How important are Southern Hemisphere wind changes for low glacial carbon dioxide? A model study



[1] The response of atmospheric CO2 to modifications in the strength and position of Southern Hemisphere westerlies is examined with the Bern3D ocean model. The model responds more sensitively to changes in the wind amplitude than to variations in latitudinal position. Depending on the model setup, a 50% reduction in wind strength leads to a CO2 drawdown of 3–34 ppm, while a 50% increase results in a rise of 10–24 ppm. A poleward shift of 5° lowers CO2 by 2–16 ppm whereas an equatorward shift of 5° induces a CO2 increase of 2–14 ppm. Physical and biological mechanisms equally contribute to the modeled changes in atmospheric CO2. Our results are in conflict with the hypothesis that Southern Hemisphere wind changes are responsible for the low atmospheric CO2 concentrations during glacial periods.

1. Introduction

[2] In spite of more than 20 years of research, the causes of glacial-interglacial fluctuations in atmospheric CO2 still remain an unresolved issue. Many potentially important mechanisms have been suggested [e.g., Archer et al., 2000a; Sigman and Boyle, 2000], but sparse paleodata and limitations of currently applied models impede an unequivocal identification of the relevant processes. It is well established, however, that the atmospheric CO2 variations are the result of the redistribution of carbon between the terrestrial, oceanic and atmospheric reservoirs.

[3] In this study, we focus on the role of the Southern Hemisphere westerly winds (SHW) in regulating the large-scale ocean circulation and the ocean's carbon storage capacity. Our goal is (1) to assess the sensitivity of the large-scale ocean circulation, implied biogeochemical changes and the response in atmospheric CO2 to a modified strength and position of the SHW and (2) to discuss their potential role in explaining glacial-interglacial CO2 fluctuations.

[4] Toggweiler et al. [2006] suggest that a feedback between SHW, ocean circulation and atmospheric CO2 is largely responsible for glacial-interglacial CO2 fluctuations. The proposed mechanism is based on the concept that the global large-scale overturning circulation can be split into two different regimes that occupy distinct domains in the ocean interior [Toggweiler et al., 2006, Figure 3]. Their northern regime corresponds to the “biologically productive North Atlantic circuit” that is filled with NADW (North Atlantic Deep Water). The southern regime is the “unproductive Southern Ocean circuit,” i.e., the domain of AABW (Antarctic Bottom Water). Sinking organic particles are remineralized at depth in both circuits. When the water of the southern circuit comes up to the surface in the Southern Ocean (SO) today, the remineralized carbon escapes fairly easily to the atmosphere. In contrast, relatively little remineralized carbon escapes from the surface waters of the biologically productive northern circuit since biological production draws down surface pCO2.

[5] Toggweiler et al.'s [2006] view of the glacial ocean is a scenario where the northern circuit remains active while the southern circuit is weak leading to a buildup of respired CO2 in the deep ocean and to less CO2 outgassing from the SO. In their hypothesis, Toggweiler et al. [2006] attach critical importance to the position of the SHW. They argue that the glacial SHW were equatorward of their present position such that upwelling next to Antarctica was suppressed and the southern circuit was essentially shut off during glacial times.

[6] Several recent studies show that SH winds modify large-scale overturning and air-sea CO2 fluxes in the SO on decadal time scales. Observations and models indicate a poleward intensification in the SHW during recent decades [Thompson and Solomon, 2002; Gillett and Thompson, 2003], which has resulted in an enhanced wind-driven upwelling around Antarctica [Russell et al., 2006; Lovenduski and Gruber, 2005; Hall and Visbeck, 2002]. Lovenduski et al. [2007] propose that this has led to an anomalous uptake of anthropogenic CO2 but to an even larger anomalous outgassing of natural CO2 over the SO. Therefore, higher stabilization levels of atmospheric CO2 would have to be expected on a multicentury time scale if the trend persists in the future [Le Quéré et al., 2007].

[7] In this study, we will revisit Toggweiler's glacial-interglacial hypothesis testing its feasibility in the Bern3D coarse-resolution model [Müller et al., 2006; Parekh et al., 2008]. It is tested whether an equatorward shift in the latitudinal position of the SHW can substantially reduce deep ocean ventilation and whether a shutdown of the southern circuit results in lower atmospheric CO2. Furthermore, Toggweiler's study is extended regarding two important issues. In their modeling study, Toggweiler et al. [2006] do not explicitly shift the SHW in order to test their hypothesis. Instead they weaken the SHW arguing that this would have essentially the same effect on upwelling in the SO as an equatorward wind displacement. Here, both the effect of changing the strength of the westerlies, as well as the response to an explicit shift in the position of the westerlies are investigated. Second, to avoid changes in nutrient utilization upon circulation changes, Toggweiler et al. [2006] simulate the production of sinking organic particles and CaCO3 by restoring surface phosphate concentrations toward a fixed field. In addition to model runs with this same procedure, we also perform simulations with a prognostic formulation of marine biogeochemistry. This enables us to quantitatively analyze contributions from physical as well as biogeochemical processes (section 3.3).

[8] A series of different wind stress scenarios spanning the entire range of available reconstructions for glacial SH winds was created to generally test the model's sensitivity to marine carbon cycling and atmospheric CO2 (section 2.3). In a first set of experiments we probe the model response to a change in the SH westerlies' amplitude and compare the results with the findings of previous studies [De Boer et al., 2008; Toggweiler et al., 2006; Toggweiler and Samuels, 1995]. In a second set we explore the effect of an explicit latitudinal shift in the SH westerlies' position to find out whether the model behaves similarly as in the case of modified wind strength.

[9] In order to examine the robustness of the model response with respect to different initial states, we impose the various scenarios on 8 model setups (Table 1) that differ with respect to the type of boundary conditions, the strength of diapycnal mixing, the rate of Atlantic meridional overturning and the strength of the carbon “pumps” [Volk and Hoffert, 1985].

Table 1. Different Model Setupsa
SetupFAtl-Pac (Sv)κdia (10−5 m2 s−1)BCNUE (%)CEE (%)
  • a

    The amount of the Atlantic to Pacific freshwater flux (FAtl-Pac), the value of diapycnal diffusivity (κdia) and the boundary conditions (BC) employed along with simulated global nutrient utilization efficiency (NUE) and global CaCO3 cycle efficiency (CEE) (as defined in section 3.3) at steady state for standard wind stress.

RBC lowdiff0.01.0restoring80.912.3
weakFw lowdiff0.01.0mixed80.812.4
strongFw lowdiff0.301.0mixed78.313.1

[10] We finally carry out additional model sensitivity tests examining the effect on atmospheric CO2 induced by (1) increased CO2 solubility, (2) reduced gas exchange in the SO due to enlarged sea ice cover, (3) CaCO3 compensation, and (4) a 500 GtC carbon input into the atmosphere. In this last set of simulations we separately and jointly impose these four mechanisms on a model state in which atmospheric CO2 is lowered in response to weakened SHW.

[11] The cost-efficient, dynamical, three-dimensional Bern3D ocean model [Müller et al., 2006] allows for an adequate representation of the large-scale ocean circulation and the cycling of major biogeochemical tracers (section 2.2). Its computational efficiency makes it ideally suited for multimillennial paleosimulations and large ensembles of sensitivity simulations.

