Seasonal variability of the carbon cycle in subantarctic surface water in the South West Pacific


  • Holger Brix,

    Corresponding author
    1. Department of Atmospheric and Oceanic Sciences, University of California, Los Angeles, Los Angeles, California, USA
    • Corresponding author: H. Brix, Department of Atmospheric and Oceanic Sciences, 5845 Slichter Hall, University of California, Los Angeles, Los Angeles, CA 90095-1567, USA. (

    Search for more papers by this author
  • Kim I. Currie,

    1. National Institute of Water and Atmospheric Research, Dunedin and Chemistry Department, University of Otago, Dunedin, New Zealand
    Search for more papers by this author
  • Sara E. Mikaloff Fletcher

    1. National Institute of Water and Atmospheric Research, Klibirnie, Wellington, New Zealand
    Search for more papers by this author


[1] Few Southern Hemisphere time series measurements of biogeochemical tracers are available, and this scarcity is a major impediment in understanding the biological and physical processes underlying the oceanic carbon and nutrient cycles in vast parts of the global oceans. We make use of bi-monthly measurements of carbonate parameters from 1998 to 2010 in upper Subantarctic Surface Water east of New Zealand's South Island at 45.85°S 171.50°E to investigate seasonal cycles and trends in these species and processes controlling their variability. This time series reveals positive trends in salinity normalized dissolved inorganic carbon (sDIC) and the partial pressure of carbon dioxide (pCO2) that are smaller than would be expected from the anthropogenic increase in atmospheric pCO2 alone, possibly due to a decrease of the average temperature over the observational period. The seasonal cycle of pCO2 is dominated by that of DIC, but is substantially modified by the influence of the annual cycle of sea surface temperature. Investigations with a δ13COC-constrained diagnostic box model suggest that net community production (NCP) is the dominant process controlling the observed seasonal variability in sDIC by removing 1.2 ± 0.7 mol C m−2 yr−1 from the mixed layer. This carbon drawdown, aided by an additional carbon removal due to horizontal transport, is balanced by vertical diffusion, entrainment, and air-sea gas exchange of CO2. Oceanic pCO2 is below atmospheric pCO2 for nearly the entire year, leading to an annual mean surface ocean pCO2 undersaturation of about 12 µatm and an annual oceanic uptake of CO2 from the atmosphere of 0.9 ± 0.1 mol C m−2 yr−1.

1 Introduction

[2] The exchange of both natural and anthropogenic carbon between the atmosphere, ocean, and terrestrial biosphere is not constant but varies spatially, seasonally, and from year to year due to variations in uptake rates by the natural carbon reservoirs and the variability of emission sources. Natural land and ocean sinks have removed, on average, carbon from the atmosphere in roughly equal proportion over the last 50 years [Le Quéré et al., 2009; Friedlingstein et al., 2010]. Historically, much of our understanding of trends and variability in the oceanic carbon sink has been based on model simulations, in part due to a limited number of observations and the difficulty in interpreting them. More recently, large-scale efforts to measure carbon-related tracers in the world's oceans have yielded fresh insight into the trends and interannual variability of ocean carbon biogeochemistry. In particular, ocean time series measurements have yielded detailed information about the biological and physical processes controlling the seasonal cycles, interannual variability, and trends in ocean carbon uptake.

[3] Long time series of upper ocean biogeochemical observations so far have been limited to the Northern Hemisphere. Measurements in the North Pacific and Sargasso Sea [e.g., Karl and Lukas, 1996; Michaels and Knap, 1996; Steinberg et al., 2001; Lomas et al., 2002; Bates et al., 2012] have helped to address major gaps in our understanding of the drivers of regional trends and carbon cycle variability. One-dimensional models have proven valuable in understanding processes governing seasonality, variability and trends in ocean biogeochemical tracers at these sites [Gruber et al., 1998; Keeling et al., 2004; Brix et al., 2004]. For instance, a one-dimensional modeling study showed that the interannual variability of the oceanic carbon cycle is primarily driven by an interplay between sea surface temperature, air-sea gas exchange, net community production, and lateral water transport at Station ALOHA, which is in the equatorial Pacific Ocean near Hawaii [Brix et al., 2004]. At the Bermuda Atlantic Time Series (BATS) site in the Sargasso Sea in the western subtropical North Atlantic, one-dimensional modeling showed that the variability in the physical oceanography of the area is the major contributor to the carbon chemistry variability. This is because the entrainment of deeper nutrient and carbon-rich waters in winter varies in strength and is possibly related to changes in the North Atlantic Oscillation. Seasonal variability of the surface CO2 concentration at the European Station for Time Series in the Ocean (ESTOC) in the northeast Atlantic subtropical gyre is dominated by biological drawdown of carbon in the period March to October and by mixing processes from October to the end of (boreal) autumn. The de-seasonalized CO2 fugacity (fCO2) shows a long-term increase [Gonzalez-Davila et al., 2003].

[4] The Southern Ocean is of particular interest to the scientific community, as it accounts for nearly a quarter of the anthropogenic carbon uptake by the global oceans [Mikaloff Fletcher et al., 2006; Gerber and Joos, 2010; Gruber et al., 2009]. Trends in pCO2 and dissolved inorganic carbon (DIC) south of Tasmania have been established combining data from multiple cruises [Yoshikawa-Inoue and Ishii, 2005; Brévière et al., 2006; Borges et al., 2008; Midorikawa et al., 2012]. Although these studies have provided valuable insight into the Southern Ocean carbon cycle, interpreting observations from periodic cruises is complicated by seasonal sampling biases and large interannual variability [e.g. Rodgers et al., 2008]. Furthermore, relatively little has been done using data from other sectors of the Southern Ocean. Understanding the relative influence of biology and physics on carbon cycle dynamics requires resolving seasonal cycles of these quantities.

[5] In 1998, a joint program between New Zealand's National Institute of Water and Atmospheric Research (NIWA) and the University of Otago began measuring tracers associated with carbon uptake and ocean biogeochemistry along a transect from Dunedin into subantarctic waters every 2 months. This data set, known as the Munida time series, reveals a trend toward gradually decreasing CO2 concentrations (expressed as pCO2) in Subantarctic Surface Water (SASW) between 1998 and 2005, followed by a period of increasing pCO2 between 2005 and the present [Currie et al., 2009]. The Munida time series indicates a change from the ocean as a source of CO2 in 1998/1999 to a sink in 2004/2005. The seasonal amplitude of pCO2 in SASW during the studied time span was about 9 µatm.

