Carbon uptake and biogeochemical change in the Southern Ocean, south of Tasmania

. Biogeochemical change in the water masses of the Southern Ocean, south of Tasmania, was assessed for the 16-year period between 1995 and 2011 using data from 4 summer repeats of the WOCE/JGOFS/CLIVAR/GO-SHIP SR03 hydrographic section (at ~140°E). Changes in temperature, salinity, oxygen, and nutrients were used to disentangle the effect of solubility, biology, circulation and anthropogenic carbon (C ANT ) uptake on the variability of dissolved inorganic carbon 10 (DIC) for 8 water mass layers defined by neutral surfaces ( γ n ). C ANT was estimated using an improved back-calculation method. Warming (~0.0352 ± 0.0170 ° C yr -1 ) of Subtropical Central Water (STCW) and Antarctic Surface Water (AASW) layers decreased their gas solubility, and accordingly DIC concentrations increased less rapidly than expected from equilibration with rising atmospheric CO 2 (~0.86 ± 0.16 μmol kg -1 yr -1 versus ~ 1 ± 0.12 μmol kg -1 yr -1 ). An increase in apparent oxygen utilisation (AOU) occurred in these layers due to either remineralization of organic matter or intensification of upwelling. The 15 range of estimates for the increases of C ANT were 0.71 ± 0.08 to 0.93 ± 0.08 μmol kg -1 yr -1 for STCW and 0.35 ± 0.14 to 0.65 ± 0.21 μmol kg -1 yr -1 for AASW, with the lower values in each water mass obtained by assigning all the AOU change to remineralization.

The Southern Ocean is a key region in terms of climate change and climate variability, influencing the Meridional Overturning Circulation and therefore modulating the global circulation and oceanic biogeochemical cycles (Sarmiento et al., 1998(Sarmiento et al., , 2004Orr et al., 2005). Deep waters, formed in the North Atlantic, spread south and enter the Southern Ocean, where they mix with deep layers of the Antarctic Circumpolar Current (ACC) and ultimately upwell between the Southern ACC Front and the Polar 5 Front. The upwelled waters are eventually transformed into bottom, intermediate and mode waters, which are exported from the Southern Ocean to ventilate the thermocline and bottom layers of the major ocean basins. Some of the Southern Ocean waters subducted into the ocean interior return to the North Atlantic to balance the southward flux of North Atlantic Deep Water (Speer et al., 2000;Lumpkin and Speer, 2007;Iudicone et al., 2008).
Within the eastward flow of the ACC major water exchange between the three ocean basins takes place. The circumpolar path 10 of the ACC consists of various narrow jets associated with sharp fronts that separate waters with different characteristics (Orsi et al., 1995;Belkin and Gordon, 1996). These jets can reach deep layers and often meander, intensify, merge and split, conditioned by the topography of the ocean floor, the stratification of the ACC, and atmospheric variability (Moore et al., 1999;Rintoul, 2002, 2009;Peña-Molino et al., 2014). Movements in the jets enhance cross-stream transports and mesoscale activity that can result in local changes of water mass properties and this may complicate the computation of long-15 term changes in water mass properties (Rintoul and Bullister, 1999;Sallée et al., 2008;Peña-Molino et al., 2014).
Water mass formation and ventilation transport heat, salt, and dissolved gases from the atmosphere to the ocean interior and other basins (Sarmiento et al., 2004), with the Southern Ocean contributing ~40% to the anthropogenic CO2 (CANT) inventory of the ocean (Sabine et al., 2004;Gruber et al., 2009;Khatiwala et al., 2009). Circulation and biological processes drive the redistribution of dissolved inorganic carbon (DIC) that ultimately affects the capacity of the waters to uptake more CO2. The 20 uptake of CO2 by the Southern Ocean presents strong spatiotemporal variability (Lenton et al, 2013) and this can lead to conflicting results for observational studies, models and atmospheric inversions depending on the methodology used (Verdy et al., 2007;Lenton et al., 2012;Fay et al., 2014). Quantifying long-term changes in the carbon system is difficult due to the scarcity of data Kouketsu and Murata, 2014;Fay et al., 2014) and the influence of biological processes.
Notably, long term trends in CANT concentration are difficult to estimate due to its small signal (~3%) with respect to that of 25 DIC in the ocean.
Once CO2 dissolves in the ocean (DIC) it begins the process of ocean acidification, i.e., decreases the pH and the saturation state of calcium carbonate (CaCO3) minerals such as calcite and aragonite (Feely et al., 2004;Bates et al., 2014), with potential to disrupt ecosystems and biological processes (Doney et al., 2009).
Numerous studies have documented warming and freshening of deep and bottom layers of the Southern Ocean in recent 30 decades (see reviews by Jacobs, 2006 andvan Wijk and. The abyssal waters of the Australian-Antarctic Basin (A-AB) show the greatest freshening in the last 40 years (van Wijk and . This freshening was accompanied by a warming of the deep-bottom layers, leading to a contraction in bottom waters by more than half their volume in the basin 3 (Purkey and Johnson, 2012;van Wijk and Rintoul, 2014). Subsurface to intermediate layers have also warmed and freshened south of Australia (Bindoff and Church, 1992;Wong et al., 1999;Aoki et al., 2005). The solubility of gases and its distribution in the ocean depends on the dynamics and properties of water masses, and the thermohaline changes that have occurred south of Tasmania ( Fig. 1) have implications in the carbon system of the region.
A reduction of the carbon sink of the Southern Ocean was observed between the 1980s and the earlies 2000s (Le Quéré et al., 5 2007;Lovenduski et al., 2008), with more recent studies suggesting a recovery and even intensification of the CO2 uptake by (Zickfeld et al., 2008Fay et al., 2014;Landschützer et al., 2015). DeVries et al. (2017), used a global inverse model to postulate that changes in circulation are responsible for most of the variability in the oceanic CO2 uptake, with the weakening of the upper-ocean circulation being responsible for the increase in oceanic carbon uptake over the past decade.
Ocean acidification has been observed in the whole Southern Ocean (Lauvset et al., 2015) and locally in the Atlantic and 10 Pacific sectors (Williams et al, 2015;Hauri, et al., 2015). South of Tasmania, McNeil et al., (2001) reported an increase in CANT uptake between 1968 and 1996 for the region 45-50°S. These authors reported, for the first time, CANT accumulation in the AABW and highlighted the importance of the formation of bottom and mode waters as a mechanism for transporting CANT to the ocean. In terms of ocean acidification, we are not aware of any study about trends in ocean acidification in the water masses south of Australia. 15 Considering the lack of observational estimates for recent biogeochemical changes in the A-AB as well as the large changes in CO2 uptake and storage suggested by recent atmospheric and surface observations in the Southern Ocean (e.g., Fay et al., 2014;Landschützer et al., 2015) there is a need to provide a full ocean depth observational perspective on how the ocean is changing. The aim of this paper is to provide the first estimates of biogeochemical change in the water masses south of Tasmania, for the period 1995-2011, disentangling the effects that solubility, circulation, biology and CANT uptake have on the 20 variability of DIC. We use data from four summer repeats of the WOCE/JGOFS/CLIVAR/GO-SHIP hydrographic section SR03 ( Fig. 1; Table 1), one of the most revisited sections in the Southern Ocean. Trends in oxygen (O2), nutrients, and the carbon system parameters, i.e., DIC, total alkalinity (TA), anthropogenic carbon (CANT), total pH (pHT) and % aragonite saturation (ΩAr) were estimated for the period 1995-2011, when both DIC and TA measurements are available. CANT estimates were obtained with a back-calculation method (Pardo et al., 2014). The changes were evaluated in the different water mass 25 layers of the section defined by neutral surfaces (γ n , McDougall et al., 1987).

