Frontal movements and property fluxes: Contributions to heat and freshwater trends in the Southern Ocean

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 116,, doi: /2010jc006832, 2011 Frontal movements and property fluxes: Contributions to heat and freshwater trends in the Southern Ocean A. J. S. Meijers, 1,2 N. L. Bindoff, 2,3,4,5 and S. R. Rintoul 1,2,3,4 Received 22 November 2010; revised 18 May 2011; accepted 31 May 2011; published 23 August [1] We examine the synoptic variability of temperature and salinity in the Southern Ocean using a static gravest empirical mode (GEM) mapping based on historical hydrography and a time evolving projection of the GEM created using satellite altimetry (satgem). The GEM and satgem projections allow the separation of observed trends into a component due to a shift of the circumpolar fronts (adiabatic) using the satgem and a component due to changes in water masses (diabatic) expressed as a temporal trend in residual between historical hydrography and the static GEM. The mean southward movement of the Antarctic Circumpolar Current (ACC) fronts drives an adiabatic warming of ± W m 2 distributed over most of the ACC and at all depths. This is strongest where the meridional temperature gradient is largest, such as in the upper 1000 dbar of the Subantarctic Front (SAF). There is a weak adiabatic freshening of 6.57 ± 0.18 mm yr 1 m 2, concentrated mainly below the Antarctic Intermediate Water and south of the SAF. Residuals between historical hydrography and the static GEM field, driven by diabatic changes in the water mass structure, have temporal trends in temperature and salinity. When these trends are integrated over the whole ACC there is a net cooling of ± W m 2 and strong freshening of ± 0.70 mm yr 1 m 2. By combining the diabatic and adiabatic components and integrating over the ACC the net increase in heat and freshwater in the region is ± W m 2 and ± 0.72 mm yr 1 m 2, respectively. The sum of the adiabatic and diabatic changes are consistent with previous trends inferred from observations. Citation: Meijers, A. J. S., N. L. Bindoff, and S. R. Rintoul (2011), Frontal movements and property fluxes: Contributions to heat and freshwater trends in the Southern Ocean, J. Geophys. Res., 116,, doi: /2010jc Introduction [2] Although observational evidence is still relatively sparse in the Southern Ocean in comparison to much of the global ocean, it is becoming clear that it exhibits long term property trends. Gille [2002, 2008] shows using PALACE and Argo observations that there has been a uniform temperature increase of up to ± C yr 1 at dbar in the Southern Ocean, with the greatest warming taking place on its northern edge near the Subantarctic Front (SAF). Willis et al. [2004] and Helm et al. [2010] similarly demonstrate that some of the strongest increases in heat content over the global ocean occur at the northern edge of the Antarctic Circumpolar Current (ACC), at around S. 1 Marine and Atmospheric Research, CSIRO, Hobart, Tasmania, Australia. 2 Wealth from Oceans Flagship, National Research Flagships, Hobart, Tasmania, Australia. 3 Antarctic Climate & Ecosystems Cooperative Research Centre, Hobart, Tasmania, Australia. 4 Centre for Australian Weather and Climate Research, Hobart, Tasmania, Australia. 5 Institute of Marine and Antarctic Studies, University of Tasmania, Hobart, Tasmania, Australia. Published in 2011 by the American Geophysical Union. There is less clear cut evidence for changes in the Southern Ocean haline structure, due to a paucity of data, but recent studies by Durack and Wijffels [2010], Helm et al. [2010], and Böning et al. [2008] show that there is a strong freshening signal over the Southern Ocean, with a freshwater trend of up to 100 mm yr 1 m 2. Aoki et al. [2005a] and Johnson et al. [2008] both observe freshening trends at depth in the Southeastern Indian Ocean, which they label as a fingerprint of anthropogenic climate change, as predicted in modeling studies [Banks and Bindoff, 2003]. [3] In the Southern Ocean the warming in the upper 700 dbar is insufficient to explain the observed rise in sea level (Figure 1), in contrast to much of the rest of the global ocean where the thermal expansion in the upper ocean dominates the sea level trend [Cabanes et al., 2001; Cazenave et al., 2003; Domingues et al., 2008]. Ishii et al. [2006] and Morrow et al. [2008] show that south of Australia the steric change down to at least 2000 dbar must be considered in order to account for the observed sea level height change. Gille [2008] suggests that these deep warming trends are inconsistent with surface driven warming fluxes and observes that the changes may be due to a coherent poleward shift of the ACC fronts. She shows that as much as 87% of the observed temperature trend in the Southern Ocean below 200 dbar over the last years may be explained by 1of17

