A rapid reduction in the Atlantic meridional overturning circulation (AMOC) can significantly disrupt the global heat transport and likely triggered abrupt climate change during the last glacial cycle. A slowdown in AMOC has long been assumed to inhibit the exchange of carbon between the atmosphere and the deep ocean and thus cause radiocarbon (14C), which is produced in the atmosphere, to accumulate in the atmosphere. Indeed previous model studies have demonstrated that a reduction in AMOC leads to higher atmospheric 14C abundance (Δ14C). However, this seems inconsistent with the observed rise in atmospheric pCO2 during Heinrich 1 and the Younger Dryas stadial events and the emerging view that this CO2 rise resulted from the deep ocean venting “old” carbon. Using an Earth system model, we offer an alternative scenario that AMOC slowdown and an accompanying dynamical response in the south (i.e., the bipolar seesaw) can in fact lead to a decline in atmospheric Δ14C. This decline is realized in the model when the bipolar seesaw and thus the flux of old carbon from the Southern Ocean are sufficiently large so as to overcome the accumulation of 14C in the atmosphere as AMOC is reduced. The bipolar seesaw we describe invokes an oceanic teleconnection, whereby a freshwater perturbation in the North Atlantic drives a southern Δ14C response, but this does not necessarily preclude an atmospheric teleconnection.
 The large glacial-interglacial variation in atmospheric CO2 over the past 800,000 years recorded in Antarctic ice core [Luthi et al., 2008; Siegenthaler et al., 2005] indicates significant changes in the global distribution of carbon. While there is as yet no consensus on the exact mechanisms of the CO2 variation, it is generally understood from an early study by Broecker  that the storage of carbon in the ocean is the key. The deep ocean in particular is important, because its carbon reservoir is far larger than those of terrestrial vegetation, soil, and atmosphere. Even small changes in the deep ocean carbon reservoir can translate into large changes in the other reservoirs, especially the smallest atmospheric reservoir. There is also broad consensus that the glacial-interglacial CO2 variation is strongly coupled to marine processes operating in the Southern Ocean, as suggested by a very high correlation of atmospheric CO2 and Antarctic temperatures on orbital and millennial timescales [Ahn and Brook, 2008; Fischer et al., 2010; Siegenthaler et al., 2005].
 The radioactivity of 14C in the atmosphere (Δ14C) is also sensitive to changes in the global carbon cycle and, in particular, the exchange of carbon with the deep ocean. The latter can make atmospheric Δ14C either young or old, depending on whether the exchange allows newly produced 14C to remain in the atmosphere or causes the deep ocean to vent “old” carbon to the atmosphere. Atmospheric Δ14C also reflects changes in the rate of 14C production, which occurs in the upper atmosphere. The production depends on the flux of cosmic rays that convert atmospheric nitrogen into 14C. The flux of cosmic rays reaching the atmosphere in turn depends on the geomagnetic field intensity and solar activity. With an independent knowledge or assumption about the production, variation in atmospheric Δ14C can be used to infer changes in the global carbon cycle.
 During the last deglaciation as the global climate transitioned from glacial conditions to the Holocene, atmospheric CO2 increased by nearly 80 ppm while atmospheric Δ14C decreased by about 400‰ (Figure 1). These changes were not monotonic, but the general trends make intuitive sense: The release of old carbon (low 14C/C ratio) from the Southern Ocean would both increase atmospheric carbon inventory and dilute the atmospheric 14C/C ratio. This notion is supported by the new Southern Ocean records of Δ14C from deep sea corals [Burke and Robinson, 2012] and atmospheric δ13C records from Antarctic ice cores [Schmitt et al., 2012]. It is important to that the deglacial decline in atmospheric Δ14C requires changes in the global carbon cycle, since changes in 14C production alone are insufficient to account for the entire decline. For example, during an early stage of the deglaciation referred to as the Mystery Interval (between 17.5 and 14.5 kyr), which is approximately equivalent to the Heinrich 1 (H1) stadial event, changes in 14C production as suggested by paleomagnetic and sedimentary 10Be reconstructions can only account for 37‰ of the 190‰ drop [Broecker and Barker, 2007].