Hydrography of the region.
The dynamical structure of the region south of Tasmania ( Fig. 1) is characterized by a number of fronts that separate the major water masses of the region (Sokolov and Rintoul 2002, 2009. At the northern end of the section, the presence of the weak Subtropical Front (STF, Fig. 1) separates warm, salty subtropical surface waters from cooler and fresher sub-Antarctic 30 surface waters (Deacon, 1937). The northern end of the section is a complex mixing zone where waters transported down the east coast of Tasmania in a series of mesoscale eddies from the East Australian Current (EAC, Fig. 1) mix into the Subantarctic 4 Zone, and also meet Zeehan Current (ZC, Fig. 1) waters transported down the west coast of Tasmania (Boland and Church 1981;Baines et al., 1983;Speich et al., 2002;Davis, 2005;Ridgway et al., 2007;Sloyan et al., 2016). The EAC transported south of Tasmania forms a zonal jet towards the southeast Indian Ocean known as the Tasman Outflow that reaches the bottom of the Tasman slope and that is maintained all year round (Rintoul and Bullister, 1999;Ridgway et al., 2007). The encounter between these currents presents high variability at the northern end of the section (Fig. 1). 5 Farther south (Fig. 1), the Sub-Antarctic Front (SAF) and the Polar Front (PF) are regions of maximum transport in the ACC (Rintoul and Bullister, 1999;Sokolov and Rintoul 2002). North of the SAF, deep winter convection generates Sub-Antarctic Mode Water (SAMW), a relatively uniform water mass that occupies subsurface layers down to ~600m deep (McCartney, 1977;Rintoul and Bullister, 1999). SAMW constitutes the main part of the upper limb of the MOC, ventilating the thermocline of all the ocean basins (e.g., Speer et al., 2000;Sloyan and Rintoul, 2001). The SAF coincides with the deepening of the salinity 10 minimum at intermediate depths (Whitworth and Nowlin, 1987), which is the signature of the Antarctic Intermediate Water (AAIW). The AAIW underlies the SAMW and also ventilates the global thermocline layers of the ocean. It is mainly formed in the southeast Pacific and it is continuously transformed on its way to the region south of Tasmania (Hanawa and Talley, 2001). South of the SAF, colder and fresher Antarctic Surface Water (AASW) covers the surface ocean. AASW originates from progressive warming of Winter Water (WW, Mosby, 1934), which can be perceived even in summer as a remnant layer 15 of cold water at the base of the AASW (Rintoul et al. 1997).
The Southern ACC Front (SACCF, Fig. 1) is a deep front located south of the PF (Orsi et al., 1995) and can coincide with the southern boundary of the ACC, which is represented by the southern limit of the oxygen minimum (Orsi et al., 1995). The oxygen minimum is related to the advection of Upper Circumpolar Deep Water (UCDW) that originates in the Indian and Pacific Oceans (Callahan, 1972) from where it spreads south, mixes with deep layers of the ACC and ultimately upwells near 20 the Antarctic continent as part of the lower cell of the MOC. Lower Circumpolar Deep Water (LCDW) is below UCDW and is identified by a salinity maximum and contributes to upwelling around Antarctica. The precursor of the LCDW is North Atlantic Deep Water (NADW), originated in the Labrador and Nordic seas of the North Atlantic polar region (Dickson and Brown, 1984) that flows southward to enter the ACC as part of the MOC (Callahan, 1972;Orsi et al., 1995;Johnson, 2008).
At the southern end of the section, the Antarctic Slope Front forms the boundary between cold and fresh shelf water and 25 relatively warm and salty waters offshore (Jacobs, 1991). Polynyas along the Adélie and George V Land coast ( Fig.1) contribute to the formation of Adélie Land Bottom Water (ALBW), which is a mixture of High Salinity Shelf Water (HSSW) resulting from brine rejection during ice formation and ultra-modified LCDW (Foster and Carmack, 1976;Rintoul, 1998;Marsland et al., 2004). ALBW constitutes ~ 25% of the total volume of water <0°C in the ocean (Rintoul, 1998). The bottom waters near the southern end of the section also contain a component of Ross Sea Bottom Water (RSBW) that originates to the 30 east and is deflected westwards towards the A-AB as it is transported down the continental slope. The RSBW is modified when it arrives at the location of the SR03 section by mixing with deep layers of the ACC and recently formed ALBW (Gordon and Tchernia, 1972;Rintoul, 1998). The ALBW and the modified RSBW together ventilate the abyssal layers of the A-AB before spreading north to ventilate the Indian and Pacific basins (Mantyla and Reid, 1995;Fukamachi et al. 2010). The SR03 hydrographic section between Tasmania and Antarctica ( Fig. 1) was occupied between 1991 and 2011.
Measurements of total alkalinity (TA) were only available from the beginning of 1995 and our evaluation of biogeochemical changes is limited to four summer sections occupied for the period 1995-2011 (Table 1). A winter cruise in 1996 was not 5 considered in this study in order to minimise seasonal biases.
Water column salinity, temperature, pressure and dissolved oxygen (O2) were collected from the conductivity-temperaturedepth (CTD) device with an accuracy of ±0.002 for salinity and temperature, ±0.015 % of full scale range for pressure and ±1 % for O2, according to WOCE standards (Joyce, 1994). Samples from the Niskin bottles were analysed for dissolved inorganic carbon (DIC) by coulometry and TA by open cell potentiometric titration (Dickson et al, 2007). Certified reference material 10 provided by A. Dickson, Scripps Institution of Oceanography, were used as reference standards for DIC and TA. The precisions of DIC and TA measurements improved slightly on more recent sections, and for all sections were better than ±2 μmol kg -1 , for both variables, based on analysis of duplicate samples and certified reference material. Samples for dissolved O2 were measured using modified Winkler titrations (Hood et al 2010), with an estimated accuracy and precision of ± 0.3% for the sections. For the 1995-cruise, sensor based O2 was used instead of sampled O2 because of the poor quality of many of the 15 Winkler measurements.
Other variables of the dissolved CO2 system were calculated from the DIC and TA measurements. We calculate pHT and ΩAr (from measured DIC and TA) using the CO2sys program from Lewis and Wallace (1998) adapted to MATLAB by van Heuven et al. (2011). We use the constants for the carbonic acid from Mehrbach et al (1973) refit by Dickson and Millero (1987), the CO2 solubility equation from Weiss (1974), dissociation constants for sulphate from Dickson (1990) and borate constant from 20 Uppstrom (1974). Aragonite saturation states (ΩAr) are calculated because it is a less stable form of CaCO3 than calcite and is the predominant biogenic form of CaCO3 precipitated by calcifying organisms.
Data presented here were interpolated to a regular grid, and the water mass layers were defined as layers between neutral surfaces (γ n ) determined using potential temperature and salinity and published literature values (Table 2, section 2). The first 50m of the water column were eliminated in order to reduce the short-time scale variability in surface properties. The use of 25 γ n to identify the water mass layers reduces the variability due to isopycnal heave caused for example by eddies and internal waves (McDougall et al., 1987;Bindoff and McDougall, 1994;Jackett and McDougall, 1997). The upper ocean layers south of the SAF were divided into the AASW layer to the north of the PF, and the AASWupw layer that is composed of surface waters south of the PF (Fig.1). ADLBW and RSBW were included in the AABW layer ( Table 2). The ACC fronts along the SR03 section (Table 3) were defined as a function of hydrographic variables following Sokolov and Rintoul (2002). 30