2 Figure 1. Mean SSH trend between 1992 and Contours are mean dynamic height (relative to 2000 dbar) approximately associated with, from the south, the saccf, spf, npf, ssaf, csaf, nsaf, and northern edge of the ACC. The regions shown in color represent the region of the Southern Ocean where the satgem is valid. The GEM relation becomes invalid at the Subtropical Front, and the southern edge is dictated by sea ice extent. Other regions may be partially recreated by the satgem, due to varying ice cover for example, but these are excluded from the calculations of property trends in section 5 to avoid temporal aliasing. Units are m yr 1. such frontal movements. Such a coherence between frontal shifts and temperature trends in the ACC was also noted in an observational study by Sokolov and Rintoul [2003]. An analysis of changes in wind stress curl by Cai [2006] notes that the trend in winds since the 1970s may potentially shift the ACC southward, resulting in an apparent increase in temperature. Sokolov and Rintoul [2009] identify just such a southward shift of the mean ACC frontal positions between 1992 and 2006, through altimetric tracking of the dynamic height contours that define the main fronts of the ACC [Sokolov and Rintoul, 2007]. This southward shift of fronts may be due to increased wind stress, and numerous climate models link the wind stress change to an increase in the Southern Annular Mode (SAM) [Cai et al., 2005, 2010; Fyfe and Saenko, 2007] which has had a coherent positive trend between the mid 1960s and 2002 [Marshall, 2003]. The increasing SAM trend is possibly driven by increased levels of anthropogenic atmospheric CO 2 [Fyfe, 2006] or ozone depletion over Antarctica [Thompson and Solomon, 2002; Cai, 2006]. [4] We use a gravest empirical mode (GEM) [Meinen, 2001] projection of the Southern Ocean to partition and quantify Southern Ocean temperature and salinity variability into that driven by frontal movement (adiabatic) and water mass changes driven by heat and freshwater fluxes (diabatic). The GEM is a mapping that takes advantage of the remarkable vertical coherence of T S profiles due to the time mean equivalent barotropic nature of the Southern Ocean [Killworth, 1992; Killworth and Hughes, 2002] to parameterize the subsurface profiles as a function of dynamic height and longitude. This effectively projects the meridional temperature and salinity profiles in stream function space and has been shown in several studies [Sun and Watts, 2001, 2002; Watts et al., 2001; Meijers et al., 2011; Swart et al., 2010] to describe over 90% of the variability of the temperature and salinity fields. Swart et al. [2010] and Meijers et al. [2011] showed that by using gridded satellite altimetric products as a proxy to estimate the in situ dynamic height, the GEM fields may be used to construct gridded, full depth temperature, salinity and velocity fields in the ACC that evolve 2of17

3 in time as the sea surface height (SSH) changes. These time varying fields are known as satgem fields. [5] Because the satgem essentially just rearranges the vertical GEM profiles horizontally as the SSH varies, any trends in temperature or salinity in the satgem fields will be attributed to the adiabatic spatial displacement of water masses, such as occurs during coherent frontal shifts. This is the horizontal analogue of the pure vertical heave of isopycnal surfaces described by Bindoff and McDougall [1994]. The trends in temperature and salinity within the water column over the satgem period ( , during which time altimetry is available) can therefore be used to quantify the contribution of adiabatic frontal shifts to the T S changes in the Southern Ocean. In addition, the temporal trend in the residuals between the static GEM and the historical hydrography contains information about the changing temperature and salinity structure of the Southern Ocean in dynamic height coordinates. These water mass changes alter the structure of the ACC and must be due to fluxes of heat and freshwater that are independent of frontal movement, and we therefore term these diabatic changes. [6] In this study we examine the mean trend of temperature and salinity in the GEM and satgem. We compare the trends to observations, showing that our estimates of the diabatic and adiabatic components agree with observations when summed, thereby quantifying the influence of frontal movement and water mass changes on observed temperature and salinity trends. In Section 2 we describe the data and the construction of the GEM and satgem fields. In Section 3 we diagnose the adiabatic trends observed in the satgem between 1992 and 2010, and in Section 4 establish the temporal trend in residuals between hydrography and the GEM fields to determine the diabatic component. By integrating both components over the ACC in Section 5 we derive bulk adiabatic and diabatic trends for heat and freshwater and their relative contribution to the total heat and freshwater budgets. In Section 6 we discuss the difficulties inherent in interpreting trends in dynamic height coordinates and possible sources for the observed property trends. Concluding remarks are made in Section Data [7] The Southern Ocean GEM fields were constructed using a combination of the Southern Ocean Database (SODB) [Orsi and Whitworth, 2001] and Argo float data. Over quality controlled profiles of temperature and salinity were used in an optimal interpolation scheme to produce gridded temperature and salinity fields on a longitude, dynamic height and depth grid. This was then combined with AVISO satellite mean sea level anomaly (MSLA) data to create a time series of temperature and salinity point profiles. This process is detailed in the next section. The AVISO MSLA product combines the gridded output of along track altimetry [Le Traon et al., 2003] from four satellite missions, significantly improving mesoscale structure recovery from the data [Pascaul et al., 2006]. The data is gridded onto an approximately 1/3 mercator grid and contains 911 weekly snapshots from 14 October 1992 to 24 March Ducet et al. [2000] demonstrate that these SSH maps resolve features in the Southern Ocean with wavelengths longer than 100 km, and that some variability is resolved at wavelengths of around 50 km, although with a reduced energy. More information on the creation of the data set can be found in the SSALTO/DUACS User Handbook [SSALTO/DUACS, 2007]. [8] The 2006 CSIRO Atlas of Regional Seas (CARS) [Ridgway et al., 2002] is used to create a steric mean dynamic topography (MDT) relative to 2000 dbar. CARS is based on several data sets including Argo, the World Ocean Database 2001, WOCE WHP 3.0, and CSIRO data holdings, and was produced using a modified Loess filter to map this data onto a regular 1/2 1/2 grid. For this study the MDT was interpolated linearly onto the 1/3 mercator grid of the SSH altimetric data from AVISO described above. This MDT was then added to the AVISO MSLA to produce a timeevolving dynamic topography (DT) over the domain. CARS was chosen over more sophisticated MDTs such as RIO05/ 09 as it is a purely baroclinic product and can be easily referenced to an arbitrarily chosen reference level GEM Creation [9] Sun and Watts [2001] demonstrated that there is a robust empirical relationship in the Southern Ocean between the dynamic height of the water column and the subsurface temperature and salinity profiles, such that for a given meridional section each observed dynamic height can be assigned to a unique T S profile. This mapping, known as a gravest empirical mode [Meinen and Watts, 2000], or GEM, reduces the dimensionality of temperature and practical salinity distributions from (x, y, z, t) and S p (x, y, z, t) to GEM (x, z, ) and S pgem (x, z, ) where x and y are longitude and latitude and dynamic height is given by 100 Z ¼ dp 2000 where d is the specific volume anomaly at some pressure p. The dynamic height coordinates take advantage of the Southern Ocean s strong meridional density gradient and its equivalent barotropic nature whereby columns of water maintain their vertical structure during horizontal translation [Watts et al., 2001; Killworth, 1992]. Because of this the mapping removes temporal and spatial mesoscale variability without oversmoothing and smearing frontal structures. A full description and validation of the GEM and satgem fields used in this study are given by Meijers et al. [2011], but a brief summary is given here for completeness. We create GEM temperature and salinity fields by objectively mapping hydrography at regular depth levels from the SODB and Argo profiles onto a regular grid in dynamic height, depth and longitude coordinates. The reference depth of 2000 dbar is chosen because it captures much of the steric variability in the Southern Ocean [Morrow et al., 2008] while still using the depth limited Argo profiles. The dynamic height integration is taken to 100 dbar rather than the surface to reduce the impact of mixed layer variability. The GEM mapping is still carried out above this depth, so to reduce the variability between summer/winter mixed layers a simple empirical model (following Watts et al. [2001]) of seasonal mixed layer temperature and salinity anomalies is then subtracted from the in situ data to effectively create seasonless profiles before calculating the GEM fields. This substantially increases the accuracy of the mapping near the surface. The ð1þ 3of17