[6] Here, we will expand the analysis of Currie et al. [2009] and focus on interpretation of trends and the seasonal cycles of a variety of properties in the Munida time series, especially with regard to identifying the respective roles of physical and biological processes on both time scales.

2 Observations

[7] The Munida time series transect is located in the South West Pacific Ocean, off the south-east coast of New Zealand. The 60 km long surface transect extends from the coast (45.77°S 170.72°E; 0 km) to (45.85°S 171.50°E) crossing the subtropical front (which is locked into place here due to topographic constraints) into Subantarctic Surface Water [Currie et al., 2009, see Figure 1]. pCO2 and associated parameters have been measured in the surface waters every 2 months since January 1998. A cast to 500 m in SASW was added to the sampling program in May 2001, however depth data are not available from all cruises.

[8] The physical oceanography of the area is such that the transect traverses three water types (Figure 1). Neritic water, typically located between 0 and 20 km along the transect, is influenced by land, riverine and coastal effects, including the Otago Harbour, the Clutha and Taieri Rivers, and the anticyclonic eddy of Blueskin Bay. Modified subtropical water [Jillett, 1969; Heath, 1985; Murdoch et al., 1990] occupies the mid-portion of the transect, generally between 20 and 45 km offshore. This water is generally of higher temperature and salinity than the adjacent SASW. The subtropical front is a circum-global front between subtropical water to the north and Subantarctic Surface Water to the south. Off the south-east coast of New Zealand, the subtropical front is generally locked to the 200 m bathymetry, thus forming a sharp and distinct boundary feature, at approximately 45 km along the transect (distance from coast) [Shaw et al., 1999; Hopkins et al., 2010]. North of the study site, the front turns eastward along the Chatham Rise. SASW is located at the seaward end of the transect at distances greater than 45–50 km offshore. Satellite sea surface images [e.g., Currie et al., 2009] clearly show the subtropical front between the subtropical waters and the cooler Subantarctic Surface Water, as this front is characterized by a large temperature gradient. SASW is in general cooler, less saline [Sutton, 2003; Currie et al., 2009], and of higher nutrient concentration [Hawke, 1989] than the water of subtropical origin. Our investigation focuses on the properties of this water mass. We will refer to the measurements taken at this end location of the transect as Sta. Munida.

Figure 1.

Water masses, currents and frontal features in the New Zealand region (modified from Heath [1985]). The box shows the location of the study region, enlarged in the inlay: Sta. Munida is marked with an X.

2.1 Data Sets

[9] Between January 1998 and December 2010, 71 cruises were conducted at approximately two-monthly intervals as part of an ongoing program. For each cruise, the data obtained from SASW (>50 km offshore) were averaged to give one value per parameter. Surface samples were typically taken at approximately 5 m depth with some ranging from 2 to 10 m. The deep casts include samples taken at depths of 0–10 m, 50 m, 100 m, 200 m, and 500 m. A detailed description of the measurements and analysis methods is given in the Supporting Information.

[10] The biogeochemical quantities used in this study include pCO2, alkalinity (Alk), and nutrients. A limited number of samples were analyzed for δ13C in dissolved inorganic carbon (DIC). The concentration of DIC was calculated from the measured pCO2 and alkalinity using the Mehrbach equilibrium constants [Mehrbach et al., 1973] as refitted by Dickson and Millero [1987]. In addition to standard surface measurements of temperature and salinity, mixed layer depths were estimated from CTD temperatures and salinities for the 26 cruises after 2001 for which depth casts for both quantities were available by applying the variable σΘ criterion of Sprintall and Tomczak [1992] with a temperature difference between the surface and the base of the mixed layer of 0.5 °C. To remove the influence of fresh water fluxes, the salinity normalized values for DIC (sDIC) and alkalinity (sAlk) were calculated using the approximate long-term mean salinity at Sta. Munida (34.3). Time series plots of the sea surface temperature, salinity, and mixed layer depth are given in Figure 2.

Figure 2.

Time series of (a) temperature, (b) mixed layer depth (MLD), and (c) salinity. Temperature and salinities have been measured at about 5 m depth, MLD has been calculated using the variable σΘ criterion of Sprintall and Tomczak [1992] for the 26 occasions when CTD values for temperature and salinity were available; for details see text and Supporting Information.

[11] Atmospheric partial pressure of CO2 (pCOinline image) was calculated from the mole fraction of CO2 in dry air (xCO2) as measured at Baring Head (located at the southern end of New Zealand's North Island, approximately 800 km NNW from Sta. Munida, the closest atmospheric carbon dioxide monitoring station) [Brailsford et al., 2012]. Atmospheric values of δ13C were measured by the same laboratory (data provided by Anthony Gomez and Gordon Brailsford, NIWA, Wellington) [Ferretti et al., 2000]. We used daily averages of wind speeds measured at Taiaroa Head at the landward end of the transect (observations made by NIWA until October 2002, then by Port Otago Ltd. (Alan Sutherland, personal communication)) Other atmospheric and oceanic data sets used in the modeling part of this study will be described in section 3.1 and are listed in Table S2 in the Supporting Information.

2.2 Trends

[12] The concentrations of sDIC, pCOinline image, and pCOinline image all show a clear, statistically significant, long-term trend (Figure 3 and Table 1), modulated by interannual variability, especially in the oceanic quantities. The average increase of sDIC from 1998 to 2009 is 1.39 ± 0.39 µmol kg−1 yr−1, oceanic pCO2 increases by 1.1 ± 0.4 µatm yr−1, and atmospheric pCO2 by 1.9 ± 0.0 µatm yr−1 (see Table 1). Unfortunately, our δ13C time series is too short to allow us to use the model simulations to interpret the trends. The upward trend of pCO2 in oceanic surface waters is not unexpected, as it possibly reflects the anthropogenic signal of increasing atmospheric pCO2 concentrations. However, the rate of atmospheric increase exceeds that of the oceanic values. The mean difference ∆pCO2 = pCOinline imagepCOinline image is 12 µatm and increases on average by about 0.9 µatm yr−1 (compare Table 1). Wind speed decreases by 0.06 m s−1 yr−1, thus weakening the effect of an increasing ∆pCO2 on the air-sea carbon flux, which shows a weak (albeit not significant) increasing trend.

