Estimating the storage of anthropogenic carbon in the subtropical Indian Ocean: a comparison of five different approaches

. The subtropical Indian Ocean along 32 ◦ S was for the ﬁrst time simultaneously sampled in 2002 for inorganic carbon and transient tracers. The vertical distribution and inventory of anthropogenic carbon (C ANT ) from ﬁve different methods: four data-base methods ( 1 C*, TrOCA, TTD and IPSL) and a simulation from the OCCAM model are compared and discussed along with the observed CFC-12 and CCl 4 distributions. In the surface layer, where carbon-based methods are uncertain, TTD and OCCAM yield the same result (7 ± 0.2 molC m − 2 ), helping to specify the surface C


Introduction
The fate of the anthropogenic CO 2 (C ANT ) emissions to the atmosphere is one of the critical concerns in our attempts to better understand and possibly predict global change and its impact on society (IPCC, 2007).Quantifying the global carbon cycle is still the subject of much scientific effort, especially where new processes and carbon pathways have to be considered (Cole et al., 2007;Duarte et al., 2005;Prairie and Duarte, 2007).While being a well-known and significant process, the uptake of C ANT by the world ocean is not well-constrained at present and its magnitude and variability may change in the future.Different approaches to estimate the global oceanic C ANT uptake have arrived at essentially the same number, about 2 Pg C yr −1 (Wetzel et al., 2005).However, despite this general agreement, questions about the reliability of these estimates remain.
The Joint SOLAS-IMBER implementation plan has identified the research priorities for ocean carbon research, among them the separation of natural from anthropogenic carbon, the oceanic storage and transport of C ANT and the effect of decreasing pH, ocean acidification, on the marine biogeochemical cycles, ecosystems and their interactions (www.imber.info/products/CarbonPlan final.pdf).Large impacts of ocean acidification are expected to occur in high latitudes, i.e. the Southern (Bopp et al., 2001;Orr et al., 2005) and the Published by Copernicus Publications on behalf of the European Geosciences Union.
The first attempts to estimate C ANT from oceanic measurements were based on the back-calculation method proposed independently by Brewer (1978) and Chen and Millero (1979).This method was reformulated by Gruber et al. (1996) and specific improvements were proposed for the Atlantic Ocean (Pérez et al., 2002) and Southern Ocean (Lo Monaco et al., 2005a).Several other methods based upon completely different concepts also arose; such as one based on water mass mixing (the MIX method; Goyet et al., 1999), another based on estimating Transit Time Distribution (TTD) or ages from transient tracers as CFCs, SF 6 or CCl 4 (TTD method; Hall et al., 2002), and one based on a composite tracer (TrOCA method; Touratier et al. 2004a, b;Touratier et al., 2007).In addition there are simulations from threedimensional Ocean General Circulation Models (OGCM) (Orr et al., 2001).Despite these efforts, no clear conclusion has been achieved about the best method after several comparison exercises (Coatanoan et al., 2001;Feely, 2001;Hall et al., 2004;LoMonaco et al., 2005b;Sabine and Feely, 2001;Wanninkhof et al., 1999;Waugh et al, 2006;Touratier et al., 2007), which are still necessary and on-going, for example, within the Integrated Project CARBOOCEAN for the Atlantic basin (Vázquez-Rodríguez et al., 2009).
The contribution of the Indian Ocean to the global C ANT storage was initially discussed in Chen (1993) and Sabine et al. (1999) and later in Sabine et al. (2004) from a global perspective.While the ocean volume of the Indian Ocean is 20% less than that of the Atlantic, the total C ANT inventory of the Indian Ocean is only half that of the Atlantic, and it contributes ∼21% to the global ocean C ANT inventory (Sabine et al., 2004).The relevant areas or processes introducing C ANT into the Indian Ocean are: 1. full equilibration of the upper mixed layer; 2. the formation of Red Sea -Persian Gulf Intermediate water (Papaud and Poisson, 1986;Mecking and Warner, 1999) in the northwestern Indian Ocean spreading equatorward; 3. the formation of Subantarctic Mode Water (SAMW) north of the Subantarctic Front (McCartney, 1977), including the large volume of SAMW being formed in the southeast Indian Ocean (e.g., Sloyan and Rintoul, 2001;Sallée et al., 2006) and transported equatorwards; and, 4. the formation of Antarctic Intermediate Water (AAIW), usually delineating the lower limit of C ANT penetration (Sabine et al., 2004) in the Indian Ocean.
The thermostad associated with SAMW is formed by deep mixing in winter on the equatorward side of the Subantarctic Front (McCartney, 1977), and is found at about 400-600 m.
SAMW is linked to AAIW in the Southeast Pacific (Mc-Cartney, 1977) circulating in this basin through subduction.AAIW of the South Atlantic and Indian oceans is produced in the confluence of the Malvinas and Brazil currents by injection of surface water into the subtropical gyre, which then circulates eastwards and northwards in the South Atlantic and Indian Oceans, where no other AAIW sources are found (Talley, 1996;Hanawa and Talley, 2001).
Deep water formation is another important mechanism sequestering C ANT into the ocean.The formation and transport of North Atlantic Deep Water (NADW) is associated with high C ANT inventories present in the North Atlantic (e.g., Álvarez et al., 2003;Sabine et al., 2004;Touratier and Goyet 2004b).Another major pathway theoretically introducing C ANT into the deep ocean could be the production of Antarctic Bottom Water (AABW).However; large discrepancies exist in estimates of the role of the Southern Ocean in the C ANT uptake and storage.Both OGCMs (Caldeira and Duffy, 2000;Orr et al., 2001) and inversion estimates based on OGCMs but constrained with data (Mikaloff Fletcher et al., 2006) find a high C ANT uptake but low C ANT storage in the Southern Ocean, with high uncertainties.These studies also find high C ANT transport northwards toward the Antarctic convergence zone.The low storage is supported by databased estimates (e.g., Poisson and Chen, 1987;Gruber, 1998;Hoppema et al., 2001) that rely on factors such as the very high Revelle factor of these waters, the relatively short contact time with the surface between upwelling and subduction and decreased CO 2 uptake due to the presence of seaice.However, these findings are contradicted by the detection and accumulation of CFCs in Antarctic deep and bottom waters (e.g., Meredith et al., 2001;Orsi et al., 2002) and the C ANT accumulation detected south of Australia (McNeil et al., 2001) and in the South Atlantic Ocean (Murata et al., 2008), or using the TTD technique for the whole Southern Ocean (Waugh et al., 2006).Recent carbon-based studies also detect significant C ANT accumulation in deep and bottom waters of the Southern Ocean (Lo Monaco et al., 2005a, b;Sandrini et al., 2007).
This study is a comparison exercise between different data-based (carbon-based and TTD) techniques for estimating C ANT applied along a transoceanic section along 32 • S in the Indian Ocean.Differences among the C ANT distributions and inventories are presented, and the strengths and weaknesses of the individual methods are discussed.To provide an independent and unrelated comparison, results from an OGCM are presented alongside the data-based techniques.The final aim is to obtain the best C ANT inventory in the subtropical Indian Ocean with new data and based on different approaches.

