Variability of the ocean carbon cycle in response to the North Atlantic Oscillation

ABSTRACT Climate modes such as the North Atlantic Oscillation (NAO), representing internal variability of the climate system, influence the ocean carbon cycle and may mask trends in the sink of anthropogenic carbon. Here, utilising control runs of six fully coupled Earth System Models, the response of the ocean carbon cycle to the NAO is quantified. The dominating response, a seesaw pattern between the subtropical gyre and the subpolar Northern Atlantic, is instantaneous (<3 months) and dynamically consistent over all models and with observations for a range of physical and biogeochemical variables. All models show asymmetric responses to NAO+ and NAO− forcing, implying non-linearity in the connection between NAO and the ocean carbon cycle. However, model differences in regional expression and magnitude and conflicting results with regard to air–sea flux and CO2 partial pressure remain. Typical NAO-driven variations are ±10 mmol/m3 in the surface concentration of dissolved inorganic carbon and alkalinity and ±8 ppm in the air–sea partial pressure difference. The effect on the basin-wide air–sea CO2 flux is small due to compensating fluxes on the sub-basin scale. Two models show a reduced carbon sink in the north-eastern North Atlantic during negative NAO phases, qualitatively in accordance with the observed decline during a phase of predominantly negative NAO. The results indicate that wind-driven dynamics are the main driver of the response to the NAO, which – via vertical mixing, upwelling and the associated entrainment of dissolved inorganic carbon and nutrients – leave an imprint on surface pCO2 and the air–sea CO2 flux as well as on biological export production, pH and the calcium carbonate saturation state. The biogeochemical response to the NAO is predominantly governed by vertical exchange between the surface and the thermocline; large-scale horizontal transport mechanisms are of minor importance.


Introduction
The North Atlantic Ocean, constituting the second largest natural carbon sink right after the Southern Ocean, is of eminent importance for the global carbon cycle (e.g.Sabine et al., 2004, Gerber andJoos, 2010).However, the correct assessment and interpretation of the North Atlantic carbon system is challenging since (1) observations are scarce and limited either in time or space and (2) the climate system in this area features significant modes of natural variability, e.g. the North Atlantic Oscillation (NAO).
Here, utilising control runs of six Earth System Models (NCAR CSM1.4 and CCSM3, GFDL ESM2.1, IPSL-CM4-LOOP, BCCR BCM-C and MPI-ESM), the natural variability of the North Atlantic carbon cycle in response to the NAO is investigated.The purpose of this study is to identify the underlying mechanisms behind the climatecarbon system in the North Atlantic which, hopefully, helps in interpreting observations, i.e. making the distinction between natural variability and anthropogenically forced climate change.
In the last 200 yr, mankind emitted large amounts of carbon dioxide through (1) combustion of fossil fuels (oil, carbon and natural gas), (2) land-use change (mainly largescale deforestation) and ( 3) industrial processes such as cement production (e.g.Hegerl et al., 2007).During this period, starting with the industrialisation and sometimes referred to as 'Anthropocene' (Crutzen and Stoermer, 2000), the global carbon cycle has been brought out of steady state.Due to the continuous carbon emissions, atmospheric CO 2 has increased from about 280 ppm (preindustrial) to 391.05 ppm (September 2012), a concentration higher than ever measured in ice core records covering the past 800 000 yr (Lu¨thi et al., 2008).In combination with other forcing agents (e.g.CH 4 and N 2 O), these CO 2 concentrations have pushed the global climate system towards conditions that have probably not occurred over the past 20 million years (Blackford and Gilbert, 2007) and at a speed unprecedented at least during the last 22 000 yr (Joos and Spahni, 2008).The atmospheric CO 2 concentrations illustrate the complex nature of the global carbon cycle Á from the 257 PgC emitted to the atmosphere between 1960 and 2007 (Boden et al., 2010), only 56% remained airborne.The estimated oceanic uptake is 33%, and the remaining 11% are assumed to be taken up by the terrestrial biosphere (Sarmiento et al., 2010).Since carbon cycle and climate system are in a state of mutual dependency and steady interaction, a sound knowledge of the processes taking place is necessary to correctly detect past and present trends and, based on this, make robust projections for the future.In this context, it is essential to take into account that long-term trends and natural variability have the potential to enhance or mask each other (e.g.Latif et al., 1997;Fro¨licher et al., 2009;Dolman et al., 2010;McKinley et al., 2011), making it sometimes difficult to draw clear conclusions.
The NAO is a large-scale climate phenomenon and the dominating mode of weather and climate variability in the Northern Atlantic.Constituting a seesaw in atmospheric mass between a subtropical high-pressure system and a subpolar low-pressure system (Bjerknes, 1964), NAO involves large changes in air surface temperature, wind, storminess and precipitation.In the ocean, NAO implies changes in heat content, gyre circulations, mixed-layer depth, salinity, high latitude deep water formation and sea-ice cover (Hurrell and Deser, 2009).A positive NAO phase (i.e. an increased pressure gradient between the two systems) is linked to positive sea-surface temperature (SST) anomalies in the subtropics, negative anomalies in the subpolar gyre (Cayan, 1992;Cullen et al., 2001) and the corresponding winter mixing: due to changes in ocean convection, regions with warmer SSTs experience less vigorous winter mixing than normal and vice versa (Dickson et al., 1996).Mechanisms and the current state of knowledge with respect to the NAO are given in review articles (e.g.Wanner et al., 2001, Pinto andRaible, 2012 andreferences therein).
A number of modelling studies have addressed the connection between NAO and North Atlantic Ocean carbon cycle utilising global ocean general circulation models (McKinley et al., 2004;Thomas et al., 2008;Ullman et al., 2009;Levine et al., 2011;Tjiputra et al., 2012), coupled ecosystem-circulation models of the North Atlantic (Friedrich et al., 2006;Lo¨ptien and Eden, 2010) and a single atmosphere-ocean general circulation model (AOGCM) (Patara et al., 2011).In accordance with observations (e.g. Bates et al., 2002), these modelling studies identified variations in wind-induced mixing and SST as the driving mechanisms of the response of ocean biogeochemistry.In the subpolar gyre, the carbon cycle in the upper ocean is mainly governed by mixing-driven entrainment of dissolved inorganic carbon (DIC) from nutrient-enriched deeper layers to the surface; airÁsea gas exchange is of minor importance.In the subtropics, changes in SST are the dominating factor.Matter of debate are the importance of the biological pump (e.g.McKinley et al., 2004, Lo¨ptien andEden, 2010) and advective transport processes during periods of persistent negative or positive NAO forcing (e.g.Thomas et al., 2008).
In the next section, the six models are described and evaluated.In the result section, first the response of ocean physics and biogeochemistry to the NAO is presented.Then, the results are compared with observations.In the discussion section, the results are summarised and compared with observations and earlier model studies.Finally, conclusions are given.