2. Methods

2.1. Model Description

[12] The physical core of the Bern3D model is a frictional-geostrophic balance ocean model [Müller et al., 2006; Edwards and Marsh, 2005]. Model grid resolution is 36 by 36 cells in the horizontal and 32 vertical layers with exponentially increasing thickness toward the ocean bottom. The model includes an isopycnal diffusion scheme and Gent-McWilliams parametrization for eddy-induced transport [Griffies, 1998]. Convection is represented in the model by shifting down statically unstable water parcels from the surface further downward in the water column according to the density of the parcel. Subsequently, tracer concentrations are homogenized over remaining regions of static instability in the water column.

[13] Sea surface temperatures (SST) and salinities (SSS) are constrained by using either restoring (RBC) or mixed boundary conditions (MBC). Restoring compels SST and SSS to remain close to modern-day observational values. MBC on the other hand prescribe freshwater fluxes on the ocean surface rather than SSS such that circulation-salinity feedbacks are allowed to develop upon perturbation of the model state [Bryan, 1986]. An additional uniform freshwater surface flux from the Atlantic to the Pacific basin is applied in order to intensify and deepen the Atlantic MOC. Forcing fields for wind stress are based on a simple analytical profile that is zonally and temporally uniform (section 2.3). Here, meridional wind stress is neglected as in the work by Toggweiler et al. [2006]. Standard values for diapycnal diffusivity κdia and for the Atlantic-Pacific freshwater surface flux FAtl-Pac have been selected to reproduce a realistic meridional overturning circulation as well as a realistic radiocarbon distribution on a basin-wide scale (Figure 1a). The application of analytical wind stress profiles in this specific study as opposed to the NCEP wind stress data applied in other Bern3D studies [Müller et al., 2008; Siddall et al., 2007; Parekh et al., 2008; M. Gerber et al., Regional air-sea fluxes of anthropogenic carbon inferred with an Ensemble Kalman Filter, submitted to Global Biogeochemical Cycles, 2008] required an upward adjustment of κdia from 1.0 to 3.5 × 10−5 m2 s−1. Standard parameter values for the physical and biogeochemical model are summarized in Table 2.

Figure 1.

Standard model state: (a–e) Basin-averaged annual mean depth profiles for natural Δ14C, phosphate, dissolved inorganic carbon, alkalinity, and silicic acid. Solid lines correspond to modeled profiles. Dotted lines are data-based estimates from GLODAP (Δ14C, DIC, alkalinity [Key et al., 2004]) and World Ocean Atlas 2001 (phosphate, silicic acid [Conkright et al., 2002]). (f) Basin-averaged profiles for ΔCgas-ex (dotted lines), ΔCgas-ex + ΔCcarb (dashed lines), and ΔCgas-ex + ΔCcarb + ΔCsoft (solid lines) following Gruber and Sarmiento [2002]. Black lines are based on model output, and gray lines are calculated from data. Global mean surface values are set to zero for modeled and for data-based profiles.

Table 2. Model Parameters With Their Standard Values and References
KIisopycnal diffusivity1000 m2 s−1Müller et al. [2006]
KGMGent-McWilliams parameter2000 m2 s−1Müller et al. [2006]
KDdiapycnal diffusivity3.5 × 10−5 m2s−1this study
zeuphoticdepth of euphotic zone75 mOCMIP-2
σDOM fractionequation imageOCMIP-2
αPOM remineralization Martin-scale0.9OCMIP-2
κPO4half-saturation constant for PO40.05 μM PO4Parekh et al. [2008]
κFehalf-saturation constant for Fe0.17 nM FeParekh et al. [2008]
τbiomass turnover time scale30 daysParekh et al. [2008]
C:PRedfield ratio C:P117Anderson and Sarmiento [1994]
N:PRedfield ratio N:P16Anderson and Sarmiento [1994]
O2:PRedfield ratio O2:P−170Anderson and Sarmiento [1994]
Fe:Pratio Fe:P5.85 × 10−4Parekh et al. [2008]
lopalopal redissolution length scale10,000 mHeinze et al. [1999]
lcalcitecalcite redissolution length scale3500 mOCMIP-2
KD,minSi 10−3 mol m−3Aumont and Bopp [2006]
KD,maxSi 7 × 10−3 mol m−3Aumont and Bopp [2006]
KD*Sihalf-saturation constant for Si/C increase5 × 10−3 mol m−3Aumont and Bopp [2006]
R*Si:P 30 mol Si/mol Pthis study
RCa:PCa:P uptake ratio in calcifiers0.3 mol Ca/mol Pthis study
MCaCO3maximum rain ratio0.3this study

[14] The biogeochemical component of the Bern3D model represents the cycling of dissolved inorganic carbon (DIC), dissolved organic phosphate (DOP), 14C, alkalinity, phosphate, oxygen, iron and silicic acid. The ocean carbon cycle is coupled to a well-mixed atmospheric reservoir of CO2. The formulations for air-sea gas exchange and carbonate chemistry in the ocean are based on the updated OCMIP-2 protocols [Orr et al., 1999; Najjar and Orr, 1999] and include adaptations as described by Müller et al. [2008]. Fully prognostic formulations are used to compute export production of organic matter, biogenic opal and CaCO3 (see auxiliary material).

2.2. Model Evaluation

[15] The cost-efficient Bern3D model is able to capture the large-scale ocean circulation (Figure 2), the characteristics of the most important water masses, the large-scale distributions of a range of ventilation time scale tracers (14C, CFCs, 39Ar, anthropogenic carbon [see Müller et al., 2006]), biogeochemical tracers, temperature and salinity (Figure 1 and Table 3), as well as the spatial patterns of POM, CaCO3 and opal export production for standard parameter values. The results presented in the following two subsections have been obtained using the standard model parameter values (Table 2) and the standard analytical wind stress profile (section 2.3).

Figure 2.

Annual mean residual overturning circulation (Sv) in the global, Atlantic, and Pacific ocean for the standard model configuration. (a–c) The circulation for the case of standard (modern) winds and (d–f) the results for SH westerlies with a fourfold reduced amplitude. (g–i) The case of SH westerlies that are shifted by 10° toward the equator. The stream functions include the contribution resulting from Gent-McWilliams transport. Continuous (dashed) contours are for positive (negative) values.

Table 3. Correlations and Relative Standard Deviations of Modeled Tracer Fields in the Standard Model Setup in Comparison to Data-Based Fields
TracerCorrelation With DataRelative Standard DeviationData Source
Temperature95.2%0.97Levitus and Boyer [1994]
Salinity79.2%1.37Levitus et al. [1994]
Δ14C94.9%1.18Key et al. [2004]
Phosphate88.9%1.03Conkright and Boyer [2002]
Preindustrial DIC91.8%1.41Key et al. [2004]
Alkalinity92.3%1.80Key et al. [2004]
Silicic acid90.4%1.30Conkright and Boyer [2002]

2.2.1. Meridional Overturning Circulation and Water Masses

[16] The global annual mean residual overturning circulation (Figure 2a) in the deep ocean displays the clockwise overturning cell in the north driven by the formation of NADW and the anticlockwise cell in the south associated with the formation of AABW. The middepth NADW cell can be interpreted as the “northern biologically productive circuit” and the deep anticlockwise overturning can be associated with the “southern unproductive circuit” as described by Toggweiler et al. [2006]. Deep convection occurs predominantly in the North Atlantic area south of Greenland (NADW) as well as in the Ross and Weddell Seas next to Antarctica (AABW). Maximum overturning in the Atlantic (Figure 2b) reaches 21.7 Sv at around 55°N comparable to the estimate of Talley et al. [2003] of 17 ± 5 Sv. 14.8 Sv of this southward flow are exported from the Atlantic into the SO at 30°S. A weak anticlockwise overturning cell with a maximum of 2.4 Sv fills the deepest parts of the Atlantic with AABW. Overturning in the deep Pacific (Figure 2c) is anticlockwise and amounts to a maximum of 9.8 Sv, 4.8 Sv of which recirculate back into the SO. Deep equatorial upwelling in the Pacific amounts 5 Sv, relatively modest for a coarse-resolution model.