Data set
During March-April 2002, cruise 139 of RRS Charles Darwin (CD139), a trans-Indian hydrosection was made nominally along 32 • S (Bryden et al., 2003).The section used here consisted of 133 full-depth stations (Fig. 1) with a typical spacing over the deep basins of 90 km and a maximum of 120 km.Over the shelf and other topography the station spacing was decreased.
CTD data were taken with a SeaBird 9/11 plus system.Discrete samples for dissolved oxygen (O 2 ) were analysed by a semi-automated whole-bottle Winkler titration unit with spectrophotometric end-point detection.Inorganic nutrient concentrations were measured using a Skalar San Plus autoanalyser, configured according to the manufacturer's specifications (Kirkwood, 1995).The overall accuracy for O 2 , nitrate, phosphate and silicate is 1, 0.1, 0.01 and 0.6 µmol kg −1 , respectively.
Chlorofluorocarbon (CFC) samples were collected from the same Niskin bottles sampled for Total Alkalinity and pH.Concentrations of CFC-11 and CFC-12 in seawater were measured in about 2100 water samples using shipboard electron capture gas chromatography (EC-GC) techniques similar to those described by Bullister and Weiss (1988).A subset (∼540) of the water bottles sampled for CFCs were also sampled and analyzed for dissolved carbon tetrachloride (CCl 4 ) on a separate analytical system using similar techniques.CFC and CCl 4 concentrations are reported in picomoles per kilogram seawater (pmol kg −1 ).The overall accuracy for dissolved CFC-11 and CFC-12 measurements was estimated to be 2% or 0.010 pmol kg −1 (whichever is greater) and 3% or 0.012 pmol kg −1 for CCl 4 measurements.
The CFC-12 age (τ ) of any water sample has been calculated following Doney and Bullister (1992) assuming 100% initial saturation.Reconstructed CFC-12 annual mean dry air mole fractions in the Southern Hemisphere were taken after Walker et al. (2000), extended with yearly mean values from the AGAGE sampling network.Note that the TTD method uses a different approach to estimate ages from CFCs.
The CD139 trans-Indian section was completely analysed for pH; 69 stations were analysed for Total Alkalinity (TA), typically every other station.Since these two variables allow the carbonate chemistry system to be fully constrained, Total Inorganic Carbon (C T ) was not routinely measured.However, C T samples were collected at 4 stations and were analysed post-cruise on land as quality control.
pH was measured spectrophotometrically following Clayton and Byrne (1993) using a seawater m-cresol purple dye solution.Replicate analysis from deep Niskin bottles shows a reproducibility of ±0.0009.pH analysis on CRM samples were also performed.For more details about the pH analyses and quality control see Appendix A1.
TA was measured using a double end-point automatic potentiometric titration (Pérez and Fraga, 1987).Concentrations are given in µmol kg-sw −1 .Determinations of TA on CRM were made during the cruise to monitor the titrator performance.At a test station, the whole set of bottles were closed at the same depth.The resulting TA standard deviation of a total of 24 analyses over 12 bottle samples was 1.04 µmol kg −1 .For more details about the TA analyses and quality control see Appendix A2.
C T samples were collected in 500mL borosilicate bottles, immediately poisoned with HgCl 2 and stored in the dark until analyzed on shore.C T was measured in the lab using a coulometer with a SOMMA (Single Operator Multiparameter Metabolic Analyzer) inlet system (Johnson et al., 1993).CRM were used to correct any offset in the analysis.
The above results indicate the high quality and internal consistency of the CD139 CO 2 data base.Despite this, we have also performed a crossover analysis using The following sections provide a basic description of the different back-calculation methodologies used in this comparison exercise: 1) C* approach: developed by Gruber et al. (1996) and applied specifically to the Indian Ocean by Sabine et al. (1999, hereinafter SAB99); 2) IPSL approach: an improved version of the backcalculation technique proposed by Lo Monaco et al. (2005a, hereinafter LM05) for the Southern Ocean; 3) C* combined ( C*Comb): several improvements were suggested to the C* approach in the North Atlantic by Pérez et al. (2002), some of them can be also applied in the Indian Ocean.

C* (SAB99) approach
where C T represents C T measurements, C bio T reflects the change in C T due to biological activity, C 280 T denotes the C T in equilibrium with the pre-industrial atmosphere and C dis T reflects the air-sea CO 2 disequilibrium when water masses are formed.The first three terms make up the quasiconservative tracer C* ( C*=C T − C bio T −C 280 T ), which reflects both the anthropogenic signal and the air-sea CO 2 disequilibrium ( C*=C ANT + C dis T ) (Gruber et al., 1996).
−106/104 × N * where: AOU is the Apparent Oxygen Utilization (oxygen saturation (Benson and Krause, 1984) minus measured oxygen); TA 0 is preformed TA; R C and R N are stoichiometric ratios C/O 2 and N/O 2 according to Anderson and Sarmiento (1994); N* is a quasi-conservative tracer used to identify nitrogen excess or deficits relative to phosphorus.N* values are converted to carbon with a denitrification carbon to nitrogen ratio of 106:-104 (Gruber and Sarmiento, 1997).
where: S is salinity; PO a conservative tracer (PO=O 2 +170×P; Broecker, 1974); θ is potential temperature; P and N are the phosphate and nitrate concentrations, respectively.

C 280
T was obtained from thermodynamic equations of C T as a function of preformed alkalinity for a pre-industrial partial pressure of CO 2 (pCO 2 ) of 280 ppm (Gruber et al., 1996).In order to keep the definition of C* conservative, Gruber et al. (1996) linearized C 280 T about the mean values of temperature, salinity and alkalinity observed in surface waters of the Atlantic Ocean, which yield an uncertainty of 4 µmol kg −1 .SAB99 used the linearized formula obtained by Gruber et al. (1996) for the Atlantic Ocean given below: Finally, SAB99 obtained C dis T following the technique proposed by Gruber et al. (1996): i) for water masses younger than 40 years they used the C*τ method where C τ T is the C T in equilibrium with the atmospheric CO 2 at the time of water mass formation: t form =t obs − τ , the water mass age τ being calculated from CFC-12 ages; the atmospheric time history for CO 2 is taken from the South Pole SIO station (Keeling and Whorf, 2005)]; ii) for old waters with CFC-12 concentrations lower than 0.005 pmol kg −1 they assumed no anthropogenic carbon so that C dis T is given by C* mean values; and iii) for waters older than 40 years with significant CFC-12 concentrations they used a combination of the two methods mentioned above.Values of C dis T were determined along σ θ intervals.One of the main assumptions of this method is that the effective disequilibrium values remain more or less constant within the outcrop region of each isopycnal surface.Consequently, we have used SAB99 disequilibrium values on this work, using their Tables 2 and 3.
The C* approach assumes that: 1. total alkalinity is not significantly affected by the CO 2 increase in the atmosphere; 2. the effective CO 2 air-sea disequilibrium has stayed constant within the outcrop region of a particular isopycnal surface; 3. water transport is mainly along isopycnal surfaces; 4. preformed O 2 is in equilibrium with the atmosphere; 5. the decomposition of organic matter follows a constant Redfield relationship.
See Matsumoto and Gruber (2005) for a discussion of these assumptions.).The improvements introduced in C bio T calculation by LM05 are i) to account for the oxygen disequilibrium in waters formed under the ice and ii) a better characterization of TA 0 by using two different relationships (depending on water mass origin) which were determined using either winter and early spring surface measurements (0-50 m) or subsurface measurements (50-150 m):

IPSL approach
where: k stands for the mixing ratio of ice-covered surface waters determined using an optimum multiparameter method (OMP, see Appendix B); and α is the mean O 2 undersaturation in ice-covered waters, 12% as justified in LM05.The stoichiometry term (1/R C + 0.5/R N ) equals 0.8 following Körtzinger et al. (2001).Linear equations for the preformed values, TA 0 and C 0,obs T were obtained from winter and early spring surface data using South Atlantic and Indian ocean data (WOCE (Key et al., 2004) and OISO (Metzl et al., 2006)  where all the terms have been previously defined.The stoichiometry terms (R N and R P ) are taken from Körtzinger et al. (2001).
In the subtropical Indian Ocean the only contribution of northern water is North Atlantic Deep Water (NADW) entering the Indian Ocean south of Africa.The mixing ratio of NADW (k NADW ) is obtained from the OMP analysis, the southern water contribution is then given by 1-k NADW .The expressions used to estimate the preformed values are as follows: C 0,REF T is calculated using a water mass formed before the industrial revolution which serves as a reference (details are given in LM05).In 2002 along the CD139 cruise, NADW was detected on the Mozambique Basin between 2000 and 4000 m (Fig. B2) as a salinity maximum, old enough to be C ANT free, where no CFC-12 or CCl 4 were found (Fig. 5b).
Following LM05 and using data with a contribution of NADW higher than 50%, the mean value for C 0,REF T is −58.6±1.4 µmol kg −1 (37 samples).This value is not significantly different from that used by Vázquez-Rodríguez et al. (2009) when using the IPSL method in the whole Atlantic Ocean.
The IPSL approach assumes that: 1. total alkalinity is not significantly affected by the CO 2 increase in the atmosphere; 2. the oxygen disequilibrium in ice-covered waters is constant in space and time; 3. the decomposition of organic matter follows a constant Redfield relationship.