4 resulting GEM temperature and salinity fields capture over 95% of the temperature variance and 92% of the salinity variance in the upper 3000 dbar over the region of the Southern Ocean where this method is valid (see colored region in Figure 1 for valid domain). When the GEM fields are compared against independent hydrography deliberately excluded from the initial field construction (approximately 9000 profiles) the rms difference approaches the a prioi noise in the hydrographic data. Here the a prioi noise is defined using the Bindoff and Wunsch [1992] methodology of rms differences between nearest neighbor hydrography (0.9 C, at the surface, reducing to less than 0.1 C and below 1000 dbar) Combination With Altimetry [10] Meijers et al. [2011] demonstrate that there is a strong positive correlation between the SSH anomaly as measured by the AVISO gridded MSLA and the dynamic height anomaly of colocated hydrography relative to the CARS MDT. They also showed that approximately half of the sea level anomaly (SLA) is due to steric changes in the upper 2000 dbar, agreeing with similar estimates made by Guinehut et al. [2006]. Using the empirically determined relation ¼ 0:52SLA ð2þ we can estimate the in situ anomaly of dynamic height at 100 dbar referenced to 2000 dbar from the altimetric SLA from the altimeter. This estimate has a rms error of dynm when compared to the known in situ dynamic height values. This rms error is due to variability in the ratio of baroclinic to deep baroclinic/barotropic contribution to SLA, temporal and spatial aliasing, as well as observational errors in the SLA. The relation given in equation (2) is estimated circumpolarly, but it does not change significantly by longitude or latitude in the study domain. Once the altimetric SLA is converted to a dynamic height anomaly it is added to the CARS mean dynamic height field, and the subsurface temperature and salinity profiles are calculated using the GEM field profiles at the estimated dynamic height and longitude. This process is carried out at each of the grid points and time intervals of the AVISO MSLA, producing gridded temperature and salinity fields in latitude and longitude at 7 day time intervals between 1992 and The simple seasonal model is then added to the surface layers to approximate seasonal changes in the mixed layer. [11] These time varying T S fields (referred to hereafter as satgem fields) are validated against independent in situ hydrographic profiles and mooring time series of Meijers et al. [2011] by linearly interpolating these fields to the position and time of the hydrographic data. The rms difference between the satgem estimates and in situ values is larger than when the in situ dynamic height is used, due to the additional error introduced by the SLA to dynamic height conversion, increasing to 1.16 C, at the surface and around 0.2 C, at 1000 dbar. This error is caused by barotropic or deep baroclinic (below the reference level) variability in the SLA that is uncorrelated with the baroclinic variability, as well as instrument and mapping errors. However, the rms error remains substantially less on average (25 50%) than estimates of the hydrography based on climatologies such as CARS06 or the World Ocean Atlas 05. In regions close to fronts or mesoscale features the relative accuracy of the satgem is even higher, due to the inclusion of altimetric data and its finer spatial resolution. 3. Trends due to Water Mass Movement [12] Sokolov and Rintoul [2009] show that there is a coherent southward displacement of fronts over the altimetric period. The satgem reproduces the water mass movements associated with the frontal shift, and the trends shown in this section are therefore functions of the horizontal water mass displacement and the associated meridional temperature and salinity gradients on pressure and density surfaces. Stronger gradients will produce greater trends for an observed displacement in the surface dynamic height. This results in strong warming and salinification trends near the SAF, Polar Front (PF) and at the southern end of the ACC, where the meridional gradient of temperature and salinity are greatest. In this section we examine the adiabatic (by definition) trends in the satgem temperature and salinity fields on both pressure and neutral density surfaces. We compare the sat- GEM trends with in situ observations, and find that the satgem fields recreate the observed spatial structure of the temperature trend but with larger magnitudes. The salinity trend is not as well recreated. This demonstrates that adiabatic water mass movement alone is insufficient to describe observed changes in the Southern Ocean, and may potentially alias a portion of the diabatic trend Adiabatic Mean Zonal Temperature and Salinity Trends in the satgem [13] We define mean trends in the satgem time series at a given latitude and longitude as Dx ¼ i¼832 X i¼1 ðx iþ52 x i Þ ð3þ where x is the temperature or salinity at a grid point, i is the time index (in weeks) and 832 is the number of 1 year differences (x i+52 x i ) in the time series between October 1992 and March The annual differences remove the seasonal signal, either introduced by the seasonal mixed layer model, or by seasonal trends in the SSH field. The relatively large number of differences in each ensemble reduces the noise caused by short period variability. These differences are taken both on constant pressure surfaces and layers of constant neutral density [Jackett and McDougall, 1997]. [14] The local trends are averaged around the ACC in bins oriented along contours of mean dynamic height to reduce aliasing caused by the meridional variation of the mean ACC position, as with Böning et al. [2008]. The observed general increase in SSH over the Southern Ocean (Figure 1) indicates a southward movement of water masses. It is clear then that the mean temperature trend is positive at almost every latitude and depth (Figure 2a). The trend is greatest (>0.04 C yr 1 ) in the upper 1000 dbar of the SAF (approximately dynm) [Meijers et al., 2011] and intensifies towards the surface where mean meridional gradients are greatest. On pressure surfaces the enhanced warming trend is associated with the movement of the Antarctic Intermediate Water (AAIW) core and Subantarctic Mode 4of17