[17] A comparison between modeled and data-based salinity and natural radiocarbon zonal mean sections in the Atlantic and the Pacific reveals that overall surface-to-deep transport and the primary flow paths of the most important water masses are represented quite realistically in the model (not shown). The observed NADW tongue of relatively salty and young waters in the North Atlantic is clearly established in the model. Compared to data-based distributions, modeled NADW seems to be too young in Δ14C by about 20 per mil. The northward flowing Antarctic Intermediate Water (AAIW) in the Atlantic is modeled as a relatively fresh water mass lying above NADW. The deep Atlantic is filled with AABW. Horizontal radiocarbon gradients in the deep Pacific are relatively well captured in the model, however, the water in the deep North Pacific is somewhat too old. AAIW penetrates the Pacific basin from surface to a depth of about 1 km.

2.2.2. Tracer Distributions and Biological Export Fluxes

[18] The Bern3D model is able to adequately represent basin-scale surface-to-deep gradients for a range of different tracers (Figure 1). The modeled tracer distributions correlate reasonably well with observed fields interpolated to the model grid (Table 3). Simulated zonal mean sections in the Atlantic and the Pacific generally compare favorably with observations (e.g., Figures 3a and 3b). Modeled export fluxes for POM (15.9 GtC a−1), CaCO3 (1.4 GtC a−1) and opal (97.6 TmolSi a−1) are compatible with observation-based estimates [Falkowski et al., 1998; Iglesias-Rogdríguez et al., 2002; Treguer et al., 1995]. Their spatial patterns (Figures 3c and 3d) display the prominent large-scale features of the observed fields. However, because of imperfections in the model circulation, POM and CaCO3 export in the North Pacific are underestimated [Schlitzer, 2000].

Figure 3.

Standard model state: Zonal mean DIC distribution (μmol kg−1) in (a) the Atlantic and (b) the Pacific. (c) Annually averaged CaCO3 export production (mol C m−2 s−1) and (d) opal export production (mol Si m−2 s−1). Total yearly export productions of 1.4 GtC a−1 for CaCO3 and 97.6 Tmol Si a−1 for opal are comparable to available estimates.

[19] Figure 1c shows that basin-wide surface-to-deep gradients in DIC are reasonably well simulated in the Bern3D model. The question remains, however, whether the three main processes responsible for shaping the spatial distribution of DIC in the ocean, namely the organic matter cycle, the CaCO3 cycle and the air-sea gas exchange, do contribute in realistic proportions to the modeled gradients. Gruber and Sarmiento [2002] introduced the three tracers ΔCsoft, ΔCcarb and ΔCgas ex in order to assess their relative importance. ΔCsoft and ΔCcarb quantify the impact of the organic matter and CaCO3 cycles in absence of gas exchange. ΔCgas ex represents the residual part of the gradients in the salinity-normalized DIC distribution. Figure 1f displays simulated and data-based depth profiles of ΔCsoft, ΔCcarb and ΔCgas ex. Both positive and negative deviations for the observation-based profiles are simulated for the individual tracers. The comparison indicates that the relative strengths of the three processes are reasonably simulated.

[20] ΔCgas ex is the sum of two opposing contributions ΔCgas exbio and ΔCgas extherm arising from biological processes and thermal fluxes, respectively. The net surface-to-deep gradients in ΔCgas ex are therefore comparatively small (see Figure 1f). Note that the surface-to-deep gradient in the ΔCgas extherm corresponds to the solubility pump as defined by Volk and Hoffert [1985]. There is an ongoing debate in the literature about how realistically the solubility pump is represented in current ocean models [e.g., Broecker et al., 1999; Archer et al., 2000b; Toggweiler et al., 2003] It seems that in OGCMs the solubility pump is subject to a rather strong kinetic limitation by air-sea gas exchange. On the other hand, in box models the kinetic limitation seems to be much weaker [Toggweiler et al., 2003]. Unfortunately, it is currently not possible to estimate ΔCgas extherm and thus the real strength of the solubility pump directly from data [Sarmiento and Gruber, 2006]. However, there are accurate data-based estimates available for the potential strength of the solubility pump, i.e., for the case with infinitely rapid gas exchange.

[21] Next, we assess how the Bern3D model represents the solubility and the potential solubility pump to place the model in the current discussion. This is done by running the model in a setup without marine biota using standard gas exchange transfer velocities in a first experiment; in a second experiment transfer velocities are increased by a factor of fifteen. The two simulations yield surface-to-deep gradients of 91 μmol kg−1 and 154 μmol kg−1, respectively. The latter value is in very good agreement with the data-based estimate of 155 μmol kg−1 by Sarmiento and Gruber [2006] for the potential solubility pump. In the Bern3D model, we simulate a deviation of 63 μmol kg−1 between the realized and the potential solubility pump. Toggweiler et al. [2003] suggest in their concluding section that this deviation should be about 50 μmol kg−1 in the real ocean. These gradients and differences in gradients may also be compared with the overall preindustrial surface-to-deep DIC gradient of about 305 μmol kg−1. Uncertainties in the model representation of the solubility pump remain, as long as its real strength cannot be well quantified by observations.

2.3. Wind Stress Scenarios

[22] The reconstruction of past wind circulation from indirect proxy indicators is a complex task. The interpretation of Late Quaternary paleoenvironmental records from South America and Australia has therefore led to conflicting claims about the latitudinal position of the SHW at the LGM. One school argues for equatorward shifted westerlies [Moreno et al., 1999; Heusser, 1989] and another supports poleward shifted westerlies [Markgraf, 1989; Harrison and Dodson, 1993]. In a review of multiple proxy indicators [Shulmeister et al., 2004] come to the conclusion that there is good evidence for enhanced westerly flow and increased wind-driven upwelling at the LGM and that glaciation-interglaciation shifts were no more than 3° to 4°.

[23] Model results are also contradictory regarding the nature of SHW at the LGM. Some of them indicate a general intensification and a poleward displacement of their zonal maximum [Wyrwoll et al., 2000], and some suggest a weakening and a slight equatorward shift [Kim et al., 2002]. Others point at no significant latitudinal shift with either a modest weakening [Rojas et al., 2008] or a slight strengthening of the wind stress over the SO [Otto-Bliesner et al., 2006]. In light of the large gaps in the quantitative and qualitative understanding of the SH wind circulation at the LGM, we created a variety of scenarios that span the entire range of available reconstructions.