 As synthesized recently by Denton et al.  and Clark et al. , much of the 80 ppm increase in CO2 during deglaciation occurred during two abrupt climate stadial events H1 and the Younger Dryas (YD). CO2 increased approximately by 50 ppm during H1 and 30 ppm during YD (Figure 1). There is broad consensus that these cold events were triggered by freshwater (FW) perturbation to the North Atlantic, although there remain uncertainties in the sources and routes of the FW perturbation [Clark et al., 2012]. The two abrupt climate events share a number of characteristics. For example, the Northern Hemisphere experienced cooling [Stuiver and Grootes, 2000], the Intertropical Convergence Zone (ITCZ) and the monsoons shifted in position and strength [Wang et al., 2001; Yuan et al., 2004], the Atlantic meridional overturning circulation (AMOC) was drastically reduced [McManus et al., 2004], the global sea level rose by 5–15 m [Yokoyama and Esat, 2011], the Southern Hemisphere experienced warming [Calvo et al., 2007, Monnin et al., 2001], deep upwelling increased in the Southern Ocean [Anderson et al., 2009], and of course atmospheric pCO2 increased [Monnin et al., 2001].
 Many of these observations have theoretical support from modeling studies. For example, in simulations with a global atmospheric model coupled to a slab ocean, an imposed anomaly of sea ice expansion, which likely occurred during H1 and YD in the North Atlantic, causes the ITCZ to shift away from the hemisphere with the imposed anomaly [Chiang and Bitz, 2005]. Also, as shown by Rahmstorf  in one of the first studies and reproduced by numerous studies since, the addition of FW to the North Atlantic stratifies the surface waters and shuts down AMOC. It is generally believed, although the details may not be entirely clear, that a shutdown can dramatically cool the Northern Hemisphere centered on the North Atlantic and warm the Southern Hemisphere, as the interruption of the global ocean heat transport effectively prevents heat from getting transported from south to north [Crowley, 1992]. There is now strong support for this so-called bipolar seesaw [Broecker, 1998; Stocker, 1998], as rapid north-south antiphasing is observed in the ocean during deglaciation [Barker et al., 2009] as well as in ice cores during much of the marine isotope stage 3 [Blunier and Brook, 2001; EPICA Community Members, 2006]. The simulated CO2 response to FW hosing is actually quite variable among models, as the oceans may absorb or vent CO2; however, it seems reasonable to conclude that the correct model response should be CO2 venting [Schmittner et al., 2007], which could account for the increase in atmospheric pCO2 recorded in ice cores.
 This study is concerned with the evolution of atmospheric Δ14C during rapid climate events triggered by FW. If indeed the two-step increase in atmospheric CO2 during H1 and the YD is a result of the Southern Ocean venting old carbon from the deep ocean, it seems that atmospheric Δ14C should decline during these two periods as is generally observed (Figure 1). This is in fact the position that we advocate in this study. However, this position is at odds with the traditional view in the literature that the shutdown of AMOC, triggered by FW forcing, causes “an accumulation of 14C in the atmosphere” [Stocker and Marchal, 2000].
 In a seminal work, Hughen et al.  report high-resolution Δ14C data from the laminated sediments of the Cariaco Basin. After accounting for changes in 14C production, they find a large and rapid rise in Δ14C that is synchronous with the onset of the YD, concluding that “changes in both Δ14C and climate were caused by the same forcing mechanism.” They use a box model to show that the Δ14C rise can be accounted for by an AMOC shutdown, which is broadly supported as noted above. A shutdown prevents the exchange of carbon between the atmosphere and the deep ocean and allows newly produced 14C to accumulate in the atmosphere. The authors have difficulty explaining the subsequent decline in Δ14C that continues for much of the remainder of the YD and calls for a vigorous North Atlantic Intermediate Water, for which there are little supporting observations.