combined with an Optimum
Multi-Parameter (OMP) analysis (Tomczak, 1981), described by Pardo et al. (2014). This technique has the advantage of considering water mass mixing and the temporal variability of the air-sea CO2 disequilibrium. The accuracy of the method is ±6 μmol kg -1 (Pardo et al., 2014). Back-calculation methods assume the ocean is in steady state for dynamical and biological 5 processes and estimate CANT (CANT_BC) as an excess of DIC in the ocean resulting from the increase of atmospheric CO2 due to anthropogenic emissions as: where is the biological contribution to DIC, and is the preformed concentration of DIC in preindustrial times.
represents the remineralization of organic matter, with the apparent oxygen utilization (AOU) defined as (AOU = OSAT -15 O2) the difference between the saturation of oxygen (OSAT) at the potential temperature (θ) and salinity of the measured O2.
The preformed preindustrial term, , is the total concentration of carbon dioxide in seawater saturated with respect to the 20 preindustrial pCO2 ( ) and corrected for an air-sea CO2 disequilibrium ( ) term: where is the current disequilibrium between the ocean and the atmosphere pCO2, and is the change in the 25 disequilibrium from preindustrial to current times.
was obtained by similar parameterizations to those used for 0 , combined with monthly mean values of atmospheric CO2 values from the NOAA network (Dlugokencky, et al., 2016) (see section A3 of the appendix and Table A3). The values were obtained from results of the 1/10° resolution carbon model OFAM3-WOMBAT (Appendix section A1).
Interior values of preformed variables ( 0 and ) and were obtained using an optimum multiparameter (OMP) 30 analysis to mix end members as described in Appendix sections A2 and A3, using the end members in Table A1. The OMP analysis is based on the assumption that a property measured in a certain point is the result of linear mixing between end members, known as source water types (SWTs). A system of equations is created for each measurement point and is solved to obtain the fractions of the different SWTs (Appendix section A2). The application of the OMP analysis requires good regional hydrodynamic knowledge as the results are strongly dependant on the definition of the SWTs (Tomczak, 1981). We used 11 SWTs to characterize the biogeochemical properties of the waters in the SR03 section and the SWT properties were assumed to be constant with time (Appendix Table A1, Fig. A1). 5 Small negative values of CANT_BC can occur due to an overestimation of the term (Eq. 4), that acts as measure of the age of the water mass and has high values in old deep layers (see Table A1 in the appendix). These negative values of CANT_BC were found at some points in deep waters of the section (mainly UCDW and LCDW layers) and were small (between 0 and -2 μmol kg -1 , i.e., less than the accuracy of the methodology) and changed to zero for our analysis.

Changes in the carbon system. 10
Changes in carbon system parameters ( , , _ , and Ω ) were estimated using linear regressions with time for the period 1995-2011 in each one of the water mass layers defined by their γ n condition ( Table 2). The trends were estimated using all the points in each water mass layer and only those linear trends with p<0.05 and r>=0.2 were considered statistically significant for discussion. We show the value of the root mean square error (RMSE or square root of the variance of the residuals), which can be interpreted in large part as unexplained variance caused by short-time scale processes and the different 15 seasonal timings of the cruises. RMSE has the same units as the response variable.
Respect to the total change of DIC ( ), our goal is to disentangle the effects that solubility, circulation, biology and CANT uptake have on the variability of DIC. The total change of DIC ( ) in a water mass is due to changes in the atmosphereocean interchange, biological processes and circulation processes. In order to account of the change in DIC due the atmosphereocean interchange and biological processes, we compare to _ and , Eq.
(2). The change in DIC not 20 explained by _ or will then be due to circulation processes.
In order to compare to _ we need to consider how the change in and ( ) (terms of Eq (1), section 3.2) affect the changes in CANT_BC.
The term can be expressed as: The terms and reflect changes in the properties of the water masses over time, primarily temperature and salinity change due to mixing and heating/cooling. and are defined at the ocean surface (in each of the SWTs: Table   A1) and are calculated at each point in the ocean interior using the OMP analysis (Appendix sections A2 and A3). Because a change in temperature and/or salinity in the water is solved by the OMP analysis as a change in the SWTs fractions, this also produces varying and . No significant trends were obtained for in any of the layers. CANT_BC (Eq. 1) is not affected by the changes in solubility occurring from one voyage to another (thus neither is _ ), since any change in temperature or salinity is cancelled out by the subtraction of with respect to DIC in Eq. (1) (section 3.2) and by O2 respect to OSAT in the term (Eq. 2). However, based on the measured DIC in the sections will be affected by changes in the solubility over time and this difference needs to be accounted for when comparing with _ to obtain 5 a better approximation for the change in CANT: = _ + .
The term can be influenced by changes with time of alkalinity due to changes in the rate of carbonate precipitation/dissolution and of AOU due to changes in the rate of remineralization and in circulation. In the present study only surface waters of the SR03 section present changes between 1995 and 2011. Numerous studies have reported a strong influence of biological communities in the seasonal cycle of dissolved O2 in surface waters (Bender et al., 1996;Moore and 10 Abbott, 2000;Sambrotto and Mace, 2000;Trull et al., 2001a). Interannual variability in O2 in upper layers of the Southern Ocean have also been related to changes in the entrainment of deeper waters into the mixed layer due to the mixed layer depth variability (Matear et al., 2000;Verdy et al., 2007;Sabine et al., 2008;Sallée et al., 2012). Although some studies found longterm decreases in O2 due to circulation in deep waters of the Weddell Sea (van Heuven et al., 2014) and for the first 1000 m of the global ocean (Helm et al. 2011), significant long-term trends in O2 due to circulation and remineralization processes 15 have not yet been reported for surface waters of the Southern Ocean. Thus, the term may also contribute to variation in _ , since part of the changes in AOU with time reflect changes in circulation that we cannot separate from those in remineralization. We consider the best approximation for the change in CANT as a range depending on the possible effect of biology and circulation processes on . If the value of is due to the variability in the remineralization rates and the change in solubility is considered, the estimate _ will be the lower limit of the range, (lower limit of = 20 _ + ). For the upper limit of the range, we consider that the value of is due to changes in circulation and the upper limit of the range is obtained by = _ + + . We assume that the changes in due to circulation do not affect the amount of DIC in the layer. This assumption is one of the caveats of the methodology, since we cannot know how much of the change in DIC is associated to changes in circulation, i.e., how much of the change in DIC is a change in non-anthropogenic DIC. We will discuss this more in section 6.2. 25 4 Results.
The different trends in biogeochemical properties are summarized in Table 4. The he biogeochemical changes between 1995 and 2011 are presented for each of the water mass layers and the effect of changes in solubility, biological processes and circulation in the estimates of and are considered along with changes in the aragonite saturation depth and CANT storage.