5 Figure 2. Mean (left) temperature and (right) salinity trends between 1992 and 2010 along mean dynamic height contours on (a and b) pressure surfaces (neutral density contours) and (c and d) neutral density surfaces (pressure contours). In Figures 2c and 2d the white line represents the temperature minimum, the blue line represents the salinity maximum, and the black line represents the salinity minimum. Units are C yr 1 and years 1. Note color scales are capped to provide maximum readability. For actual maximum values see text. Water (SAMW) (g n < 27.6 kg m 3 ). The practical salinity trend in the upper AAIW and SAMW is towards increasing salinities, and is greatest (>0.005 years 1 ) at the same point in the upper 1000 dbar of the SAF as the temperature trend. [15] At greater pressures the warming trend is at a minimum in the Upper Circumpolar Deep Water (UCDW, 28 > g n > 27.4 kg m 3 ) where the trend is less than C yr 1, but increases below this in the Lower Circumpolar Deep Water (LCDW, 28.2 > g n >28kgm 3 ), with values of up to 0.01 C yr 1, increasing towards the south. The salinity trend in pressure coordinates has a maximum negative trend in the UCDW of to 0.01 years 1, extending across 5of17

6 the whole domain, and intensifying to the south. In the LCDW there is a slight increase in salinity (<0.001 years 1 ). The Antarctic Bottom Water (AABW) generally has non statistically significant changes, but the densest layers (g n > 28.3 kg m 3 ) freshen slightly (<0.001 years 1 ). [16] On isopycnal surfaces (Figure 2c) the trends caused by frontal movement have a different pattern due to the weaker tracer gradients along neutral density surfaces, and there is only weak cooling/freshening below 1000 dbar outside of the SAF. The dominant feature is a strong warming/ salinification signal concentrated in the upper 500 dbar where the density surfaces outcrop. North of 1.75 dynm there is cooling/freshening (<0.01 C yr 1 ) on neutral density surfaces, strongest in the upper 100 dbar, but extending through the full water column Regional Adiabatic Temperature and Salinity Trends in the satgem [17] The zonal spatial pattern of adiabatic temperature and salinity trends on pressure surfaces (Figure 3) is broadly uniform circumpolarly. There is statistically significant (at the 95% level) warming over most of the ACC. This warming is greatest at the northern edge of the ACC around the approximate position of the north SAF. The strongest warming (up to 0.1 C yr 1 at 900 dbar) is concentrated in the Indian Ocean between 0 and 50 E, 110 and145 E. Although the warming trend is dominant, there are scattered and spatially variable regions of mean cooling distributed around the ACC. The most obvious cooling (< 0.02 C yr 1 at 900 dbar) occurs between30 and 60 E, extending from the southern SAF to the south edge of the satgem domain. [18] The salinity trends (Figure 3b) have a more complex spatial signal, due mainly to their more complex vertical structure, particularly around the AAIW low salinity tongue. There is generally freshening south of the SAF, with increasing salinity north of this. Again the strongest trends (salinity increasing at up to 0.01 years 1 ) occur around the northern SAF in the Indian/Australian sectors of the Southern Ocean. There are similarly strong freshening trends south of the PF in the Australian sector. The region immediately west of Kerguelen Plateau shows very weak or no freshening trends, in contrast to the regions east and west of it where there is strong freshening. [19] Trends on the neutral density surface g n 27.9 kg m 3 (Figures 3c and 3d) are generally weaker than those observed on pressure surfaces due to the weaker tracer gradient in this coordinate frame. There are still statistically significant changes, however, notably at the southern edge of the isopycnal surface where it outcrops into the mixed layer. There is relatively strong warming/salinification of up to 0.05 C yr 1 /0.005 yr 1 south of the PF in both the Australian and Atlantic sectors, and either weak or no cooling and freshening north of the PF in the Australian and Pacific sectors. The Atlantic sector exhibits mostly weak (<0.01 C yr 1 ) warming north of the PF Comparison With Observations [20] The spatial distribution of temperature trends on pressure surfaces agrees well with the observationally based estimates by Gille [2008] who observe positive trends in the upper 1000 dbar of the ACC, intensifying and deepening at the northern edge associated with the SAF. Gille [2002] specifically identifies the SAF as the strongest region of warming, as well as the coherent region of cooling between 30 and 50 E also observed in this study. Importantly, however, the magnitude of the mean circumpolar satgem warming signal on pressure surfaces of up to 0.04 C yr 1 is approximately twice that observed by Gille [2008]. This is also the case with regional values, where temperature trends can be up to 0.1 C yr 1, twice that described by Gille [2002] and much greater than the C yr 1 observed by Aoki et al. [2003] in the Indian sector above 900 dbar. The mean circumpolar adiabatic salinity trends are less consistent with observations. Boyer et al. [2005] observe a longterm increase in salinity in the upper 500 dbar north of 50 S, with freshening south of and below this, but does not observe an increase in salinity along the LCDW, nor strong surface salinification as far south as the Polar Front (approx. 1 dynm) as is seen in this study. The increase in salinity in the SAMW and AAIW in the satgem is the opposite of that observed by Böning et al. [2008], although the freshening trend along the mean depth of neutral density surfaces below 27.4 kg m 3 is consistent with that study. [21] On neutral density surfaces we see further evidence that the adiabatic satgem alone is insufficient to explain the entirety of the observed Southern Ocean trends. Böning et al. [2008] observe warming/salinification of densities greater than 27.4 kg m 3, and cooling/freshening north of the PF in the lighter SAMW and AAIW. This agrees with the fingerprint of anthropogenic warming suggested by Banks and Bindoff [2003], and observed by Aoki et al. [2005a] in the Indian Ocean sector of the ACC. In contrast the satgem trends on neutral density surfaces are for warming/salinification across all water masses, greatest on the surface outcropping layers, with warming/salinification extending to below 1000 dbar only near the SAF. There is no cooling/freshening observed in the SAMW/AAIW except at the northernmost edge of the domain. The satgem does not extend as far north as some of the studies mentioned, so it is possible that the SAMW/AAIW changes observed in the other studies may be outside of the satgem domain. 4. Trends due to Changes in the GEM Structure [22] The satgem interprets any movement of dynamic height contours as adiabatic water mass movement. However, if the dynamic heights are changed diabatically, by the warming of the upper ocean or addition of freshwater for example, the satgem will alias this diabatic change into the adiabatic trends. However, Sokolov and Rintoul [2009] show that ACC fronts, defined by their hydrographic structure, are consistently associated with a unique dynamic height over the altimetric record. This suggests that diabatically driven changes to the vertical structure of the ACC are either small or largely density compensating, and so the aliasing of the diabatic trend into the adiabatic trend discussed in the previous section is probably small. [23] In this section, we investigate the influence of diabatic trends, and quantify the temporal change in ACC water mass structures due to the fluxes of heat and freshwater. These diabatic trends appear in the time series of residuals between the GEM and the hydrographic observations used 6of17