 The accuracy of the Cariaco data during the YD as a record of atmospheric Δ14C has recently been questioned because of issues with variable surface reservoir age and was thus excluded from the community's standard atmospheric Δ14C calibration IntCal09 between 12.55 and 12.9 kyr [Reimer et al., 2009]. IntCal09 is based on tree ring data back to 12.55 kyr. Nevertheless, the initial Δ14C rise, first identified in the Cariaco data, appears in IntCal09 (Figure 1). Therefore, the 30‰ rise from 12.7 to 12.5 kyr, followed by a larger and longer decline, is likely a robust feature of the atmospheric record.
 All model studies following Hughen et al.  confirm that AMOC shutdown causes atmospheric Δ14C to rise. In one of the first studies to use a dynamical model, Stocker and Wright  force their zonally averaged ocean model with FW. They also prescribe atmospheric pCO2 to follow prescribed paths, some constant and others increasing in order to mimic the observed pCO2 rise. They find that the AMOC shutdown causes atmospheric Δ14C to rise by about 35‰. Marchal et al.  use the same model and also note that the direct impact of an AMOC shutdown is a rise in atmospheric Δ14C. Again, using FW forcing and prescribed CO2 paths but applied to a three-dimensional ocean model, Butzin et al.  find that atmospheric Δ14C rises by about 30‰. Similarly, Ritz et al.  find that atmospheric Δ14C increases by about 15‰ in their three-dimensional model with FW forcing and constant atmospheric CO2. In a more realistic setup, Meissner  applies FW to the UVic Earth system model of intermediate complexity (EMIC) while allowing atmospheric CO2 to vary freely; in a simulation where AMOC collapses, atmospheric Δ14C rises by 44‰ from 200‰ to 244‰. Using a different EMIC, Singarayer et al.  also report that shutting down AMOC causes atmospheric Δ14C to rise by 30‰.
 These earlier studies are able to explain to various degrees of success the initial Δ14C rise at the outset of the YD with an AMOC shutdown, but they leave unexplained the much larger and longer decline that continues until the end of the YD. The decline in atmospheric Δ14C during H1 is even larger, thus more difficult to explain. In this study, we build on these studies and evaluate the additional impact of the bipolar seesaw on atmospheric Δ14C. Using an EMIC, we demonstrate that FW forcing triggers the bipolar seesaw and CO2 outgassing from the Southern Ocean causes it to accumulate in the atmosphere. More importantly, if the bipolar seesaw is sufficiently strong, atmospheric Δ14C in our model declines, as observed and expected if indeed 14C-depleted CO2 is upwelled and vented out of the Southern Ocean [Anderson et al., 2009; Denton et al., 2010]. We also confirm that the transient response of the surface ocean Δ14C becomes decoupled from that of the atmosphere due to variation in the surface reservoir age as AMOC shutdown alters the residence time Atlantic surface waters [Meissner, 2007; Ritz et al., 2008; Singarayer et al., 2008]. Our proposal that atmospheric Δ14C should decline as the bipolar seesaw is activated clarifies the role that the ocean plays in controlling atmospheric Δ14C during times of rapid climate change over the deglaciation.
2 Earth System Model
 The EMIC employed here is based on MESMO [Matsumoto et al., 2008], which consists of a dynamical model of the ocean, a dynamic and thermodynamic model of sea ice, an energy and moisture balance model of the atmosphere, and a prognostic model of ocean biogeochemistry. The model is successfully used in a number of process studies of the global carbon cycle [Chikamoto et al., 2008; Lee et al., 2011; Matsumoto et al., 2010; Sun and Matsumoto, 2010; Ushie and Matsumoto, 2012] as well as in community-wide model intercomparison projects [Archer et al., 2009; Cao et al., 2009]. For this study, we have coupled MESMO to a model of terrestrial biosphere ENTS [Williamson et al., 2006] and included carbon isotopes. In terms of the compartments representing the Earth system or model dynamics, the model employed here is similar or comparable to EMICs employed previously for investigation of atmospheric Δ14C [Meissner, 2007; Singarayer et al., 2008].