Upper ocean layers (STCW, AASW and AASWupw). 5
In the STCW layer, DIC increased between 1995 and 2011 (Fig. 2a, b) at a rate of 0.86 ± 0.07 μmol kg -1 yr -1 ( , Table 4), leading to a decrease of pHT of -0.0027 ± 0.0001 yr -1 ( , Table 4, Fig. 2e, f). The trend in pHT is similar to the one found by Lauvset et al. (2015) between 1991 and 2011 for the IO-STPS (Indian Ocean subtropical permanently stratified) biome (-0.0027 ± 0.0005 yr -1 ). We found a decrease of ( = -0.34 ± 0.06 μmol kg -1 yr -1 , Table 5) in the STCW layer due to a negative trend of resulting from a decrease in solubility that resulted from a temperature increase (calculated from the 10 section data) in the STCW layer of 0.0335 ± 0.0130 °C yr -1 (not shown). The increase in temperature agrees with the warming trend observed south of Tasmania of 0.2 to 0.3 °C decade -1 obtained from satellite data (Armour and Bitz, 2015) and from combined data and models (0.5 °C / 30 yr -1 , Aoki et al., 2015). For θ=16 °C and S=35.1 (definition of SWTSTW16 in the OMP analysis, Table A1), a change in temperature of 0.03 °C yr -1 would lead to a decrease in of -0.27 μmol kg -1 yr -1 , which is similar to the value obtained for in the STCW layer ( Table 5). The difference between these trends is related to the 15 mixing of the different SWTs fractions within the STCW layer established from the OMP analysis (see section 3 and section A2 of the Appendix). When the solubility change is incorporated into _ , i.e. _ + ( Table 5, Fig. 2c, d), we obtain a value of 0.71 ± 0.08 μmol kg -1 yr -1 .
There is an increase of in the STCW layer ( , Table 5), that also affects the estimates of _ . The increase of is due to an increase in AOU (no changes were found in TA, Eq. (2) in section 3.2) due to a decrease of O2 in the 20 layer. We cannot separate the effects of circulation and biology on the AOU change and in Table 4 should be considered a range. If the changes in AOU are only due to the variability in the remineralization rates, the calculated lower limit of is 0.71 ± 0.08 μmol kg -1 yr -1 (Table 4, = _ + in Table 5). If the changes in AOU are due to changes in circulation, the upper limit value of 0.93 ± 0.11 μmol kg -1 yr -1 ( = _ + + ) will explain the increase of DIC in the STCW layer ( ≈ , Table 4). The increase of CANT found in this layer is comparable to the 25 range of increase (0.8 -1.3 μmol kg -1 yr -1 ) found by Carter et al. (2017) in the Pacific Ocean (P16 WOCE, CLIVAR and GOSHIP lines) for the two past decades (1990s-2000s and 2000s-2010s).
Changes in the AASW layer are summarised in Table 4. DIC increased at a similar rate of 0.85 ± 0.14 μmol kg -1 yr -1 to the STCW layer and the trend is similar to the values found by Williams et al. (2015) for the AASW layer in the Pacific sector of the SO (12-18 μmol kg -1 for the period 1992-2011 and 3-5 μmol kg -1 for the period 2005-2011). The increase of DIC in the AASW layer results in a pHT decrease of -0.0035 ± 0.0002 yr -1 , close to Williams et al. (2015) estimates for surface waters (~ -0.0023 ± 0.0009 yr -1 ) and Lauvset et al. (2015) estimates of -0.0021 ± 0.0002 yr -1 for the Southern Ocean seasonally stratified, 5 SO-SPSS, biome. The AASW layer for our sections warmed at a similar rate (0.0369 ± 0.0109 °C yr -1 ) to the STCW layer, reducing the solubility and influencing (Table 5) due to changes in . The also increased with time (0.50 ± 0.16 μmol kg -1 yr -1 , Table 5) due to an increase of AOU. Following the same reasoning as for the STCW layer and considering the trend in CANT of 0.70 ± 0.06 μmol kg -1 yr -1 obtained by the back-calculation method ( _ ; Table 5), the best estimation of in the AASW layer is a range of 0.35 ± 0.14 to 0.85 ± 0.22 μmol kg -1 yr -1 (Table 4) DIC in the AASWupw layer increased at a rate of 0.61 ± 0.10 μmol kg -1 yr -1 , and the pHT decreased -0.0015 ± 0.0004 yr -1 (Table   4). We were not able to detect a statistically significant trend in (i.e., solubility) or CANT from the estimates of the back-15 calculation method ( _ Table 5). However, we found an increase of of 0.42 ± 0.28 μmol kg -1 yr -1 ( Table 5) that is due to an increase in AOU. Considering the different drivers of the AOU increase (biology/circulation), the optimal estimation of for this layer is a value between 0 and 0.42 ± 0.28 μmol kg -1 yr -1 (Table 4).

Mode waters and intermediate layers (SAMW and AAIW).
The increase in DIC in the SAMW layer (1.10 ± 0.14 μmol kg -1 yr -1 ) for the period 1995-2011 is higher than that of upper 20 ocean layers and pHT decreases over the same period at -0.0031 ± 0.0003 yr -1 (Table 4). The DIC increase is explained almost entirely by _ of 0.92 ± 0.09 μmol kg -1 yr -1 . No significant trend was found in or (i.e, = _ ).
In the AAIW layer the DIC trend of 0.40 ± 0.15 μmol kg -1 yr -1 results in a pHT decrease of -0.0017 ± 0.0002 yr -1 and is also explained by the increase of CANT (0.42 ± 0.06 μmol kg -1 yr -1 , = _ , Tables 4 and 5). As with SAMW, no changes in solubility ( ) or biology/circulation processes ( ) were detected in the AAIW layer. The values found in the 25 SAMW and AAIW layers are very similar to the mean decadal changes found by Murata et al. (2007) between the 1990s and the 2000s in the subtropical Pacific Ocean (~1 μmol kg -1 yr -1 for the SAMW layer and 0.4 μmol kg -1 yr -1 for the AAIW). Waters et al., (2011) used data from the P18 line along ~110°W and estimated an increase in CANT of 0.89 ± 0.4 μmol kg -1 yr -

Deep-bottom layers (UCDW, LCDW and AABW).
The UCDW layer shows an increase of DIC of 0.29 ± 0.02 μmol kg -1 yr -1 between 1995 and 2011 and a change in pHT of -0.0013± 0.0001 yr -1 and are similar to the change in DIC (0.20 ± 0.02 μmol kg -1 yr -1 ) and pHT (-0.0012± 0.0002 yr -1 ) for LCDW. No statistically significant changes in time were detected of CANT_BC or in the and terms for any of these two layers. 5 The AABW layer also shows an increase of DIC (0.24 ± 0.02 μmol kg -1 yr -1 ) during the period 1995-2011 with an associated decrease of pHT of -0.0013 ± 0.0002 yr -1 (Table 4). The increase in CANT ( _ ) of 0.07 ± 0.01 μmol kg -1 yr -1 is low and this trend indicates an increase in CANT of ~ 1 μmol kg -1 , which is less than the accuracy of the back-calculation method (± 6 μmol kg -1 ).