C* combined approach
where: C bio T is calculated as in SAB99 (Sect.3.1); TA 0 is calculated following the LM05 approach (Sect.3.2, Eq. 12); and C 280 T is calculated as a function of θ, S, TA 0 and pCO 280 2 using the constants from Lueker et al. (2000) instead of the linearized Eq. ( 5).pCO 280 2 includes the water vapour correction term as indicated by Pérez et al. (2002).C dis T is calculated as in SAB99 (Sect.3.1) but using the corresponding modified TA 0 and C 280 T .The C* combined approach shares the same assumptions as C* described above.
Regarding the uncertainty of each method a detailed error assessment is given in the corresponding publications.Typical uncertainties converge to a common value of 6-10 µmol kg −1 .www.biogeosciences.net/6/681/2009/  Biogeosciences, 6, 681-703, 2009 This carbon-based method uses the semi-conservative parameter TrOCA (Tracer combining Oxygen, inorganic Carbon and total Alkalinity).A detailed description of the TrOCA approach is given in Touratier and Goyet (2004a, b) and further improvements in Touratier et al. (2007): where C ANT is calculated as the difference between current (Eq.15b) and pre-industrial TrOCA (TrOCA 0 , Eq. 15c) according to Touratier et al. (2007) divided by a stoichiometric coefficient, a.
TrOCA 0 and the coefficient a were adjusted using 14 C and CFC-11 data to identify water masses with particular ages.The parameter values used are a=1.279±7.3×10−3 , b=7.511±5.2×10−3 , c=−1.087×10 −2 ±2.5×10 −5• C −1 and d=−7.81×10 5 ±2.9×10 4 (µmol kg −1 ) 2 .The TrOCA approach assumes that: below the mixed layer, the decomposition of organic matter follows a constant Redfield relationship and today's air-sea CO 2 disequilibrium is the same as in pre-industrial times.No explicit assumptions are made about the preformed values for alkalinity or inorganic carbon.The estimated uncertainty for the TrOCA approach to estimate C ANT is about 6 µmol kg −1 (Touratier et al., 2007).

TTD method
The Transit Time Distribution (TTD) method is a formal way of describing the history of the individual components (e.g.water molecules) making up a water sample.For any water sample collected at a given location in the ocean, the various water molecules making up the sample will have travelled different pathways to reach that point, with each molecule having its own "age", i.e. time since it was last in contact with the atmosphere.The distribution of all these ages comprises the TTD of a water sample.Once the TTD is established, in principle the concentration of any other passive tracer (e.g.anthropogenic CO 2 ) entering the ocean at the surface can be calculated.In several previous studies (following Waugh et al., 2004;Waugh et al., 2006) the TTDs have been assumed to have an inverse Gaussian shape, with the mean age ( ) and the width ( ) of the TTD as fundamental descriptors.In these studies and in this work it is also assumed that the ratio / =1, i.e. the mean age is equal to the width of the TTD.This is found to be a realistic assumption of the relation between advective and diffusive transport in the Ocean (Waugh et al., 2004(Waugh et al., , 2006;;Tanhua et al., 2008).
The TTD method used here to estimate C ANT concentrations is that described by Waugh et al. (2004Waugh et al. ( , 2006)).We assume that C ANT is an inert passive tracer (with a well known atmospheric history), and that the transfer of inorganic carbon from the atmosphere to the ocean can be determined by using the empirical relations between surface salinity and alkalinity (e.g.Brewer et al., 1986) and the inorganic carbon chemistry.Thus, with only observations of salinity, temperature and tracer, the oceanic C ANT input function for each water sample can be determined.We used CFC-12 data to determine the TTDs of the water samples using the timedependent saturation described in Tanhua et al. (2008) and we have assumed that the disequilibrium of carbon between the atmosphere and the surface ocean did not change during the last few hundred years.The latter assumption is possibly the single largest single source of error for the C ANT TTD calculation; other sources of errors are discussed in Waugh et al. (2006) and Tanhua et al. (2008).For instance, uncertainties in the / ratio propagates to uncertainties in C ANT TTD and is dependent on CFC concentrations and / ratio; the C ANT TTD estimate is relatively insensitive to errors in the / ratio for CFC-12 levels higher than 0.5-0.6 pmol kg −1 and to errors in the / ratio for moderate to large mixing ( / ≥0.75).The TTD method is also sensitive to uncertainties on the CFC saturation state at the time of water mass formation; the biasing effect is larger for CFC-12 concentrations larger than about 450 ppt due to the low atmospheric increase rate in recent times (Tanhua et al., 2008).

General ocean model C ANT
The model used here is OCCAM (Ocean Circulation and Climate Advanced Modelling), a global, medium-resolution, primitive equation ocean general circulation model (Marsh et al., 2005, describe a high-resolution version).OCCAM's vertical resolution is 66 levels (5 m thickness at the surface, 200 m at depth), with a horizontal resolution of typically 1 degree.OCCAM's prognostic variables are temperature, salinity, velocity and free-surface height.The model includes an Elastic Viscous Plastic sea-ice scheme, a K-Profile Parameterization mixed layer and Gent-McWilliams eddy parameterisation.Advection is 4th order accurate, and the model is time-integrated using a forward leapfrog scheme with a 1 h time-step.Surface fluxes of heat and freshwater are not specified but are calculated empirically using NCEP-derived atmospheric boundary quantities (Large and Yeager, 2004).In this way, simulations are forced for the period January 1958 to December 2004, and repeat cycles of this 47-year forcing are used to spin-up the model.OCCAM incorporates a NPZD (Nitrate Phytoplankton Zooplankton Detritus) plankton ecosystem model (Oschlies, 2001;Yool et al., 2007) which drives the biogeochemical cycles of nitrogen, carbon, oxygen and alkalinity.Air-sea fluxes of CO 2 and CFC tracers (for watermass age) make use of the protocols developed for Biogeosciences, 6, 681-703,   the OCMIP-2 project (see Dutay et al., 2002;Matsumoto et al., 2004).The simulation shown here was initialised from rest using physical and biogeochemical climatological fields (Conkright et al., 2002;Key et al., 2004), and underwent an initial 47 year cycle of pre-industrial spin-up.After this, the model used three 47 year cycles to simulate the period 1864 to 2004, during which model atmospheric pCO 2 followed the historical record.A duplicate C T tracer that was not exposed to this record was used to both separate the natural cycle of carbon from the anthropogenic perturbation and control for simulation drift.

C* distributions
C* is calculated here by two methods: SAB99 (Sect.3.1) and a combined (Comb) approach (Sect.3.4).The difference between these stems from their estimations of TA 0 , included in the C bio T term, and C 280 T , also dependent on TA 0 .In SAB99, TA 0 is taken from Eq. ( 3) and adjusted as a function of Indian Ocean data shallower than 60 dbar.C 280 T is calculated from Eq. ( 5) which was obtained by Gruber et al. (1996) for the Atlantic Ocean.In the case of the combined approach, TA 0 discerns waters with southern and northern origins, taking into account the mixing, and C 280 T is obtained using a thermodynamic formula to calculate C 280 T as a function of pCO 280 2 and TA 0 (see Sect. 3.3).These details in the C* calculation have a significant impact on estimates of C ANT .Figure 2 shows the vertical distributions of C* SAB99 and C* Comb: below 1000 dbar C*SAB99 is lower, more negative, than the C*Comb by about 10 µmol kg −1 , with maximum negative differences (SAB99-Combined C*) found in the NADW core (Fig. 3).This difference inverts in the upper 1000 dbar, C*Comb is higher by up to 5 µmol kg −1 (Fig. 2 and 3).A closer look at the contribution from C bio T and C 280 T to the C* difference shows (Fig. 3) that the C 280 T is primarily responsible for the differences between SAB99 and Comb C* in the upper 1000 dbar.Here the mean±STD C 280 T contribution to the mean±STD C* difference is −4.5±1.9 compared to 3.1±2.5;while below 1000 dbar both terms have a similar contribution, 4.7±1.8from C 280 T and 4.2±1.7 µmol kg −1 from C bio T ( C* difference −8.9±2.4 µmol kg −1 ).This result points to the importance of using accurate approximations for preformed values, especially where mixing of extreme origin water masses is occurring, and also that the complete thermodynamic equation for C 280  T should be used in preference to linearized functions obtained for different basins.