 The MESMO-ENTS is appropriate for investigating atmospheric Δ14C, because its interior ocean ventilation is well calibrated against modern observations including natural and bomb 14C and chlorofluorocarbon gases [Matsumoto et al., 2004]. Also, simulations of 14C typically require many thousand years of integration, and the computational efficiency of MESMO-ENTS meets the requirement. We use MESMO-ENTS in its well-calibrated, preindustrial state for the most part. However, we also confirm the robustness of our results by running the model under full glacial boundary conditions for ice sheets, atmospheric pCO2 (180 ppm), and orbital parameters for insolation at 20 kyr before present. In all sensitivity simulations, 14C production is diagnosed from the preceding equilibrium simulations and kept constant.
 Details of the model components are described elsewhere [Matsumoto et al., 2008; Williamson et al., 2006], so some of the salient features are briefly noted here. The dynamics of the three-dimensional ocean model is based on the frictional geostrophic equations [Edwards and Marsh, 2005]. The model has a 36 × 36 equal-area horizontal grid with 10° increments in longitude and is uniform in sine of latitude. There are 16 levels in the vertical with the top two layers being 100 m thick. It includes the Gent-McWilliams eddy mixing parameterization [Griffies, 1998] that reduces excessive mixing. Key physical model parameters have been optimized objectively using atmospheric and oceanic observations as constraints.
 The atmospheric model is based on an energy and moisture balance model [Fanning and Weaver, 1996]. There is no atmospheric dynamics and hence no wind feedbacks. The vertical distribution of thermodynamic energy is decreased with e-folding atmospheric height and determined by latent, sensible, and radiative heat transfer between the ocean and the atmosphere. The balance of evaporation or sublimation and precipitation determines moisture. The two prognostic variables in the atmospheric module are surface air temperature and surface specific humidity.
 In the sea ice model, the heat flux from the atmosphere is assumed to equal the vertical heat flux through the ice, which is determined by the difference of surface temperature and ice temperature, heat conductivity, and ice thickness [Edwards and Marsh, 2005]. Dynamic equations are used to predict the fraction of the ocean surface covered by sea ice and the average height of sea ice. Surface temperature of the sea ice is solved using a diagnostic equation.
 The model of ocean biogeochemistry includes a simple prognostic export production that depends on Michaelis-Menten nutrient uptake kinetics, light, estimated phytoplankton biomass, temperature, and mixed layer depth. Remineralization of sinking particles is based on a tuned sinking rate and temperature-dependent remineralization rates. Seasonal variation in insolation causes seasonal variation in export production, as light availability, mixed layer depth, and vertical nutrient supply change. The export production of particulate organic carbon and CaCO3 in MESMO is, respectively, 10.6 and 0.9 Pg C yr−1.
 Predictions of CFC-11 uptake and anthropogenic carbon uptake by MESMO during the 1990s are, respectively, 0.69 × 109 mol and 118 × 1015 g C, which compare well to data-based estimates of 0.55 ± 0.12 × 109 mol [Willey et al., 2004] and 118 × 1015 g C [Sabine et al., 2004], respectively. These tracers typically reflect ocean ventilation by intermediate waters in the upper ocean and by the relatively rapid North Atlantic Deep Water. MESMO is also well calibrated on longer, centennial timescale with respect to natural 14C in the deep ocean: MESMO predicts Δ14C of −153‰ for Circumpolar Deep Water (CDW) and −216‰ for North Pacific Deep Water. The equivalent data-based estimates used as metrics of model evaluation in the Ocean Carbon Cycle Model Intercomparison Project are −155 ± 12‰ and −226 ± 14‰ [Matsumoto et al., 2004].
 The efficient numerical terrestrial scheme (ENTS) is a simple prognostic model of terrestrial biosphere that calculates the exchange of energy, moisture, and carbon between land and the atmosphere [Williamson et al., 2006]. Global fluxes of carbon from photosynthesis, plant and soil respiration, and leaf litter drive carbon stocks of land vegetation and soil. Photosynthesis has dependence on atmospheric CO2, water stress, air temperature, and biomass (or vegetation fraction). Land vegetation is expressed as fractional coverage with corresponding albedo based on vegetation, soil cover, and soil type. Prognostic variables include vegetation and soil carbon as well as land surface albedo and temperature. The original ENTS did not have carbon isotopes, which were added here.
