreement with that seen in the SST variations for regions W and E shown in Fig. 6a. There appears to have been a fairly rapid transition following 1965, when the mode water changed from being mostly denser than 26.4 kg m–3 to mostly lighter.
The redistribution of water in density classes is reflected in changes in the density of the stability (or PV) minimum (middle gray dashed curve in each panel of Fig. 13). In the east of the section, this varied over the range 26.86–26.69 kg m–3 (0.17 kg m–3) and, in the west, over 26.3–26.2 kg m–3 (0.1 kg m–3). This degree of variation needs to be reconciled with the results of M05, who diagnosed a change of 0.07 kg m–3 in the west of the section between 1987 and 2002 but very little change (–0.01 kg m–3) in the east, implying density compensating changes in the mode water of that sector. To resolve the discrepancy, we have compared observed and modeled changes in the density of the stability minimum for 1965–87/88 and 1987/88–2002 in Table 2. The modeled changes are mostly a little larger (i.e., less compensating) than the observed, which may be due to the temperature dominance in the forcing in experiment 4; however, they are of the same sign and vary in proportion to them. In particular, the modeled change for the eastern sector in the 1987/88–2002 period agrees with the cruise data in being near zero. The smallness of this change can be seen from Fig. 13b to be due to the stability minimum reaching its lowest density for the 55-yr simulation in 1992, in the middle of this time interval. If the model results are any guide, the smallness of the observed 1987–2002 change should not be taken as typical of decadal variations in the eastern part of the South Indian Ocean gyre or as an indication that temperature and salinity changes in the formation region are normally density compensating.
Changes also occurred in the value of the stability at its minimum. This was measured as the thickness of the layer lying between 0.05 kg m–3 on either side of the density of the minimum (Table 2, lower part). Again, model changes were greater than those observed but were in the same direction; in each case, the layer thickness decreased from 1965 to 1987/88 and then increased again to 2002. The variation in thickness with time is shown by the distance between the outer gray dashed curves in Fig. 13. In the 80°–105°E sector, the layer thickness was 330–440 m before 1976, 180–230 m between 1979 and 1992, and >330 m after 1998. The small layer thicknesses in 1979–92 correspond to the period in which the denser mode waters were being replaced by lighter ones, causing a crowding of isopycnals (Fig. 13). A contraction of layer thicknesses and transition to lower mode water densities also occurred in the 40°–70°E sector, but the tabulated model thickness changes do not represent the time interval in which this occurred (1967–73) and are due mostly to interannual variations.
Below the limit of SAMW influence, which extends to about 26.4 kg m–3 in the western half of the section and 26.9 kg m–3 in the eastern half, changes in isopycnal depths and temperatures were small, smoothly varying, and delayed by a decade or more with respect to mode water changes. The implication is that these denser waters are not formed locally but are the result of advection by the subsurface currents and gradual subsidence in the gyre.
The similarity in form of the 1987–2002 and 1965–87 difference patterns, whether modeled or observed, in Fig. 9, and the consistency of the time variations with depth in Fig. 13 suggest that most of the variability on the section may be explained by a few principal modes. The sample (two difference sections) is too small to test this with the data, but the first and second EOFs of the model section temperature series (not shown) were found to explain the major part of the variation (40% and 23%) and closely resembled the difference patterns for the intervals (1965–87 and 1987–2002), respectively. Their amplitude time series were dominated by the 50-yr variation remarked upon earlier, the first being roughly in phase with the surface forcing and the second 90° out of phase.
e. Anomaly propagation
Because subsurface velocities vary little by year or season and tend to follow isopycnals, it was decided to approximate trajectories by streamlines using climatological horizontal velocities interpolated to an appropriate isopycnal surface. Velocity fields for 1 January were used because the isopycnals of interest do not outcrop in the formation region at this season. Figure 14 shows 10-yr streamlines integrated backwards on the σθ = 26.75 kg m−3 surface from points at 5° longitude intervals along the 15 section. In the model, the section is located north of the axis of the gyre, and subsurface water arriving at the section at all points east of the center of the recirculation at 35°E approaches from the southeast. Points on the σθ > 26.75 kg m−3 level in the eastern part of the section receive water that has come from south of Australia and as far east as Tasmania but that will have been recently modified upon passing through the convective region southwest of Western Australia.