Changes in the aragonite saturation ( ) and CANT storage. 10
There are statistically significant decreases of ΩAr in the STCW, AASW and SAMW layers (~ -0.010 ± 0.001 yr -1 , Table 4) similar to the trends observed at open-ocean time series sites in recent decades (Bates et al., 2014). The decrease of ΩAr found for the AASW layer (-0.61 ± 0.19 % yr -1 , Table 4) is also similar to the values obtained by Williams et al. (2015) for the Pacific sector of the Southern Ocean (-0.47 ± 0.10 % yr -1 for the period 1992 -2011 and -0.50 ± 0.20 % yr -1 for the period 2005-2011).
Accompanying the decrease of ΩAr with time along SR03, is the shoaling of the aragonite saturation depth (ASD, ΩAr = 1, Fig.  15 3) at a mean rate of -13 ± 3 m yr -1 . The shoaling of the ASD is not uniform over the section. North of the PF, the ASD shoals at a rate of -6 ± 4 m yr -1 while the rate is 3.5 times greater south of the PF (-21 ± 4 m yr -1 ). North of the PF the shoaling mostly affects the AAIW layer (Fig. 3a). South of the PF from ~62°S, the movement of the ASD follows the upwelling path of the UCDW layer (Fig. 3) with a shoaling of ~340 m over the 16-year period.
The storage rate of CANT (Table 6) for the surface and intermediate water mass layers is obtained from (Table 4) with 20 the most storage in SAMW and AAIW due to both their greater thickness and values. The rate of increase of the CANT storage in the whole longitude band of the SR03 section is 0.30 ± 0.24 mol m -2 yr -1 , calculated by computing the mean of the storage rates of the layers weighted by the mean volume occupied by each of the layers for the period 1995-2011 (Table 6) 5 Discussion.
Our results are indicative of a scenario of increased transport of deep waters into the section and enhanced upwelling at high 25 latitudes for the period between 1995 and 2011 linked to strong westerly winds. Several studies have reported a trend in the Southern Annular Mode (SAM) toward its positive phase from the 1960s until the 2000s (Thompson and Solomon, 2002;Marshall, 2002Marshall, , 2003Lenton and Matear, 2007;Sallée et al., 2008). According to these studies, the positive phase of the SAM is correlated with an intensification and southward movement of the subpolar westerly winds that ultimately lead to the enhancement of northward Ekman transport, meridional overturning and upwelling south of the ACC. Also, surface warming and more intense and frequent pulses in the extension of the EAC at long-time scales have been related to a poleward movement of the westerly winds (Rintoul and Sokolov, 2001;Ridgway, 2007;Hill et al., 2011). From the 2000s on, the SAM index no longer presents a positive trend but, although exhibiting considerable interannual variability (Fig. A2 in the supplementary material), the SAM index remains in its positive phase, favouring strong winds over the region. 5 In the northern part of the SR03 section, the area occupied by the STCW has high variability due to the encounter between the EAC and the ZC in the North of the section (Ridgway et al., 2007;Herraiz-Borreguero and Rintoul, 2011;Sloyan et al., 2016).
The warming of the STCW layer found in this study (0.0335 ± 0.0130 °C yr -1 ) could be linked to variability in the extension of subtropical waters but it could also be related to atmospheric warming. Aoki et al. (2015) related the 30-year warming found north of the SAF in the South Pacific and Indian oceans to the intensification of the subtropical gyres, which promote the 10 arrival of warmer waters. In the AASW layer that extends approximately between the SAF and the PF we found a similar warming (0.0369 ± 0.0109 °C yr -1 ) to that of the STCW. This could indicate that the increase of temperature found in the upper layers of the section could be most likely due to ocean heat uptake and atmosphere warming.
Due to the surface warming, the increase of DIC found in the STCW layer (Table 4) is lower than expected from the increase in atmospheric CO2 (~ 1 ± 0.12 μmol kg -1 yr -1 ). Nevertheless, at least 83% of the increase of DIC in the STCW layer is 15 explained by the increase in CANT (Table 4). As for the AASW layer, our results indicate that temperature does affect the estimate of , but the effect of the increase in (due to an increase in AOU) overweigh that of solubility ( Table 5).
The seasonal to interannual variability of the AASW layer is also influenced by the variability of the positions of the SAF and PF (Fig. 1, Table 3), that is highly conditioned by the flow of the ACC over the South-East Indian Ridge (Fig. 1). A close relationship between phytoplankton blooms and regions where the ACC fronts interact with large topographic features has 20 been noticed (Moore et al., 1999;Moore and Abbott, 2000). A variability in the remineralization rates due to phytoplankton blooms variability could explain the changes in observed in the AASW layer. Nevertheless, no changes in nutrients (nitrates or phosphates) are measurable in this layer that could indicate intense biological activity.
Furthermore, the AASW layer is also affected by the upwelling of deep waters south of the PF, and an intensification of the upwelling could increase the content of low-O2 DIC-rich waters in the AASW layer leading to an increase in AOU. The 25 increase in found in the AASWupw layer (Table 4), south of the PF, is similar to the increase obtained for the AASW layer, which indicates the likelihood that the upwelling of deep waters results in the increase in AOU. The increase in in the AASWupw layer coincides with an increase of salinity of 0.0029 ± 0.0001 yr -1 (not shown), that is consistent with increased transport of saltier waters from the deep ocean to subsurface layers. Besides, we also found an increase in dissolved silicate of 0.36 ± 0.06 μmol kg -1 yr -1 ( 4 Table 4) that could be related to the upwelling enhancement as well (Tréguer, 2014). 30 The influence of the upwelling on the DIC budgets (as non-anthropogenic DIC) is clearer in the AASWupw layer than in the AASW layer. For the AASW layer, the lower limit of (i.e., the change in is assumed to be due to biological processes, Table 4) indicates that at least 41% of the increase of DIC in the layer is explained by the increase of CANT while the lower limit of is zero for AASWupw (Table 4), meaning that the effect of the upwelling over AASW is lower than over AASWupw. Matear and Lenton (2008) using carbon models, concluded that the uptake of CO2 by the waters north of the PF is more influenced by the wind variability than by other processes such as the upwelling. An intensification of the winds (due to a positive phase in SAM) could contribute to the increase in CANT found in the AASW layer. Considering the upper limits of in both layers (i.e., the change in is assumed to be due to circulation processes), the increase of CANT 5 in the AASWupw layer represents no more than 69% of the increase in DIC (upper limit of , Table 4) while the upper limit of for the AASW equals the increase of DIC. Thus, AASWupw layer, at least ~30% of the increase in DIC (~ 0.18 μmol kg -1 yr -1 ) is still not explained and is most probably related to the upwelling of DIC-rich waters. The increase in nonanthropogenic DIC could be even higher, since we assume that the change in due to circulation does not affect DIC (see section 3.3). 