Models
This study is based on simulations performed with six fully coupled 3-D atmosphere-ocean climate models, each of them including additional carbon cycle modules covering the terrestrial and oceanic biosphere.An overview is given in Table 1.2.1.1.NCAR CCSM3-BEC.The Community Climate System Model Version 3 (CCSM3), based on the framework of the Community Climate System Model (CCSM) project (Blackmon et al., 2001), was developed by the National Center for Atmospheric Research (NCAR) in Boulder, USA.The 'standard' model includes components for atmosphere, ocean, sea ice and land surface, all linked by a coupler exchanging fluxes and state information between the components (Collins et al., 2006a).The atmospheric component, Community Atmosphere Model version 3 (CAM3; Collins et al., 2006b), has a horizontal resolution of approximately 3.758 and 26 vertical levels.The land component, Community Land Surface Model version 3 (CLM3; Oleson et al., 2004), possesses the same horizontal grid as the atmosphere.The ocean component is represented by the Parallel Ocean Program (POP) ocean general circulation model (Smith and Gent, 2004) and integrated on a dipole grid.The horizontal resolution is on average 3.68 in longitude and 0.88 to 1.88 in latitude, and the vertical dimension features 25 unevenly distributed levels (5 levels for the top 100 m, 13 for the top 1000 m) extending to a depth of 4750 m.The sea-ice component, Community Sea Ice Model version 5.0 (CSIM5; Briegleb et al., 2004), features the same horizontal grid as the ocean model.Ocean biogeochemistry is represented by the Biogeochemical Elemental Cycling Model (BEC), consisting of modules for the upper ocean ecology (Moore et al., 2004) and full-depth biogeochemistry (Doney et al., 2006).The marine ecosystem component is based on traditional phytoplankton-zooplankton-detritus-nutrient food web models (e.g.Fasham et al., 2001) and includes (1) the nutrients nitrate, ammonium, phosphate, dissolved iron and silicate, (2) four groups of phytoplankton, (3) zooplankton, (4) dissolved organic matter (DOM) and ( 5) sinking particulate detritus.CaCO 3 production is included in phytoplankton.The biogeochemical module comprises the cycling of carbon, oxygen, nitrogen, phosphorus, silicon and iron.
2.1.2.NCAR CSM1.4-carbon.The CSM1.4-carbon climate model (Fung et al., 2005;Doney et al., 2006) was developed by the NCAR in Boulder, USA.The fundament of the model is a modified version of the NCAR CSM1.4 coupled physical model, consisting of ocean, atmosphere, land and sea-ice components linked by a flux coupler (Boville et al., 2001).The atmospheric component CCM3 has a horizontal resolution of 3.758 and 18 vertical levels (Kiehl et al., 1998).The ocean component is the NCAR CSM Ocean Model (NCOM) with 25 levels in the vertical and a resolution of 3.68 in longitude and 0.88 to 1.88 in latitude (Gent et al., 1998).Similar to the CCSM3, the components for sea ice and ocean as well as land surface and atmosphere run on the same horizontal resolution.The CSM1.4-carbon model comprises a modified version of the terrestrial biogeochemistry model CASA (Carnegie-Ames-Stanford Approach; Randerson et al., 1997) and a prognostic version of the OCMIP ocean biogeochemistry model (Najjar et al., 2007).The ocean biogeochemical model includes the main processes of the organic and inorganic carbon cycle within the ocean and airÁsea CO 2 flux.The biological productivity within the ocean model is controlled by temperature, solar irradiance, mixed-layer depth, phosphate and iron.(Delworth et al., 2006).In a fully coupled state, the model includes components for ocean (MOM-4;Griffies et al., 2004), atmosphere (AM2.1) and terrestrial biosphere (LM3; Anderson et al., 2004).AM2.1 has a horizontal resolution of 2.58 longitude by approximately 28 latitude and a vertical resolution of 24 levels.LM3 is based on potential vegetation, so no land use is included.MOM-4 has 50 vertical levels (22 10-m levels in the upper 220 m), its spatial resolution is nominally 18 globally, with a higher resolution of 1/38 near the equator.Ocean biogeochemistry is represented by TOPAZ (Dunne et al., 2005;Dunne et al., 2007) and includes a number of major nutrients (N, P, Si and Fe), both labile and semi-labile dissolved organic pools and parameterisations to represent the microbial loop (Henson et al., 2010).The ocean ecosystem is based on three classes of phytoplankton, and biological productivity is modelled as a function of chl:C ratios and limited by nutrients and light.Also a simplified version of the ocean iron cycle is included, comprising biological uptake and remineralisation, particle sinking and scavenging and adsorption/desorption.

MPI-ESM.
The Earth System Model run at the Max-Planck-Institute for Meteorology (MPIM) in Hamburg, Germany, consists of the atmospheric model EC-HAM6 with embedded Jena Scheme for BiosphereÁ Atmosphere coupling in Hamburg (JSBACH) terrestrial biosphere model and the MPIOM physical ocean model, which further includes a sea-ice model (Marsland et al., 2003).Biogeochemistry is represented by the Hamburg Ocean Carbon Cycle (HAMOCC) 5.1 marine biogeochemistry model (Maier-Reimer et al., 2005), which is implemented into the MPIOM ocean model and integrated on a coordinate system with a nominal resolution of 1.58.The North Pole is located over Greenland, allowing a good coverage of the Nordic Seas.The vertical resolution is 40 levels, with higher resolution in the upper part of the water column (10 m at the surface to 13 m at 90 m).The coupling between the different components is done by using the OASIS coupler (Terray et al., 1995).The concept behind HAMOCC5.1 is to address large-scale and longterm features of the marine carbon cycle, and it is not designed to give a complete description of the marine ecosystem.Marine biological productivity is temperature dependent following Eppley (1972) and is limited by the availability of phosphorous, nitrate, iron and light.The export production is split between opal and calcite producers depending on the availability of silicate.
2.1.5.BCCR BCM-C.The Bergen Earth system model (BCM-C; Tjiputra et al., 2010) was developed at the Bjerknes Centre for Climate Research (BCCR) in Bergen, Norway.The atmosphere is represented by the spectral atmospheric general circulation model ARPEGE from Me´te´o France (Furevik et al., 2003) with a horizontal resolution of about 2.88 )2.88.A total of 31 vertical levels is employed, ranging from the surface to 0.01 hPa (20 layers in the troposphere).The ocean component is an updated version of the Miami Isopycnic Coordinate Ocean Model (MICOM; Bleck et al., 1992), featuring horizontal grid spacing of approximately 2.48 )2.48 and 34 isopycnic vertical layers.For a better resolution of the dynamics near the equator, the horizontal spacing in the meridional direction is gradually decreased to 0.88 along the equator.The sea-ice model is the original model integrated into the MICOM ocean model, consisting of one ice and one snow layer assuming a linear temperature profile in each layer.All the components are, similar to MPIM, linked using the OASIS coupler.The ocean biogeochemistry component is the same as in the MPIM model, HAMOCC5.1 (Maier-Reimer et al., 2005), but adopted to and re-tuned for use on isopycnal coordinates and the flow field of the MICOM dynamical model (Assmann et al., 2010).The terrestrial biogeochemistry is represented by the Lund-Postdam-Jena Model (LPJ; Sitch et al., 2003), a large-scale terrestrial carbon cycle model with a horizontal resolution of approximately 2.58)2.58.It implements dynamical vegetation (including terrestrial photosynthesis, respiration, resource competition, tissue turnover, dynamic vegetation population with 10 plant functional types (PFTs), soil organic and litter dynamic as well as natural fire occurrence) but does not take land-use change into account.
2.1.6.IPSL-CM4-LOOP.The IPSL-CM4-LOOP model was developed at the Institut Pierre-Simon Laplace (IPSL), Paris, France.Its component units are the Laboratoire de Me´te´orologie Dynamique atmospheric model (LMDZ-4) with a horizontal resolution of approximately 38)38 and 19 vertical levels (Hourdin et al., 2006), coupled to the OPA-8 ocean model with a horizontal resolution of 28 )28 and 31 vertical levels and the LIM sea-ice model (Madec et al., 1998).The land biosphere is represented by the global vegetation model ORCHIDEE (Krinner et al., 2005) and the marine carbon cycle by the PISCES model (Aumont et al., 2003).PISCES includes the cycling of carbon, oxygen and the major nutrients, two phytoplankton size classes (nanophytoplankton and diatoms) and two zooplankton size classes (microzooplankton and mesozooplankton).Phytoplankton growth is limited by the availability of nutrients, temperature and light.