Time–distance sections constructed along back streamlines give an indication of the time of subduction. Diagrams of 1 January temperature and isopycnal depth from the transient simulation have been plotted in Fig. 15 as functions of calendar year and position on the 26.75 kg m−3 back streamline ending at 80°E on the section. The “position” on the horizontal axis is measured in years relative to the time of arrival at the section assuming spatially variable but temporally constant velocities taken from the 1 January solution from the equilibrium simulation. Signal propagation can be recognized in the upward sloping troughs and ridges in the contour patterns, but this applies only after (to the right of) the point at which the water is subducted. Ideally these axes should be inclined at the slope of one year of date per year of streamline distance indicated by the bold arrow.
The axes delineate the maxima and minima of 5–10-yr time-scale anomalies. The episodes of warmer isopycnal temperature and therefore cooler surface temperatures (bold lines in Fig. 15a) tend to correspond to shallower isopycnal depths (dashed lines in Fig. 15b). In the years of coldest surface forcing, the mixed layer density would have been ≥26.75 kg m−3 at some time during the winter; detrainment during the spring would have left this isopycnal level at the base of the summer thermocline, which accounts for the minimum depths of between 100 and 200 m that occurred in (January) 1958, 1967, and 1974 (Fig. 15b). Minimum depths (circles on dashed lines), lie at approximately −7 yr and near the centers of the shaded areas indicating the deepest winter mixed layers; they therefore occur at positions where the mixed layer had been deepest, but not necessarily those where the 26.75 kg m−3 isopycnal was last in contact with the mixed layer. The latter are better indicated by the temperature maxima (circles on solid lines in Fig. 15a), which lie near the 26.7 kg m−3 mixed layer density contour, which marks the boundaries of the shaded areas. This density is lower than that of the streamline, but the contour is only for 1 September, so it may give an indication of the positions at which a density of 26.75 kg m−3 could have been reached at some time during winter. The temperature maxima are mostly at position −4 yr on the chosen streamline but vary in position, depending on the strength of the cold episode, the positions being most advanced (closer to 0 yr) in the years of coldest SST, when deep mixed layers were closest to the section line. The temperature minima, marking subduction of water in periods of maximum SSTs, are located farther back along the streamline.
The expansion and contraction in the area of the deeply mixed region are shown in the MLDs for two extreme seasons (Fig. 16). The first was 1965, when maximum depths of 500 m developed only two years’ travel from the section. The second was 1987, when maximum MLDs were ≈350 m, well south of the section, and the 26.75 kg m−3 isopycnal was at 500 m (see Fig. 15), below the level of winter exposure.
4. Discussion
A major issue in conducting this type of study was that of providing a suitable time series of forcings. The basic source of interannual variation was provided by the NCEP fluxes and wind stresses; however, because of the problem of model drift caused by limited model resolution and systematic errors in the forcings, some relaxation to observed surface properties was essential. This damps interannual variations unless monthly mean restoration fields are used, but these are only available for SST. As a practical compromise, a hybrid relaxation to monthly mean SSTs and climatological SSSs was used. This overemphasizes the thermal variations relative to the freshwater forcing variations and produces unbalanced contributions to the buoyancy forcing, but it may be appropriate for the many parts of the world’s oceans where temperature has the dominant effect on density. However, there are indications that the relative effects of temperature and salinity variations may be more comparable in parts of the SAZ. Rintoul and England (2002), from observations in six surveys of the WOCE SR3 section between Tasmania and Antarctica during the 1990s and from the results of a coupled climate model (Manabe and Stouffer 1996), found year-to-year T–S variations to be large but nearly density-compensated in the SAZ; they showed that this was consistent with variations in the advection of cool fresh sub-Antarctic water across the Antarctic Circumpolar Current (ACC), but not with the correlations between heat and freshwater fluxes, which would have tended to produce T–S correlations of opposite sign.
The degree of compensation can have an important bearing on the interpretation of changes in interior properties. For monitoring these changes, it is desirable to suppress variations due to the passage of mesoscale eddies, Rossby waves, and current displacements; this is commonly done by analyzing changes on surfaces of constant density (or some other state variable). We have shown that the temperature change on isopycnals may be of the same or opposite sign to that of the surface source region, depending on the degree to which temperature and salinity changes are compensated. When compensation is complete or substantially so, isopycnal temperature changes will be of the same sign as surface changes, but when the ratio of the temperature to the salinity change is greater than that of the water mass into which the water is subducted, the sign will be opposite.