Figure 3.

Time series of (a) dissolved inorganic carbon normalized to a salinity of 34.3 (sDIC), (b) salinity normalized alkalinity (sAlk), (c) δ13C (scale is inverted), and (d) atmospheric (solid line) and oceanic pCO2 (circles) at Sta. Munida from 1998 to 2009. The straight lines in Figures 3a and 3b and the dashed straight line in Figure 3d represent (de-seasonalized) linear trends that have been fitted to the time series of sDIC, sAlk, and oceanic pCO2, respectively.

Table 1. Long-Term Data Trends for Sta. Munidaa
  1. a

    Uncertainties (±) represent one standard deviation. The p-values have been computed using Student's t-test; bold face highlights statistical significance >95%. Values are from 1998–2009 unless otherwise stated. The trends were calculated from de-seasonalized data (obtained by subtracting harmonic fits).

  2. b

    No trend is given for δ13Coc as the time series is too short for meaningful values.

  3. c

    δ13Catm and pCOinline image data from Baring Head [Brailsford et al., 2012].

  4. d

    pCO2 = pCOinline image − pCOinline image, as determined for all instances of oceanic measurements.

  5. e

    Time series for mixed layer depth from 2000–2009.

sDICµmol kg−12086.81.39 ± 0.390.001
DICµmol kg−12094.60.43 ± 1.170.718
sAlkµmol kg−12285.30.45 ± 0.280.111
Alkµmol kg−12285.30.14 ± 0.230.543
δ13Coc b1.522± 
δ13Catm c–8.104–0.027 ± 0.0020.000
pCOinline imageµatm358.21.1 ± 0.40.004
pCOinline image cµatm367.71.9 ± 0.00.000
pCO2dµatm12.00.9 ± 0.40.021
T°C10.32–0.06 ± 0.020.007
Spract. scale34.306–0.004 ± 0.0030.175
hem54.79–1.68 ± 2.120.453
Wind speedm s−17.06–0.06 ± 0.020.000

[13] The partial pressure of CO2 in seawater is determined by physical and biological processes and can be described by a function of DIC, alkalinity, temperature, and salinity. The temperature sensitivity of pCO2 is given by Sarmiento and Gruber [2006] to be 13 µatm per °C. For salinity, it is approximately 9 µatm per salinity unit change, both given constant DIC and alkalinity. The observed temperature decrease of −0.06 ± 0.02 °C yr−1 could therefore account for approximately −0.8 ± 0.3 µatm yr−1 of the pCO2 change. The effect of salinity changes for the observed salinity trends would be negligible. The observed increases in DIC and Alk have opposite effects on pCO2. Increases in DIC elevate pCO2, while Alk increases lower pCO2. It can therefore be speculated that the weaker increase in oceanic pCO2 (compared to the atmospheric pCO2 increase) could be due to the temperature decrease while DIC and Alk changes tend to cancel each other out.

[14] Freshwater fluxes do have an influence on DIC and Alk, which is clear when the trends of these species are compared with salinity normalized values (not shown). For both sAlk and sDIC, the trends are about threefold the non-salinity normalized values. Following the analysis of Keeling et al. [2004], we decompose the total change in DIC over a given period of time (∆DIC) as the sum of changes of sea surface temperature (∆T), salinity (∆S), salinity normalized alkalinity (∆sAlk), atmospheric pCO2 (inline image), and the air-sea pCO2 difference (inline image), using S0 = 34.3 to normalize salinity.

display math(1)

[15] This formulation accounts separately for salinity-induced changes in the dissociation constants in the second term on the right-hand side of the equation and for changes due to the freshwater effect (third term). The partial derivatives in (1) are computed using the approximation that the change in each parameter is small compared to its long-term mean value. To determine ∆ values, the trends reported in Table 1 were multiplied by the respective time periods. For details of the derivation of this equation, see Appendix B of Keeling et al. [2004].

[16] The observed trend in DIC has a large error margin (0.43 ± 1.17 µmol kg−1 yr−1, Table 1). Within these large margins, the calculated DIC change of 1.04 ± 0.75 µmol kg−1 yr−1 (Table 2) is consistent. This result is not statistically significant, but the decomposition of this possible trend can allow us to gain insight into the factors that may influence the carbon budget at Sta. Munida. As might have been expected, increases in DIC are dominated by rising atmospheric CO2 (compare Table 2). Decreasing temperatures and increasing sAlk add to the positive DIC changes. The changes in salinity at Sta. Munida (through the dissociation constants and dilution effects) exert a minor effect. The increasing air-sea difference in pCO2 works against the observed DIC increase. This term reflects the fact that DIC is increasing more slowly than would have been the case if it were in equilibrium with T, S, Alk, and atmospheric pCO2. Besides the interpretation of a temperature dominance of this change as given above, the possibility of a change in the chemical and biological state of the upper ocean carbon system cannot be ruled out, which could be interpreted as a change in the characteristics of the dominant water mass at Sta. Munida.

Table 2. Contribution of Different Variables to the Observed Long-Term Trend of DIC at Sta. Munidaa
  1. a

    Values are in µmol kg−1 yr−1. Uncertainties have been calculated using 2σ errors from Table 1, applying quadratic error propagation.

  2. b

    The symbols represent the entire term, including the partial derivatives (see equation (1)).

  3. c

    The ∆∆pCO2 = inline image term is multiplied with −1 to reflect its contribution to the DIC budget correctly.

  4. d

    Σ is the sum of all five terms, ∆DICobs is the observed trend in DIC.

  5. e

    The partial derivatives were computed around the long-term mean values of the different parameters, while keeping all other parameters constant. The following values were obtained: ∂ DIC/∂ T = −7.42 µmol kg−1 (°C)−1; inline image = 0.48 µmol kg−1 μatm− 1; ∂ DIC/∂ S = −4.46 µmol kg−1; (sAlk/S0)(∂ DIC/∂ Alk) = 57.6 µmol kg−1; and ∂ DIC/∂ Alk = 0.87 µmol kg−1 (µmol kg−1)−1.