Model evaluation
Since the focus of the study is on the intrinsic variability of the climate-carbon system, the analysis is based on control runs; that is, CO 2 emissions are set to zero and other forcings are set to constant preindustrial levels.A prerequisite when working with control runs is the assumption that the models are in quasi-steady state.Since the spin-up time is not necessarily an appropriate measure to verify this assumption, we evaluated the drift of the model runs for each grid cell by linear regression.In our area of interest, the North Atlantic (108 to 808N and 808W to 208E), the models show maximal surface trends in DIC (mmol/m 3 /decade) and SST (8C/decade) of 0.04/0.00(CCSM3), 0.05/0.00(CSM1.4),0.04/0.00(GFDL), 0.56/0.02(BCCR), 0.00/ 0.00 (MPIM) and 0.00/0.00(IPSL).The comparably high values for BCCR originate from a few grid cells at the northern boundary of the region while the drift is much smaller in most of the North Atlantic.Overall the drift is negligible.For a subset of the models, an evaluation of historical simulations can be found in Schneider et al. (2008) and Roy et al. (2011).Figure . 1 shows the annual mean differences between control simulations and observations of the variables DIC, alkalinity and PO 4 .In this context the absence of forcing in the model data has to be accounted for.This holds in particular for the concentrations of DIC, which have been increasing over the anthropocene due to the uptake of anthropogenic CO 2 .Also to be considered has the fact that the observational data is likely not representing real annual averages.Both points confine a simple one-toone comparison.Compared to observations of DIC and alkalinity (Key et al., 2004), four models match the observed pattern with only small deviations.Two models (BCCR and IPSL) tend to overestimate the surface concentrations of these two variables in large parts of the North Atlantic.For PO 4 (Garcia et al., 2010), four out of six models show too low concentrations in the northern North Atlantic.Towards the south, five out of six models overestimate PO 4 .One reason for this discrepancy between the models is the implementation of the iron cycle.In CSM1.4 and MPIM, the positive offset of PO 4 is caused by a too strong iron limitation of biological production (Schneider et al., 2008;Steinacher et al., 2010).However, all models reasonably capture the observed large-scale patterns of the concentrations.
To investigate the interannual variability, the standard deviation along the annual means was calculated (not shown).For all models and all three variables the highest variability can be found in the subpolar gyre and the adjacent northern regions.However, the magnitude of these large-scale variations differs significantly between the models.The standard deviations of DIC (mmol/m 3 ), Alk (mmol/m 3 ) and PO 4 (mmol/m 3 ) reach values of up to 20/25/0.06 (CCSM3), 25/30/0.09(CSM1.4),6/10/0.05(GFDL), 10/15/0.02(BCCR), 20/25/0.05(MPIM) and 20/ 25/0.12 (IPSL).For the two NCAR models, the annual means of the control runs were compared with historical simulations (years 1980Á2009).The spatial patterns of all three variables are similar in both simulations.The same holds true for the concentration of PO 4 in the CCSM3, and the CSM1.4 however shows lower historical PO 4 levels in the subpolar gyre (up to (0.1 mmol/m 3 ).Surface DIC is higher in both historical simulations, up to 60 mmol/m 3 in the subtropical and 35 mmol/m 3 in the subpolar gyre.Historical surface Alk concentrations are slightly higher (up to 25 mmol/m 3 ) in the subtropics.In the subpolar gyre, the models are not in accordance (CCSM3: '10 mmol/m 3 ; CSM1.4: up to (35 mmol/m 3 ).
The NAO is defined by an empirical orthogonal function (EOF) analysis of sea level pressure (SLP) in the North Atlantic area (308 to 808N and 1008W to 408E).Prior to the EOF analysis, the SLP field is area-weighted to account for the convergence of meridians on a latitudeÁlongitude grid (Baldwin et al., 2009).The leading EOF patterns of the six models (Fig. 2c) capture reasonably well the spatial structure of the NAO, yet the explained variability varies significantly.The NAO is known to be most pronounced during winter.The observation-based leading EOFs of winter (DJF), spring (MAM), summer (JJA) and autumn (SON) explain 39.44, 32.92, 27.21 and 23.17% (www.cgd. ucar.edu/cas/jhurrell/indices.info.html,access: 08.05.2012).The models yield explained variances of 52.3,47.7,40.9 and 38.1% (CCSM3),42.5,39.1,42.7 and 32.6% (CSM1.4),37.6,37.7,36.7 and 32.5% (GFDL),36.6,30.1,40.2 and 31.3% (BCCR),35.9,31.8,31.3 and 33.4% (MPIM) and 37.7,37.6,31.4 and 39.4% (IPSL), missing the seasonal variability of the NAO signal.The observation-based NAO index shows some low-frequency variability, e.g. a period of mainly negative NAO from approximately 1960Á1980 followed by a NAO ' phase during the 1990s.Compared to this, the NAO time series of the six models (Fig. 2a) are relatively unstable.The autocorrelation of the winter NAO time series (Fig. 2b) indicates some decadal-scale variability which is not captured by the models.This lack of variability, a known and prevalent issue in climate modelling (Parker et al., 2007), is possibly due to the usage of control runs; based on a multi-model study, Osborn (2004) suggests that the observed variability of the NAO is not only intrinsic, but at least partly driven by external greenhouse gas forcing.However, for the most part the autocorrelation of the observations is within the model range.
Based on the two NCAR models, the required time period of model data was evaluated.In each case 200 yr of monthly data were investigated, and the results proved to be stable for time periods of approximately ]60 yr (CCSM3) and ]45 yr (CSM1.4).So, for good measure as well as computational reasons, it was decided to perform the analysis on 100 yr of monthly data.Furthermore, it was decided to focus on the winter (DJF) season since (1) the NAO is known to be most pronounced during winter and (2) tests based on annual data resulted in the expected closely related, yet weaker response of the ocean.Despite the different explained variances of the EOF patterns, the results support the assumption that the findings of this study are valid and not the consequence of a misrepresentation, i.e. over-or underestimation of the NAO.A number of analysis techniques were applied, including cross correlation, linear regression and composite analysis.Since the results showed consistency, it was decided to restrict the presented results to correlation and composite analysis.In a similar spirit, the following presentation strategy is adopted to make it easier for the reader to follow the arguments and to limit the number of figures to a reasonable  2004), and PO 4 (mmol/m 3 ) Á taken from Garcia et al. (2010).Below: The difference between the simulated concentrations and the observations.DIC and Alk, originally mmol/kg, were converted using the annual salinity and SST fields of Garcia et al. (2010).
amount.First, our findings are discussed using results of a single model and a broad set of variables.Then, the range of models is compared for a more restricted set of key variables to illustrate robustness of findings and variations among the models.We note that the more complete analysis was carried out for all models.The CCSM3 simulation is used to introduce results since it was performed by the group of the first authors (Steinacher et al., 2010) and relatively easy access to also less common output is possible.(2001).This anticyclonic (clockwise) anomaly, covering a good part of the subtropical and subpolar Atlantic, is framed by cyclonic anomalies in the north and the south.

Ocean physics
SST is basically the inverse of wind stress and MLD, indicating large-scale stratification in the (especially western) subtropics.Surface salinity (SSS) features large-scale positive anomalies in the subpolar region.Since this pattern is closely related to MLD, DIC and PO 4 (see next section), it is presumably caused by entrainment of water masses from deeper layers to the surface.The composite patterns for NAO ' and NAO ( states are approximately reversed.However, differences concerning spatial patterns as well as signal-strength indicate non-linearity in the response of ocean physics to the NAO.Overall the results agree with the available observations as summarised in Hurrell and Deser (2009).
Figure .4 presents the composite patterns (NAO !NsN) for the winter season of MLD and upwelling for the six models.Generally speaking, all models are consistent in showing a similar physical response as discussed for the CCSM3, in particular when recalling the differences in explained SLP variance by the NAO for the different models.MLD and SST (not shown) clearly illustrate the seesaw pattern between the subpolar and subtropical regions while BSF (not shown) and the corresponding upwelling exhibit the imprint on the gyre circulation.However, the models differ concerning strength and spatial pattern of the signal.Striking is the discrepancy between the mean states of NAO ' and NAO ( , which is evident when the absolute anomalies are compared.With regard to MLD, the strongest response can be found locally in the subpolar Atlantic (CCSM3: 400m/ (280 m for NAO ' /NAO ( ; CSM1.4: 160 m/(175m; GFDL: 500 m/ (700 m; BCCR: 160 m/(80 m; MPIM: 280 m/ (400 m; and IPSL: 250 m/ (300 m), the main centres of action around southern Greenland (GFDL, MPIM) or in the west (BCCR)/east (CCSM3, CSM1.4,IPSL) of it.The inverse anomalies in the subtropics are less prominent, not exceeding (40 m/60 m for NAO ' /NAO ( (MPIM).Variations of SST (not shown) range between 918C in the subtropical (CCSM3, CSM1.4,MPIM) and, more localised, between (28C for NAO ' (CCSM3) and '1.4 8C for NAO ( (CSM1.4) in the subpolar gyre.In addition to this, all models show localised anomalies (up to 1.758C (CCSM3)/ (1.28C (BCCR) for NAO ' /NAO ( ) in both the North and Norwegian Sea.
A further distinction between the models comes from the intra-annual variability of the response to NAO forcing.An investigation of the four seasons (DJF, MAM, JJA SON) illustrates that ocean dynamics (i.e.MLD and upwelling) show a response throughout the year, yet most pronounced during winter and spring.With regard to thermodynamics (e.g.SST) the models are not in accordance.The majority shows a more or less constant response throughout the year while single models show the strongest anomalies in winter/spring (BCCR) and spring/summer (CSM1.4).A comparison with the strength of the respective winter NAO signal shows that the model with the largest explained variance, the CCSM3 (52.3% explained by EOF1), has not necessarily the strongest response to NAO forcing.Despite this significant overestimation of the NAO (observation-based EOF1: 39.44%), the response of ocean physics is comparable in strength as well as significance with the models GFDL (37.6%) and BCCR (36.6%).In conclusion, a generally consistent response to winter NAO is found for ocean physics across the range of models.