10 In terms of the change in oxygen, Helm et al. (2011) found an average decrease in the concentration of O2 between 100 and1000 m from 1970 to 1992 of ~ -0.23 μmol l -1 for the Southern Ocean (27% of the estimated global average change, -0.93 ± 0.23 μmol l -1 ) . Considering the volume of the first 1000 m of the water column of the Southern Ocean to be 19400 · 10 -9 l (obtained using ETOPO1 doi:10.7289/V5C8276M) and the volume of the first 1000 m of the SR03 section to be 2700 · 10 -9 l, the decrease of O2 found by Helm et al. (2011), if constant in time, would correspond to a decrease of ~ -1.7 μmol l -1 yr -1 . We only 15 found changes in oxygen within the surface water mass layers (STCW, AASW and AASWupw) that approximately fill the first 300 m of the water column of the SR03. Then, the decrease of ~ -1.7 μmol l -1 would correspond to an average change of O2 of ~ -0.32 μmol kg -1 yr -1 for surface waters of the SR03. This means that values of ~ 0.20 μmol kg -1 yr -1 due to circulation processes can be expected in for surface waters, which is comparable to the average of our findings (Table 5), 0.32 ± 0.24 μmol kg -1 yr -1 and could indicate that the change in O2 is related to circulation processes. 20 The variability of the SAMW and AAIW layers south of Tasmania has been related to variability in the northward Ekman transport that drives the northward movement of AASW (Rintoul and England, 2002;Sallée et al., 2006Sallée et al., , 2012. A scenario of intensification of the upwelling near the Antarctic Divergence would lead to an increase in the northward Ekman transport, conditioning the properties of these water mass layers and particularly for SAMW, which is mostly form north of the SAF. There is a significant freshening of the SAMW layer (-0.0026 ± 0.0001 psu yr -1 , not shown) between 1995 and 2011 that could 25 be related to higher inputs of AASW into the SAMW layer and consistent with the increase in Ekman transport. Besides, an intensification of the winds due to the positive trend of the SAM favours the ventilation and thus the increase in CANT uptake by both water mass layers (Matear and Lenton, 2008). Our results indicate that the change of DIC in the SAMW and AAIW layers is driven mostly by the uptake of atmospheric CO2 ( ≈ , Table 4). The increase of CANT in the SAMW layer is higher than that found for the upper ocean layers and closer to the expected from the increase of atmospheric CO2 (~1 μmol 30 kg -1 yr -1 ). The smaller increase of CANT in the AAIW layer compared to the SAMW layer (Table 4) agrees with lower ventilation of the AAIW layer south of Tasmania (see section 2) due to the fact that this layer carries recently ventilated waters mixed with older waters ventilated far out the SR03 section. The lack of measurable long-term changes in and in both AAIW and SAMW layers indicate that circulation and biological processes do not have a large effect on .
Deep to bottom layers of the section show significant trends for DIC that are not explained by the increase of CANT. These trends are most likely due to the advection of old and DIC-rich waters. Concretely for deep waters (UCDW and LCDW), the trends could result from an intensification of upwelling at high latitudes being offset by enhanced transport of old and CO2-5 rich waters to replace the upwelled waters, since the increase of DIC follows the upwelling path of the UCDW and LCDW layers (Fig. 4). We separated the UCDW layer into two latitudinal sectors: north and south of the SAF (Fig. 4). The increase of DIC in the UCDW layer north of the SAF is 0.44 ± 0.04 μmol kg -1 yr -1 while south of the SAF is smaller at 0.26 ± 0.04 μmol kg -1 yr -1 (not shown), consistent with a greater supply of waters from the north at depth. A decrease of O2 in the UCDW to the north of the SAF occurs mostly in the upper to middle parts of the UCDW layer (Fig. 4), and this is not observed south 10 of the SAF. The decrease of O2 is also in agreement with the arrival of waters from Indian-Pacific origin since these waters provide the characteristic oxygen minimum zone that defines UCDW (Callahan, 1972;Talley 2013). Another feature that agrees with the hypothesis of upwelling intensification is the shoaling of the ASD following the path of upwelling of the UCDW layer. This feature was also described by Bostock et al. (2013) in an oceanic climatology of ΩAr and could be due to the naturally lower buffer capacity of the UCDW layer (low value of TA/DIC ≈ 1.043) with respect to upper layers (TA/DIC 15 ≈ 1.06 in the AASW layer). However, the greatest shoaling of the ASD in the UCDW layers compared to the AAIW layer (Table 6) is consistent with the upwelling of UCDW, as both water masses have similar TA/DIC ratios (TA/DIC ≈ 1.043 for the UCDW layer and TA/DIC ≈ 1.042 for the AAIW layer). Furthermore, the increase of SiO4 found in deep-bottom layers (Table 4) could also indicate the arrival of old waters to the section that are progressively enriched in SiO4 (e.g., Callahan, 1972). 20 Statistically significant decreases in pHT with time were observed in all water mass layers (Table 4), with the greatest change in surface water masses, coinciding with the greatest DIC changes. The decrease of pHT in the STCW and SAMW layers is related to the increase in the uptake of CANT, while for the AASW and AASWupw layers the pHT change appears to be linked to the upwelling of DIC-rich waters at high latitudes. At deep layers, the tongue of water of pHT =7.9 off the shelf is reduced in 2011 compared to 1995 (Fig. 2e,f), which is consistent with the advection of DIC-rich waters in the section due to the 25 enhanced upwelling. The different rates of pHT change in the water masses is in part related to the buffering capacity of the waters. AASW layer has lower temperature than the STCW layer (~2.3 °C for the AASW layer compared to ~11.0 °C for the STCW layer, mean values for the period 1995-2011) and lower buffer capacity than the STCW (TA/DIC ≈ 1.058 for the AASW layer versus TA/DIC ≈ 1.095 for STCW). For similar increases in DIC ( Table 4) the decrease of pHT in the AASW is expected to be higher than in the STCW layer. 30 In terms of carbon, previous studies concluded that the intensification of the upwelling (as a consequence of the SAM variability) caused a reduction in the uptake of CO2 by the Southern Ocean between 1980s and 2000s due to the outgassing of CO2 near the Antarctic Divergence (Le Quére et al, 2007;Lovenduski et al., 2008). Landschützer et al. (2015) showed that the efficiency of the Southern Ocean CO2 sink declined through the 1990's, and the trend reversed from about 2002, although the reversal in the sink efficiency was not zonally uniform. The results from Landschützer et al. (2015) are consistent with a carbon sink influenced by the upwelling of DIC-rich waters at high latitudes and superimposed on this is the near surface response to atmospheric forcing that modifies the sink efficiency and could mask longer term trends in the upwelling of DIC-rich waters at high latitudes. A comparison of our results with those of Landschützer et al. (2015) is problematic as their data is restricted 5 to surface waters and our analysis is on long-term trends in water mass properties below 50m depth. Both data sets do show continued uptake of CO2 throughout the period of study and indicate the importance of the circulation in influencing the regional carbon sink, which has also been established by recent model results (DeVries et al., 2007).
Our results also agree with the conclusions from different model simulations done by Matear and Lenton (2008), who established that intense wind regimes (associated to a positive phase in the SAM) favour the uptake of CANT and ventilation of 10 the SAMW and AAIW layers. These authors highlighted the complex response of the uptake of CO2 by the Southern Ocean due to the diverse forcing acting on upper layers, which can be also seen in our results (e.g., differences in the biogeochemical changes in the AASW and AASWupw layers). Matear and Lenton (2008) also noticed the complex relationship between the upwelling and subduction areas of the Southern Ocean, with the same drivers acting in opposite direction for the changes in non-anthropogenic DIC with respect to the changes in CANT uptake. 15