7 Figure 3. Mean (left) temperature and (right) salinity trends between 1992 and 2010 on (a and b) 900 dbar and (c and d) g n 27.9kgm 3. Units are C yr 1 and years 1. Trends that are not significant at 95% are left blank. Contours are mean dynamic height relative to 2000 dbar. to create it. This effectively represents the time evolution of the GEM field, and consequently the ACC density structure. This has previously been treated as static in the Southern Ocean [Sun and Watts, 2001], on the assumption that the time evolution of the ACC density field is longer than decades [Callahan, 1971; Clarke, 1982]. Here we demonstrate, for the first time, the presence of trends in the Southern Ocean GEM temperature and salinity fields, and show that while the diabatically driven change in net dynamic height may be small, there are still significant trends in temperature and salinity Observed Trends in Residuals to GEM [24] Meijers et al. [2011] demonstrated that the residuals between the historical hydrography and the GEM field are evenly distributed around a zero mean in space, but we observe here that there are statistically significant trends in the residuals with time. In order to achieve suitable temporal 7of17

8 Figure 4. Trends significant at 95% confidence in (left) salinity and (middle and right) temperature residuals to the mean GEM field by time for observations from (a c) 1950 to 2009, (d f) 1980 to 2009, and (g i) 1992 to Figures 4a, 4b, 4d, 4e, 4g, and 4h are given on pressure coordinates and Figures 4c, 4f, and 4i are given on g n surface. Units are years 1 and C yr 1. resolution to observe these trends, all residuals between historical hydrography from the 1920s to 2009 and the time invariant GEM field (not the satgem created by combining the GEM with altimetry) are binned by their synoptic dynamic height and depth for three time periods and combined into a circumpolar stream following mean. Linear least squares regressions of the residuals onto time are then fitted to each height/depth bin to 4500 dbar (Figure 4). The trends are calculated using data from three periods; from 1950 to 2009, 1980 to 2009, and 1992 to All periods show fairly similar distributions of temperature and salinity trends in pressure and density coordinates. The same calculations were performed excluding all Argo data (not shown) to test for possible data biases, but no significant changes were found in the distribution of the trends, although variability was considerably higher due to lower data density. This is evident at depths below 2000 dbar where there is some variability between the three periods due to the paucity 8of17