C dis T values
The next step in the C* back-calculation approach is to obtain the C dis T values by σ θ intervals (Sect.3.1).C dis T values should ideally be calculated along a wide regional or age range, therefore, databases with ample coverage are needed.These also enable the identification and treatment of different water mass end-members.The database used by SAB99 covered the whole Indian Ocean, where most of the waters below the mixed layer have a southern origin except intermediate waters formed in the Arabian Sea.Taking this into account and the assumption of a constant C dis T with time we used SAB99 C dis T values.No Arabian Sea endmember was considered.However, to check the impact of using different TA 0 and C 280 T approximations on C dis T we recalculated new values for the effective air-sea disequilibrium: using our own CD139 data set for C* and C*τ i) from Sect.3.1, the SAB99 approach; and ii) from Sect.3.3, the combined approach.Figure 4    own data for σ θ <26.8 should be disregarded due to the reduced density of samples with these characteristics (situated above 600 dbar along the CD139 section).Data between 27.55>σ θ >27.25 should be weighted means between C* and C*τ ; however, our own data has no points that fulfil the conventions in SAB99, and consequently they were linearly interpolated.C dis T values between 27.2>σ θ >26.8 are obtained from C*τ values and they show a clear consistency between the SAB99 (SAB99 Tables and our own data) and combined (for our data) approaches.For the C*τ calculation the main source of error is the estimation of the age, calculated equally on the three C*τ approaches, here the effect of a different preformed TA 0 (for the combined method) is minor (Matsumoto and Gruber, 2005).In deep layers, σ θ ≥27.6, the C dis T disagreement is obvious between the two SAB99 methods and the combined one, the two SAB99 estimations coincide as expected, with values from −11 to −18.6 µmol kg −1 while with the combined approach, C dis T varies from −4 to −6.5 µmol kg −1 .This difference stems from the C 280 T and TA 0 (Fig. 3), included in the C* formula.

C ANT distributions
The vertical pCFC-12, pCCl 4 and C ANT distributions along with some reference neutral density levels are shown in Figs. 5 and 6.In deep layers below γ =27.7, no C ANT is expected according to CFC-12 levels (Fig. 5a), here the C* method is used to calculate C dis T , and C ANT SAB99 values are practically null, 0±3 µmol kg −1 , which is less than the limit of detection of the method.However, according to CCl 4 levels C ANT is expected in this layer.In this sense, C ANT IPSL and TrOCA estimates range from 0 to 10µmol kg −1 , with slightly lower values estimated by the TTD method and values below 5 µmol kg −1 simulated by OCCAM.The western NADW core (Fig. B2) between 2000-3000,dbar presents consistently negative values for the SAB99 method (Fig. 5b).A combination of processes leads to these negative values: predominantly the erroneous estimation of TA 0 in the biological correction, and also the application of a C dis T too negative and more representative of southern waters, dominant in the rest of the section.The TrOCA, IPSL and TTD methods do detect this core while the OCCAM results are quite homogeneous for deep waters (Figs. 5 and 6).The influence of water formed under ice is clearly detected by the C ANT -IPSL method below 4000 dbars (Fig. B2) where C ANT slightly increases towards the bottom (Fig. 5f) as pCCl 4 does (Fig. 5b).The TrOCA and TTD methods also show this slight increase (Fig. 6b, d) despite being based on completely different assumptions.
In the region between roughly 1000 and 1500 dbar the SAB99 method uses weighted means of C*τ and C* to estimate C dis T (Fig. 4), and a steep gradient in C ANT is detected here (Fig. 5c), with a clear discontinuity with pressure not apparent in any of the other approaches or physical, biogeochemical or tracer variables (not shown).Similar results were found by Waugh et al. (2006) in the Atlantic Ocean.The other C ANT methods with a smoother C ANT gradient in this pressure range point to a deeper penetration of C ANT below the AAIW limit.
In thermocline waters, upper 1000 dbars, all methods, except TTD, show C ANT values increasing eastwards, near to the formation region of SAMW (McDonagh et al., 2005).Although the distributions are similar, absolute values differ as will be commented using Fig. 7.
Figure 7 shows the mean±STD vertical C ANT profile difference referred to SAB99.Any of the regional plots show a sharp change around 1300 dbar due to the discontinuity in Biogeosciences, 6, 681-703, 2009 www.biogeosciences.net/6/681/2009/C ANT SAB99 where all the methods yield higher C ANT values compared to SAB99 up to a maximum of 8 µmol kg −1 with the TrOCA method in the middle part of the section (Fig. 7c), where the salinity minimum is clearly noted.Surprisingly, the IPSL, TrOCA and TTD differences have similar values and distributions below 3500 dbar, with consistently increasing C ANT values towards the bottom, especially in the western and central portion of the section (Fig. 7b, c), where younger AABW arrives at the section from the Weddell Sea.In the upper 1000 dbar, TrOCA and SAB99 are practically in agreement, while OCCAM presents lower values compared to SAB99 especially in the western part of the section (Fig. 7b), and TTD differences continuously increase towards the surface and reach up to 15 µmol kg −1 .In this depth range, IPSL values are consistently higher compared to any other method.Between 1500 and 3500 dbars, as a whole (Fig. 7a), TrOCA and TTD, compared to SAB99, give higher results than OCCAM and IPSL, but differences arise within regions.

C ANT inventories
Studying the C ANT specific inventories by water mass domains shows again clear discrepancies and similarities.We took the neutral density layers definition by Robbins and Toole (1997) to define five layers (Fig. 5     Deep Water, CDW, and NADW on the western end) deep waters; v) bottom water, below roughly 3500 m with an Antarctic origin.A similar approach was used by McDonagh et al. (2008) to constrain the velocity field along this section.
The initially calculated C ANT values are randomly modified by ±5 µmol kg −1 .A set of 100 perturbations are done for the five methods, finally a mean and standard deviation for the total and layer C ANT inventory is calculated, the standard deviation for each layer is weighted by the layer contribution to the total section area.Inventories are shown in Fig. 8 and Table 1.
The SAB99 method estimates the lowest total inventory compared with any other method (Fig. 8, Table 1), even with OCCAM which seems to underestimate C ANT in deep and bottom layers (Fig. 7).Discrepancies and similarities arise when inventories are studied by layers.Biological processes in the upper mixed layer (comprised within the surface layer here defined) occurring during the cruise prevent the use of carbon-based methods (SAB99, TrOCA and IPSL), for ex-ample when AOU is negative.These methods neither resolve seasonal variability in the mixed layer.TTD and OCCAM do provide C ANT estimates in this layer by circumventing the direct use of biogeochemical variables: TTD relies on CFC ages that are more precise in upper, younger waters; OCCAM accounts for surface circulation and air-sea CO 2 equilibrium, and its upper waters are less affected by uncertainties in model physics and chemistry.Interestingly, the mean specific inventories for the upper layer from TTD and the OCCAM are similar, around 7 molC m −2 (Table 1).
Within the SAMW layer, all methods, except IPSL, agree within ±2 molC m −2 .Within AAIW, IPSL is again high, TTD and TrOCA agree and OCCAM and SAB99 are lower.In deep waters, TTD and TrOCA give similar results significantly higher than any of the other methods, even IPSL.In the bottom layer, TrOCA, IPSL and TTD provide similar results with a significant C ANT accumulation in this layer, while OCCAM, SAB99 and Iwithout the oxygen undersaturation correction (IPSL-Zero) show no accumulation.