Individual terms
T+0.45 ± 0.30
S+0.02 ± 0.03
S (dilution)–0.23 ± 0.35
sAlk+0.39 ± 0.49
pCOinline image+0.92 ± 0.01
–∆∆pCO2c–0.51 ± 0.34
Sum of terms versus observed
Σd+1.04 ± 0.75
ΔDICobse+0.43 ± 1.17

2.3 Seasonal Cycles

[17] In order to explore the mechanisms controlling the mean seasonal cycle of DIC and pCOinline image in our time series (compare Figure 4), we removed the linear trend from each of our species and aggregated the resulting data into one composite year.

Figure 4.

Composite annual cycles of (a) dissolved inorganic carbon normalized to a salinity of 34.3 (sDIC), (b) salinity normalized alkalinity (sAlk), (c) δ13C (scale is inverted), and (d) atmospheric (dotted line) and oceanic pCO2 (circles and solid line) at Sta. Munida. Note that atmospheric pCO has a seasonal cycle with a minimum in February, a maximum in September and an amplitude of about 2 µatm. The (e) inlay shows the contributions of temperature and DIC changes to the mean seasonal cycle of pCO2. The dotted curve represents the temperature contribution (pCO2 at constant DIC), while the dot-dashed curve represents the DIC contribution (pCO2 at constant T). The solid curve is the observed mean seasonal cycle of pCO2. For details see text.

[18] The average seasonal cycle of sDIC (Figure 4a) has its maximum in the austral winter (July and August), declines in spring, and reaches its minimum in December and January, with a relatively sharp drop in most years, followed by a slow and steady increase in values between February and July. Using a third-order harmonic fit, we find an amplitude of approximately 40 µmol kg−1 between summer and winter values. Salinity normalized alkalinity (Figure 4b) does not show much of a seasonal cycle. For δ13C (Figure 4c), we find slightly lower values in the second half of the year.

[19] The minimum of the harmonic fit for oceanic pCO2 (Figure 4d) occurs in February, and it peaks in September, with an amplitude of approximately 17 µatm. The amplitude of the atmospheric pCO2 seasonal cycle is on average about 2 µatm and therefore plays only a minor role in the seasonal cycle of ∆pCO2. During most of the year, the oceanic CO2 is undersaturated relative to the atmosphere with an average ∆pCO2 of about 12 µatm. We find single events of oversaturation between May and January. Between August and October, undersaturation and oversaturation cancel each other out on average. This observation has a bias as the difference in trends between pCOinline image and pCOinline image leads to the occurrence of most of these over-saturation events in the first few years of our time series.

[20] Again following the methodology used by Keeling et al. [2004], we analyze the contribution of temperature and DIC effects on the seasonal cycle of pCO2 by splitting the observed pCO2 into an isochemical term and an isothermal term [Keeling, 1993]:

display math(2)

with the over-bars denoting annual means. The isochemical changes of the mean pCO2 (i.e., changes at constant DIC caused by temperature variations) are described by the second right-hand term, using the experimentally determined temperature sensitivity factor after Takahashi et al. [1993], ∂ lnpCO2/∂ T = 0.0423°C− 1. The third term describes isothermal pCO2 (i.e., constant temperature, with changes being caused by DIC variations. either due to biological or physical processes), which has been determined as the remainder of the other terms.

[21] The isothermal pCO2 is out of phase with the isochemical component, showing a maximum in August, a minimum in February, and an amplitude of about 65 µatm. In contrast, the isochemical component peaks in February, has its minimum in late July/early August, and has an amplitude of approximately 50 µatm (inlay of Figure 4e). Thus, the annual cycle of pCO2 is dominated by the summer through winter increase of DIC (including the influence of biological processes), but substantially modulated by the gradual warming in the same period. Metzl et al. [1999] and Metzl [2009] report similar opposing controls in the subantarctic zone (40°–42°S) of the Indian Ocean, and Midorikawa et al. [2012], although only using summer data, argue that counteracting temperature and biological effects could reduce interannual variations in pH (and pCO2) in the Pacific sector of the Southern Ocean.

[22] Mixed layer depths (MLDs) are shallow in summer (less than 50 m between November and March, Figure 2b) and gradually deepen toward a maximum in late autumn and early winter. In spring, the picture is somewhat mixed with MLDs between 10 and 80 m. This annual cycle is consistent with observations of mixed layer depths of subantarctic water at other sites such as the Chatham Rise [Nodder et al., 2005] and south of Tasmania [Rintoul and Trull, 2001; Dong et al., 2008].

[23] The mean seasonal cycles of sDIC and δ13Coc are out of phase, suggesting that net community production (NCP) might play an important role in drawing down carbon in spring and summer. This reduction of sDIC is likely counteracted by uptake of CO2 from the atmosphere as the ocean at Sta. Munida is in an almost constant state of undersaturation. Other processes, like mixing and vertical and horizontal transport, could also have a substantial impact on the upper ocean carbon budget. A quantification of these processes from observations is a difficult task due to limitations arising from the availability of data and missing parameterizations that could be used to determine, say, an NCP budget.

[24] In the following section, we make use of the availability of simultaneous data sets of δ13Coc and other chemical and physical measurements to quantitatively analyze the components of the carbon budget at Sta. Munida with a diagnostic box model.

3 Modeling

3.1 Model Description

[25] The 13C-based model is a modified version of the diagnostic box model described by Gruber et al. [1998]. The model consists of two oceanic boxes representing the surface mixed layer and the underlying waters of the thermocline as well as an atmospheric box without a prescribed capacity, which provides boundary conditions for the exchange of CO2 and 13CO2 across the air-sea interface. The depth of the boundary between the two oceanic boxes is variable: it moves up and down, as prescribed by observations. Horizontal transport processes depend on prescribed lateral gradients and inferred magnitude and direction of water flow. Tracer concentrations throughout the atmospheric and mixed layer boxes are uniform, but vary linearly with depth in the thermocline box.

[26] The model represents five processes: (1) air-sea gas exchange at the sea surface, (2) vertical diffusion across the boundary with the thermocline, (3) entrainment of DIC from the thermocline whenever the mixed layer deepens, (4) horizontal transport, and (5) net community production. Additional information on the detailed configuration of this model is given in the Supporting Information.