Because of the uncertain role of the freshwater forcing, we performed four transient simulations using surface restoration prescriptions that produced different degrees of damping of interannual variations in model SSTs and SSSs, and hence in mixed layer depths and sectional properties. However, all simulations showed a reversal of the sign of temperature changes when analyzed on isopycnals; all showed the large changes to be in the SAMW layer; all showed the same patterns of change and time variation of response on the 15 section; and all showed the same correlation between isopycnal depth and temperature.
Our results are not necessarily contradictory to those of Rintoul and England (2002). Their analysis was focused on the SAZ south of Tasmania and over the much shorter time period of 5 yr, from 1991 to 1996. In the Tasmanian region, we also found meridional T–S variations to be almost completely density compensated in the model winter mixed layers and found the major axis of temporal variations to lie parallel to isopycnals: in the Indian Ocean, the meridional density variations were greater, and, importantly, temporal variations in the source region had a greater temperature dominance than that of the water mass into which the water was subducted. We do not know from data what surface density changes occurred in the southeast Indian Ocean, but some indication of their likelihood is provided by the observed changes in the density of the stability minimum and the thickness of the 0.1 kg m−3 density interval surrounding it. These changes are smaller than those modeled but are qualitatively similar, indicating a redistribution of density classes and implying changes of sea surface density similar to those modeled, though somewhat more compensated. The thickness changes further suggest that the asymmetric redistribution of densities found in the model may be true of the real ocean.
The question of analysis period is also relevant. Rintoul and England (2002) were mainly considering year-to-year variations. It may well be that on this time scale cross-frontal advection is more effective than surface fluxes in producing large property changes but that the balance is different over longer periods, as has been suggested by M05. The reasons for these differences warrant further investigation.
5. Conclusions
This study has used an ocean model driven by monthly mean forcings to simulate the changes observed at 32°S between the cruises of 1965, 1987, and 2002. In spite of some biases resulting from errors in boundary conditions, finite resolution, and model physics, the model simulations have reproduced the important features present in the observed sections and in the 1965–87 and 1987–2002 section differences. In both the model and observations, the largest subsurface changes were in the SAMW, which showed a cooling and deepening of isopycnals during the first period and a warming and shallowing during the second. In the model, the changes in this water mass were unrealistically concentrated on the eastern side of the section and became peaked in a lighter density class than observed on the western side; however, the patterns of the changes were the same in the two periods (except for the change in its sign after 1987), as also was the case in the observations.
The model allows us to relate these changes to the subsurface variations of which they formed a part and to the surface variations forcing them. Because of the need for data reconstruction, limited reliance can be placed on year-to-year variations in SST datasets in the South Indian Ocean; however, we can have more faith in the longer-period variations. During the period 1948–2003, observed and modeled SSTs included a multidecadal oscillation with a period of 50–60 yr and interannual variations with periods of less than 10 yr. On the eastern side of the ocean, winter surface conditions were coldest in about 1965 and warmest in 1987; in the western half, the minimum of the cycle occurred earlier. Model SSS variations differed from those of SST, but we do not have observations with which to compare them. Because both SST and SSS are used to compute restorative components of the surface fluxes, there is some uncertainty in the density forcing. This has been approached by conducting experiments with various forcing prescription and has been discussed in section 4. In all simulations, SST variations were found to have the controlling effect on mixed layer density and depth.
Model isopycnal temperature variations in the SAMW on the 15 section are remarkably similar in form to SST variations in the source regions of the SAZ. The times of maxima and minima are consistent with subduction and transport from the SAZ within several years. Although interannual subsurface temperature variations are more peaked during the period of cold SSTs than at other times, both low- and high-frequency components of the surface forcing are represented in the sectional time series. The majority of the model variation is explained by the multidecadal cycle, which has principal component patterns that are similar in form to those seen in the difference sections. This and the fact that the 1965 and 1988 cruises roughly coincided with the turning points of the cycle, increase our confidence that the observed difference patterns are characteristic of interdecadal variability at 32°S. The simulations provide evidence that the reversal of what appeared to be a cooling trend before 1987 is part of a multidecadal variation related to observed SSTs and heat fluxes; however, the apparent 50–60-yr period of this variation, which is that of the simulated time series, is a property of the forcing data, and it is not suggested that the cycle is one that will necessarily continue.