Discussion
In most of the comparative studies about C ANT estimation in the ocean no clear conclusion about the best method is given because all of them are subject to uncertainties (Coatanoan et al., 2001;Feely, 2001;Hall et al., 2004;LM05;Sabine and Feely, 2001;Wanninkhof et al., 1999;Waugh et al, 2006).In this work with the help of transient tracers, we discuss backcalculation, TrOCA, TTD and OGCM C ANT methods, trying to assess their strengths and caveats, and finally which may provide the optimal range of estimations.

Disequilibrium values on the C* method
The assumptions of the C* method and its main sources of uncertainty are thoroughly discussed in Matsumoto and Gruber (2005).They conclude that the change in air-sea CO 2  disequilibrium over time is the single most important contribution to the bias in C*-based C ANT estimations.This method assumes a constant disequilibrium over time.However, the uptake of C ANT by the ocean is occurring with an increasing more negative disequilibrium.Consequently, C ANT will be overestimated especially in upper and younger waters, causing a positive bias of 5 PgC for the whole ocean (Matsumoto and Gruber, 2005).
Disequilibrium values obtained here from C*τ values are equal, as the main source of uncertainty is the age estimate, while biases from using different expressions for TA 0 and C 280 T cancel out (Fig. 4).When disequilibrium values are calculated from C* values in waters with no C ANT expected, the disagreement is clear (Fig. 4).Here, TA 0 and C 280 T estimates do matter (Fig. 3) causing high discrepancies in the final disequilibrium estimate.C dis T estimated by the SAB99 method become more negative with increasing density, while in the combined approach they remain slightly negative and practically constant (Fig. 4).
Which C dis T values are more reasonable with our current knowledge about CO 2 dynamics in the upper ocean?The temporal evolution of the disequilibrium can only be obtained from OGCM (e.g., Matear et al., 2005;Matsumoto and Gruber, 2005) or transient tracers transit-time distributions (Hall and Primeau, 2004), while current values of the total (natural plus anthropogenic) disequilibrium can be approximated from measured (Takahashi et al., 2002) or empirical (McNeil et al., 2007) winter-early spring surface ocean-atmosphere gradients in pCO 2 .The models provide the change in C dis T from the preindustrial era till the 1990s, showing small changes, less than −5 µmol kg −1 , in the subtropical Indian ocean, and moderate changes, −5 to −10 µmol kg −1 south of 60 • S. Consequently, current pCO 2 gradients can be compared with C dis T values obtained from the C* approach.
Figure 9a shows the mean surface pCO 2 gradient sorted by surface density in the winter Indian Ocean from the Takahashi et al. (2002) climatology and the empirical approach by McNeil et al. (2007).Comparing Figs. 4 and 9b discrepancies are evident, first in the shape of the curves and second in the range of values estimated.The discontinuity in C dis T values from SAB99 appears to be an artefact derived from the approximations used in the method.
Deep waters, around σ θ =27.4 have C dis T values around −8 µmol kg −1 according to SAB99, but values around −3 µmol kg −1 by Takahashi et al. (2002) or −5 to −10 µmol kg −1 according to McNeil et al. (2007).If we consider the latter work more reliable for waters south of 60 • S, the C dis T SAB99 are reasonable in deep waters and C ANT values as well.
Nevertheless, we have to question whether it is sensible to compare surface winter air-sea C T disequilibrium values from the whole Indian Ocean for water masses formed in distinct regions of the Southwest Atlantic (AAIW), the Southeast Indian (SAMW) oceans, or the Weddell Sea (WSDW).The data of McNeil et al. (2007) show a large variability in waters denser than σ θ =26.5 (found south of 55 • S) suggesting the difficulty in defining a C dis T value for intermediate and deep waters along the CD139 section.
Disequilibrium values obtained from the SAB99 method seem to be flawed; in upper and intermediate waters they lead to a C ANT underestimate, while in deep and bottom waters cancel any C ANT accumulation.The other carbonbased methods without any a priori disequilibrium assumption, IPSL and TrOCA, predict 1.34 and 1.13 times higher C ANT specific inventories in surface, SAMW and AAIW waters; while TTD and OCCAM estimates are 1.29 and 1.18 times higher.Transient tracers such as CFC-12 and CCl 4 provide useful information on oceanic ventilation and transport times and therefore on the uptake and storage of C ANT in the ocean.The increase of C ANT in the atmosphere began earlier than for these transient tracers.The presence of a significant CFC-12 or CCl 4 concentration indicates that the water parcel was exposed to anthropogenic CO 2 in the atmosphere.However, a region free of these tracers may still have significant amounts of C ANT , leading to C ANT underestimation (e.g., Goyet and Brewer, 1993;Matsumoto and Gruber, 2005;Tanhua et al., 2004).Significant concentrations of CCl 4 are detected in the atmosphere after 1940 compared to CFC-12, detected after 1960, so CCl 4 can be used to trace C ANT via the C* method (Holfort et al. 1998;Wallace, 2001).
Another important assumption that can cause significant biases in the C*-based C ANT estimations is that CFCs would provide accurate ventilation ages (Matsumoto and Gruber, 2005).This assumption is only true if several conditions are fulfilled: first that preformed tracers were saturated and second, that transport is mainly advective.Mixing biases on age are practically compensated when different but significant tracer concentration waters mix within a linear trend in the atmospheric time evolution (e.g., Haine and Hall, 2002).According to Matsumoto and Gruber (2005) using singletracer ages in the C* method only apply to CFC ages less than 30 years causing limited biases in the C ANT estimation as both tracers, CFCs and C ANT , increase roughly linearly in the 1990s.
To assess which method provides a more robust C ANT estimate we can study the relationship between the partial pressures of , CCl 4 (pCCl 4 ) (calculated assuming 100% saturation and using solubility equations from Warner and Weiss (1985) and Bullister and Wisegarver (1998), respectively) and the theoretical upper and lower limits of oceanic C ANT .Assuming a mainly advective transport, with low mixing, the theoretical time evolution of oceanic C ANT can be calculated for two different types of surface waters, with Antarctic (TA=2280 µmol kg −1 , θ=4 • C, Sal=34, PO 4 =2 µmol kg −1 and SiO 2 =20 µmol kg −1 , Rev-elle=13) and subtropical (TA=2340 µmol kg −1 , θ=17 • C, Sal=35.5, PO 4 =0.2 µmol kg −1 and SiO 2 =3 µmol kg −1 , Rev-elle=10) origins, using the atmospheric evolution of CO 2 in the southern hemisphere and a preindustrial CO 2 value of 280 ppmv, and CO 2 constants from Lueker et al. (2000).Physical and chemical characteristics for surface waters in the subtropical Indian Ocean and the Indian sector of the Southern Ocean were taken from the GLODAP atlas and Metzl et al. (2006), Revelle factors are in agreement with Sabine et al. (2004).The theoretical curves are only valid under time-constant temperature, salinity and alkalinity, an assumption that is compromised by global warming in the ocean (e.g., Levitus et al., 2001), the likely alkalinity changes due to ocean acidification (e.g., Sarma et al., 2002), and  salinity increases (e.g., McDonagh et al. 2005).Mixing should be relevant in deep and bottom waters according to the formation mechanism of CDW and AABW.Consequently, the calculated curves should only be considered as indicative of the possible upper and lower C ANT evolution limits in the upper water of the subtropical Indian Ocean since both will be overestimations in deep and bottom waters, where mixing overcomes advection.
The relationship between pCFC-12 and C ANT in upper waters is shown in Fig. 10, along with the theoretical curves.IPSL stands out with high C ANT values compared with the upper (Revelle=10) theoretical limit.Despite being based on different approaches the TrOCA and the SAB99 C ANT values are very similar in between the two theoretical limits.C ANT TTD values approach the upper limit towards the surface (where most waters have a subtropical origin) and the lower one towards the AAIW layer.OCCAM is the opposite of IPSL, with lower than any expected values.However, despite the fact that CFC-12 penetration in OCCAM compares well with our data, temperature and salinity data seem to be lower than expected, so consequently pCFC-12 in OCCAM is underestimated and the OCCAM pCFC-12 vs. C ANT relationship shown in Figure 10   is decomposed at higher temperatures (Huhn et al. 2001).C ANT values in deep waters are generally lower than the theoretical value.This is not surprising as before 1960 the relationships between CFC-12 or CCl 4 and C ANT are very nonlinear, there is a strong mixing bias and some water masses incorporate ice-shelf waters formed with a 45% undersaturation in CFC-12 or CCl 4 (Huhn et al., 2001).Significant tracer concentrations are found in deep Indian waters which points to a likely exposure to atmospheric C ANT levels of these waters when formed.Higher levels of pCFC-12 and pCCl 4 below 4000 dbar were found on the western part of the section (Fig. 5) where younger AABW arrives to the section (Orsi et al., 1999).The SAB99 and OCCAM mean concentrations in Fig. 11b are 0.4±2.8 and 0.4±0.6 µmol kg −1 , while respectively for IPSL, TrOCA and TTD methods it is 1.5±3, 3±3, and 3±2 µmol kg −1 .Neither Fig. 10 nor 11 show the lower part of the AAIW layer, but here TrOCA and TTD yield similar C ANT values, SAB99 is compromised by its discontinuity and IPSL estimates are 10 µmol kg −1 higher in the upper AAIW than in the lower AAIW.