[27] The major source of uncertainty in these model simulations is the relative sparseness of δ13C measurements and the lack of sufficient data to establish horizontal gradients of sDIC and δ13C reliably. Only one relevant cruise-track, World Ocean Circulation Experiment (WOCE) cruise P15S, north-south along approximately 170°W, was available to determine dsDIC/dx and dδ13C/dx, thus restricting our ability to determine an appropriate error margin. Comparing our input data with regional model output [NZ-Regional Ocean Modelling System (ROMS) model using biogeochemistry from the Pelagic Interaction Scheme for Carbon and Ecosystem Studies (PISCES) (Paulo Calil and Mark Hadfield, personal communication)], we found the values that we used to be within the range produced by the model. To account for uncertainty in our input variables, we performed Monte Carlo-type experiments. For the calculation of seasonal and annual means, we choose to assign an additional 50% uncertainty to the horizontal sDIC and δ13C gradients to account for the lack of observational data.

3.2 Model Results and Discussion

3.2.1 Carbon Source and Sink Terms

[28] Figure 5 shows results from our model calculations for the source terms of air-sea gas exchange, diffusion, entrainment, net community production, and horizontal transport. The thin lines represent one standard deviation derived from the Monte Carlo experiments. For the purpose of the following interpretation, we have chosen seasons that slightly deviate from the standard definition. They are defined in Tables 3 and 4 and are marked with black triangles in Figure 5.

Figure 5.

Annual cycles of the sDIC source terms due to (a) air-sea gas exchange, diffusion, and entrainment, and (b) horizontal transport, net community production, and the sum of all five, which has been constrained to be identical to the observed annual cycle. Positive values represent fluxes into the mixed layer. The thin lines denote the uncertainty intervals as determined from Monte Carlo simulations. The triangles on the time axis denote the seasons as defined in Tables 3 and 4.

Table 3. Temporally Integrated Source Terms, inline imagea
  1. a

    As calculated by the diagnostic box-model. Units are µmol kg−1. For uncertainty calculations see text.

  2. b

    Calendar day 1 to 90.

  3. c

    Calendar day 91 to 213.

  4. d

    Calendar day 214 to 288.

  5. e

    Calendar day 289 to 365.

Inferred from model
inline imageex11.6 ± 1.33.1 ± 0.41.7 ± 0.34.7 ± 0.520.8 ± 2.4
inline imagencp–33.4 ± 11.123.0 ± 1.8–9.8 ± 4.3–34.4 ± 6.6–53.8 ± 17.8
inline imagediff17.8 ± 9.50.9 ± 0.51.8 ± 1.09.4 ± 5.029.5 ± 15.8
inline imageent4.3 ± 1.42.5 ± 0.81.7 ± 0.50.0 ± 0.08.5 ± 2.7
inline imagehtr3.6 ± 4.51.3 ± 1.6–5.9 ± 3.4–4.0 ± 3.3–4.8 ± 6.3
Total inferred from model
inline image3.930.9–10.5–24.30.2
Table 4. Integrated Fluxes, inline imagea
  1. a

    As calculated by the diagnostic box-model. Units are mol  m−2. For uncertainty calculations see text.

  2. b

    Calendar day 1 to 90.

  3. c

    Calendar day 91 to 213.

  4. d

    Calendar day 214 to 288.

  5. e

    Calendar day 289 to 365.

  6. f

    This term is balanced over the annual cycle by a covariance term between DIC and h, i.e., inline image.

Inferred from model
inline imageex0.28 ± 0.030.25 ± 0.030.10 ± 0.020.22 ± 0.030.85 ± 0.10
inline imagencp–0.68 ± 0.252.60 ± 0.21–0.84 ± 0.27–2.26 ± 0.38–1.17 ± 0.71
inline imagediff0.37 ± 0.200.07 ± 0.040.13 ± 0.070.44 ± 0.241.01 ± 0.54
inline imageent0.12 ± 0.040.20 ± 0.060.13 ± 0.040.00 ± 0.000.45 ± 0.14
inline imagehtr0.11 ± 0.110.22 ± 0.21–0.35 ± 0.21–0.34 ± 0.24–0.35 ± 0.28
Inferred from model
inline image0.203.34–0.83–1.940.78f

[29] The entrainment source term (Jent, Figure 5a) is zero from June into August, when the mixed layer decreases in size, and from October to January, when it remains stationary. It reaches its maximum at the end of January, with 0.12 µmol kg−1 d−1 after the first onset of mixed layer deepening when the vertical temperature gradient at the base of the mixed layer and, by inference, the DIC gradient is large (not shown). The seasonal cycle of entrainment has two intermediate maxima of 0.07 µmol kg−1 d−1 in April and 0.04 µmol kg−1 d−1 in September. We estimate the uncertainties in Jent to be up to 0.04 µmol kg−1 d−1 at the time of the entrainment maximum in January.

[30] Vertical diffusion (Jdiff, Figure 5a) has very small values in autumn and winter. It peaks at the transition of spring and summer with values of up to 0.42 ± .23 µmol kg−1 d−1. Its value is determined using the vertical diffusivity coefficient at the base of the mixed layer, which is estimated from the vertical density gradient just below the mixed layer. The maximum in diffusion at the turn of the year can therefore be attributed to the strong vertical temperature and density gradient during this time of the year. Its substantial uncertainty is due to the high variability of this density gradient caused by short-term events like passing storms or eddies.

[31] In winter (August to October), atmospheric and oceanic pCO2 have almost identical values (Figure 4d). The air-sea gas exchange source term (Jex, Figure 5a) is therefore near zero. In late October, the ocean becomes undersaturated (with regard to CO2) and Jex begins to rise slowly. It peaks reaching 0.21 µmol kg−1 d−1 at the same time as Jdiff, that is, during the transition to summer at the beginning of the year. The continued growth of ΔpCO2 through February and March is partially compensated for by the onset of the deepening of the mixed layer. The mixed layer keeps thickening gradually, which leads, in concert with declining ΔpCO2, to the drop of Jex until autumn. The terms controlling air-sea gas exchange are well known, which makes Jex the source term with the smallest uncertainty (up to 0.02 µmol kg−1 d−1).