Isopycnal depth is positively correlated with SST and negatively with isopycnal temperature, as was also seen at lower temporal resolution in the sectional differences. BM noted that isopycnal depths had increased during the period 1965–87 for all neutral densities, γn < 27.77 kg m−3, with maximum changes occurring in the range 26.7–27.0 kg m−3; they concluded that the volume of SAMW had increased while that of AAIW had decreased. In fact, they may have overestimated the depth interval over which the isopycnal depth changes occurred as a consequence of including nonsynoptic data. By contrast, we found a reasonably sharp maximum at σθ = 26.7 kg m−3 (equivalent to γn = 26.8 kg m−3) in the observed section differences and changes of the same sign extending between about 26.3 and 26.9 kg m−3. This is important, because it implies that the total volume of SAMW did not vary much over the study period, but that its density distribution did. The confinement of the major thickness changes to the SAMW has also been found in the model when account is taken of the somewhat greater longitudinal variation in the density of the pycnostad across the section in the model than in the cruise sections.
The time–depth plots of density for the eastern half of the section (Fig. 13b and especially Fig. 13c) indicate that progressively denser classes of SAMW were subducted during the period of cold surface forcing up until the 1970s, when mixed layers were deepening and becoming denser, but that a rather more rapid transition to the subduction of lighter classes occurred in the warmer period of the 1980s. This somewhat asymmetric progression resembles that found in a diurnal or seasonal mixed layer cycle but is the result of sampling by subduction and transport rather than local mixing and detrainment.
Although final subduction of SAMW in the southeast Indian Ocean occurs in deep mixed layers southwest of Western Australia, changes in advection by the subsurface currents exposed in this region may contribute to variability in the properties achieved. Some of the water arriving at the eastern end of the 15 section originates in the region west of Tasmania and passes through the South Australian basin. Barker (2004) has diagnosed maximum changes in mode water properties of −0.8°C and −0.18 psu at 26.9 kg m−3 between 37° and 41°S at 132°E between the USNS Eltanin cruise of 1969/70 and the RV Franklin cruise of 1994: these are very similar to the maximum changes of −0.7°C and −0.15 psu at 26.7 kg m−3 that we found for the 15 section between 1965 and 1987. Recent work (Rintoul and Bullister 1999; Barker 2004) suggests that the ocean southwest of Tasmania may be an important region for the formation of SAMW. Barker proposes that this region, where the interaction of the Tasman Outflow and the anticyclonic gyre in the South Australian Basin gives rise to a saddle point in the streamflow, is particularly favorable to water mass formation in the mode water and intermediate water ranges. The variation in the strength of the Tasman Outflow therefore could be expected to be an important contributor to changes in the properties of the mode and intermediate waters found in the South Australian Basin and ultimately in the South Indian Ocean gyre.
Acknowledgments
This work was supported by the Cooperative Research Centres Programme through the Antarctic Climate and Ecosystems Cooperative Research Centre and by the Antarctic Science Advisory Committee of the Australian Antarctic Division.
The NCEP-1 data were provided by the National Center for Atmospheric Research, which is supported by grants from the National Science Foundation. Sea surface temperature data were made available by the Hadley Centre for Climate Prediction and Research, Bracknell, United Kingdom (HadISST), and by the National Climatic Data Center, Asheville, North Carolina (Reynolds et al. Extended Reconstructed SST). World Ocean Atlas data and section data for the Atlantis II (1965) and WOCE Darwin (1987) cruises were obtained from the National Oceanographic Data Center. Data from the Darwin 2002 cruise were supplied by the British Oceanographic Data Centre, Bidston (now Liverpool), United Kingdom.
Dr. Murray will be much missed by friends and colleagues.
REFERENCES
Banks, H. T., R. A. Wood, J. M. Gregory, T. C. Johns, and G. S. Jones, 2000: Are observed decadal changes in intermediate water masses a signature of anthropogenic climate change? Geophys. Res. Lett., 27 , 2961–2964.
Barker, P. M., 2004: The circulation and formation of water masses south of Australia and the interannual wind variability along the southern Australian coast. Ph.D. thesis, University of Melbourne, 351 pp.
Bindoff, N. L., and J. A. Church, 1992: Warming of the water column in the southwest Pacific Ocean. Nature, 357 , 59–62.
Bindoff, N. L., and T. J. McDougall, 2000: Decadal changes along an Indian Ocean section at 32°S and their interpretation. J. Phys. Oceanogr., 30 , 1207–1222.