9 Figure 5. Temperature trends in residuals to the mean GEM fields between 1950 and 2009 binned by frontal region for (a) sssaf PF, (b) PF csaf, (c) csaf nsaf, and (d) nsaf STF. Error bars give 95% confidence level. Units are C yr 1. of data and consequent susceptibility to noise in the time series. [25] On pressure surfaces there is statistically significant (at 95% confidence) warming/salinification following the density contours kg m 3, extending from the southern limit of the domain to around 1.6 dynm. Above and below this layer there is cooling, with a maximum in the upper 1000 dbar north of 1.4 dynm (approximately the SAF). The distribution of the salinity trend is similar to the temperature trend, with freshening above the 27.4 kg m 3 density surface intensifying towards the surface, and a tongue of increasing salinity extending along isopycnals between 27.4 and 28.1 kg m 3. [26] The magnitude of the temperature and salinity trends are small in comparison to those driven by frontal shifts in Section 3.1, and are generally less than C yr 1 and years 1 except in the AAIW and SAMW where the cooling and freshening is greater than 0.01 C yr 1 and years 1 (significantly different from zero at 95% confidence). The rate of change of temperature and salinity is increasing with time, particularly in the upper 500 dbar and north of the SAF. This acceleration agrees with the analysis by Böning et al. [2008] who observe a similar increase in the rate of warming of the Southern Ocean from 1980 onwards. [27] In isopycnal coordinates there is a strong warming/ salinification of the lower AAIW and UCDW in the kg m 3 range of over C yr 1, from the southern edge of the ACC to around the central SAF (Figures 5a and 5b), and strong cooling/freshening of the SAMW/AAIW (up to C yr 1 ) north of 0.9 dynm and the south PF (Figures 5b and 5c). This signal agrees with the fingerprint of anthropogenic warming [Banks and Bindoff, 2003; Aoki et al., 2005a] and is consistent with observations [Bindoff and Church, 1992; Bryden et al., 2003; Wong et al., 1999] and climate change studies [Banks and Bindoff, 2003]. There is also a clear cooling/freshening trend centered in the LCDW south of the spf, and extending to the AABW where the trend is around C yr 1 (Figure 5a), supporting the observational evidence of cooling and freshening bot- 9of17

10 tom waters [Aoki et al., 2005b; Rintoul, 2007] and suggests that this trend observed in the Indian/Australian sector has a circumpolar signal in dynamic height coordinates. Northofthis,inthecoreoftheSAF,thereisevidencefor deepwarmingofupto0.001 Cyr 1 below28kgm 3.The warming/salinification trendintheucdwisreversedat the northern end of the section, and we see either cooling/ freshening or no significant trend north of the nsaf (Figure5d).ThisisincontrasttoBöning et al. [2008] who see warming/salinification in these density classes as far north as 40 S. This difference is possibly because Böning et al. [2008] uses a Eulerian coordinate frame, and so includes changes due to both diabatic and adiabatic (i.e., frontal shifts) causes, while the trend observed in the residuals to the GEM field are purely diabatic. The fact that this difference occurs in the deeper layers near the SAF supports this hypothesis, as there are strong T S gradients along neutral surfaces at the SAF, and so an adiabatic southward frontal shift will induce an apparent warming/ salinification on g n surfaces (see Figures 2c and 2d). A strong warming/salinification adiabatic component observed near a front may then obscure the diabatic cooling/freshening component if observed in Eulerian rather than dynamic height coordinates Interpretation of Trends in Dynamic Height Coordinates [28] Trends in the residuals to the mean GEM fields are difficult to link to temperature and salinity trends at fixed points in the water column due to the feedback that temperature and salinity changes have in dynamic height coordinates. This is demonstrated in Figure 6 where the upper layers of the mean circumpolar GEM is artificially warmed (by 0.05 C at the surface, linearly decreasing to 0 C at 500dbar)andthesalinityfieldisheldconstant.Whenthe dynamic heights are recalculated for the warmed field, the profiles are effectively shifted to the right on the x axis by the reduced density in the upper 500 dbar. The difference between the original GEM T S fields and the new warmed fields show a much deeper and more complex signal in both temperature and salinity than just a simple warming in the upper 500 dbar. Notably there is apparent cooling at depths below about 500 dbar, and strong freshening on the northern side of the S pmin AAIW layer and increasing salinity south of this, despite the fact that the salinity and temperature distribution below 500 m has not changed. By definition any change in density/specific volume in the upper ocean in dynamic height coordinates must be compensated by changes of the opposite sign in the deeper ocean, so as to keep the vertical integral of the density field (the dynamic height) unchanged. [29] Sokolov and Rintoul [2009] shows however, that hydrographically defined fronts in the ACC remain very closely associated with a unique dynamic height over the altimetric record. This indicates that there is relatively little diabatic contribution to net dynamic height change and consequently little aliasing of the diabatic trend into the adiabatic, which would produce apparent deep changes of the type shown in Figures 6c and 6d. This is supported by Figures 4a and 4b where the observed trends between 1950 and 2009 are only significant above 2000 dbar, and are strongest near the surface and along the AAIM tongue where water masses may have had relatively recent contact with the mixed layer. Additionally, these changes are largely density compensated between temperature and salinity, and so have relatively small impacts on the net dynamic height. This contrasts with Figure 6 where the temperature and salinity trends clearly do not density compensate each other. Therefore it seems likely that the displacement of dynamic height contours in the Southern Ocean are largely due to adiabatic water mass movement, while net diabatic contributions to MSLA changes may be relatively small in comparison. 5. Implied Heat and Freshwater Fluxes [30] In order to quantify the components of change in heat and freshwater driven by the respective adiabatic and diabatic trends identified in the previous sections, we add the temporal diabatic trend to the time evolving adiabatic sat- GEM fields and calculate the resulting change in heat and freshwater content in the satgem latitude/longitude domain (Figure 1). To more accurately determine the diabatic trend (the binned linear regressions in Figure 4 do not exactly conserve dynamic height) we construct two GEM fields; one using hydrography up to the Argo period (2002), and one using observations made after this. These are created as twodimensional stream following circumpolar means to increase the accuracy of the trend, as there is insufficient data density for three dimensional GEM fields in different temporal periods. The difference between these GEMs (Figure 7) is analogous to the trends observed in Figures 4d 4f, with the temporal mean of the first GEM being approximately 1982, and the second Only summer (January March) hydrography is used, to avoid seasonal biases, so the difference between the the two GEM fields is not necessarily directly comparable to Figure 4, although they do show very similar signals. [31] At each weekly time step in the satgem the linear diabatic trend determined from the difference in GEMs is incrementally added to the satgem temperature and salinity profiles at the appropriate depths and dynamic height. Note that because the later mean GEM was created using Argo data, the diabatic component of change only extends to 2000 dbar. Figure 4 suggests that the majority of the diabatic trends occur in the surface 2000 dbar, at least north of the spf, so this limitation is not a particularly restrictive condition in this case. The resulting temperature and salinity fields are essentially satgems where the vertical profiles diabatically evolve in time as well as adiabatically translate horizontally. [32] The total Eulerian heat content (Q) of the domain is calculated at each weekly time step as the volume integral Z Z Z Q ¼ C p dxdydz ð4þ where C p is the specific heat capacity and r is the in situ density. The total freshwater content is similarly defined as the volume integral of the negative salt anomaly S ¼ S p S p0 S p0 ð5þ 10 of 17