 In the control run of the coupled MESMO-ENTS model, vegetation and soil carbon stocks are, respectively, 535 and 1310 Pg C, which compare well to modern estimates of 451 Pg C [Olson et al., 1985] and 1306 Pg C [Batjes, 1995]. These modern estimates include postindustrial land use changes, which resulted in a release of about 120 Pg C by vegetation since 1850 [Houghton, 1999]. The spatial distribution in carbon vegetation (Figure S1 in the supporting information) compares favorably to observations: peak carbon storage in tropical rainforests and secondary maximum in boreal forests. It also captures the main desert regions of the world. Soil carbon distribution is also reasonable in showing relatively high values in boreal regions, where low temperatures limit soil respiration, and values in tropical regions are low, where temperatures and thus soil respiration rate are high. As expected, vegetation and soil are relatively close to being in equilibrium with the atmosphere in terms of Δ14C. Greater soil Δ14C values are found in colder regions such as Alaska, where respiration rates and thus carbon turnover rates are low.
 Following standard practice, FW hosing is applied to the North Atlantic to slow down AMOC in MESMO-ENTS. In four separate simulations each lasting 2500 years, forcings of 0.1, 0.2, 0.3, and 0.4 Sv (Sv = 106 m3 s−1) are applied for 1000 years between years 50 and 1050. These FW fluxes were chosen to achieve a range of AMOC response in our model. A small salinity flux is applied uniformly to the world ocean outside the North Atlantic so as to compensate for the FW perturbation and maintain salinity globally.
 In all four simulations, the forcing causes reduction in AMOC and increase in atmospheric CO2, and the magnitude of response depends on the size of the FW forcing (Figures 2a and 2b). For 0.1 Sv forcing, AMOC is reduced by a third from 15 Sv, and CO2 increases by only 5 ppm, which is quite small compared to observed changes during H1 and the YD. For 0.4 Sv forcing, AMOC is reduced much more strongly to 2 Sv, and CO2 increases by 23 ppm. In all cases, AMOC is reduced quickly and remains in the reduced state while hosing is applied. The AMOC resumption at the end of FW hosing does not require an artificial negative FW flux, which some earlier studies of Δ14C needed [Meissner, 2007; Ritz et al., 2008].
 The response of atmospheric Δ14C to the FW forcing is generally a reduction (Figure 2c). The exception is the simulation with the smallest 0.1 Sv forcing, in which case atmospheric Δ14C actually increases albeit with very small amplitude of less than 5‰. This response offers a way to relate to earlier studies, which all indicate a rise in atmospheric Δ14C as already noted, and has interesting implications. Therefore, we will return to discussing this case after first examining the 0.4 Sv forcing case, which arguably has more reasonable changes in atmospheric pCO2 and AMOC and offers a new, alternative interpretation of Δ14C during H1 and the YD. The simulations with 0.2 and 0.3 Sv of forcing have smaller responses to the 0.4 Sv simulation but are qualitatively the same.
 The simulation with 0.4 Sv forcing exhibits the typical bipolar seesaw temperature anomaly (Figure 3a). The near-collapse of AMOC causes the Northern Hemisphere centered on the North Atlantic to cool dramatically, while the south experiences warming. The 23 ppm rise in atmospheric pCO2 during the AMOC reduction (Figure 2b, solid line) is caused by CO2 degassing from the ocean since land biosphere is absorbing CO2 (Figure 3b). Most of this degassing occurs in the Southern Ocean. During the hosing period, CO2 flux from the Southern Ocean is generally 0.2–0.3 Pg C yr−1 (Figure 3c, solid line), which accounts for much of the global ocean outgassing (Figure 3b, solid line), after accounting for some compensating flux from tropical Pacific and Indian Oceans (Figure 3c, dashed line).