Bryan, K., 1984: Accelerating the convergence to equilibrium of ocean–climate models. J. Phys. Oceanogr., 14 , 666–673.
Bryden, H. L., E. L. McDonagh, and B. A. King, 2003: Changes in ocean water mass properties: Oscillations or trends? Science, 300 , 2086–2088.
Chen, D., and L. M. Rothstein, 1994: A hybrid vertical mixing scheme and its application to tropical ocean models. J. Phys. Oceanogr., 24 , 2156–2179.
Conkright, M. E., R. A. Locarnini, H. E. Garcia, T. D. O’Brien, T. P. Boyer, C. Stephens, and J. I. Antonov, 2002: World Ocean Atlas 2001: Objective analyses, data statistics, and figures: CD-ROM documentation. National Oceanographic Data Center, Silver Spring, MD, 17 pp.
Cox, M. D., 1987: Isopycnal diffusion in a z-coordinate ocean model. Ocean Modell., 74 , 1–5.
Gordon, C., C. Cooper, C. A. Senior, H. Banks, J. M. Gregory, T. C. Johns, J. F. B. Mitchell, and R. A. Wood, 2000: The simulation of SST, sea ice extent, and ocean heat transports in a version of the Hadley Centre Coupled Model without flux adjustment. Climate Dyn., 16 , 147–168.
Griffies, S. M., 1998: The Gent–McWilliams skew flux. J. Phys. Oceanogr., 28 , 831–841.
Johnson, G. C., and A. H. Orsi, 1997: Southwest Pacific Ocean water-mass changes between 1968/69 and 1990/91. J. Climate, 10 , 306–316.
Kraus, E. B., 1990: Diapycnal mixing. Climate-Ocean Interaction, M. E. Schlesinger, Ed., Kluwer Academic, 269–293.
Levitus, S., 1982: Climatological Atlas of the World Ocean. NOAA Prof. Paper 13, 173 pp. and 17 microfiche.
Manabe, S., and R. J. Stouffer, 1996: Low-frequency variability of surface air temperature in a 1000-year integration of a coupled ocean–atmosphere–land surface model. J. Climate, 9 , 376–393.
Marsh, R., J. G. Nurser, A. P. Megann, and A. L. New, 2000: Water mass transformation in the Southern Ocean of a global isopycnal coordinate GCM. J. Phys. Oceanogr., 30 , 1013–1045.
McCartney, M. S., 1977: Subantarctic Mode Water. A Voyage of Discovery: George Deacon 70th Anniversary Volume, M. V. Angel, Ed., Pergamon, 103–119.
McCartney, M. S., 1982: The subtropical recirculation of Mode Waters. J. Mar. Res., 40 , 427–463.
McDonagh, E. L., H. L. Bryden, B. A. King, R. J. Sanders, S. A. Cunningham, and R. Marsh, 2005: Decadal changes in the South Indian Ocean thermocline. J. Climate, 18 , 1575–1590.
Murray, R. J., and C. J. C. Reason, 2001a: A curvilinear version of the Bryan-Cox-Semtner Ocean Model and its representation of the Arctic circulation. J. Comput. Phys., 171 , 1–46.
Murray, R. J., and C. J. C. Reason, 2001b: A curvilinear ocean model using a grid regionally compressed in the South Indian Ocean. J. Phys. Oceanogr., 31 , 2809–2823.
Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V. Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 2003: Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. J. Geophys. Res., 108 .4407, doi:10.1029/2002JD002670.
Rintoul, S. R., and J. L. Bullister, 1999: A late winter hydrographic section from Tasmania to Antarctica. Deep-Sea Res. I, 46 , 1417–1454.
Rintoul, S. R., and M. H. England, 2002: Ekman transport dominates local air–sea fluxes in driving variability of Subantarctic Mode Water. J. Phys. Oceanogr., 32 , 1308–1321.
Smith, T. M., and R. W. Reynolds, 2003: Extended reconstruction of global sea surface temperatures based on COADS data (1854–1997). J. Climate, 16 , 1495–1510.
Smith, T. M., and R. W. Reynolds, 2004: Improved extended reconstruction of SST (1854–1997). J. Climate, 17 , 2466–2477.
Warren, B. A., 1981: Transindian hydrographic section at Lat. 18°S: Property distributions and circulation in the South Indian Ocean. Deep-Sea Res., 28 , 759–788.