11 Figure 6. Mean circumpolar (a) temperature and (b) salinity GEM fields. The surface 500 dbar of the GEM is artificially warmed (see text for details) and the dynamic heights for the resultant T S fields are recalculated, producing a warmed GEM temperature and salinity field. The difference between the warmed and original GEM (c) temperature and (d) salinity fields are show. Units are C and practical salinity units. in place of, where S p is the in situ practical salinity and S p0 is a reference practical salinity (35). The physical volume used here is the circumpolar region for which there is sat- GEM coverage at every weekly interval between October 1992 and March 2010, as defined by the valid dynamic height range of dynm. The exact region is further restricted by bathymetry, as the satgem fields are invalid in water shallower than 2000 m and in areas of sea ice extent that prevents SSH observation in the altimetric data. The integration domain is shown in Figure 1, and has an area of m 2. [33] The resultant trends in heat and freshwater content of the satgem domain over the altimetric period are shown in Figure 8 and their linear regressions are given in Table 1. The major influence on heat content variability is the changing volumetric extent of each dynamic height bin. This is largely driven by the seasonal adiabatic signal and higher frequency mesoscale activity (shown at 1 standard deviation), but there is still an unambiguous trend towards increasing heat in both the purely adiabatic heat content in the unmodified satgem, and the satgem + diabatic trend. The increase in adiabatic heat content is driven by the southward movement of the frontal positions over the altimetric period [Sokolov and Rintoul, 2009]. The increase in the purely adiabatic heat content is ± PW, while the diabatic contribution to this (concentrated largely 11 of 17

12 Figure 7. (left) Temperature and (right) salinity differences between GEMs with temporal means of 2006 and 1982 on (a and b) pressure surfaces and (c and d) neutral density surfaces. In Figures 7c and 7d the black line represents the temperature minimum, the blue line represents the salinity maximum, and the white line represents the salinity minimum. Note depths are only to 2000 dbar due to the usage of Argo data. Units are C yr 1 and years 1. in the AAIW, SAMW and AABW) is a net cooling of ± PW, resulting in an overall warming of ± PW. There is also a clear difference between the adiabatic and adiabatic + diabatic trend freshwater content. Although the adiabatic freshwater volume mean trend (0.010 ± Sv) is relatively small, the contribution by the diabatic flux is much larger (0.044 ± Sv). This result demonstrates that although the movement of water masses is slightly increasing the net ACC freshwater content, there is a significant additional input of freshwater to the region and that this diabatic effect is larger than the adiabatic contribution. [34] The differences in the diabatic/adiabatic trends explain the inconsistencies between the adiabatic trends caused by 12 of 17

13 Figure 8. Total volume integrals of (a) heat and (b) freshwater over the satgem domain by time (1 year running mean). The trend for the (black) adiabatic satgem alone and (blue) adiabatic satgem with diabatic GEM residual trend are given for both heat and freshwater. Colored areas denote 1 standard error and are caused by short period variability. 13 of 17