Evaluation of the five C ANT methods
Taking into account i) that most of the water masses in the subtropical Indian Ocean have been formed in the Southern Ocean, a particularly challenging area for data-or modelbased C ANT estimates, and ii) the uncertainties inherent for the set of five C ANT methods here evaluated; we attempted to summarise the consistency of each method using the knowledge about water mass formation and the relation between C ANT estimates and transient-tracers in five water mass layers defined along the 32 • S Indian Ocean section.
Upper surface waters, shallower than about 200 dbars, occupy a small fraction of the whole water column.Carbonbased methods are unable to properly correct for biological activity and therefore unable to estimate C ANT (Fig. 8, Table 1), while TTD is based on tracer distributions independent of biological activity, but affected by seasonality; and OCCAM assumes CO 2 air-sea equilibration, which is not always true.Despite the uncertainties, we consider TTD and OCCAM more reliable in the surface layer, and therefore the inventory here would be 7±0.2molC m −2 .
At the SAMW layer, between roughly 200-700 dbars, carbon-based, TTD and OCCAM show relatively good agreement in C ANT values (to within ±6µmol kg −1 ) and inventories (±2 molC m −2 ).CFC-12 levels are higher than 0.6 pmol kg −1 in this layer and the relation between surface CO 2 and CFC saturation is one (Fig. 6 in Matear et al., 2003), so the TTD estimates offer highly reliable support to the other methods, except IPSL which appears too high.Thus the inventory in this layer would be 8.9±0.5 molC m −2 (mean from SAB99, TTD, TrOCA and OCCAM).
In the AAIW layer, ≈700-1500 dbars, the situation is more complicated.CFC levels below 1000 dbars sharply change to values lower than 0.5 pmol kg −1 , and CO 2 /CFC saturation varies between 0.8-0.9 in the formation area of AAIW, at the Brazil-Malvinas current confluence, consequently, the uncertainty in C ANT TTD increases.OCCAM, predicts a CFC-12 penetration in agreement with observations, and C ANT values in agreement with the other methods, despite this, relatively low inventories are obtained because of the density field after the misleading temperature and salinity fields.C ANT distributions obtained from TrOCA, TTD and OCCAM are quite similar (Figs. 6 and  7), while SAB99 has the discontinuity below 1200 dbar and, IPSL estimates are 10 µmol kg −1 high in the upper AAIW (Fig. 5).No definitive support for any data method can be derived from the relation with tracers between 1000-1500 dbars where CFC-12 is too low and CCl 4 is affected by decomposition.However, SAB99 can be disregarded because of the unreasonable discontinuity and OCCAM because of its inaccurately simulated density field.Thus an inventory of 9.4±0.1 molC m −2 can be assigned to the AAIW layer (mean from TTD and TrOCA).
Here OCCAM fails to predict any CFC-12 or CCl 4 signal in these layers while data shows an increase with depth in their values, especially CCl 4 below 3500 dbars.SAB99 C ANT estimates distribute randomly around zero at these layers.As a result, a low confidence is attributed to SAB99 or OC-CAM in these layers.The TTD, TrOCA and IPSL methods predict significant and consistent C ANT values, from 0 to 10 µmol kg −1 , except at the NADW core where they agree with C ANT ≈ 0 µmol kg −1 .Relying on TTD, TrOCA and IPSL on both layers, the mean inventory would be 1.5±0.8 and 1.3±0.1 molC m −2 in deep and bottom waters, respectively.
Summing up, we would suggest a best estimate for the water column specific inventory in the subtropical Indian Ocean of 28±2 molC m −2 , which is significantly higher than the C* value, 24±2 molC m −2 .

Conclusions
This work investigates the C ANT penetration and inventory in the subtropical Indian Ocean along 32 T values are inconsistent with the current airsea CO 2 disequilibrium found in current Indian Ocean winter C dis T values.Although this could also be misleading taking into account that water masses are usually formed in particular times and areas of the ocean.
Previous estimates of C ANT inventory in the subtropical Indian Ocean with the C* method appear to be underestimates.Considering the C ANT estimates derived from those methods consistent with the tracer distributions and the knowledge about water masses, our best estimate for the mean C ANT specific inventory is 28±2 molC m −2 which compares with 24±2 molC m −2 from C*, 17% higher.
These conclusions, so far, apply only to the Indian Ocean subtropical gyre.Despite this, our conclusions may have important implications not only for quantifying the uptake and storage of C ANT in the Indian Ocean basin, but also for predicting the consequences of acidification on the local carbon cycle and marine biota.Although tedious, comparison exercises within other ocean basins are still necessary and revealing.Combining them with time-evolution approaches from repeat sections or time-series analysis will further help to constrain how, where and how much C ANT is penetrating into the ocean and also the data and model-based C ANT estimation methods.Fig. A1.Spectrophotometric pH25T measurements on the CRM batch 55 during the cruise.Each set of analyses consists of 7 to measurements from the same CRM bottle.The first six sets of CRM pH analysis (black dots) were discarded in the calculation of the final mean and standard deviation (STD), 7.9169±0.0018.Note that the pH values contain the R and 0.0047 corrections.