Wong, A. P. S., N. L. Bindoff, and J. A. Church, 1999: Large-scale freshening of intermediate waters in the Pacific and Indian oceans. Nature, 400 , 440–443.
Wong, A. P. S., N. L. Bindoff, and J. A. Church, 2001: Freshwater and heat changes in the North and South Pacific Oceans between the 1960s and 1985–94. J. Climate, 14 , 1613–1633.
Wyrtki, K., 1971: Oceanographic Atlas of the International Indian Ocean Expedition. National Science Foundation, 531 pp.
MLDs as at 1 Sep in the equilibrium simulation [contour interval (CI) is 50 m, with shadings for mixed layers greater than 350 and 450 m]. Also indicated are the tracks of the 1987 I5 and 1994 I8S WOCE cruises (bold), the 1965 Atlantis II cruise (bold dashes), and the 2002 Darwin cruise (dots). The 2002 cruise followed the I5 track in the western Indian Ocean but diverged somewhat from it eastward of 80°E.
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
Temperatures (faint contours) and densities (bold contours) (a) from the December 1994 cruise along the I8S WOCE section, which runs north–south between 45° and 30°S (½° smoothing horizontally) and from the model for (b) 1 Jan 1995 and (c) 1 Sep 1994, interpolated to the WOCE section (CI is 1°C for temperature and 0.1 kg m−3 for density). The dashed lines indicate MLDs.
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
(a) The EOF pattern for the first principal mode (variance 41%) of model SST computed over the contoured area shown for the 1 Sep model solutions over the period 1948–2002 (CI is 0.1°C; negative contours are dashed), and (b) its amplitude time series. The bold outlines indicate two regions in the unshaded area of reduced positive variability (0°C < EOF1 < 0.5°C) associated with the band of deep mixed layers in the SAZ—W between 40° and 75°E, and E between 75° and 115°E; these are used for averaging of the time series in Figs. 4, 5, 6 and 12.
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
Time variation of (a) August and September SSTs from the HadISST and Reynolds et al. monthly average datasets and (b) NCEP March–August average net heat and evaporative (evaporation minus precipitation) fluxes, in each case averaged over region E indicated in Fig. 3. The dashed curves in (b) are the fluxes time-smoothed using a diffusive filter with a 4-yr time scale.
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
Time series of 1 Sep sea surface (a) temperature, (b) salinity, and (c) density from expts 1–4, averaged over region E; (a) also shows the Aug–Sep average Hadley SSTs for the same region (bold dashed line).
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
Time variation of (a) SST and (b) prognostic MLD (values increasing downward) from expt 4 (standard experiment, 30-day HadISST restoration, solid curves) and expt 5 (same as for expt 4, but with time-smoothed forcings, dashed curves) averaged over regions W (faint) and E (bold).
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
Temperatures (faint) and densities (σθ ≤ 26.0 kg m−3; bold contours) from (a) the December 1987 WOCE 15 cruise and from the model for (b) 1 Jan 1988 and (c) 1 Sep 1987, interpolated to the WOCE section (CIs are 1°C for temperature and 0.1 kg m−3 for density). The shading is for static stabilities N 2 < 7.5 × 10−6 s−2 [equivalent to PV = |ƒ| N 2/g < 59 × 10−14 cm−1 s−1 at 32°S, which is close to the 60 × 10−14 cm−1 s−1 used by McCartney (1982)].
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
Unsmoothed differences of (a) isopycnal depth and (b) temperature on isopycnals between the 1987 and 2002 cruise sections, plotted after interpolation to standard levels and a ½° longitude grid (CIs are 20 m and 0.1°C; negative differences are shaded). The dashed curve in (b) is the 200-m isobath (averaged over the two cruises and smoothed).
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
Horizontally smoothed observed differences of isopycnal depth for the periods (a) 1965–87 and (b) 1987–2002, and of temperature on isopycnals for (c) 1965–87 and (d) 1987–2002; and (e)–(h) the (unsmoothed) model differences for the respective periods and variables (CIs are 20 m for depth and 0.1°C for temperature; negative differences are shaded). (Note that solutions for 1 Jan 1965, 1988, and 2002 were used for the model solutions rather than the actual cruise dates.) The bold dashed curves enclose mode water with a static stability N 2 < 10 × 10−6 s−2; the stability in (a)–(d) was averaged on isopycnal coordinates from the three smoothed cruise sections and in (e)–(h) was computed directly from the 1 Jan solution of the equilibrium simulation.