 Figure 2b shows a small negative pulse of pCO2 at the onset of the FW hosing and another small positive pulse at the termination for all simulations. A closer examination of the 0.4 Sv simulation shows that these are caused by large changes in terrestrial carbon flux (Figure 3b, dashed line). At the onset, cooling in the north suppresses respiration and causes soil carbon to grow. This draws down CO2 from the atmosphere and causes the first negative pulse. At the termination of the hosing, the quick resumption of AMOC triggers warming centered on the North Atlantic. The warming in turn enhances soil respiration and releases CO2 to the atmosphere, causing the second positive pulse. The response of respiration to changes in temperature is opposite to but much larger than the response of photosynthesis, because photosynthesis is limited not only by temperature but also by other factors such as nutrients [Matsumoto, 2007]. In order to put these net fluxes into perspective, we note that they are 1 order of magnitude smaller than global emissions of anthropogenic CO2 in recent decades and 2 orders of magnitude smaller than the global fluxes of gross terrestrial carbon uptake and respiration. The impact of these terrestrial changes on atmospheric Δ14C is limited, because Δ14C of vegetation, soil, and atmosphere is similar.
 In the simulation with 0.4 Sv forcing, venting of oceanic carbon reduces atmospheric Δ14C in the model by 27‰ during the time of AMOC shutdown (Figures 2c and 4). The response of the surface ocean Δ14C varies by region (Figure 4). As noted previously [Meissner, 2007; Ritz et al., 2008; Singarayer et al., 2008], the Atlantic waters become more in equilibrium with the atmosphere because their residence times at the surface increase. Therefore, in our simulation, the Atlantic surface water Δ14C increases initially and approaches the atmospheric value, reducing its 14C reservoir age. The closer the surface water is to the site of the North Atlantic Deep Water formation, the greater the change in the residence time and hence the reservoir age. The impact on Pacific surface waters is limited, thus leading to decoupled Δ14C trajectories taken by the Atlantic and Pacific surface waters [Singarayer et al., 2008]. Following the initial rise, all surface waters follow the atmospheric decline until AMOC recovers.
 Figure 5 shows changes to deep ocean Δ14C in the 0.4 Sv forcing simulation after 1000 years. Prior to FW perturbation, the deep North Atlantic is well ventilated and thus significantly younger in Δ14C than the deep Southern Ocean and deep North Pacific (Figure 5, solid lines). The subsequent near-collapse of AMOC largely isolates the deep Atlantic from the atmosphere, thereby making the deep North Atlantic Δ14C almost as old as the deep North Pacific one (Figures 5a and 5c). In contrast, the deep Southern Ocean generally becomes somewhat younger as it exchanges carbon more vigorously with the atmosphere (Figure 5b). As the deep North Pacific is heavily influenced by CDW, it follows the Southern Ocean trend and becomes younger as well (Figure 5c).
 Finally, we have one additional simulation where the same 0.4 Sv was applied to MESMO-ENTS under glacial boundary conditions. The glacial model response is generally smaller than the equivalent preindustrial model response but qualitatively the same. At maximum, AMOC weakens by 12 Sv, atmospheric pCO2 increases by 9 ppm, and atmospheric Δ14C declines by 12‰. The transient response of the glacial model is very similar to the simulations with 0.2–0.4 Sv of FW forcing shown in Figure 2.
 We are not the first to demonstrate with an EMIC that an FW forcing in the North Atlantic triggers the bipolar seesaw, which in turn leads to the Southern Ocean venting deep ocean CO2. Using the UVic model, Schmittner et al.  present similar results and argue that the enhanced vertical mixing in the Southern Ocean is driven by salinity redistribution. It has previously been called to explain a different kind of seesaw, the Atlantic-Pacific seesaw, triggered also by an FW perturbation to the North Atlantic [Saenko et al., 2004]. In the bipolar seesaw and according to Schmittner et al. , salinity injection into the deep ocean is reduced by a shutdown of AMOC, making the upper ocean relatively more saline and the deeper ocean fresher. The vertical density gradient thus becomes smaller. In the Southern Ocean, the stratification weakens, the mixed layer depth deepens, and the sloping isopycnals become steeper. Some of these physical changes were previously reported by Mikolajewicz . In the UVic model, these changes facilitate the venting of CO2 from depth with the largest loss of oceanic carbon occurring around 1500 m [Schmittner et al., 2007]. All indications from our bipolar seesaw simulations point to the same mechanism, as we too see negative salinity anomaly originating from the North Atlantic and causing similar changes in the vertical density gradient, mixed layer depth, and isopycnals. In our model, the largest reduction of carbon occurs around 1800 m in the Southern Ocean.