[24] All the wind stress scenarios are based on an analytical profile. For simplicity, the meridional wind stress component is neglected and the zonal component τλ is zonally and temporally uniform. The profile's mathematical expression is a function of two parameters ε and μ that allow for the scaling of the SH westerlies' amplitude and for a latitudinal shift in the SH wind stress profile, respectively:

display math

[25] Figure 4a shows that the standard profile (ɛ = 1.0, μ = 0) reasonably reproduces the zonally and seasonally averaged zonal component of the NCEP reanalysis derived data [Kalnay et al., 1996].

Figure 4.

The wind stress scenarios and the associated Ekman pumping: (a and b) scaling scenarios and (c and d) shift scenarios. The vertical solid lines at 51°S indicate the location of the tip of South America in the model geometry which corresponds to the northern border of the Southern Ocean.

[26] We created two ensembles of scenarios by modifying either ɛ or μ with respect to their standard values. In the first ensemble (scaling scenarios) the SHW are weakened and intensified without altering their location by setting ε to 0.25, 0.5, 0.75, 1.5, 2.0 and 3.0 (Figure 4a). In the second ensemble (shift scenarios) we shift the position of the SH westerlies' peak poleward as well as equatorward (Figure 4c). Maximal latitudinal displacements are 5° toward the South Pole (μ = −5) and 10° toward the equator (μ = 10). The intermediate displacement scenarios are spaced by 1°.

[27] The Ekman pumping strength quantifies wind-driven upwelling, which corresponds to the degree of divergence of wind-driven horizontal flows in the sea surface. Wind-driven upwelling in the open ocean is generally limited to a surface layer some hundreds of meters thick. In the SO, however, the absence of the geostrophic balance above topographic ridges leads to a situation where deep water is brought to the surface through the action of winds. The Ekman pumping strength, integrated from 90°S, is depicted in Figures 4b and 4d for the different wind stress scenarios.

2.4. Model Setups and Spin-Up

[28] Although the performance of the Bern3D model regarding the representation of the modern ocean is quite reasonable (section 2.2), there still remains some degree of uncertainty due to the limited spatial coverage of the observations needed to test the model and due to the interpretation of indirect measurements. For this reason, the wind stress scenarios are imposed on 8 different setups to assess the robustness of the model response. As summarized in Table 1 the model setups differ in the type of boundary conditions (restoring versus mixed), the value for diapycnal diffusivity κdia and the intensity of the anomalous surface freshwater flux FAtl-Pac from the Atlantic to the Pacific basin.

[29] The driving forces of the meridional overturning circulation (MOC) are still a matter of intense investigation. Presently, two major mechanisms are discussed [Kuhlbrodt et al., 2007]. The first one is turbulent mixing of heat from the surface into the ocean's interior, the second is wind-driven upwelling in the Southern Ocean. From this perspective, the choice of diapycnal diffusivity κdia controls the relative importance of thermohaline versus wind-driven MOC forcing in a given model setup.

[30] The sinking of dense, salty water in the North Atlantic is an important organizing feature of the modern deep ocean circulation. Modeled deep water mass properties, ventilation time scales and tracer distributions in different ocean basins are quite sensitive to the modeled rate of deep water formation in the North Atlantic. The anomalous surface freshwater flux FAtl-Pac permits a systematical increase (decrease) in Atlantic sea surface salinities and a strengthening (weakening) of the Atlantic MOC.

[31] For every model setup a steady state has been computed in the following manner: First, the physical model is spun-up for 10,000 model years using RBC. Subsequently, air-sea freshwater fluxes are diagnosed and averaged over the last 1000 years to provide mixed boundary conditions. The model is now switched to mixed boundary conditions and anomalous freshwater fluxes FAtl-Pac are included. In the case of the setups that employ restoring, by definition RBC are maintained and no additional freshwater flux FAtl-Pac is applied. Then, a biogeochemical spin-up phase with a fixed atmospheric CO2 mixing ratio of 278 ppm is run for another 10,000 years. Thereafter, atmospheric CO2 is allowed to evolve freely. Annual means are used for the analysis.

2.5. Overview of Experiments

[32] In the first set of experiments we impose the scaling scenarios on a spun-up model state by instantaneously switching from the standard to a modified wind stress profile. The model is then run into a new steady state over 5000 model years. The gas transfer velocity field is kept unchanged upon wind stress perturbation. The second set consists of corresponding experiments employing the shift scenarios. Each of the wind stress scenarios is applied to the 8 different setups of the model. The CO2 response in simulations where the SHW are both scaled and shifted simultaneously is essentially the superposition of the responses that are induced by the wind scaling and the shift separately. Such experiments with changes in both position and strength will not be further discussed. In a third set of experiments we investigate the additional effects of enhanced glacial CO2 solubility, increased sea ice cover to gas exchange, CaCO3 compensation and terrestrial carbon release with respect to atmospheric CO2. A 500 GtC carbon input from the terrestrial biosphere [e.g., Shackleton, 1977] is represented by instantaneously increasing atmospheric CO2 by 236 ppm. Enhanced CO2 solubility is modeled by imposing a reconstruction of LGM SST and SSS [Paul and Schäfer-Neth, 2003] on the carbonate chemistry routine that calculates the speciation of DIC and pCO2 in the sea surface. Reduced gas exchange from increased sea ice cover is represented by linearly scaling down piston velocities according to a monthly map of fractional sea ice cover for the LGM [Paul and Schäfer-Neth, 2003]. To simulate CaCO3 compensation the CO32− concentration in each bottom ocean cell below 2000 m is restored toward the CO32− value diagnosed in the unperturbed state. This approach allows for both sediment dissolution and accumulation leading to an addition or removal of alkalinity and DIC in a 2:1 ratio. Starting from the initial state lowdiff, we perform another five simulations where, in addition to a halving of the SH westerlies' amplitude, the four additional mechanisms are taken into account separately and jointly. Each configuration is run into a new steady state during a period of 10,000 model years. A more detailed description of how the additional mechanisms are represented in this study can be found in the auxiliary material.

3. Model Results

3.1. Scaling Experiments

3.1.1. Ocean Circulation and Ventilation

[33] To compare our model results with those of previous studies, we first examine the dependence of the southward water flow leaving the Atlantic at 30°S on the amplitude of the SHW (Figure 5). An almost perfectly linear relationship is simulated for the setups with RBC (solid/dashed black lines). When MBC are applied instead of RBC (all other lines), circulation-salinity feedbacks are allowed to develop which tend to counteract the effect of the changes in wind forcing. The response of the Atlantic outflow to changes in the amplitude of SHW is thus largely damped in these setups (red, blue and green solid/dashed lines). However, NADW formation and Atlantic outflow shut down when weak winds are applied in the setups with low AMOC (red solid/dashed lines).

Figure 5.

The “Drake Passage effect” [Toggweiler and Samuels, 1995]: The Atlantic outflow at the latitude of the Cape of Good Hope as a function of the northward Ekman transport at the tip of South America. With restoring boundary conditions (black lines) we model an almost perfectly linear relationship. When using mixed boundary conditions (green, blue, and red lines) we find a substantially weaker coupling between the Ekman transport and Atlantic outflow.