14 Table 1. Volume Integrated Heat and Freshwater Trends From 1992 to 2010 Over the ACC a Heat Trend (PW) Heat Trend (W m 2 ) Freshwater Trend (Sv) Freshwater Trend (mm yr 1 m 2 ) Diabatic ± ± ± ± 0.70 Adiabatic ± ± ± ± 0.18 Combined ± ± ± ± 0.72 a Trends are given for purely adiabatic changes, as well as diabatic and the sum of the two components. Trends normalized by area use the satgem domain area of ± m 2. One standard error is given. frontal movement and the in situ observations discussed in Section 3.3. The mean adiabatic warming trend is 1.20 ± 0.09 W m 2, approximately twice that estimated by Gille [2008] once the discrepancy in observational areas is accounted for. Similarly it is 2 3 times greater than the mean global estimates of Bindoff et al. [2007] (0.5 ± 0.18 W m 2 ) and Levitus et al. [2005] (0.42 ± 0.18 W m 2 ) for the period Once the diabatic cooling trend is added to the adiabatic trend, however, there is a much closer agreement between the heating trend (0.570 ± W m 2 ) and these other studies. The net adiabatic freshwater trend has a domain mean of only 6.57 ± 0.18 mm yr 1 m 2. This small value is largely due to the near balance between salinification of the SAMW/AAIW, and freshening below and to the south of these water masses in the satgem fields. However, this does not agree with observational evidence for freshening over the whole Southern Ocean [Boyer et al., 2005; Helm et al., 2010; Böning et al., 2008; Durack and Wijffels, 2010]. The diabatic trend obtained from the time evolving residuals to GEM, however, accounts for this discrepancy, driving a strong increase in the net volume of freshwater. This is equivalent to an additional increase in freshwater from advection into the domain or by changed surface fluxes of approximately ± Sv (30.27 ± 0.70 mm yr 1 m 2 ) and agrees with observations much more closely. Additionally, this freshening occurs primarily in the AAIW and SAMW north of the PF on pressure surfaces, largely canceling out the adiabatic salinification from the satgem estimates due to frontal shift and thus bringing the Eulerian trend into agreement with the estimate by Böning et al. [2008]. 6. Discussion 6.1. Attribution of Property Trends [35] The cause of the observed net diabatic cooling and freshening over the ACC is difficult to ascertain. As discussed in Section 4.2 the meridional distribution of changes is complex due to the dynamic height coordinate system. However, the concentration of trends near the surface and in ventilated water masses may suggest two likely causes; increased precipitation or increased sea ice/glacial melt. Reduced mixing or eddy transport from north of the ACC may also be possible, but as eddy transport is thought to be enhanced during positive SAM [Fyfe and Saenko, 2007; Screen et al., 2009] and there has been a positive SAM trend between 1950 and 2000 [Marshall, 2003], it would seem unlikely that this is driving the observed cooling and freshening. Both increased precipitation and ice melt will freshen the surface layers around Antarctica and be transported northward by Ekman transport, subducting and freshening the SAMW and AAIW layers. Increased air/sea heat loss could drive increased cooling, particularly in the well mixed SAMW, as the frontal systems penetrate further south. Sokolov and Rintoul [2009] show, however, that the mean southward displacement of the fronts is approximately 60 km between 1992 and 2007, significantly less than the atmospheric synoptic scale, so it seems unlikely that this shift would be sufficient to drive large enough changes in ocean/atmosphere fluxes to produce the observed cooling. An increase in precipitation, as observed by Durack and Wijffels [2010], possibly caused by an increased southward penetration of warm moist air due to a positive SAM [Hall and Visbeck, 2002], would drive enhanced freshwater fluxes at the surface and in the subducting AAIW and SAMW as observed. However, this does not also explain the observed diabatic cooling on pressure surfaces in the SAMW/AAIW outcropping layers, nor the cooling and freshening of AABW. [36] Fyfe and Saenko [2007] show in a coupled ocean/sea ice model that Antarctic ice melt may be advected north via Ekman fluxes and cool the upper ocean layers. Enhanced ice melt is also consistent with the cooling and freshening of AABW, something that is observed both in situ [Rintoul, 2007; Aoki et al., 2005a], in climate change studies [Helm et al., 2010], as well as in this paper. However, the additional melt required to produce the observed diabatic freshening in this study is around 1387 ± 63 Gt yr 1. This is of a similar magnitude to the net freshwater flux out of Antarctica (1621 ± 32 Gt yr 1 for 2006) and an order of magnitude greater than increase in outflow of 75 ± 44 Gt yr 1 between 1996 and 2006 [Rignot et al., 2008]. Therefore, it would appear that ice loss alone is insufficient to explain the observed changes in the ACC freshwater content. Speculatively, the observed southward shift of the ACC fronts and net warming could provide a possible mechanism by which increased melt may occur through enhanced export of CDW onto the continental shelf with subsequently increased basal melting of ice shelves. The factors driving the diabatic trends remain unclear and will no doubt be the subject of future investigations Comparison With Trends Observed in Eulerian Coordinates [37] An interesting result of the satgem trends is the demonstrated ability of adiabatic frontal movements to produce trends on density surfaces in fronts and outcropping regions when viewed in an Eulerian coordinate system. This is clear when the adiabatic and diabatic components of the satgem temperature trend and their sum are compared with the observational trend in Eulerian coordinates (subsequently averaged along circumpolar streamlines) calculated by Böning et al. [2008] in Figure 9. While the GEM diabatic component (Figure 9a) appears qualitatively similar to the Eulerian study (Figure 9d), the diabatic warming trend 14 of 17

15 Figure 9. Mean annual temperature trend on neutral density surfaces for the (a) diabatic, (b) adiabatic, (c) diabatic and adiabatic components, and (d) the mean trend calculated in Eulerian coordinates (reproduced from Böning et al. [2008]). Trends above 100 dbar and below 2000 dbar have been excluded. Units are C yr 1. in the CDW south of the PF has a magnitude of only half of what is observed, while the diabatic cooling and freshening of the AAIW and SAMW (g n < 27.4 kg m 3 ) in the SAF is substantially greater than in observations. In a similar vein the in situ warming trend at kg m 3 north of 50 S does not appear in the GEM diabatic component. These differences are due to the adiabatic warming caused by the southward frontal displacement present in the Eulerian observations but not in the purely diabatic case observed in dynamic height coordinates. The addition of the adiabatic component (Figure 9b) to the diabatic component gives an overall trend (Figure 9c) quantitatively in agreement with the in situ observations due to the increased warming in the surface outcropping layers above 500 dbar, particularly south of 54 S, and in the SAF near 50 S down to around 1000 dbar. The fact that the diabatic and adiabatic trends were estimated using independent methodologies and produce a net trend closely in agreement with observations gives us confidence in the methods and results presented here. [38] This comparison illustrates that while observing trends on isopycnal surfaces at fixed geographic locations reduces the effect of adiabatic water mass movement, it will not remove it all together. Therefore adiabatic movement should be considered as a factor when estimating heat/freshwater changes in regions where there are strong T S gradients on 15 of 17