Ocean biogeochemistry
Figure. 5 shows the response of ocean biogeochemistry to NAO forcing based on the CCSM3.As for ocean physics, the result is a seesaw between subtropical and subpolar gyre.Of particular prominence is the similarity between DIC and the nutrient PO 4 , both featuring patterns closely related to MLD.This consensus identifies on-site entrainment as the main driver behind these anomalies rather than large-scale horizontal transport and, in case of DIC, indicates the minor importance of airÁsea CO 2 exchange.This is supported by an investigation of the column inventories of DIC and PO 4 (not shown).In response to NAO ' forcing, both variables show significant positive (negative) surface anomalies in the subpolar (subtropical) gyre.The positive anomalies in the subpolar ocean are barely visible in the column inventory, which again points towards the dominance of vertical relocation processes within the water column rather than horizontal transport or, in case of DIC, gas exchange.In contrast to this, significant negative anomalies of both inventories can be found in the (mainly eastern) subtropics.This can be attributed to downwelling processes in that area, diluting the concentrations of DIC and PO 4 by the pumping of low-nutrient surface waters to deeper ocean layers.Further evidence is supplied by Hovmo¨ller diagrams (not shown) of meridional transects crossing the North Atlantic which show that the increased concentrations of DIC and PO 4 in the subpolar surface ocean are indeed accompanied by negative anomalies in the subjacent ocean layers.Besides, there is no clear indication of either latitudinal or meridional horizontal transport.
The altered surface DIC and PO 4 concentrations go hand in hand with corresponding variations of (1) export production (POC; transport of particulate organic carbon out of the euphotic zone, in this case at a depth of 75 m) and (2) pCO 2 and, consequently, pH and OmegaA (the saturation state of calcite provided similar results: spatial patterns and correlation coefficients are exactly the same, the composite anomalies slightly higher Á varying VARIABILITY OF THE OCEAN CARBON CYCLE between 90.1 for OmegaA and 90.16 for OmegaC).Point (1) is interesting in the sense that, although the patterns are calculated for the winter season, POC shows a significant response even in the subpolar gyre Á the increased supply of nutrients results in an increase of biological production.
Compared to this, calculations on an annual basis (not shown) provide a partly different response: while the increased supply of nutrients in the subpolar gyre is still accompanied by a rise in production, the decrease of nutrients in the subtropics goes hand in hand with a rise in production as well.A possible explanation for this increase in biological production, despite the reduced nutrient concentrations, could be the warming of the surface ocean due to prominent stratification.Regarding point (2), the large-scale differences between DIC and pCO 2 in the subtropics can be mainly attributed to variations in SST.
Figure .6 presents the composite patterns (NAO!NsN) of the variables DIC and PO 4 .Overall, the response of ocean biogeochemistry is consistent across all models.Typical NAO-driven variations are 910 mmol/m 3 in the surface concentration of DIC, and the response of Alk is of comparable magnitude.Surface DIC shows the strongest reaction in the subpolar North Atlantic (CCSM3: 24/ (28 mmol/m 3 for NAO ' /NAO ( ; CSM1.4: 16/(24; GFDL: 5/ (5; BCCR: 3/(3; MPIM: 12/(6 and IPSL: 12/(12), and the less prominent anomalies in the subtropics stay within (15 (CCSM3, CSM1.4) and 12 (IPSL) for NAO ' / NAO ( .In addition to this, a positive response to NAO forcing can be found in the Nordic Seas (all models except BCCR) and the Labrador Sea (all models except IPSL).For GFDL, BCCR and MPIM the latter can be explained with the deepening of the MLD (Fig. 4), which is supported by the corresponding positive anomalies of PO 4 .Upwelling is calculated at a depth of 75 m, the composites are given as anomalies and checked areas indicate statistical significance (t test, 5% level).For BCCR, an isopycnic model, upwelling is not available.
The composites of the two NCAR models indicate no connection between NAO and MLD in this area, yet it can be found in the correlation patterns (CCSM3: see Fig. 3).The positive anomaly in the Nordic Seas is likely caused by the inadequate or even missing representation of freshwater fluxes within the models.A salinity normalisation of DIC to a reference salinity of 35 psu (not shown) conserves the main feature, i.e. the seesaw pattern between subtropical and subpolar gyre.Yet, the localised anomaly in the Nordics Sea turns from positive to negative in all six models.Since the overall effect on DIC turned out to be small and the insight associated with the procedure is debated (e.g.Robbins, 2001, Friis et al., 2003), it was decided to focus on the actual concentrations.Again the models differ concerning strength and spatial patterns of the response to the NAO.This is illustrated by the DIC composites of two NCAR models, CCSM3 and CSM1.4.The CCSM3 composite for NAO ' shows significant positive (negative) anomalies in large parts of the subpolar gyre (the western subtropics).The NAO ( composite roughly provides the reversed picture.Yet, the significant anomalies in the subpolar gyre are shifted in a south-eastern direction towards the European continent and the anomalies in the western subtropics feature hardly statistical significance.In contrast to this, the CSM1.4 shows no statistical significance in the subpolar gyre at all.However, significant negative (positive) anomalies can be found in the eastern (western) subtropics for NAO ' (NAO ( ).Another difference between the models concerns the seasonal cycle.As expected the majority of the models show the best correlations between NAO and surface biogeochemistry during winter and spring, with two exceptions: the MPIM model shows correlations of comparable strength throughout the year, and IPSL features the highest values during summer and autumn.When taking the strength of the respective winter NAO signal into account it appears that, concerning DIC and PO 4 , the two models with the largest explained variability (CCSM3: 52.3%; CSM1.4: 42.5%) display the strongest, statistically significant response.As for ocean physics, this is not the case for DpCO 2 and the CO 2 flux, supporting the assumption that the results are valid and not the consequence of a misrepresentation, i.e. overestimation of the NAO.Next, we turn to the links between winter NAO and surface ocean pCO 2 (pCO o 2 ), the gradient in pCO 2 between surface ocean and atmosphere (DpCO 2 ) and airÁsea CO 2 flux.pCO o 2 is not conserved during transport, but rather a function of DIC, Alk, temperature and salinity.Changes in these variables exert often opposite influences on pCO o 2 , interfering the attribution of pCO o 2 changes to mechanisms and drivers.For instance, upwelling usually increases both DIC and Alk.However, while a rise in DIC increases pCO o 2 , a positive change in Alk shows the opposing effect, and consequently the overall change in pCO o 2 can be either positive or negative.Changes in DpCO 2 are further influenced by variations in atmospheric pCO 2 (pCO a 2 ) due to atmospheric transport and airÁsea fluxes, and the exceptions are the BCCR and GFDL models where pCO a 2 is prescribed.However, we note that the response of DpCO 2 to the NAO is very similar to that of pCO o 2 in all models.The airÁsea carbon flux does not only vary with pCO o 2 and DpCO 2 , but also due to changes in the primarily turbulence-dominated gas transfer velocity (e.g.Wanninkhof, 1992), parameterised as a function of wind speed in the models.Given all these complexities, we expect that the links between NAO and pCO o 2 and related variables are not as clear and more heterogeneous across the model range as for example the links found for DIC or PO 4 .Usually the pCO 2 gradient between the surface ocean and atmosphere (DpCO 2 0pCO o 2 Á pCO o 2 ) is the main driver of the airÁsea CO 2 flux, and a rise in pCO o 2 and DpCO 2 results in a decrease of the airÁsea CO 2 flux and vice versa.This is indeed the case in the western subtropical gyre, where NAO ' -induced changes of MLD and SST trigger large-scale stratification (see the CCSM3-composites in Fig. 7).This state, suppressing the entrainment of water from deeper, nutrient-enriched ocean layers, results in negative anomalies of DIC and pCO o 2 and, in turn, an increase of the airÁsea CO 2 flux.The same, yet with changed signs, is found for large areas in the northern North Atlantic.In contrast, there are also some regions where the anomalies of DpCO 2 and airÁsea CO 2 flux point in the same direction.Thus, the change in flux is not dominated by a change in the pCO 2 gradient, but by the change in the gas transfer velocity.An example can be found off Norway where, during NAO ' , significant positive anomalies of DpCO 2 and airÁsea CO 2 flux match with increases in wind and MLD.
Figure .7 presents the composite patterns (NAO!NsN) of the variables DpCO 2 and airÁsea CO 2 flux.Here, it has to be mentioned that, due to the usage of control runs, pCO 2 and DpCO 2 go largely hand in hand.The models BCCR and GFDL are even run with constant atmospheric CO 2 concentrations, resulting in DpCO 2 being simply oceanic pCO 2 minus atmospheric pCO 2 (BCCR: 287.4 ppm; GFDL: 286 ppm).As for the CCSM3 model, the airÁ sea CO 2 flux is mainly driven by DpCO 2 which, in turn, largely follows DIC.This system of dependencies is valid for all models; differences between the three variables are caused by the additional effects of wind stress (CO 2 flux) and SST, salinity, alkalinity and the formation of calcite (pCO 2 ).Similar to DIC, DpCO 2 features the strongest anomalies in the subpolar North Atlantic (CCSM3: 16/(12 ppm for NAO ' /NAO ( ; CSM1.4: 10/ (8; GFDL: 6/(5; BCCR: 6/(6; MPIM: 2/ (4 and IPSL: 8/(8).The response in the subtropics ranges within96 ppm for all models.Yet it is not possible to identify a distinct seesaw pattern as for DIC and PO 4 , the direction of the anomalies varies on a small scale.In addition, there are large differences between the models.The same holds true for the corresponding airÁsea flux of CO 2 .For instance, during NAO ' conditions increased carbon uptake occurs in the western subtropics (CCSM3, GFDL), south of Greenland (CCSM3, CSM1.4,MPIM) and the Labrador Sea (BCCR, MPIM) while a decrease in uptake is found in the eastern subpolar gyre (CCSM3, GFDL, BCCR, IPSL) and the eastern (CCSM3, CSM1.4), western (MPIM) or entire subtropics (BCCR).For the six models, the airÁsea exchange of carbon is largest off Norway (CCSM3: 0.005/ (0.002 mmol/m 3 cm/s for NAO ' /NAO ( ), south of Iceland (CSM1.4:0.005/ (0.007), south and southeast of Greenland (GFDL: (0.002/0.003;IPSL: 0.003/(0.003)and in the Labrador Sea (BCCR: 0.004/ (0.004; MPIM: 0.003/ (0.004).An integration of the airÁsea CO 2 flux over the North Atlantic (10: 808N; (80: 208E) and the entire winter season (90 d) results in flux anomalies of CCSM3: (0.0068/0.0078PgC for NAO ' /NAO ( , CSM1.4: (0.0055/0.0025,GFDL: (2.54 )10 (5 /0.0028, BCCR: (0.0050/0.0057,MPIM: (0.0015/0.0053 and IPSL: (0.0028/0.0050.In relation to the winter mean state, these anomalies account for up to 5%.Globally, the integrated anomalies range between (0.024 (CCSM3) and 0.037 PgC (MPIM) for NAO ' and between (0.014 (MPIM) and 0.051 PgC (IPSL) for NAO ( . The importance of vertical entrainment of carbon for variations in surface DIC and air-to-sea carbon fluxes is underlined by the model CSM1.4.Due to the comparably simple formulation of the ocean biogeochemistry module, changes in DIC due to marine productivity and the remineralisation of organic matter and calcite (DDIC bio ) are linearly linked to variations in phosphate (DPO 4 ) and alkalinity (DAlk) according to the Redfield ratios (Redfield et al., 1963).This linear relationship allows to assign changes in DIC to (1) variations in the marine biological cycle and (2) airÁsea gas exchange driven by CO 2 concentrations only (see Plattner et al., 2001, Gruber and Sarmiento, 2002, Fro¨licher and Joos, 2010): (1) In eq. ( 1), the terms on the right-hand side represent reorganisation processes of (1) the organic matter cycle linked to the carbon-phosphate Redfield ratio and (2) the calcite cycle.The difference between DDIC total and DDIC bio [eq.( 2)] can be assigned to gas exchange.An investigation of the DIC composite patterns shows that DIC total (Fig. 6) and DIC bio (not shown) virtually feature the same pattern, DIC bio with slightly higher positive as well as negative anomalies in the subpolar gyre (NAO ' : DIC total varies between914 mmol/m 3 , DIC bio be-tween917 mmol/m 3 ).This small discrepancy can be explained by DIC gas (NAO ' : variations between95 mmol/m 3 ) which shows positive and, on a very local scale, negative anomalies.Beyond the subpolar gyre DIC gas features negative anomalies in the very eastern subtropics; however, a large part of the North Atlantic shows no response at all.In conclusion, this points out the dominance of biological processes over airÁsea gas exchange.Since the former are limited by the supply of nutrients, this again indicates the importance of ocean dynamics.