[34] These findings confirm previous studies [Bugnion et al., 2006; Rahmstorf and England, 1997; Toggweiler and Samuels, 1995]. The tight coupling between SHW and the AMOC is an artifact of unrealistic restoring boundary conditions. We conclude furthermore that the strength of the coupling (i.e., the slope of the lines in Figure 5) significantly depends on the value of the diapycnal mixing parameter κdia. The setups with low κdia respond more sensitively to changes in wind forcing as compared to the setups with standard diapycnal diffusion.

[35] Figure 6 shows the strength of AMOC and deep SO overturning (below a depth of 1000 m) as a function of the scaling parameter ɛ. These quantities can be interpreted as the intensity of the northern and the southern circuit, respectively [Toggweiler et al., 2006]. We analyze overturning in the deep SO as the interest here is on the link between the surface SO and the deep waters of the world ocean and not on shallow overturning.

Figure 6.

Circulation response to changes in SHW amplitude: (a) Maximum Atlantic overturning (Sv), (b) deep Southern Ocean overturning (Sv), (c) basin mean Δ14C signatures (per mil) for the Southern Ocean (below 640 m), (d) the upper Pacific (640–2100 m), and (e) the deep Pacific (2100–5000 m) as a function of the scaling factor ɛ.

[36] Across all the setups a significant trend is that the strength of AMOC (northern circuit) is positively correlated with the amplitude of the SHW. Modeled basin mean Δ14C signatures in the Atlantic (not shown) reflect that the simulated changes in Atlantic ventilation are modulated by the changes in AMOC. This is no surprise since the Atlantic basin is mainly ventilated by NADW.

[37] In the SO the interplay between wind stress, overturning and ventilation seems less evident. There is a tendency, but no univocal trend, that the SHW strength controls the intensity of deep SO overturning (Figure 6b). The frequency and depth of convective events in the SO are clearly linked to the amplitude of SHW with stronger winds leading to more vigorous convection and vice versa (not shown). At the same time there is an evident correlation between basin mean Δ14C signatures in the SO and the SHW strength (Figure 6c). The amplitude of ventilation changes in the SO also depends on the diapycnal mixing parameter κdia (Figure 6c).

[38] The clockwise overturning cell in the upper North Pacific is very sensitive to an intensification of the SHW. In all the model setups this cell considerably increases in strength and extent for strong winds which is manifest in the Δ14C signature of the upper part of the Pacific basin (Figure 6d). Circulation-salinity feedbacks increase SSS in the North Pacific under MBC and amplify this sort of a response. In the case of weakened winds in contrast, the overturning cell becomes slower and shallower. A similar model response in the Pacific has also been found by De Boer et al. [2008], where an OGCM coupled to an energy moisture balance and a dynamic ice model is used to examine the circulation's sensitivity to changes in the amplitude of SH winds.

[39] The simulated Δ14C signatures in the deep Pacific (Figure 6e) also indicate the tendency that the rate of ventilation increases for stronger SHW and decreases for weaker SHW. In some of the model setups, a stagnant water mass at the bottom of the Pacific basin appears when the winds are weakened. In these cases, sharp drops in the deep Pacific Δ14C signature are simulated (blue and green solid/dashed lines).

[40] In conclusion, these experiments cover a wide range with respect to SO circulation, convective activity and deep water ventilation and thus allow us to explore the carbon cycle response to a wide variety of physical changes.

3.1.2. Biogeochemistry and Atmospheric pCO2

[41] Our results indicate that atmospheric CO2 is rather sensitive to the amplitude of the SHW (Figure 7a). Intensified winds consistently induce a CO2 rise in all the setups. The simulated changes are in the range of 35–60 ppm for a threefold strengthening (ɛ = 3.0). The biogeochemical rearrangements in response to a wind intensification are quite similar for all the setups. The modeled CO2 rise in the atmosphere can be primarily attributed to an increase in sea surface pCO2 in the SO.

Figure 7.

Biogeochemical response to changes in SHW amplitude: (a) Atmospheric CO2 (ppm) and its variations split into (b) a physical and (c) a biological contribution as a function of the scaling factor ɛ. (d) Nutrient utilization efficiency in the Southern Ocean (%) (south of 34°S) and (e) the global CaCO3 cycle efficiency (%).

[42] In the case of weakened winds however, the various setups respond differently. Setups with rather weak AMOC (black and red lines in Figure 7a) show a substantial drawdown in CO2 of 30 to 55 ppm for a fourfold reduction in the strength of the SHW (ε = 0.25). The setups standard and strongFw, on the other hand, have very small responses of −2 to −7 ppm (blue and green solid lines). Finally, lowdiff and strongFw lowdiff display a nonmonotonic CO2 response with respect to the wind-scaling factor ε (blue and green dashed lines). The modeled reorganizations of the DIC and alkalinity distributions as well as of sea surface pCO2 reveal that qualitatively different biogeochemical responses occur when weak winds are imposed on the various model setups. The attribution of changes in CO2 to individual mechanisms will be further explored in section 3.3.

3.2. Shift Experiments

3.2.1. Ocean Circulation and Ventilation

[43] Figure 8a shows that within the range of our shift scenarios for most setups we do not find any significant dependence of AMOC on the latitudinal position of the SHW. An exception to this are the two setups with MBC and weak AMOC (red dashed/solid lines). In both setups NADW formation collapses upon wind shifts in both directions.

Figure 8.

Model response to shifts in the latitudinal position of the SHW: (a) Maximum Atlantic overturning (Sv), (b) deep Southern Ocean overturning (Sv), (c) basin mean Δ14C signature (per mil) in the Southern Ocean (below 640 m), (d) atmospheric CO2 (ppm), and (e) the global CaCO3 cycle efficiency (%) as a function of the shift parameter μ. Positive values of μ correspond to equatorward wind shifts, and negative values of μ relate to poleward wind shifts.

[44] Equatorward shifts tend to reduce overturning in the SO whereas poleward shifts lead to an increase in SO overturning (Figure 8b). In the setups with MBC the typical structure of the deep SO overturning cell with a distinct maximum at around 40°S disappears when large equatorward shifts are applied. These results confirm the hypothesis that equatorward shifts tend to reduce the intensity of the southern overturning regime, as Toggweiler et al. [2006] suggested.

[45] However, the collapse of the SO overturning cell results in a decrease in SO ventilation only in the setups with low diapycnal diffusivity (Figure 8c). The corresponding setups show a decrease in Δ14C of 50 to 77 per mil upon large equatorward shifts (red, blue and green dashed lines). In the other setups there is almost no change in SO ventilation when the SHW are shifted in position. Convective activity in the SO is rather unresponsive to shifts in wind stress for all the setups.

[46] As in the case of wind scaling, the Pacific circulation responds quite sensitively to latitudinal shifts in the SHW in all the setups. Equatorward displacements reduce the rate of upwelling from the deep to the surface and enhance the extent of the clockwise overturning cell in the upper North Pacific. The most prominent change caused by a poleward displacement is an increase in deep equatorial upwelling in the Pacific. However, the changes in deep Pacific overturning and deep Pacific ventilation are rather small (not shown).

3.2.2. Biogeochemistry and Atmospheric pCO2

[47] The sensitivity of atmospheric CO2 to latitudinal displacements of the zonal wind maximum is significantly smaller compared to its dependence on the amplitude of the SHW. The simulations indicate that poleward displacements of the winds induce a slight CO2 drawdown whereas equatorward shifts tend to elevate atmospheric CO2 (Figure 8d). The concentration does not drop below 260 ppm when poleward shifts are imposed. Upon equatorward displacements of up to 10° we simulate a maximum CO2 concentration of 315 ppm. CO2 in the RBC and RBC lowdiff cases responds the least to wind shifts (black lines in Figure 8d). The setups with rather strong AMOC (blue and green lines) also display a rather low sensitivity. Most responsive are weakFw and weakFw lowdiff (red lines), the only setups showing major circulation rearrangements upon wind shifts.