 The main contribution of this work is demonstrating the impact of the FW-driven bipolar seesaw on atmospheric Δ14C. Our simulations indicate that when the AMOC reduction and CO2 outgassing are sufficiently large to cause an increase in atmospheric pCO2 by more than a few parts per million, atmospheric Δ14C declines. This outgassing occurs primarily in the Southern Ocean (Figure 3c), which is consistent with observations of enhanced Southern Ocean upwelling during H1 and the YD, coinciding with AMOC reduction and atmospheric pCO2 rise [Anderson et al., 2009; Burke and Robinson, 2012; Denton et al., 2010; Skinner et al., 2010]. At the same time, our simulations indicate both an initial rise and a subsequent long decline in Atlantic surface water Δ14C (Figure 4), a transient behavior that resembles the Cariaco Δ14C record from the subtropical North Atlantic [Hughen et al., 1998].
 It is important to note that the Southern Ocean is the region with the greatest CO2 outgassing not only in the 0.4 Sv forcing case (Figure 3c) but also in all simulations presented here. For example, even in the 0.1 Sv forcing simulation, the Southern Ocean CO2 flux (~0.09 Pg C yr−1) is almost 1 order of magnitude larger than the fluxes from all other oceans (not shown). The difference is that the bipolar seesaw and the Southern Ocean outgassing flux are just weaker with smaller FW forcing. This means that even with 0.1 Sv forcing, the Southern Ocean still vents old carbon, causing an overall increase in atmospheric CO2 (Figure 2b); yet atmospheric Δ14C unexpectedly rises (Figure 2c, long dashed line). The reason is that the effect of diluting atmospheric 14C/C by outgassing low oceanic 14C/C is not large enough in the 0.1 Sv simulation to overcome the opposing effects of 14C production and accumulation as AMOC is weakened. This may explain why some earlier models predicted that atmospheric Δ14C should rise when FW perturbation shuts down AMOC. In the limit that there is no outgassing of old oceanic carbon by the bipolar seesaw, which would be the case in box models [Goslar et al., 2000; Hughen et al., 1998], atmospheric Δ14C will definitely rise.
 Another reason why some previous studies predicted a rise in atmospheric Δ14C in response to AMOC reduction seems to be related to their experimental design, in which atmospheric pCO2 was prescribed to follow a certain path [Butzin et al., 2005; Marchal et al., 2001; Ritz et al., 2008; Stocker and Wright, 1996]. Some had prescribed an increasing pCO2 path to mimic the YD trend. This experimental design amounts to modeling the dynamic behavior of 14C but not 12C, but both isotopes are in fact needed to properly quantify the variation in their ratio. In order to evaluate the impact of this experimental design, we conducted two additional simulations with MESMO-ENTS, where atmospheric pCO2 was prescribed to increase or decrease by 20 ppm over 1000 years but allowing Δ14C to vary freely. In these runs, the net flux of CO2 across the air-sea interface is driven predictably by the pCO2 gradient: into the ocean in the case of prescribed pCO2 increase and out of the ocean in the case of prescribed decrease. Prescribing an increase suppresses degassing of 14C-depleted CO2 from the ocean and thus causes atmospheric Δ14C to rise by about 10‰. The opposite is true when pCO2 is prescribed to decrease. Note that this occurs in the absence of an AMOC slowdown or any other change to circulation.