Time scale of response
To investigate the interannual component of the ocean response to the NAO, Fig. 8 shows the anomalies of wind stress, BSF, MLD, SST and DIC of five consecutive winters dominated by NAO ' for the model CCSM3.The increase of surface wind stress instantaneously results in corresponding anomalies of gyre circulation and MLD.An exception can be found at year 4 in the subpolar gyre where, in spite of negative anomalies of wind stress, MLD is deeper than usual.Since the pattern of this anomaly is equal to that of year 3 and the scale of magnitude is significantly smaller (year 3: up to 900 m; year 4: less than 300 m), it is presumably a remainder from the very strong anomaly in the previous year.The five SST patterns are consistent in showing the expected positive (negative) anomalies in the subtropical (subpolar) gyre.In addition, all years feature positive SST anomalies in the high north.Likewise, DIC is persistent in constituting a seesaw between negative anomalies in the subtropical and positive in the subpolar gyre and the northern North Atlantic.Neither ocean dynamics nor biogeochemistry indicate any significant large-scale horizontal transport.-8 -6 -4 -2 0 2 4 6 8 -0.003 -0.0015 0 0.0015 0.003 IPSL Fig. 7. Model comparison: Composite mean winter patterns for prominent NAO (NAO !NsN) of the gradient of pCO 2 between atmosphere and ocean (DpCO 2 ) and airÁsea CO 2 flux (mmol/m 3 cm/s).Calculations were done on surface level, the composites are given as anomalies and checked areas indicate statistical significance (t test, 5% level).The unit of DpCO 2 is ppmv (CCSM3, CSM1.4,MPIM) and matm (GFDL, BCCR, IPSL), respectively.For the integrated annual airÁsea CO 2 flux (mol m (2 yr (1 ), multiply by 315.36.instantaneous ( B3 months).The two locations differ with regard to the direction of the response.In the subpolar gyre both variables are significantly correlated with the NAO (correlation r: 0.70 and 0.25 for MLD and DIC, respectively).In the subtropics MLD/DIC and NAO are anticorrelated (r: (0.34 and (0.38 for MLD and DIC, respectively), reflecting the inverse wind stress anomaly compared to the subpolar gyre.The anomalies of MLD are smaller in the subtropical than in the subpolar gyre, ranging between (35/40 m and (70/125 m, respectively.The range of DIC anomalies, however, shows hardly any difference, with values between (19/15 mmol/m 3 (subpolar gyre) and (16/14 mmol/m 3 (subtropics).Regarding the instantaneous response of the ocean to NAO forcing, further support is provided by time series on a monthly basis (not shown) which underline especially the close relationship between NAO and MLD.In this context it can be stated that the anomalies of MLD in the subpolar gyre are largest during winter and spring irrespective of the strength of the NAO.Something alike is not detectable in the subtropics and in neither case for DIC.A shift of the time series of MLD and DIC in relation to the NAO shows that (1) significant correlations can be found with lags up to 4 yr, yet (2) the highest correlation in each case is present at lag 0. This can be illustrated on the basis of DIC, with lagged correlations of 0.25 (lag 0), 0.07 (lag 1), (0.08 (lag 2), (0.21 (lag 3), (0.15 (lag 4) and 0.03 (lag 5) in the subpolar and correlations of (0.38, (0.28, 0.08, 0.32, 0.26 and (0.03 in the subtropical gyre.Striking is the high anticorrelation (relative to lag 0) at lag 3.However, this signal is put into perspective when the autocorrelation of the NAO index is taken into account (lag 1: 0.16; lag 2: (0.21; lag 3: (0.25; lag 4: (0.12; lag 5: 0.15).As can be seen, the lagged signal is not necessarily a physical response.In this case, it is likely caused by the temporal variability of the NAO.