3.3. Decomposition of CO2 Changes Into Physical and Biological Components

[48] To further understand the mechanisms controlling the CO2 response, the simulated CO2 changes are split into a “physical” and “biological” contribution as described below.

[49] Sinking organic particles transport carbon and nitrate to depth and thus deplete the sea surface in DIC while enriching it alkalinity. The organic matter cycle hence lowers sea surface pCO2 and increases the ocean's carbon storage capacity compared to a world without marine biota. Nutrient utilization efficiency (NUE) is a frequently utilized measure to characterize the impact of the organic matter cycle on marine carbon cycling [Sigman and Boyle, 2000]:

display math

equation imagesurface and equation imagedeep are the average PO4 concentrations in the surface and the deep ocean respectively.

[50] In contrast to the organic matter cycle, the production, sinking and redissolution of CaCO3 shells act to deplete sea surface alkalinity and to increase sea surface pCO2. To assess the effect of the CaCO3 cycle on CO2 partial pressures in the surface ocean we define similarly to NUE a CaCO3 cycle efficiency (CCE):

display math

PAlk stands for potential alkalinity which is defined following Brewer et al. [1975]:

display math

[51] The second term on the right hand side of equation (4) represents a correction term to remove the influence of the organic matter cycle on the alkalinity distribution such that the distribution of PAlk is only affected by the cycling of biogenic CaCO3.

[52] Toggweiler et al. [2006] restore surface PO4 concentrations toward a fixed field to determine export production. However, reduced upwelling of remineralized carbon is accompanied by decreased nutrient supply to the surface and a likely change in NUE. Here, prognostic formulations of biological cycling are used (see auxiliary material) allowing surface nutrient fields and NUE to alter upon wind stress perturbations.

[53] The impact of changes in NUE and CEE on atmospheric CO2 is quantified by an additional factorial run. Euphotic zone concentrations of PO4 and potential alkalinity (PAlk) are instantaneously restored toward seasonally varying fields that were previously diagnosed in the initial steady state. In doing so, NUE and CEE remain largely unchanged. Note that equation (4) implies that this procedure also keeps sea surface alkalinities essentially unmodified when the model state is perturbed.

[54] Here, we arbitrarily term the modeled changes in atmospheric CO2 resulting from the simulations with PO4− and PAlk-restoring “physical” (ΔpCO2,phys). These are induced by changes in the interplay of circulation, CO2 solubility and carbon chemistry, air-sea gas exchange, and the cycling of organic material, calcite and silicic acid for maintained NUE and CEE.

[55] The difference in ΔpCO2atm between a nutrient restoring simulation (Δ pCO2,phys) and the corresponding run with fully prognostic biology is associated with Δ pCO2,bio such that:

display math

ΔpCO2,bio thus reflects CO2 changes associated with altered NUE and CEE.

3.3.1. Scaling Experiments

[56] When the SHW are intensified physical and biological mechanisms play equally important roles in controlling CO2 changes (Figures 7b and 7c). In the case of a threefold intensification (ɛ = 3.0) the physical component ΔpCO2,phys explains 15 to 40 ppm of the CO2 rise, largely because of altered circulation in the upper Pacific. A large fraction of this ocean region is ventilated by the clockwise overturning cell in the upper North Pacific when subjected to strong SHW (section 3.1.1). Cold and carbon-rich water masses stemming from the deep ocean get replaced by warm and well-ventilated waters that originate from the Pacific surface.

[57] ΔpCO2,bio contributes 6 to 40 ppm to the modeled CO2 increase in the case of very strong winds (ɛ = 3.0; Figure 7d). A decrease in NUE in the SO (south of 34°S) is mainly responsible. The rate of exchange between deep and surface waters in the SO increases and thus the supply of nutrients to the surface is enhanced which results in increased export production. Yet only a fraction of the increase in nutrient supply is utilized by marine biology and thus sea surface PO4 and DIC concentrations rise. This implies that the marine organic matter cycle becomes less efficient in reducing sea surface pCO2 when SHW are intensified. Changes in CaCO3 cycle efficiency are small (Figure 7e) such that the overall biological response is dominated by the changes in the organic matter cycle which lead to an increase in atmospheric CO2.

[58] As mentioned earlier (section 3.1.2), the various setups respond differently to weakened SHW. The simulations leading to a substantial CO2 drawdown display a significant decrease of surface pCO2 in the SO (RBC, RBC lowdiff, weakFw and weakFw lowdiff; black and red lines in Figure 7a). Such behavior is absent in the scenarios with a modest CO2 response (standard, lowdiff, strongFw and strongFw lowdiff; blue and green lines in Figure 7a).

[59] Most setups display a very small physical CO2 response when weak SHW are applied (Figure 7b). Only the RBC simulations (black lines) show a distinctly negative ΔpCO2,phys of around −18 ppm for ε = 0.25. This CO2 drawdown is associated with enhanced carbon storage in the upper Pacific and in the deep North Atlantic which results from a slowing and shoaling of the respective overturning cells.

[60] NUE in the SO increases monotonically in all the setups when the SHW are reduced in strength (Figure 7d). However, only in the setups RBC, RBC lowdiff, weakFw and weakFw lowdiff this NUE increase leads to a substantially negative Δ pCO2,bio of 25 to 45 ppm for ε = 0.25 (black and red lines in Figure 7c). In the other setups (standard, lowdiff, strongFw and strongFw lowdiff) an increase in CEE counteracts the effect of enhanced NUE resulting in a modest ΔpCO2,bio of −13 to +2 ppm when ε is set to 0.25 (blue and green lines in Figure 7c). In these simulations, stagnant water masses with low Δ14C (blue and green lines in Figure 6e, section 3.1.1) emerge in the deep Pacific which induce major rearrangements in the cycling of DIC and alkalinity.

[61] In our model more than 85% of particulate organic matter is remineralized in the top 1000 meters, whereas about half of the CaCO3 shells dissolve below a depth of 2500 m. For this reason, the CaCO3 cycle is more strongly affected by the deep stagnant water masses than the organic matter cycle. The enhanced dissolution of CaCO3 leads to an increase in alkalinity at depth and, as a consequence, to a depletion of alkalinity at the surface and in the thermocline, and thus to higher values of CEE when SHW are weakened (blue and green lines in Figure 7e). Low κdia leads to a more efficient alkalinity accumulation in the deep since upward diffusive transport is limited.

[62] In summary, an increase in SHW strength consistently leads to an increase in atmospheric CO2 for all model setups in the range of 20 to 40 ppm per 100% change in wind stress strength. Decreased nutrient utilization efficiency and enhanced ventilation increase atmospheric CO2 over the range of model setups. In contrast, the CO2 response to weakened winds is sensitive to the model setup. Simulated changes are in the range of −2 to −55 ppm for a fourfold reduction in wind stress.