Appendix A
Further information on the CO 2 data quality assessment A1 Quality assessment on the pH measurements Method details: pH was measured spectrophotometrically following Clayton and Byrne (1993) using a seawater mcresol purple dye solution.Samples were analyzed in optical cells with a 10-cm path length thermostated at 25±0.2 • C in a CECIL 3041 spectrophotometer.Following determination of the blank, absorbance measurements were done.pH values are reported on the total scale at 25 • C (pH T 25 , mol kg-sw −1 ).
Accuracy: practically every other day during the CD139 cruise (24 over 45 days), seven to eight samples from a CRM bottle (batch 55, certified chemical characteristics for salinity, 33.506; silicate, 2.7 µmol kg −1 ; nitrate, 0.46 µmol kg −1 ; nitrite, 0.0 µmol kg −1 ; phosphate, 0.41 µmol kg −1 ; total alkalinity, 2227.85±0.54µmol kg −1 ; and total inorganic carbon, 2012.06±0.34µmol kg −1 ) were drawn carefully to avoid bubbles and analysed for pH using the spectrophotometric method.Those chemical characteristics combined with the dissociations constants from Lueker et al. (2000) and an Excel-based CO 2 software program developed by M. Álvarez and F. F. Pérez, yield a theoretical pH T 25 value of 7.917.The mean and standard deviation of the seven or eight samples for each day is shown in Fig. A1.The first six sets of CRM pH analysis were too high compared to the theoretical value.There is no clear explanation for this; perhaps the transfer procedure from the CRM bottle into the 10 cm cell was not optimized at the beginning.Taking into account 16 of the 24 sets (white circles in Fig. A1) the mean value for the CRM measurements was 7.9169±0.0018.This value takes into account the double addition correction (see below)   and the 0.0047 addition proposed by DelValls and Dickson (1998) and Millero (2007).
Reproducibility: several cells from the same Niskin bottle were collected along the cruise to check the reproducibility of our measurements.Table A1 shows the characteristics of these deep water samples and the mean, standard deviation (STD) and number of cells collected from each bottle.The mean of the STD is 0.0009 which could be considered as the reproducibility of pH measurements during the cruise.
Dye addition correction: the injection of the indicator solution into seawater perturbs the sample pH slightly (Clayton and Byrne, 1993).The magnitude of this correction depends on the different acidity between the sample and the indicator solution.Consequently it should be quantified for each batch of dye solution.During the CD139 cruise, one batch of indicator solution done in seawater (1mM) was used, 100 µl of this solution were added to the sample and the absorbance ratio R 1 =(578A−730A)/(434A−730A) is calculated.From  a second addition of the dye solution another R 2 is also calculated.This operation is done over a wide pH range using the upper and deep CD139 samples.Figure A2 shows the relationship between R (=R 2 -R 1 ) and R 1 .The correction equation is the following: R = −0.0087±0.0007× R 1 + 0.0099±0.0009,(A1) Consequently, the final R (R ok ) value is calculated as: where R is calculated with Eq.A1 as a function of R 1 .
A2 Quality assessment on the TA measurements TA was measured using automatic potentiometric titration with 0.1M HCl to a final pH of 4.4, with a Metrohm 6.0233.100combination glass electrode and a Pt-100 probe to check the temperature (Pérez and Fraga, 1987).The electrode was standardised using a 4.4 buffer made in CO 2 free seawater (Pérez et al., 2000).Concentrations are given in µmol kg-sw −1 .The 0.1 N hydrochloric acid was prepared mixing 0.5 mol (18.231 g) of commercially HCl supplied by Riedel-deHaën (Fixanal 38285) with Milli-Q water into a graduated 5L beaker at controlled temperature conditions and referred to 20 • C (Table A2).The variation caused by salinity after the titration is lower than 0.1 units, which is taken into account in the final TA calculation.
CRM analyses were performed in order to control the accuracy of our TA measurements (Fig. A3).Accordingly, the final pH of every batch of analyses was corrected to obtain the closest mean TA on the CRM analyses to the certified value.Usually, each sample was analyzed twice for alkalinity.Table A2 shows the average difference of the replicates analyzed during each batch of analysis.This difference was about 1.0 µmol kg −1 .
Surface seawater was used as "quasi-steady" seawater substandard, it consists in surface seawater taken from the nontoxic supply and stored in the dark into a large container (25 l) during 2 days before use.This substandard seawater was analyzed at the beginning and at the end of each batch of analyses to control the drift in the analyses for each batch.

A3 Comparison between calculated and measured C T
As explained in the Data set section, salinity-normalized C T calculated from pH T 25 and TA (NC T calc) using Lueker et al. (2000) constants compared to normalized coulometric C T (NC T coul) with a linear relationship: NC T calc=1.006±0.007×NCT coul−14±15 (r 2 =0.998, n=51, 0±4 µmol kg −1 , mean ±STD of the residuals).Figure A4 shows this relationship.

A4 Crossover analysis for CO 2 variables
Another approach to evaluate the quality of physical or chemical data from a particular cruise is to compare the distribution of the tracers with already calibrated cruises at crossing or overlapping positions, at density or pressure levels where no temporal changes are expected.The final aim is to integrate the new cruise within the merge-calibrated data set.The Global Ocean Data Analysis Project (GLO-DAP, http://cdiac.esd.ornl.gov/oceans/glodap/Glodaphome.htm) (e.g., Key et al., 2004) was a major effort of the oceanographic community to produce calibrated and uniform data bases for the different ocean basins.Here, we use the GLO-DAP calibrated cruises for the Indian Ocean crossing the CD139 cruise (Fig. A5).
The CD139 western section from Africa to the Madagascar basin completely overlaps with the 1995 I5W section (Donohue and Toole, 2003) (Fig. A5).North Atlantic Deep Water (NADW) is confined by the Madagascar and Daves ridges to the Natal Valley and Mozambique basin (Toole and Warren, 1993).This water mass is the oldest in the Indian Ocean were no temporal trends in tracers should be found if we assume that hydrography and biogeochemical processes have been in steady state for the period covered by the data (1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002).NADW is characterized by a salinity maximum around θ=2 • C and nitrate and C T minimum.
the threshold value used in the GLODAP comparison exercises (6 µmol kg −1 , Lamb et al., 2002;Wanninkhof et al., 2003); in the case of C T , the 2002 data seem to be 4 µmol kg −1 higher, but this value is still near the threshold value (4 µmol kg −1 , Lamb et al., 2002;Wanninkhof et al., 2003).No corrections are applied based in this first crossover.
The CD139 transindian section also intersects with other GLODAP lines in the Indian Ocean (Fig. A5).The quality and consistency of all the tracers measured on these cruises have been assessed in GLODAP and further information regarding the CO 2 data can be found in SAB99.Here we present results for TA, C T , CFC-11 and CFC-12, the approximation used to calculate the differences between data sets follows the convention agreed for the synthesis effort on post WOCE cruises currently on course within the framework of the European CARBOOCEAN integrated project.The methodology is as follows: at least three stations from two cruises are selected falling within a limit of 100 km, tracer data below 1500 m are selected and plotted against σ 4 , a mean profile for each cruise is obtained at standard densities using a piecewise cubic hermite interpolation function, con-sequently a mean and standard deviation profile is obtained for each cruise, after subtracting those two profiles, a mean and standard deviation difference value is calculated for every pair of cruises at each crossover.Differences are considered significant when the mean value is higher than twice the standard deviation.The thresholds limits for correction are 0.005 for CFCs, 4 µmol kg −1 for C T and 6 µmol kg −1 for TA.
Table A4 and Fig. A6 show the mean and standard deviation differences between any cruise and the CD139 one.At the light of these results CD139 CFCs and TA data are consistent with the other GLODAP cruises but CD139 C T data seems to be higher in about 3 or 4 µmol kg −1 .We decreased the calculated CD139 C T data in 4 µmol kg −1 .

Fig. 1 .
Fig. 1.Positions of the CTD stations occupied during the CD139 cruise crossing the Indian Ocean.Stations 59 and 113 are marked.

Fig. 1 .
Fig. 1.Positions of the CTD stations occupied during the CD139 cruise crossing the Indian Ocean.Stations 59 and 113 are marked.
Fig. 2. ∆C* (µmol kg -1 ) vertical distribution along the CD139 section estimated according to SAB99 (a, b) and the combined method (c, d).White lines correspond to σ θ isopycnals used as references.

Fig. 4 .
Fig. 4. Air-sea CO 2 disequilibrium (µmol kg -1 ) by σ θ density intervals.Blue points correspond to the values taken directly from SAB99 work, light blue and orange points correspond to the values estimated using the CD139 biogeochemical data, following SAB99 (orange) and combined (light blue) methods.Interpolated points are highlighted with black open circles.

Fig. 4 .
Fig. 4. Air-sea CO 2 disequilibrium (µmol kg −1 ) by σ θ density intervals.Blue points correspond to the values taken directly from SAB99 work, light blue and orange points correspond to the values estimated using the CD139 biogeochemical data, following SAB99 (orange) and combined (light blue) methods.Interpolated points are highlighted with black open circles.

Fig. 8 .
Fig. 8. Mean ± standard deviation C ANT specific inventories (molC m -2 ) for each water mass layer (see text) and for the whole section on the right.The total specific inventory is calculated as the sum of the SAMW, AAIW, Deep and Bottom contributions plus 7 molC m -2 , i.e., the TTD and OCCAM mean specific inventory for the upper layer.

Fig. 8 .
Fig. 8. Mean±standard deviation C ANT specific inventories (molC m −2 ) for each water mass layer (see text) and for the whole section on the right.The total specific inventory is calculated as the sum of the SAMW, AAIW, Deep and Bottom contributions plus 7 molC m −2 , i.e., the TTD and OCCAM mean specific inventory for the upper layer.