 In our simulations, FW forcing weakens AMOC and immediately triggers the bipolar seesaw, so there is no appreciable rise in atmospheric Δ14C in the model (Figure 4) that resembles the 30‰ rise observed at the outset of the YD in IntCal09 (Figure 1). The rise could be explained if there is a time lag in the bipolar seesaw, so that the initial Δ14C rise was due to an AMOC shutdown but the subsequent long-term Δ14C decline was due to a delayed response of the bipolar seesaw [Broecker, 1998]. Alternatively, there could have been a change in the strength of the bipolar seesaw. For example, the AMOC slowdown could have been modest initially (like our 0.1 Sv case), causing cooling in the north but a rise in atmospheric Δ14C. A subsequent drop in atmospheric Δ14C could be explained, if the AMOC slowdown became much stronger (like our 0.4 Sv case).
 Because our model does not have a dynamical atmosphere, the north-south teleconnection (southern Δ14C response to northern FW forcing) presented here occurs entirely within the ocean. However, this does not preclude the possibility that an atmospheric teleconnection could also drive a southern Δ14C response. For example, Lee et al.  show that a cooling in the North Atlantic, which presumably would occur with FW forcing, strengthens the southern westerlies via tropical atmospheric dynamics. Stronger Southern Ocean winds in turn can drive a southern Δ14C response [Rodgers et al., 2011] with the same sign as this study. It seems that the two teleconnections are not mutually exclusive, and in fact, both may have acted synergistically [Denton et al., 2010]. It so happens that Lee et al.  quantified the CO2 response of their atmospheric teleconnection with the same model that is used here, and it was an increase of 20 ppm, which is roughly similar to the CO2 response in this study. We simply state this as a side note, since the same model was used in both studies, and do not imply that the atmospheric versus oceanic teleconnection is equally important. There is no obvious way to evaluate strength of the cooling forcing used in Lee et al.  and the FW forcing used here.
 There are some Δ14C observations from the deep ocean that bear on this study. As shown in Figure 5, our model predicts a large change toward older ages in the North Atlantic (~100‰) and modest changes toward younger ages in the Southern Ocean (10–20‰) and the North Pacific (20–40‰). Consistent with the model prediction, measurements of Δ14C in deep sea corals and benthic-planktonic foraminifera from the midlatitude North Atlantic [Robinson et al., 2005] show marked changes of similar magnitude to older ages during H1 and the YD. Also, benthic-planktonic Δ14C data from high latitude North Atlantic show >200‰ shift to older ages [Thornalley et al., 2011]. In comparison, Δ14C data from the Southern Ocean are not entirely consistent with each other but generally indicate that the deep Southern Ocean became better ventilated over the course of deglaciation [Burke and Robinson, 2012; Skinner et al., 2010]. This is qualitatively similar to our model results (Figure 5b) and means that the vertical stratification became weaker, allowing more carbon to escape the deep ocean into the atmosphere.
 We believe that the scenario presented here (atmospheric Δ14C should decline) is a viable alternative to the traditional scenario (atmospheric Δ14C should rise) as a response to FW-triggered shutdown of AMOC. A point of departure of this work from previous modeling studies of atmospheric Δ14C is recognizing that the bipolar seesaw offers a convenient mechanism for the Southern Ocean to vent old carbon and reduce atmospheric Δ14C while AMOC is reduced. The new mechanism has some distinct advantages. It is consistent with the observations that atmospheric pCO2 increased and atmospheric Δ14C decreased during H1 and the YD and with the understanding that these changes were driven by enhanced Southern Ocean ventilation. The initial rise in atmospheric Δ14C at the outset of the YD could be explained if there was a delayed activation of the bipolar seesaw or a change in its intensity. Finally, the mechanism we proposed invokes an oceanic teleconnection, whereby an FW perturbation in the North Atlantic drives a southern Δ14C response, but this does not preclude an atmospheric teleconnection from acting in concert.
 Support for KM was provided by visiting professorship from the Atmosphere and Ocean Research Institute of The University of Tokyo and sabbatical support from the University of Minnesota. Support for YY was provided by JSPS (NEXT program GR031). Numerical computation was carried out using resources at the University of Minnesota Supercomputing Institute.