3.3.2. Shift Experiments

[63] The response in the carbon cycle to shifts in the position of the SHW is diverse for the different setups. The model setups with RBC show almost no sensitivity in atmospheric CO2 when the SHW are shifted. In the setups with MBC, however, quite a robust trend emerges in the sense that poleward shifts induce a slight reduction in atmospheric CO2 whereas equatorward shifts lead to a modest CO2 increase (Figure 8d). Both physical and biological mechanisms contribute toward the simulated changes in atmospheric CO2. ΔpCO2,phys and Δ pCO2,bio are generally of similar magnitude and of the same sign (not shown).

[64] The results show that the position of the Deacon cell [Döös and Webb, 1994] is tightly attached to the westerlies' zonal maximum. Equatorward wind shifts lead to northward shifts of this cell and of the associated fronts and hence increase the outcrop area. Accordingly, poleward displacements reduce the area where upwelled deep waters are exposed to the atmosphere. These circulation rearrangements in the SO are also manifest in modeled surface PO4. The area of elevated PO4 concentration in the SO decreases in the case of poleward wind shifts and increases for equatorward wind shifts.

[65] Preindustrial air-sea fluxes of CO2 are characterized by outgassing in the SO (south of 44°S), uptake at midlatitudes of both hemispheres (18–44°S and 18–49°N) and strong outgassing in the tropics (18°N–18°S). This pattern is relatively well captured by means of inverse estimates in a number of ocean models including the Bern3D model [Mikaloff Fletcher et al., 2007]. In the wind shift experiments the most prominent changes in sea surface pCO2 occur in a latitudinal belt at around 40°S, the location of the northern edge of the modern Deacon cell. Shifts toward the equator elevate pCO2 and displacements toward the pole lead to a pCO2 reduction in that region. The model results thus suggest that the zone of CO2 outgassing in the SO contracts upon poleward wind shifts and expands upon equatorward shifts, explaining the simulated changes in ΔpCO2,phys.

[66] The biological contribution ΔpCO2,bio is dominated by NUE changes in the setups with standard diapycnal diffusivity κdia and by CEE changes in the setups with low κdia. In the former an increase in the Deacon overturning under equatorward shifts increases both nutrient supply and export production in the SO by about 20%. The result is a slight reduction in NUE in the SO (not shown). In the latter reduced SO ventilation and decreased deep upwelling in the Pacific enhance CEE when the SHW are shifted toward the equator (Figure 8e).

3.4. Additional Model Sensitivities

[67] To examine further model sensitivities, in addition to halving the SH westerlies' amplitude, we separately apply to the setup lowdiff the effects of (1) increased glacial CO2 solubility resulting from changes in SST and SSS, (2) hampered air-sea gas exchange in the SO due to enhanced sea ice cover, (3) CaCO3 compensation, and (4) a 500 GtC carbon input into the atmosphere stemming from the terrestrial biosphere [e.g., Shackleton, 1977] (Table 4, Exp2, Exp3, Exp4 and Exp5; for details of these simulations see auxiliary material). Finally, we take into account all of these mechanisms simultaneously (Table 4: Exp6). Table 4 summarizes the simulations and the results. The changes in the DIC distribution due to a halving of the SHW (Exp1-control) are illustrated in Figures 9a and 9b, the DIC changes in response to the additional processes (Exp6-Exp1) are depicted in Figures 9c and 9d.

Figure 9.

Changes in zonal mean DIC (μmol kg−1) (a and b) due to a halving of the SHW amplitude and (c and d) in response to modified CO2 solubility, sea ice cover, CaCO3 compensation, and a 500 GtC carbon input into the atmosphere. Zero lines are shown as thick black contours.

Table 4. Additional Carbon Cycle Sensitivitiesa
ScenarioSH Westerlies ×0.5CO2 SolubilitySea Ice CoverCaCO3 Compensation500 GtC Terrestrial InputAtmospheric CO2 (ppm)POM Export (GtC a−1)CaCO3 Export (GtC a−1)NUE (%)CEE (%)Net CaCO3 Flux From Sediments (GtC)
  • a

    The mechanisms considered in the different simulations are listed along with the results obtained at steady state.

control     27810.960.9879.412.69-
Exp1x    2567.540.6884.612.68-
Exp2xx   2327.660.6984.312.51-
Exp3x x  2617.620.6884.412.65-
Exp4x  x 2537.620.6884.4-74
Exp5x   x2947.560.6884.612.68-
Exp7xxxx 2337.620.6884.4-87

[68] When the four additional effects are taken into account simultaneously (Exp6) we achieve a total CO2 drawdown of 22 ppm to 256 ppm. Increased biological efficiencies (ΔpCO2,bio = −24 ppm), enhanced CO2 solubility (−24 ppm) and CaCO3 compensation (−15 ppm) are mainly responsible for the CO2 reduction. On the other hand, the terrestrial carbon input (+38 ppm, without CaCO3 compensation), physical rearrangements (ΔpCO2,phys = +2 ppm) and hampered gas exchange in the SO due to extended sea ice coverage (+5 ppm) elevate CO2.

[69] The halving of the SH westerly winds' amplitude induces an AMOC reduction of about 8 Sv and deteriorates the ventilation in the upper Pacific and the Southern Ocean (blue dashed lines in Figures 6a, 6c, and 6d for ε = 0.5). Reduced nutrient supply to the surface decreases POM export production by about 30%, but increases global NUE by 5% (Table 4). Export of biogenic CaCO3 also decreases, but CEE remains virtually unchanged. The induced changes in marine carbon cycling lead to a drawdown of 22 ppm (Exp1). In this simulation, deep ocean alkalinity and DIC increase roughly in concert such that their difference, which approximates the CO32− concentration, changes only marginally. For this reason, we simulate only a small additional response of −3 ppm when CaCO3 compensation is considered in addition to the halving of the SHW (Exp4).

[70] The modeled drawdown of 24 ppm due to the solubility effect (Exp2) is very close in magnitude to the estimate of 23.5 ppm given by Sigman and Boyle [2000] which was calculated using the CYCLOPS ocean box model [Keir, 1988] and a rough guess of glacial SST and SSS. As found in other GCMs [Archer et al., 2003], increased sea ice cover in the SO leads to higher atmospheric CO2 (Exp3). Enhanced sea ice cover excludes surface waters with low pCO2 from gas exchange such that the new atmospheric CO2 equilibrium is shifted toward a higher pCO2. The input of 500 GtC carbon into the atmosphere induces a CO2 increase of 38 ppm without CaCO3 compensation and 23 ppm with it. The first value is in good agreement with an estimate of 37 ppm for the uncompensated case [Archer et al., 2000a], the latter is somewhat higher than the 17 ppm estimate reported in that same paper for the compensated case.

[71] In a further experiment (Exp7), all of the additional effects described above were considered, but the terrestrial carbon input was excluded. In this case CO2 was 233 ppm. The total simulated 45 ppm drawdown is of similar magnitude as the 43 ppm reduction modeled by Brovkin et al. [2007] in a comparable model configuration. In their study glacial SST, SSS, ocean circulation and sea ice cover were obtained by prescribing a complete set of glacial forcings to the CLIMBER-2 intermediate complexity Earth system model.