Sensitivity of the results to underlying assumptions.
This section considers the sensitivity of assumptions used to calculate temporal changes in CANT, including errors associated with the assumption of steady state in the oceanic circulation and remineralization processes, and the sensitivity to stoichiometric ratios for the biological processes.

Comparison of CANT changes using other methods. 20
We compared the changes of CANT obtained in our study with the results from two regression-based methodologies (Table 7); the extended multiple linear regression (eMLR) method (Friis et al., 2005) and the two-regression method (Thacker, 2012).
These methods use repeat hydrodynamic sections to quantify the temporal change in CANT.
The eMLR method (Friis et al., 2005) estimates the change in CANT between two repeats of a hydrodynamic section by establishing MLRs for each section and relating the observed DIC for each observation to a set of other measured oceanic 25 variables: where ( ) are the coefficients of the fit between DIC and the n observed variables (P1,….Pn) chosen for the fit, all measured at the time (t) of the survey.
Taking the difference between DIC at two times, t1 and t2, gives an equation for the change in CANT over the time period 30 between the two hydrographic surveys (ΔCANT): ∆ = 0( 2) − 0( 1) + ( 1( 2) − 0( 1) ) 1( 2) + ⋯ + ( ( 2) − 0( 1) ) ( 2) The two-regression method was introduced by Thacker (2012) as an improvement in regression-based methods. The region of study is first divided into sub-regions since the empirical relationships between DIC and other environmental variables vary spatially (Thacker, 2012). MLRs are investigated between DIC and other measured variables (predictors) using a stepwise technique. The procedure is applied for each sub-region using all data from the repeat surveys within the period to be 5 investigated, resulting in an optimal MLR for each sub-region (similar to Eq. 6). A linear regression with time is established for the residuals (observed DIC -predicted DIC) of the regional fits, which directly gives ΔCANT averaged over the space-time in each sub-region. The purpose of the first MLR is to remove the natural variability of DIC, leaving the anthropogenic signal and noise (random variability) in the residuals and the second MLR is used to separate the anthropogenic signal from the noise.
We applied both methodologies within the different water mass layers separated by γ n used as sub-regions. The predictor 10 variables of θ, S, σ0, nitrate (NO3), SiO4 and AOU were used for the MLR procedures. The three methodologies estimate similar rates of increase in CANT for most water mass layers (Table 7). In the STCW layer, the value of ΔCANT (eMLR) is higher than our maximum estimate of and the value obtained from the two-regression method. The eMLR method is less suitable for the upper layers of the ocean subject to high seasonal to interannual variability, such as the STCW layer, resulting in large residuals that bias the regression (Friis et al., 2005). For the AAIW layer, the increase of CANT estimated by the two-15 regression method is half the increase that is established by our method and the eMLR method. The lower value of the trend estimated by the two-regression method is due to the fact that the two-regression method uses stepwise MLR. This means that the two-regression method only considers those predictors that give the best fit while the eMLR method is forced to consider all the predictors in the fit. This is also the cause of the low RMSE obtained with the two-regression method compared to both our trends and those obtained by the eMLR method. For deep to bottom layers, the two-regression method estimates a small 20 increase of CANT similar to the one found in this study for the AABW layer ( Table 7) that can be considered negligible given the resolution of the back-calculation method (± 6 µmol kg -1 ). The eMLR method finds increases of CANT (with relatively high uncertainties) higher than the two-regression method that over the 16-year period also give values of CANT change lower than the resolution of our back-calculation method (although close to it for the LCDW layer, ~ 5 µmol kg -1 for the 16-year period).

Circulation and biological processes at steady state. 25
The back-calculation method assumes the circulation of the ocean and the biological processes are in steady state. The contribution of non-linear mixing is unknown and some of the changes in DIC in the water mass layers could be erroneously included in the estimates of CANT rather than as a non-anthropogenic change in DIC. The non-steady state of the circulation in our analysis is included to some extent through the changes in (Eqs. (3,4), which is solved by the OMP analysis, which is subjected to the limitations of quantifying the mixing mostly through thermohaline changes in the water masses (section A.2 30 of the Appendix).
For biological processes, remineralization rates are usually considered to be in steady state (Sarmiento et al., 1992). Climate change has been suggested as potentially driving changes in carbon fixation and export that can influence the uptake of CO2 by the oceans (Falkowski et al., 1998). Pahlow and Riebesell, (2000) first suggested that decadal changes in remineralization rates occurred in the deep waters of the Northern Hemisphere, although this is still a matter of debate (e.g., Li and Peng, 2002;Najjar, 2009). 5 Metzl et al. (1999) and Shadwick et al. (2015) observed that the uptake of CO2 over the sub-Antarctic zone (SAZ, between the SAF and the STF) in summer is mostly controlled by biological processes. If a change in remineralization rates has occurred, i.e. the changes in are due to biological effects, a change in nutrient concentrations of the water masses would be expected. The detection of long-term trends of nutrients at upper layers of the ocean can be masked by short time scale physical processes such as changes in the mixed layer depth, mesoscale activity and advection (Sambrotto and Mace, 2000;Rintoul 10 and Trull, 2001;Sallée et al., 2010). We did not find a measurable trend in nutrient concentrations for any of the layers in the period 1995-2011, except for an increase of SiO4 over the Antarctic Divergence ( Table 4) that is most likely due to the upwelling of SiO4-rich deep waters. We cannot confidently assign the changes of and its impact on in upper layers ( Table 5) to any particular process and, instead we provide a range of values for (Table 4). For the scenario of intensification of the Antarctic upwelling, the increase in low-O2 and DIC-rich waters would increase the content of DIC in 15 subsurface waters, leading (at least for the AASW and AASWupw layers) to values of closer to the lower limit of the range (i.e., >> , Table 4).
In deep layers of the section, the increase of DIC is not explained by the long-term change of any of the terms in Eq. (1), which is other implication of considering the circulation in steady state. The differences found in the increase of DIC in the UCDW layer north and south of the SAF (section 4.3, Fig. 4) add consistency to the idea of the advection of older waters to the section. 20 Considering these differences we can assign a change in DIC of at least ~0.20 ± 0.02 μmol kg -1 yr -1 (lower rate of increase of DIC found in deep-bottom layers, Table 4) due to the upwelling intensification. Since the AASW and AASWupw layers are the most affected by the upwelling we can correct the values of in these layers for this effect (trends with ** in Table 4).