Fig. 9 .
Fig. 9. a) Late winter-early spring pCO2 gradient (ocean-atmosphere, in µatm) for the Indian ocean south of about 40ºS by σθ density from the Takahashi et al. (2002) climatology: raw data (crosses) and mean ± STD (green line); and empirical data from McNeil et al. (2007): raw data (blue points) and mean ± STD (red line).b) Equivalent CT disequilibrium values calculated from ∆CT dis = ∆pCO2/359 • 2020/R, 359 is the mean atmospheric pCO2 and 2020 the mean surface ocean CT for 1995, R the Revelle factor, taken as 9 and 13, data from Takahashi et al. (2002) are the black and green lines, data from McNeil et al. (2007) in pink and red.

Fig. 9 .
Fig. 9. (a) Late winter-early spring pCO 2 gradient (oceanatmosphere, in µatm) for the Indian ocean south of about 40 • S by σ θ density from the Takahashi et al. (2002) climatology: raw data (crosses) and mean±STD (green line); and empirical data from McNeil et al. (2007): raw data (blue points) and mean ± STD (red line).(b) Equivalent C T disequilibrium values calculated from C dis T = pCO 2 /359×2020/R, 359 is the mean atmospheric pCO 2 and 2020 the mean surface ocean C T for 1995, R the Revelle factor, taken as 9 and 13, data from Takahashi et al. (2002) are the black and green lines, data from McNeil et al. (2007) in pink and red.

Fig. 10 .
Fig. 10.Partial pressure of CFC-12 (ppt) and C ANT estimates (µmol kg -1 ) fo potential temperature higher than 5ºC, pressure higher than 200 dbar and CFC years.Also shown in black is the atmospheric evolution of CFC-12 and C ANT u of 10 and 13 (see text for details), some time markers are shown.

Fig. 10 .
Fig. 10.Partial pressure of CFC-12 (ppt) and C ANT estimates (µmol kg −1) for upper waters with potential temperature higher than 5 • C, pressure higher than 200 dbar and CFC-12 age less than 30 years.Also shown in black is the atmospheric evolution of CFC-12 and C ANT using Revelle factors of 10 and 13 (see text for details), some time markers are shown.
Fig. 11.a) Partial pressure of CFC-12 (ppt) and b) partial pressure of CCl 4 (ppt) versus C ANT (µmol kg -1 ) estimates for deep waters with potential temperature lower than 5ºC, pressure higher than 200 dbar and CFC-12 age higher than 40 years.In the case of CCl 4 , the temperature limit is 3ºC.Also shown in black are the atmospheric evolution of CFC-12, CCl 4 and C ANT using Revelle factors of 10 and 13 (see text for details), some time markers are shown.

Fig. 11 .
Fig. 11.(a) Partial pressure of CFC-12 (ppt) and (b) partial pressure of CCl 4 (ppt) versus C ANT g(µmol kg −1 ) estimates for deep waters with potential temperature lower than 5 • C, pressure higher than 200 dbar and CFC-12 age higher than 40 years.In the case of CCl 4 , the temperature limit is 3 • C. Also shown in black are the atmospheric evolution of CFC-12, CCl 4 and C ANT using Revelle factors of 10 and 13 (see text for details), some time markers are shown.
Fig. A1.Spectrophotometric pH25T measurements on the CRM batch 55 during the cruise.Each set of analyses consists of 7 to 8 measurements from the same CRM bottle.The first six sets of CRM pH analysis (black dots) were discarded in the calculation of the final mean and standard deviation (STD), 7.9169±0.0018.Note that the pH values contain the ΔR and 0.0047 corrections.

Fig. A2 .
Fig. A2.Perturbation of sample pH induced by addition of indicator, expressed as R (=R 2 -R 1 ) as a function of R 1 .R 1 is the first addition and R 2 the double addition.R is the ratio between absorbances ((578A-730A)/(434A-730A)).
Fig. A4.Relationship between measured and calculated (from pH and alkalinity) salinity normalized total inorganic carbon (NCT in µmol kg -1 ).

Fig. A5 .
Fig. A5.Indian Ocean map showing the intersections or crossovers between the CD139 a GLODAP cruises.Each crossover is identified with a number.

Fig. A5 .
Fig. A5.Indian Ocean map showing the intersections or crossovers between the CD139 and GLODAP cruises.Each crossover is identified with a number.
Fig. A6.Mean and standard deviation between the CD139 and GLODAP cruises at each crossover for (a) CFC-11; (b) CFC-12; (c) CT and (d) TA.Each crossover is identified with a number as in Figure A5.Results also shown in TableA4.

Fig. A6 .
Fig. A6.Mean and standard deviation between the CD139 and GLODAP cruises at each crossover for (a) CFC-11; (b) CFC-12; (c) C T and (d) TA.Each crossover is identified with a number as in Fig. A5.Results also shown in TableA4.
GLODAP cruises overlapping our cruise (see Appendix A4) and concluded that the calculated CD139 C T data should be reduced by 4µmol kg −1 .
www.biogeosciences.net/6/681/2009/Biogeosciences, 6, 681-703, 2009 WOCE-3 Back-calculation methods to estimate C ANT presents the three C dis , estimated with the SAB99 and Comb.methods.The NADW core is highlighted in black.All in µmol kg -1 .For clarity every 4 th sample was represented.SAB99 and Comb.methods.The NADW core is highlighted in black.All in µmol kg −1 .For clarity every 4th sample was represented.

Table 1 .
Mean±standard deviation C ANT specific inventory (molC m −2 ) by water masses in the subtropical Indian Ocean for the different methods here evaluated.Mean and standard deviation values were obtained randomly modifying the initially calculated single C ANT values by ±5 µmol kg −1 .A set of 100 perturbations were done for the five methods.The standard deviation for each layer is weighted by the layer contribution to the total section area.See text for the acronyms.N/D stands for not determined.Values between brackets correspond to the IPSL method assuming a 100% oxygen saturation in Eq. (7), α=0.The total specific inventory is calculated as the sum of the SAMW, AAIW, Deep and Bottom contributions plus 7 molC m −2 , i.e., the TTD and OCCAM mean specific inventory for the upper layer. *
• S calculated with data collected in 2002.Five different methods are compared and discussed: three carbon-based methods ( C*, IPSL and TrOCA), TTD and a simulation from the OCCAM global model.Comparatively, the C* method seems to yield too shallow penetration of C ANT , with no or very low C ANT detected in deep or bottom waters, mainly due to the formulation and assumptions of C dis T .Our results indicate that SAB99 C dis TableA2shows the pH ( pH) correction applied to each batch and the mean value of the CRM determinations after applying the former correction.

Table A1 .
Location and physical characteristics of the samples analyzed for pH replicates.The mean, standard deviation (STD) pH 25T and number of samples drawn from each bottle is shown.Note that the pH values contain the R and 0.0047 corrections.

Table A2 .
Alkalinity analysis supplementary information for each batch of analysis: N HCl is the normality referred to 20 • C of the hydrochloric solution used; pH is the pH correction applied to refer the TA determinations on the CRM to the corresponding nominal value (batch 55 with a certified TA of 2227.85±0.54µmol kg −1 ).The mean value of the TA measurements on the CRM samples is also shown (Fitted TA±standard deviation (number of analysis)).The average of the difference (Av.Dif.and number of duplicates) in the duplicate's analyses is shown.

Table A3 .
NADW characteristics (1.8≤θ ≤2.3 • C and salinity≥34.81)from the 1995 I5W and 2002 CD139 cruises.Mean plus/minus standard deviation and number of samples (n) considered for the physical and chemical variables.

Table A4 .
Mean and standard deviation (STD)difference between a GLODAP cruise and the 2002 CD139 CFC, C T and TA data for each crossover (Xover number in Fig.A5).Samples are taken below 1500 dbar.The minimum number of stations for each cruise in each crossover is three.Except for the I5 cruise in 1987, the others were done in 1995.Xover CruiseCFC-11 pmol kg −1 CFC-12 pmol kg −1 C T µmol kg −1 TA µmol kg −1