Models vs. observations
A comparison with observational data is rather challenging since carbon measurements are limited in either time or space.Long-term time series are available only at a few locations while the data originating from measurement cruises is providing only a temporarily fragmented snapshot.We utilise four different datasets.The Bermuda Atlantic Time Series (BATS; e.g.Bates, 2007) is located in the western subtropics (32850?N, 64810?W) and provides monthly resolved observations since 1988.Here, we use measurements from the surface ocean (depth 58 m).The data was taken from http://bats.bios.edu/index.html,access: 20.01.2012.The subpolar gyre is represented by three datasets.One originates from the SURveillance de l'ATLANTique programme (SURATLANT; e.g.Corbie`re et al., 2007;Metzl et al., 2010), measured on cruises crossing the subpolar gyre between Iceland and Newfoundland.The data has a temporal resolution of roughly 3 months and was collected from 1993 to 1997 and 2001 to 2008.We utilise surface data from 50: 608N and 40: 208W.The other two datasets are the Irminger and Iceland Sea time series (Olafsson, 2007a, b), and both were downloaded from the CARINA database (Key et al., 2010) Tables 2 and 3 show the correlation r between NAO and the variables DIC, PO 4 and SST for the four datasets and all six models.Due to the sparse and irregular sampling of the observations, we refrained from doing a statistical test.The model data is averaged over areas representing the subpolar (50: 608N, 40: 208W) and the subtropical gyre (29: 348N, 67: 618W).The variables are correlated on two temporal scales, monthly and for the winter season.For the latter, the observations comprise the available monthly values for January, February and December while the models are represented by seasonal (DJF) means.All datasets and five out of six models show higher correlations in winter than throughout the year, underlining the fact that the NAO is dominating the winter climate in the North Atlantic region.The exception is the IPSL model which, as discussed before, shows the highest seasonal correlations between NAO and surface biogeochemistry during summer and autumn.
In the western subtropics at BATS, both DIC and PO 4 are anticorrelated with the NAO, a finding consistent with all models.The only exception is IPSL which tends against zero in both cases.In contrast to the observations, the correlations for NAO/PO 4 exceed NAO/DIC for all models, with values ranging from (0.21 (GFDL) up to (0.55 (BCCR).The positive correlation of NAO/SST supports this finding, showing that the NAO-induced reduction of wind stress results in surface stratification and, consequently, less vertical nutrient supply.This is supported by the high correlation between DIC and PO 4 both in the dataset and especially in the models, putting into perspective the importance of airÁsea gas exchange for the ocean DIC inventory.However, as for NAO/PO 4 , the link between NAO/SST seems to be overestimated by the models.
In the subpolar gyre, the comparison with observational data is complicated by the small number of observations, especially for the winter season.For instance, in the SURATLANT dataset PO 4 is represented by only seven values, six of them obtained during a negative phase of the NAO.Nonetheless, observations and models are consistent in showing positive correlations for NAO/DIC and NAO/ PO 4 and negative ones for NAO/SST.The only exception is the Iceland Sea dataset where, on a monthly scale, the correlation for NAO/SST is slightly positive.Taking the scarce number of observations into account, the correlations between NAO/DIC and NAO/PO 4 are of comparable strength.The models reasonably capture (CCSM3, CSM1.4,GFDL, BCCR) or tend to underestimate (MPIM, IPSL) this link.Again DIC and PO 4 are highly correlated.The negative correlation for NAO/ SST is evidence for the influence of the NAO on ocean dynamics, in this case reduced vertical mixing.
The direct impact of the NAO on the ocean carbon cycle was further evaluated conducting linear regressions between NAO index and surface DIC concentrations.Here, the regression slopes correspond to a change of one standard deviation s in the NAO index.In the subtropics at BATS, a one s increase is linked to a decrease in surface DIC of (2 mmol/m 3 (monthly) and (5 mmol/m 3 (DJF), respectively.On the monthly scale, the models capture (CCSM3, CSM1.4) or slightly underestimate the observation-based values.For winter the models are consistent in underestimating the response to the NAO, with values ranging from (1 mmol/m 3 (GFDL, MPIM and IPSL) to (3 mmol/m 3 (CCSM3, CSM1.4).
and Iceland Sea time series both feature values of approximately 3/3 mmol/m 3 (monthly/DJF).The models show the same tendency as SURATLANT, with larger values during winter.Yet there are differences.The two NCAR models feature values of 4/6 mmol/m 3 (monthly/DJF), exceeding the observations on both time scales.GFDL, MPIM and IPSL show anomalies of 2/4 mmol/m 3 (monthly/DJF), the weakest response can be found for the BCCR model with 0/2 mmol/m 3 (monthly/DJF).