Changes in the rates of export of particulate organic carbon and silicate from surface layers.
We assume that the export of particulate organic carbon (POC) from upper ocean layers (STCW, AASW and AASWupw) and 25 remineralization in the water column was constant between 1995 and 2011. The high-latitudes are considered important in terms of POC export, mostly because these areas are dominated by large phytoplankton, in particular diatoms (Buesseler, 1998;Sambrotto and Mace, 2000), and rapid carbon export to deep waters from phytoplankton blooms is possible (DiTullio et al., 2000;Lourey and Trull, 2001). The POC exported is remineralized to DIC below the mixed layer (Wassman et al., 1990;Asper and Smith, 1999;Trull et al, 2001a;Fripiat et al., 2015). 30 Our results show that the increase of DIC in mode and intermediate waters is fully explained by the uptake of atmospheric CO2, which could indicate that there was no detectable change in the rate of export of POC over the 1995-2011 period.
Estimates of POC export in the SAZ (~3800 m), the SAF (~3100 m) and the PFZ (~1500 m) using moored sediment traps near the section (Trull et al., 2001b) were 0.5, 0.8 and 1.0 g C m -2 yr -1 , respectively. If all this POC is fully remineralized in the UCDW layer (with a mean thickness of 2000m), we obtain a range of 0.02 -0.04 μmol kg -1 yr -1 for the maximum increase of 5 DIC due to the export of POC. This increase is close to the uncertainty of the total DIC increase estimated for the UCDW layer, which means that in order to generate an increase in DIC similar to that found in the UCDW layer, the rate of POC export should be ~10 times higher than the observed rates. This change should be certainly noticeable in in surface waters but most probably in deep waters as well, which we do not see.
The observed increase of SiO4 in deep and bottom layers of the ocean is consistent with the transport of SiO4-rich older waters 10 to the section. For UCDW, the increase of SiO4 north of the SAF is higher than at southern latitudes (0.31 ± 0.08 μmol kg -1 yr -1 respect to 0.19 ± 0.04 μmol kg -1 yr -1 , not shown). Nelson et al., (1995) observed a lower dissolution rate of diatom dominated SiO4 exported in high-latitudes regions compared to lower latitudes. Nelson et al. (1995) estimated a mean silica production rate of 0.7 -1.2 mol Si m -2 yr -1 for regions over diatomaceous sediments and concluded that 15 -25 % of the silica produced in the upper ocean accumulates in the seabed. Of the silica produced in the mixed layer of the upper ocean layers at least 50% 15 is believed to dissolve in the upper 100 m of the water column (e.g. Nelson et al., 1991;DeMaster et al., 1992). For deep waters, the production rates of Nelson et al. (1995) could result in a mean increase of SiO4 of 0.08 -0.14 μmol kg -1 yr -1 in the water column (~3400 m, mean depth). The maximum value of this increase could explain the trends of SiO4 found for the AABW layer (Table 4), but 32-48 % of the increase of SiO4 in deep bottom layers is not explained by the remineralisation of exported silica and is most likely the result of the advection of older SiO4-rich waters to the section. 20

Stoichiometric ratios for biological processes.
The back-calculation method and the OMP analysis assume constant stoichiometric ratios for remineralization. The theoretical Redfield ratios (Redfield, 1934;1958) are usually considered as a mean for the whole ocean, although they can vary from the theoretical value due to changes in phytoplankton species composition, the food-web structure and nutrient availability (Martiny et al., 2013). 25 We carried out a sensitivity analysis on the Redfield ratios following Álvarez et al., (2014), to obtain values of stoichiometric ratios for the section. A battery of OMP analyses were done with varying values of between 9 and 10 in increments of 0.2, between 120 and 145 in increments of 5, and between 0 and 8 in increments of 2. For each variation in the stoichiometric ratios, an OMP analysis was made for each section in order to determine best-fit R values and if there were differences in time for the stoichiometric ratios along the sections. The smallest residuals (differences between the nutrients measured and the 30 estimated by the OMP analysis) were obtained for = 9 and = 125. The residuals of SiO4 did not change significantly for any value of and we consider SiO4 as a conservative variable. These results indicate = 13.8, in agreement with values obtained for the region ( ϵ [8][9][10][11][12][13][14][15], Lourey and Trull, 2001).

Conclusions.
The results of our analysis south of Tasmania over the 1995-2011 period support a scenario of intensification of upwelling in the vicinity of the Antarctic Divergence due to an increase in the westerly winds at high latitudes most probably linked to the 5 variability of the SAM. The intensification of the upwelling favours the advection of older waters to deep-bottom layers of the section where we found net increase in DIC over the 16-year period. The enhanced upwelling causes the eventual entrainment of low-O2 and DIC-rich waters into upper layers, explaining the trends of decreasing O2 and increasing non-anthropogenic DIC found in surface waters close to the Antarctic Divergence. This scenario also implies the intensification of the convergence north of the SAF, implying a more efficient ventilation of the SAMW and AAIW layers and thus an efficient uptake of 10 atmospheric CO2 by these layers. The enhanced upwelling lowers the uptake of CANT in the AASW layer but the effect of ventilation more than compensates that of the upwelling allowing the increase in CANT in this layer. The atmospheric warming reduces the dissolution of CO2 in upper layers north of the PF, presenting increases in DIC lower than expected from the atmospheric CO2 increase.
Our results rely on a limited number of sections spread every 3-7 years and can only provide a long term (decadal) average 15 view of changes in water masses. More surface observations and repeat deep ocean sections are needed to help resolve interannual changes in the Southern Ocean carbon sink and to determine the main drivers and feedback to the carbon-climate system. The effort to maintain hydrographic sections with CO2 system measurements would also benefit from additional direct measurements of more variables of the carbon system, e.g. pH, which has not been measured on the SR03 section.

Data availability 20
The section data are available through the Global Ocean Data Analysis Project (http://cdiac.ornl.gov/oceans/GLODAPv2; Key et al., 2015;Olsen et al., 2016). The original data for the different cruises were corrected following the QC recommendations in GLODAPv2.   Table 2. Definition of the water mass layers between neutral surfaces (γ n ) and references used to accordingly decide the limits of the layers. We also consider limits in salinity, depth and position of the ACC fronts (see Table 3) to differentiate better some of the layers. Table 3. Location of the ACC fronts south of Tasmania for each of the cruises following the definitions of Sokolov and Rintoul (2002) for hydrographic data. * The range in the location of the SAF for the 2011-cruise could be related to a diversification of the PF or SAF into different jets (Sokolov and Rintoul, 2009) but also to crossing the meander of the front twice. 25 Table 6. Approximated rates of CANT storage based on the trends from Table 4. Table 7. Comparison between from the present study and the values of ΔCANT from the two-regression and eMLR methods.

B Figures
RMSE= root mean square error. ** The value of in the AABW layer is considered negligible because it falls below the accuracy 5 of the back-calculation method.