[72] Changes in oceanic DIC concentrations resulting from the wind weakening and associated changes in biological efficiencies (Figures 9a and 9b) are relatively large and spatially heterogeneous. The concentrations increase in the upper Pacific and in the Atlantic at intermediate depths (2–3 km). In the intermediate and deep Pacific, as well as in the upper Atlantic, carbon storage is reduced. Changes in the DIC distribution induced by increased solubility, enhanced SO sea ice cover, CaCO3 compensation and a 500 GtC carbon input into the atmosphere are generally higher in the ocean interior than near the surface (Figures 9c and 9d). Most of the additional carbon that is taken up by the ocean in response to these four mechanisms is transported by NADW.

4. Summary and Conclusions

[73] The focus of this study is a hypothesis that calls for reduced upwelling of CO2-rich deep waters in the SO resulting from an equatorward shift of the SHW to explain low glacial CO2 levels in the atmosphere [Toggweiler et al., 2006]. We perform a systematic sensitivity analysis of the large-scale ocean circulation, marine carbon cycling and atmospheric CO2 to changes in the strength and position of the SHW to probe this mechanism in the Bern3D coarse-resolution ocean model.

[74] Mechanisms of CO2 variations are quantified by targeted sensitivity simulations and analyses. The wide range of model setups and wind stress scenarios applied resulted in a variety of simulated physical and biogeochemical responses. These include strongly reduced deep SO overturning and low ventilation of the deep ocean, corresponding to the weak southern circuit suggested by Toggweiler et al. [2006]. The results, in combination with available evidence on glacial winds, suggest that SH wind changes played a limited role in modulating glacial-interglacial atmospheric CO2 concentrations. In addition, and in further conflict with Toggweiler et al. [2006], an equatorward shift in SHW results in a CO2 increase and not in the hypothesized decrease.

[75] The Bern3D model is able to adequately represent ocean circulation and the cycling of major biogeochemical tracers (see section 2.2). Yet, as with any modeling study, a number of caveats apply to our results. Although the preindustrial distribution of DIC is reasonably simulated compared to data-based estimates (see Figure 1c and Table 3), there remains some uncertainty related to the role of gas exchange and the solubility pump (see section 2.2). Instead by an interactive atmosphere, SST and SSS are constrained by applying restoring or mixed boundary conditions. However, our results are in good agreement with a recent study by De Boer et al. [2008] who examine the response of the meridional overturning circulation to changes in the SHW strength using an OGCM coupled to an energy moisture balance model for the atmosphere. Furthermore, the authors emphasize the important role of winds in stirring salinity gradients in the ocean. In our study, salinity-circulation feedbacks are accounted for under changes in wind stress when mixed boundary conditions are applied.

[76] Eddy fluxes are believed to play a more central role in the dynamics of the SO than in other areas of the world ocean [Rintoul et al., 2001]. Eddies are not represented explicitly in the Bern3D model. Furthermore, ocean-ice interactions in the SO are thought to be crucial for the formation of deep waters around Antarctica and convection is only represented in a highly parameterized way.

[77] A reduction in the strength of SHW leads to a decline in simulated atmospheric CO2. Ventilation in the SO is reduced by a decrease in the frequency and depth of convective events. In addition, biological mechanisms contribute toward the modeled CO2 drawdown which is up to 34 ppm for a 50% reduction in SHW strength. A 50% increase on the other hand leads to a rise in atmospheric CO2 of up to 24 ppm.

[78] When the SHW are shifted in their latitudinal position the meridional overturning circulation, marine carbon cycling and atmospheric CO2 respond relatively insensitively. A poleward shift of 5° lowers CO2 by up to 16 ppm whereas an equatorward shift of 5° induces a CO2 increase of maximum 14 ppm. The wind shifts do not substantially alter ventilation in the SO. Rather they affect atmospheric CO2 by changing the outcrop area of deep waters. As in the wind-scaling simulations changes in marine biology tend to enforce the CO2 response that is induced by physical mechanisms.

[79] In additional sensitivity runs, a limited set of mechanisms for low CO2 is quantified. We model a drawdown of atmospheric CO2 to 256 ppm from a preindustrial value of 278 ppm when prescribing reduced wind stress in the SO, enhanced CO2 solubility, reduced gas transfer velocities by sea ice expansion and a 500 GtC carbon input into the atmosphere due to reduced terrestrial carbon storage in glacial times. An increase in biological efficiencies (−24 ppm) and the enhanced CO2 solubility (−24 ppm) are mainly responsible for the CO2 reduction. CaCO3 compensation in response to reduced wind stress alone is small (−3 ppm). However, the reaction of CaCO3 sediments reduces the atmospheric CO2 rise due to the 500 GtC carbon input from 38 ppm to 23 ppm. The simulated amplitudes of the additional effects are generally in good agreement with those reported in previous studies [e.g., Brovkin et al., 2007; Archer et al., 2000a; Sigman and Boyle, 2000]. Additional mechanisms than the limited set considered here are required to account for the full glacial-interglacial CO2 change [see, e.g., Broecker and Henderson, 1998; Sigman and Boyle, 2000; Archer et al., 2000a; Kohfeld et al., 2005].

[80] Physical as well as biological oceanic controls on CO2 should be taken into account simultaneously in scenarios of low glacial atmospheric CO2. Decreased ventilation in the SO reduces the supply of nutrient-rich deep waters to the surface potentially increasing the efficiency of the organic matter cycle in drawing down surface pCO2. In contrast, a reduction in deep ocean ventilation might also lead to a substantial increase of deep ocean alkalinity in response to a more efficient CaCO3 cycle. This would reduce surface alkalinity and hence increase surface pCO2. These effects of the organic matter and CaCO3 cycle are antipodal with respect to atmospheric CO2 and their respective amplitudes depend on the details of the glacial circulation. We have assessed these biological implications on atmospheric CO2 using the concepts of nutrient utilization efficiency (NUE) and CaCO3 cycle efficiency (CCE).

[81] The idea of changes in deep ocean ventilation as being an important reason for increased CO2 storage in the glacial ocean [Toggweiler, 1999] appears tempting given the available data [e.g., Duplessy et al., 1988; Adkins et al., 2002; Hodell et al., 2003; Marchitto et al., 2007]. However, our results indicate that mechanisms other than wind changes were probably more relevant for altering the ventilation of the ocean's deep layers. For instance, Watson and Naveira Garabato [2006] propose that mixing in the SO was slower because denser deep waters were formed and a decreased buoyancy flux at the surface suppressed upwelling. Sigman et al. [2004] suggest that the cooling caused increased stratification by weakening the role of temperature in polar ocean density structure. If variations in the SHW are invoked as the main mechanism for the glacial CO2 drawdown a reduction in SHW strength of at least 50% would be required. Such a weakening in SO winds appears rather unlikely given the available reconstructions discussed in section 2.3 [e.g., Shulmeister et al., 2004; Otto-Bliesner et al., 2006; Rojas et al., 2008]. These conclusions are supported by a recent study that is based on the LOVECLIM earth system model of intermediate complexity [Menviel et al., 2008].


[82] This study was funded by the Swiss National Science Foundation and the EU projects CARBOOCEAN (511176-2) and EUROCEANS (511106-2). P.P. acknowledges support from the Marie Curie International Incoming Fellowship. We thank N. R. Edwards, S. A. Müller, and T. F. Stocker for helpful comments. Critical reviews from V. Brovkin, J. R. Toggweiler, and one anonymous reviewer are acknowledged.