Discussion
The variability of the ocean carbon cycle in response to the NAO has been analysed utilising six Earth System Models.
The results are consistent over all models, yet relative and absolute magnitudes differ among models and regions.The response of the North Atlantic carbon cycle, a seesaw pattern between the subtropical gyre and the subpolar Northern Atlantic, was found to be instantaneous ( B3 months).The results indicate wind-driven dynamics as the main driver of the response, which Á via vertical mixing, upwelling and the associated entrainment of DIC Á leave an imprint on surface pCO 2 and, consequently, the airÁsea CO 2 flux.While there is ample evidence that the NAO impacts large-scale ocean circulation (e.g.Curry and McCartney, 2001;Raible et al., 2001;Marshall et al., 2001;Flatau et al., 2003;Jakobsen et al., 2003;Visbeck et al., 2003), till now there has been no definite answer if this connection has a significant influence on the ocean carbon cycle.We find no lead in this direction, even when analysing consecutive years with persistent NAO.Our results suggest that it is of relatively minor importance compared to wind-driven mixing processes.
The results proved to be stable, independent of investigated time period (DJF versus annual values) or analysis technique (correlation, regression versus composite patterns).For a NAO ' phase, the results indicate the following chain of events (summarised in Fig. 10).First, the increase in wind causes a positive anomaly in gyre circulation which covers a substantial part of the subtropical and subpolar Atlantic.This anticyclonic anomaly, resembling the 'intergyre' gyre introduced by Bjerknes (1964) and Marshall et al. (2001), is framed by negative, cyclonic anomalies in the North and South.In the entire North Atlantic, the gyre circulation is accompanied by the corresponding upwelling.Second, the increased surface wind stress in some (more northern) areas leads to a deepening of MLD and, following this and partly in accordance with increased upwelling, entrainment of DIC, Alk and PO 4 .The higher nutrient supply to the euphotic zone causes an increase in export production.The higher DIC concentrations tend to rise pCO 2 and to decrease the net airÁsea CO 2 flux, as well as pH and OmegaA.However, the links between DIC and surface water pCO 2 and the net airÁsea CO 2 flux are not always coherent as changes in surface ocean pCO 2 reflect the balance of partly opposing processes.In contrast to higher DIC, higher Alk and lower temperatures tend to decrease pCO 2 and, consequently, to increase the airÁsea flux.In addition, the airÁsea gas transfer velocity increases with higher wind speeds.In the western part of the subtropical gyre, characterised by a positive anomaly in SST and a decrease in wind stress and MLD, the reduced entrainment due to stratification results in negative anomalies of DIC and pCO 2 which, in turn, increases the airÁsea CO 2 flux.The sign of response to NAO ( is generally the opposite to the response to NAO ' .However, the responses of all six models to NAO ' versus NAO ( forcing are not strictly symmetric, implying non-linearity in the connection between NAO and the ocean carbon cycle.Typical NAOdriven variations are 910 mmol/m 3 in the surface concentration of DIC and alkalinity and 98 ppm in the airÁsea partial pressure difference.Yet, despite the substantial impact of the NAO on the North Atlantic carbon cycle, the effect on the basin-wide CO 2 flux is small.The integration of the airÁsea CO 2 flux over the North Atlantic and the entire winter season results in NAO-related anomalies of up to 90.007 PgC, which account for a maximum 5% of the winter mean state.This small total effect is the consequence of compensating fluxes on subbasin scales (e.g.McKinley et al., 2004;Raynaud et al., 2006).Based on a multichannel singular spectrum analysis (MSSA) of hindcast simulations, Raynaud et al. (2006) show that the subpolar gyre, the intergyre region (the transition area between the subpolar and subtropical gyres) and the subtropical gyre contribute with multipolar anomalies at multiple frequencies that tend to cancel one another in terms of the basin-wide airÁsea CO 2 flux.Earlier suggestions of a large NAO-related variability in basinintegrated airÁsea flux (Gruber et al., 2002), based on the interpolation of observations from a single location, appear unjustified.These findings highlight the difficulties linked to the regional or even basin-wide extrapolation of local measurements and points out the need for further observation efforts.Generally speaking, the results of this study are in accordance with observations.The relevance of thermodynamics and vertical convection, especially strong winter mixing, and the associated entrainment of nutrients for the ocean carbon cycle is well established (e.g.Lu¨ger et al., 2004;Tanhua et al., 2006;Olsen et al., 2008;Hartman et al., 2010).Consequently, numerous studies (e.g. Bates Response of ocean physics and biogeochemistry in the North Atlantic to NAO ' forcing.To maintain a certain clarity, the scheme is simplified.For instance, pCO 2 depends not only on DIC and Alk, but also on SST and salinity. et al., 2002;Gruber et al., 2002;Gonzalez-Davila et al., 2007;Corbie`re et al., 2007;Schuster and Watson, 2007;Perez et al., 2008;Bates, 2012) provide evidence for a connection between NAO and ocean biogeochemistry in the North Atlantic.For the last decade, a decrease of the North Atlantic carbon sink has been reported (e.g.Lefe`vre et al., 2004;Lu¨ger et al., 2006;Bates, 2007;Corbie`re et al., 2007;Schuster et al., 2009;Takahashi et al., 2009;Metzl et al., 2010).While sea-surface pCO 2 in the subtropics closely followed its raising atmospheric counterpart, in the north, the increase in sea-surface pCO 2 outpaced the atmospheric trend.This resulted in a reduction of DpCO 2 and, consequently, the net airÁsea flux of CO 2 .Schuster and Watson (2007) estimated a large-scale decline of the carbon sink up to !50% from the mid-1990s to the period 2002Á2005, in particular pronounced in the north-east.According to Schuster et al. (2009), the net airÁsea CO 2 flux decreased in the area north of approximately 458N while, in the south, the change was small or even zero.In contrast to the more northern areas, DpCO 2 measured at two locations in the subtropical gyre, ESTOC (e.g.Santana-Casiano et al., 2007;Gonzalez-Davila et al., 2010) and BATS (e.g.Bates, 2001Bates, , 2007)), has remained more or less constant over time.The initial impulse launching research on the connection between NAO and the ocean carbon cycle was the finding that, during this specific period, the NAO was mainly in its negative phase (Corbie`re et al., 2007;Schuster and Watson, 2007).
The origin of the observed decline in the subpolar carbon sink is still not entirely clear.Proposed explanations include a decrease of biological productivity (Lefe`vre et al., 2004), rising SSTs (Corbie`re et al., 2007), a slowdown of the North Atlantic Current (Thomas et al., 2008), the interplay of the changing buffer capacity of the ocean water and NAO-driven variations in SST, vertical mixing and sea-surface circulation (Schuster and Watson, 2007;Schuster et al., 2009) and changes in surface DIC concentrations due to convective processes (Metzl et al., 2010).Around 20% of the change in pCO 2 observed from the mid-1990s to the early 2000s in the eastern parts of the temperate and subpolar Northern Atlantic are attributable to the observed variability in SST (Schuster and Watson, 2007).In this context, it is interesting to qualitatively compare the subpolar North Atlantic in the NAO ( composites of the six models with the anomalies observed during the period from 1990 to 2006.However, we need to recall that the NAO explains only part of the variability; observed anomalies result from a combination of internal modes, noise, and external forcings.Nevertheless, with regard to ocean physics (i.e.MLD and SST), our NAO ( composites and observations are in qualitative displaying warm anomalies and shallow MLD in the subpolar North Atlantic.Similarly, low DIC and nutrient concentrations are found in the model composites and the observations, apparently the result of decreased entrainment from below.The observed decrease in export production during this NAO ( period is also to some extent mirrored in model composites.Concerning the airÁsea flux of CO 2 , the picture is not that uniform.Four of the models (CCSM3, GFDL, BCCR, IPSL) indicate a strengthening of the carbon sink in the respective area while two (CSM1.4,MPIM), qualitatively consistent with observations, indeed show a weakening.
A number of modelling studies have addressed the connection between NAO and North Atlantic Ocean carbon cycle.Most studies utilised global ocean general circulation models (McKinley et al., 2004;Thomas et al., 2008;Ullman et al., 2009;Levine et al., 2011) or coupled ecosystem-circulation models of the North Atlantic (Friedrich et al., 2006;Lo¨ptien and Eden, 2010).Concerning the isolation of an NAO-only signal, such ocean-only simulations have a handicap.Since they are forced with observed surface conditions, the forcing imposed on the ocean model includes variability in addition to NAO, e.g. the Tropical Atlantic variability (TAV) or the East Atlantic pattern (EA; Barnston and Livezey, 1987).Moreover, the NAO mode does not evolve freely and in a self-consistent way as in coupled ocean-atmosphere models, hindering the establishment of a clear quantification between NAO forcing and biogeochemical response.On the other hand, the simulated biogeochemical response should closely reflect the observed tracer and airÁsea flux evolution in simulations driven by reconstructed surface forcings.These ocean-only model studies are broadly in accordance with the results presented here, although differences remain.The authors identified variations in wind-induced mixing and SST as the main drivers of the response of ocean biogeochemistry.In the subpolar gyre, mixing-driven entrainment of DIC from nutrient-enriched deeper layers to the surface ocean outweighs airÁsea gas exchange.In the subtropics, changes in SST and the associated stratification are the dominating factor.Matter of debate are the importance of the biological pump (e.g.McKinley et al., 2004;Lo¨ptien and Eden, 2010) and advective transport processes (e.g.Thomas et al., 2008;Raynaud et al., 2006), the latter especially during periods of persistent negative or positive NAO forcing.Raynaud et al. (2006) find the largest correlation between NAO and airÁsea CO 2 flux in the subtropical and subpolar gyres near lag zero (r: (0.56 and (0.31, respectively) while, in between, r increases from 0.25 at lag zero to 0.43 at lag 2.5 yr, suggesting that NAO-related advective transport may play a role for the airÁsea flux in the intergyre region.
Special attention has been given to the observed decline of the North Atlantic carbon sink.Based on historical simulations with the coupled ocean carbon cycle module of the CCSM3, Thomas et al. (2008) propose a NAO-induced weakening of the North Atlantic Current and the associated decrease in transport of warm, saline subtropical waters to the north-east as the main cause.In accordance with our results, Ullman et al. (2009) utilised the MIT Ocean General Circulation Model to identify changes in SST (subtropics) and dynamical processes, primarily entrainment of DIC (subpolar gyre), as the main drivers of ocean pCO 2 .However, they were not able to explain the observations.Instead, their model indicated an increase in carbon uptake in the North Atlantic Á as do four out of six of our models during NAO ( .Further evidence for the importance of vertical mixing is provided by recent studies by Levine et al. (2011) and Patara et al. (2011).Regarding the link between NAO and advective transport, it is interesting to note that the observed NAO index varies significantly with season and, e.g. during winter 2007/2008 (2008/2009), switched from positive (negative) DJF values to a negative (positive) MAM state.
Only recently, the biogeochemical response to NAO has been isolated using a single AOGCM (Patara et al., 2011).These authors argue that the drivers behind the response of biogeochemistry differ, depending on the timescale considered.They state that, while interannual variations are dominated by vertical mixing, ocean circulation gains impact on decadal time scales.