[32] The calculated horizontal transport source term (Jhtr, Figure 5b) is a source for sDIC from January to June with a maximum of 0.07 µmol kg−1 d−1 in May. For the remainder of the year, horizontal transport is a sink of up to −0.11 µmol kg−1 d−1 in early August. The calculated uncertainty ranges up to 0.05 µmol kg−1 d−1. As mentioned before, the horizontal transport term in our calculation is dependent on the horizontal sDIC gradient (dsDIC/dx), which was constrained by the one available measurement (WOCE cruise P15S) and is consistent with a regional model simulation (NZ-ROMS, PISCES (Paulo Calil and Mark Hadfield, personal communication)). Sutton [2003] estimates that the Southland Current comprising 90% subantarctic water flows northward at a rate greater than 0.2 m s−1 in the vicinity of Sta. Munida. In addition, the varying strength of the frontal zone [Shaw and Vennell, 2001; Hopkins et al., 2010] affects water mass mixing along the transect and therefore horizontal mixing of nutrients, trace metals, and carbon. Our results are consistent with these observations.

[33] The calculated biological source term, net community production (Jncp, Figure 5b), varies substantially over the course of the year, being negative from August to March with an extreme value of −0.80 µmol kg−1 d−1 at the turn of the year. This maximum in net community production happens at the onset of summer. This finding is consistent with sediment trap data that indicate heightened organic and biogenic silica fluxes in subantarctic water south of the Chatham Rise in spring and summer [Nodder and Northcote, 2001], indicating that this region and Sta. Munida are located in a high-nutrient, low-chlorophyll (HNLC) region, where macro nutrients are abundant, but phytoplankton growth is iron, light, and silicate co-limited [Hutchins et al., 2001; Boyd et al., 1999]. In the same region, south of Chatham Rise, chlorophyll measurements show an increase from winter to spring [Bradford-Grieve et al., 1997] and pigment data put the maximum of production in spring and summer [Nodder and Gall, 1998]. Remotely sensed ocean color data indicate a low-magnitude annual cycle of chlorophyll abundance peaking in early autumn [Murphy et al., 2001].

[34] In autumn, defined here as April to July, net community production in our diagnostic model is a source term for sDIC with a maximum value of approximately 0.32 µmol kg−1 d−1 in late May. Positive Jncp values represent respiration dominating over biological production, probably driven by a combination of the deepening of the mixed layer and iron-limited biological productivity. The erratic signal in Jncp in January is an artifact of our calculation method. The abrupt onset of entrainment needs to be compensated by the model to maintain the sum of all source terms being equal to observations. The maximum uncertainty of our calculation for net community production lies at 0.21 µmol kg−1 d−1. Despite this relatively strong uncertainty in the signal, the model results clearly show that the seasonal variations in sDIC at Sta. Munida are dominated by the biological term, while all other terms only serve to modulate the results.

3.2.2 Carbon Budget

[35] Despite the apparent dominance of the NCP source term for the carbon cycle at Sta. Munida, it is worth taking a closer look at the seasonally and annually integrated source terms (inline image) and their uncertainties as listed in Table 3.

[36] The annual integral of the source terms shows a balance between a dominating NCP sink with −53.8 ± 17.8 µmol kg−1 and two source terms: diffusion (29.5 ± 15.8 µmol kg−1) and air-sea gas exchange (20.8 ± 2.4 µmol kg−1); entrainment and horizontal transport are small and contribute less than 10% each (Table 3). The depth integration of these source terms over the mixed layer (Fi = ρ0hJi, with ρ0 the average density of sea water) shows a similar picture (Table 4): air-sea gas exchange, diffusion, and entrainment add carbon to the mixed layer (0.85 ± 0.10, 1.01 ± 0.54, and 0.45 ± 0.14 mol m−2, respectively), while NCP and horizontal transport remove it over the course of the year (−1.17 ± 0.71 and −0.35 ± 0.28 mol m−2). The flux budget can only be closed by taking into account an imbalance due to the covariance between the observed temporal change of sDIC with time and the mixed layer depth (for details, see Keeling et al. [2004]). For Sta. Munida this imbalance is 0.78 mol m−2.

[37] We return to the question of the driver(s) of the seasonal cycle of sDIC raised in section 2.3. Aided by the carbon budget calculations from the diagnostic box model, we can now evaluate the importance of the processes controlling the carbon budget in more detail.

[38] The average winter to spring drawdown of the sDIC content of the mixed layer (defined here as August to December) of 34.8 µmol kg−1 is primarily caused by NCP, which removes 44.2 µmol kg−1. In addition, horizontal transport processes remove 9.9 µmol kg−1. This loss is partially compensated by vertical diffusion and uptake of atmospheric carbon (11.2 µmol kg−1 and 6.4 µmol kg−1, respectively). Entrainment plays only a minor role (see Table 3). NCP and the sum of the other terms almost completely cancel each other out over the course of summer (January to March). In autumn, all source terms are positive (with NCP with 23.0 µmol kg−1 being the strongest carbon source) and act together to replenish carbon to the mixed layer at Sta. Munida.

3.2.3 Carbon Uptake From the Atmosphere

[39] Over the course of the year the average annual uptake of carbon from the atmosphere is calculated to be 0.9 ± 0.1 mol C m−2 yr−1. The seasonal variations in this value are minor, except that during winter the uptake drops to 0.1 ± 0.0 mol C m−2 (compare inline image in Table 4). This seasonal variability is of course modulated over the 11 year time series by the increasing difference between atmospheric and oceanic pCO2 (compare section 2.2 and Table 1).

[40] The continuous uptake of carbon dioxide during the course of the year is consistent with published values derived from both observations and models. McNeil et al. [2007] estimate a Southern Ocean (south of 50°S) sink of 0.4 ± 0.25 PgC yr−1 and a subantarctic sink of 1.1 ± 0.6 PgC yr−1 (40°–50°S) from observations. The Takahashi et al. [2009] air-sea flux climatology estimates the flux in the Southern Ocean (south of 50°S) to be −0.04 PgC yr−1. The 4° × 5° grid box in the Takahashi et al. [2009] climatology that contains Sta. Munida has an annual mean ocean uptake of 1.8 mol C m−2 yr−1, referenced to the year 2000 (compared to our average 0.9 ± 0.1 mol C m−2 yr−1 for the 1998–2010 period). Due to the coarse resolution of this climatology and the sparse data that underlie it in the Southern Ocean, it is not capable of capturing features like the course of the subtropical front off the coast of New Zealand's South Island. The climatology's adjacent grid boxes in meridional direction show lower values (down to 1.2 mol C m−2 yr−1 at 52°S, situated in Subantarctic Surface Waters). An estimate of the contemporary air-sea flux from ocean interior data and models (nominal for the year 1995) [Gruber et al., 2009] and an estimate from a biogeochemical forward model calculation (Assmann et al. [2010] for the year 2000) both show higher values than Takahashi et al. [2009] and our study for the particular 4 × 5° grid box containing Sta. Munida.