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
Zonally averaged differences of isopycnal (a) depth, (b) temperature, and (c) salinity for the periods 1965 to 1987/88 (faint) and 1987/88 to 2002 (bold). The solid curves are for the cruise data and the dashed curves are for the model. The small undulations with depth in the profiles (here and the sectional plots) are an artifact of linearly interpolating data to isopycnal surfaces, which lay at different depths on different dates.
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
Time series of model isopycnal (a) temperature and (b) depth on the σθ = 26.7 kg m−3 surface (expts 1–6) and on the σθ = 27.0 kg m−3 surface (expt 4, bold gray curve), in each case zonally averaged over the eastern part of the I5 section (80°–105°E). Averages of these quantities for σθ = 26.7 kg m−3 from the 1965, 1987, and 2002 cruise sections are indicated by stars. Note the descending scales for isopycnal depth and the offset right-hand scales for the σθ = 27.0 kg m−3 curves.
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
(a) A schematic T–S curve (CC′) showing the displacement caused by various temperature and salinity changes (vectors originating at O) and the consequent changes on the isopycnal II′ (dashed gray vectors OU and OZ). The shaded area indicates the quadrant in which warming is density-compensated by salinification. (b)–(e) The T–S points, connected chronologically, for 1 Sep model surface properties averaged over region E for expts 1–4, respectively. The coordinate axes (labeled in psu and °C) have been scaled on the page to match distances corresponding to equal changes in a α δT and β δS, resulting in isopycnals (dotted lines) having a 45° slope. The thin, almost vertical, line is a least squares fit to the major axis of variation of the T–S points. The bold dashed gray line in each panel gives vertical T–S variation averaged over 80°–105°E on the I5 section from the equilibrium simulation (approximately the time mean), with the stars marking the points (from top to bottom) at the 410-, 545-, and 710-m model levels.
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
Time series of variation with depth of model density (bold) and temperature (faint) zonally averaged over the (a) western (40°–70°E) and (b) eastern (80°–105°E) parts of the I5 section from the standard simulation (expt 4) and (c) the eastern part of the section from expt 2. CIs are 0.1 kg m−3 and 1°C, respectively, with additional dashed contours for σθ = 26.75 and 26.85 kg m−3 in (b). The gray thick dashed lines give the depth of the static stability (or PV) minimum (middle line) and the depths at which the density is 0.05 kg m−3 greater or less than that of the stability minimum (lower and upper lines, respectively).
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
Streamlines integrated backward on the 26.75 kg m−3 isopycnal surface from points at 5° longitude intervals along the I5 section (bold line) using 1 Jan velocities from the seasonal equilibrium simulaton, showing positions (dots) at 1-yr intervals. The surface outcrop of this isopycnal is shown for 1 Jan (dashed contour at 50°S) and for 1 Sep (solid contour at the boundary of the shaded area).
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
Time-distance sections of (a) temperature and (b) depth for 1 Jan series on the back streamline from longitude 80°E on the I5 section following the σθ = 26.75 kg m−3 isopycnal (CIs 0.05°C and 25 m). The vertical axis represents the calendar date and the horizontal axis, the position along the streamline in units of years relative to the time of arrival at the section line (based on climatological 1 Jan model velocities). Selected troughs and ridges of the contoured variables are indicated by bold solid and dashed lines (slightly smoothed) and maxima and minima by circles on the solid and dashed lines, respectively. The arrow at the top is the slope expected from pure advection. The shading gives the positions on the streamline at which the density at the base of the mixed layer was >26.7 kg m−3 on 1 Sep of the year given on the time axis.
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
MLDs at 1 Sep 1965 and 1987 in the transient simulation (CI is 50 m, with shadings for MLDs greater than 350 and 450 m). Also indicated is the 15 WOCE cruise track (bold).
Citation: Journal of Climate 20, 13; 10.1175/JCLI4160.1
Model experiments, where NCEP 1 = NCEP 1 monthly forcings; Clim SST/SSS = restoration to climatological (monthly) values (Hadley SSTs and World Ocean Atlas SSSs), relative to a time scale of 30 days (1/6 refers to 180-day restoration); Mon-SST = restoration to Hadley monthly SSTs; Clim flux = climatological (monthly) restoration flux diagnosed from the equilibrium simulation; low freq = smoothed time series; and high freq = anomaly from smoothed time series + climatological (monthly) values.
Observed and modeled properties of the stability minimum.