Conclusion
The response of the ocean carbon cycle to the NAO has been analysed based on control runs of six Earth System Models.The results are consistent, though relative and absolute magnitudes differ among models and regions.The response of ocean biogeochemistry in the North Atlantic is a seesaw pattern between the subtropical and subpolar gyre.This pattern emerges instantaneously (B3 months) and is mainly driven by vertical exchange between the surface and the thermocline.The overall effect on the basin-wide CO 2 flux is small.Reasons are compensating fluxes on the sub-basin scale.All models show non-linear behaviour in response to positive and negative states of the NAO.
Our study shows that the NAO has significant impact on the ocean carbon cycle in the North Atlantic, which is in accordance with a recent observation-based study (McKinley et al., 2011) quantifying the time period necessary for a long-term trend to overcome decadal-scale variability with 25 yr.Since observations are scarce and limited in time, this highlights (1) the importance of a sound consideration of the NAO in the interpretation of observations and (2) the necessity of model studies to thoroughly investigate such connections between climate and ocean carbon cycle.An issue of interest is the hypothesised future shift of the NAO to a more positive phase (e.g.Hegerl et al., 2007;Bader et al., 2011), which might have a substantial impact on the future evolution of the North Atlantic carbon sink.
Fig. 2. a) Standardised winter (DJF) NAO indices for the 20CR dataset (years 1911Á2010; Compo et al. 2011) and all six models.b) Autocorrelation r of the winter NAO index for 20CR (red) and the model range (grey).c) The corresponding EOF 1 patterns of winter sea level pressure in the North Atlantic region (30: 808N, 1008W: 408E).

Figure. 3
Figure. 3  shows the response of ocean physics to the NAO based on the CCSM3 model.Both correlations and composites represent the winter season, and the composites are calculated for the NAO exceeding one standard deviation (NAO !NsN).NAO ' induces an increase (decrease) in surface wind stress in the subpolar (subtropical) Atlantic, resulting in corresponding variations of (1) mixedlayer depth (MLD) and (2) upwelling.The latter also leaves an imprint on the Atlantic Meridional Overturning Circulation (AMOC; not shown) and is the result of the emergence of a large-scale circulation (barotropic stream function, BSF) anomaly resembling the 'intergyre' gyre introduced byBjerknes (1964) andMarshall et al. (2001).This anticyclonic (clockwise) anomaly, covering a good part of the subtropical and subpolar Atlantic, is framed by cyclonic anomalies in the north and the south.SST is basically the inverse of wind stress and MLD, indicating large-scale stratification in the (especially western) subtropics.Surface salinity (SSS) features large-scale positive anomalies in the subpolar region.Since this pattern is closely related to MLD, DIC and PO 4 (see next section), it is presumably caused by entrainment of water masses from deeper layers to the surface.The composite patterns for NAO' and NAO ( states are approximately reversed.However, differences concerning spatial patterns as well as signal-strength indicate non-linearity in the response of ocean physics to the NAO.Overall the results agree with the available observations as summarised inHurrell and Deser (2009).Figure.4presents the composite patterns (NAO !NsN) for the winter season of MLD and upwelling for the six models.Generally speaking, all models are consistent in showing a similar physical response as discussed for the CCSM3, in particular when recalling the differences in explained SLP variance by the NAO for the different models.MLD and SST (not shown) clearly illustrate the seesaw pattern between the subpolar and subtropical regions while BSF (not shown) and the corresponding upwelling exhibit the imprint on the gyre circulation.However, the models differ concerning strength and spatial pattern of the signal.Striking is the discrepancy between the mean states of NAO ' and NAO ( , which is evident when the absolute anomalies are compared.With regard to MLD, the strongest response can be found locally in the subpolar Atlantic (CCSM3: 400m/ (280 m for NAO ' /NAO ( ; CSM1.4: 160 m/(175m; GFDL: 500 m/ (700 m; BCCR: 160 m/(80 m; MPIM: 280 m/ (400 m; and IPSL: 250 m/ (300 m), the main centres of action

Fig. 4 .
Fig. 4. Model comparison: Composite mean winter patterns for prominent NAO (NAO !NsN) of mixed-layer depth (MLD; m) and upwelling (mm/s).Upwelling is calculated at a depth of 75 m, the composites are given as anomalies and checked areas indicate statistical significance (t test, 5% level).For BCCR, an isopycnic model, upwelling is not available.

Figure 9
compares time series of winter NAO with anomalies of DIC and MLD for the model CCSM3, and DIC and MLD are averaged over areas representing the subpolar (408 to 608N, 508 to 208W) and subtropical gyre (298 to 348N, 678 to 618W).All three variables vary simultaneously, confirming that the NAO-induced response of ocean dynamics and, in turn, ocean biogeochemistry is CCSM3

Fig
Fig. 10.Response of ocean physics and biogeochemistry in the North Atlantic to NAO ' forcing.To maintain a certain clarity, the scheme is simplified.For instance, pCO 2 depends not only on DIC and Alk, but also on SST and salinity.

Table 1 .
Major model characteristics The ESM2.1 earth system model was developed by the Geophysical Fluid Dynamics Laboratory (GFDL) in Princeton, USA, the physical fundament of the model being the CM2.1 coupled climate model

Table 2 .
Correlation r between monthly and seasonal (DJF) time series of NAO index and the surface variables dissolved inorganic carbon (DIC; mmol/kg), PO 4 (mmol/kg) and SST (8C) for four observational datasets

Table 3 .
Correlation r between monthly and seasonal (DJF) time series of NAO index and the surface variables dissolved inorganic carbon (DIC; mmol/m 3 ), PO 4 (mmol/m 3 ) and SST (8C) for the six models The model data is averaged over areas representing the subpolar (50: 608N, (40: (208E) and the subtropical gyre (29: 348N, 67: 618W), the NAO is defined by an EOF analysis of sea level pressure in the North Atlantic (30: 808N, 1008W: 408E).The variables are correlated on two temporal scales, monthly and for the winter season.For the latter, the models are represented by seasonal (DJF) means.