[41] To explain the uptake of pCO2, we consider the contributions of heat and fresh water fluxes, changes in DIC and Alk due to biological activity, and the influence of horizontal advection, that is, the large-scale ocean circulation. Air-sea buoyancy fluxes for the years 2005–2007 as computed by model reanalyses and state estimates [Cerovečki et al., 2011] show basically no influence of freshwater fluxes in our study region. The spatial structure of their heat flux maps follows the course of the subtropical front and shows on average moderate heat fluxes into the ocean. This net warming of the surface ocean leads to thermally induced outgassing of CO2. Our model results and observations suggest that there is a continuous transport of water from the south-west toward Sta. Munida, that is, from a region that is rather neutral with regard to heat fluxes or is cooling [Cerovečki et al., 2011]. Given the long time it takes for surface DIC to equilibrate with atmospheric CO2 after SST has been perturbed by heat fluxes and given the main transport characteristics at Sta. Munida, we estimate the thermally induced CO2 uptake to be of either minor importance or to create a slight CO2 uptake.

[42] Biological processes have a substantial influence on the carbon budget of the mixed layer. We exclude biological processes associated with Alk, as this quantity behaves conservatively at our study site. The biological dominance of the seasonal cycle of sDIC, on the other hand, is likely to influence the uptake of pCO2 at Sta. Munida. The complexity of the oceanic carbon system dictates that our NCP term of 1.2 mol C m−2 yr−1, together with the horizontal transport term of 0.4 mol C m−2 yr−1, does not automatically cause carbon uptake from the atmosphere. The influence of biological processes on air-sea gas exchange depends on the balance between the upward supply of waters enriched in DIC through remineralization and the net fixation and export of carbon converted from DIC into organic matter [Gruber and Sarmiento, 2002; Keeling et al., 2004]. In principle, if the net biological fixation of CO2 exceeds the supply of DIC (vertically and horizontally), the ocean tends to take up CO2 from the atmosphere. In our case, the mixed layer loss of DIC to NCP and horizontal transport is very nearly compensated by entrainment and diffusion over the course of the year (difference of approximately 0.1 mol C m−2 yr−1). We interpret this small difference in combination with the physical effect as the likely mechanisms that lead to the uptake of atmospheric CO2. The relative weakness of these mechanisms might also serve as an explanation as to why the oceanic uptake of CO2 at Sta. Munida cannot keep up with the atmospheric increase (Figure 3d).

[43] If the CO2 sink at our study site is indeed strongly influenced by local biological production, a question arises of how nutrients are provided. Nutrient budgets at Sta. Munida do not show depletion in either nitrate or phosphate during the course of the year (not shown), which is not surprising for an HNLC region. By the time the onset of major biological production occurs at the end of winter, (micro) nutrients that had been used during earlier blooms would need to have been re-supplied. Vertical processes that could provide these nutrients are strongest toward the height of biological production at the turn of the year, whereas other possible sources are horizontal transport during the first half of the year or atmospheric deposition.

3.2.4 Comparison With Other Time Series Sites

[44] The model analysis we used for Sta. Munida has been applied before to two other study sites with comparable data records: the Sta. ALOHA/Hawaii Ocean Time series program (HOT) in the North Pacific subtropical gyre [Keeling et al., 2004] and the Station “S”/Bermuda Atlantic Time series Site (BATS) in the North Atlantic subtropical gyre [Gruber et al., 1998, 2002]. Although these sites are both located in the subtropics and can therefore be expected to have different physical and biogeochemical characteristics compared to Sta. Munida, it is instructive to compare our results in this broader context. In the following discussion, we also need to keep in mind that the time spans covered by these analyses are different: 1983–2001 for BATS, 1988–2002 for HOT, and 1998–2009 for Sta. Munida.

[45] The annual cycles of sDIC are shifted in phase (Figure 6a): while the Northern Hemisphere stations both peak in boreal spring (March) and have their minimum in autumn (September for BATS, October for HOT), Sta. Munida has its maximum sDIC concentration in austral winter (July/August) and reaches its lowest value in early summer (January). The amplitudes of the seasonal cycles differ substantially between approximately 15 µmol kg−1 at HOT, 32 µmol kg−1 at BATS, and 38 µmol kg−1 at Sta. Munida. Our station is similar to HOT in that it is undersaturated with regard to CO2 almost year-round (Figure 6b), while BATS alternates between phases of undersaturation and supersaturation. The timing of the maxima and minima of the seasonal pCO2 cycle at BATS and HOT is almost 180° out of phase with their respective sDIC cycles, while at Sta. Munida there is only a shift of about 2 months between the extrema of both quantities. Mixed layer depths at the three stations have similar seasonal patterns with maxima in winter and minima in summer (not shown). The mixed layer amplitudes of these stations are in roughly the same relation as their sDIC amplitudes, with Sta. Munida and BATS both being much more pronounced and HOT trailing.

Figure 6.

Harmonic fits of the annual cycles of (a) observed sDIC, (b) observed oceanic (solid) and atmospheric (dotted) pCO2 (c) modeled air-sea gas exchange, and (d) modeled NCP source terms for Sta. Munida (black), HOT (red), and BATS (blue).

[46] Although the relationship between strong mixed layer depth and sDIC variability at BATS and Sta. Munida versus HOT is relatively linear and would make an easy argument, it is not sufficient to explain the dynamics of the carbon cycle. The annual biological contribution, NCP (Figure 6d), at the three sites differs between −3.4 ± 0.8 mol C m−2 yr−1 for BATS, −2.3 ± 0.8 mol C m−2 yr−1 for HOT, and −1.2 ± 0.7 mol C m−2 yr−1 at Sta. Munida. Keeling et al. [2004] pointed out that the difference between the two subtropical stations can be explained by the “large difference in seasonally varying chemical behavior at the two sites”: At BATS, the seasonal behavior of NCP and air-sea gas exchange reinforce each other in creating the strong DIC drawdown in summer, while NCP and gas exchange oppose each other at HOT. At Sta. Munida, the situation is similar to HOT with the air-sea gas exchange source term having its positive maximum when NCP is most negative.

[47] The average annual net uptake of CO2 from the atmosphere (positive values are fluxes into the ocean; Figure 6c) also shows strong differences between the sites (1.9 ± 0.2 mol C m−2 yr−1 for BATS, 1.0 ± 0.1 mol C m−2 yr−1 for HOT, and 0.9 ± 0.1 mol C m−2 yr−1 for Sta. Munida). The discrepancies are rooted in the processes governing air-sea gas exchange at the respective sites. For BATS, Follows et al. [1996] and Gruber et al. [1998] argued that the uptake of natural CO2 is controlled by net cooling due to horizontal advection and subduction processes. At HOT, biological processes dominate [Keeling et al., 2004]. As for Sta. Munida, we make the case that the ocean uptake is governed by biological processes, with some modulation due to horizontal transport processes.

[48] Although the timespans reported from the sites in this comparison differ, we can perform a qualitative comparison of average trends. Out of the three locations, Sta. Munida shows the strongest average sDIC trend with 1.39 ± 0.39 µmol kg−1 yr−1. The trends at HOT and BATS are 1.22 ± 0.08 µmol kg−1 yr−1 and 0.64 ± 0.05 µmol kg−1 yr−1, respectively. In contrast, the strongest trend in pCO2 is observed at HOT, with 2.5 ± 0.1 µatm yr−1 substantially greater than 1.5 ± 0.1 µatm yr−1 observed for BATS, and 1.1 ± 0.4 µatm yr−1 for Sta. Munida. Sta. Munida has the biggest uncertainties for both quantities due to the smaller number of observations. BATS is closest to the trend in pCOinline image, interpreted by Keeling et al. [2004] as a lack of natural decadal variability and regime shift in contrast to HOT, where evidence of such variability exists. Sta. Munida's oceanic pCO2 trend is smaller than its atmospheric counterpart, as argued before, likely due to a limitation of the capacity to remove carbon from the mixed layer.

[49] We can conclude that all three time series sites that have been analyzed with this diagnostic box model show very dynamic carbon cycles that vary substantially on seasonal time scales and have clear trends associated with the increase of atmospheric pCO2. There are distinct differences between the three sites, characterized by location specific interactions between physics and biogeochemical processes.

4 Summary and Conclusions

[50] Bi-monthly measurements of physical and biogeochemical quantities have been taken at Sta. Munida in upper subantarctic surface water east of the New Zealand's South Island from 1998 to the present. Analysis of these data sets reveals that their seasonal cycles and trends are driven by the interaction of physical and biological processes, some of them local and some connected to regional dynamics. We used a modified version of a diagnostic box model of the upper ocean carbon cycle, which has been applied previously for similar investigations of carbon budgets at Stations ALOHA and BATS. The model makes use of the availability of both sDIC and δ13C data sets to create a carbon budget for the mixed layer by quantitatively analyzing air-sea CO2 exchange, vertical diffusion, vertical entrainment due to the seasonally deepening mixed layer, horizontal transport, and net community production of organic carbon.

[51] The records of oceanic and atmospheric pCO2 show an almost constant CO2 undersaturation at Sta. Munida that our diagnostic model analyzes to be 0.9 ± 0.1 mol C m−2 yr−1. The number of measurements available for this study does not allow us to make detailed statements about interannual variability but the differing trends in atmospheric and oceanic pCO2 over the course of our time series point to an increase of this uptake over the last decade. This is consistent with a smaller than expected trend in DIC. Two possible explanations for the oceanic increase not keeping up with the atmospheric growth are the decrease of the average temperature over the observational period and a limitation of the ability of the mixed layer to remove sufficient carbon by either biological or horizontal transport processes. The latter explanation is supported by the diagnosed seasonal balance of the carbon budget terms.

[52] The seasonal cycle of sDIC is dominated by the NCP term that removes inorganic carbon from the mixed layer over two thirds of the year with an annual integral of 1.2 ± 0.7 mol C m−2 yr−1. The subtropical time series station ALOHA near Hawaii has about twice as much NCP, while the uptake of carbon from the atmosphere is about the same. Our model results lead us to attribute these differences to the role of horizontal transport and diffusion. Over the annual cycle, horizontal transport at Sta. Munida removes carbon from the mixed layer, while at HOT it adds inorganic carbon. Diffusion, albeit adding carbon at both stations, is about double the size near New Zealand. The balance between NCP and horizontal transport versus diffusion and entrainment is completed by the steady but relatively small amount of carbon taken up through the air-sea interface. We speculate that this balance has remained stable over the course of our observations leading the observed slower increase in oceanic pCO2 compared to atmospheric values.

[53] Our comparison with other time-series stations leads to the conclusion that it is not sufficient to label them as “subtropical” versus “subantarctic” but that the locations of the particular stations have a huge influence on the interpretation of the results in a larger context. The trend in surface ocean pCO2 compared to pCOinline image, for instance, is stronger near Hawaii, similar near Bermuda, and weaker at our site. What all these sites share, though, is the persistent uptake of anthropogenic CO2 and upward trends in the mixed layer carbon content.


[54] We are grateful to the captain and crew of R/V Munida (cruises from 1998 to March 2006) and R/V Polaris II (Aug 2006 onward). Keith Hunter and Malcolm Reid are integral to the Munida time series program. We also wish to express our thanks to Mark Hadfield, Phillip Sutton, and Cliff Law for scientific discussions and plenty of advice. We thank Robert Key and Andreas Schmittner for providing data, Niki Gruber for permission to use his diagnostic box model, Paulo Calil and Mark Hadfield for assistance with model calculations, Hartmut Frenzel for making model analysis tools available, and two anonymous reviewers for their constructive comments. KIC and SEMF's contributions to this manuscript were funded by the NIWA National Centre for Atmosphere's core research funding.