Abrupt Bølling‐Allerød Warming Simulated under Gradual Forcing of the Last Deglaciation

During the last deglaciation, a major global warming trend was punctuated by abrupt climate changes, likely related to Atlantic meridional overturning circulation (AMOC). One problem is that an abrupt increase in the AMOC during the Bølling‐Allerød (BA) transition occurred when the melting of Northern Hemisphere ice sheets was significant, which tended to weaken the AMOC. Here, from transient simulations of the last deglaciation using an atmosphere‐ocean general circulation model, we show that an abrupt increase in the AMOC during the BA transition could occur without reduction in glacial meltwater. The abrupt increase in the AMOC accompanied abrupt warming in Greenland and sea ice retreat in the North Atlantic, consistent with proxies and previous modeling studies. The results imply that abrupt BA warming during the middle stage of the last deglaciation was a response to gradual warming under the presence of meltwater from continental ice sheets.


Introduction
The transition from the Last Glacial Maximum (LGM) to the present interglacial occurred approximately 21,000 to 9,000 years ago, known as the last deglaciation and the last ice age termination Denton et al., 2010). The melting of continental ice sheets raised the global mean sea level during the last deglaciation Carlson & Clark, 2012). One remarkable point is that abrupt and large climate changes occurred during the middle stage of the last deglaciation, superimposed on gradual global warming Shakun & Carlson, 2010). One event was the Bølling-Allerød (BA) transition (about 14.7 to 14.2 ka, , characterized by abrupt and drastic warming in Greenland (10°C within a few decades, Buizert et al., 2014). During the BA interval (about 14.7 to 13 ka), the warming trend in Antarctica and the Southern Hemisphere stopped and turned into a cooling trend (Pedro et al., 2016). Reconstructions indicate that there was significant warming in the Southern Hemisphere during Heinrich 1 (H1,~18 to 14.7 ka) just before BA . Reconstructions also indicate that the Atlantic meridional overturning circulation (AMOC) was weaker and that North Atlantic Deep Water formation was shallower during H1 and rapidly increased in the BA transition (McManus et al., 2004;Roberts et al., 2010). The AMOC is commonly invoked to explain abrupt climate changes through the "bipolar seesaw" mechanism associated with the meridional heat transport of the AMOC (Broecker, 1998;Clark et al., 2002;Rahmstorf, 2002).
Transient simulation is suitable to understand the causes and mechanisms of the abrupt climate changes and the phase relationships of the last deglaciation. Recent studies have used atmosphere-ocean coupled general circulation models (AOGCMs) and conducted transient simulations of the last deglaciation. In these simulations, abrupt increase in the AMOC was simulated in response to cessation of meltwater input into the North Atlantic (He et al., 2013;Liu et al., 2009;Menviel et al., 2011). In contrast, gradual warming of the last deglaciation has been shown to have caused abrupt increase in the AMOC as the warming increased salinity transport from the South Atlantic, based on an ocean general circulation model (Knorr & Lohmann, 2003). In previous studies, abrupt increase in the AMOC has been explained by gradual meltwater input into the Southern Ocean reducing the density of bottom water mass originating from the Southern Ocean (Weaver et al., 2003), and the response to gradual orbital forcing under the presence of meltwater in the North Atlantic (Ganopolski & Roche, 2009), based on an Earth system model. Recently, AOGCM experiments showed that abrupt climate changes can be induced by gradual change in CO 2 concentration under the presence of meltwater and intermediate ice sheet (Zhang et al., 2017). It was shown that the state of the AMOC depends on the background climate (Brown & Galbraith, 2016;Kawamura et al., 2017;Klockmann et al., 2018). However, abrupt increase in the AMOC in response to gradual forcing of the last deglaciation has not yet been shown by AOGCMs, and the respective roles of forcings in the abrupt climate changes should be clarified. Recently, a working group of the Paleoclimate Model Intercomparison Project Phase 4 (PMIP4) collected boundary conditions for climate models and proposed a protocol for transient simulation of the last deglaciation (Ivanovic, Gregoire, Kageyama, et al., 2016). In the present study, we show that from transient simulation of the last deglaciation using MIROC AOGCM mostly following the PMIP4 protocol, an abrupt recovery of the AMOC was caused by gradual boundary condition changes during the last deglaciation. We discuss the respective roles of climate forcings by analyzing the time series of simulated climate changes and by conducting sensitivity experiments.

Model and Experimental Design
We used MIROC 4m AOGCM, the same model as that used in a previous study (Kawamura et al., 2017). MIROC 4m is based on MIROC 3.2 (Hasumi & Emori, 2004), which contributed to the Coupled Model Intercomparison Project phase 3 and PMIP2. The resolution of the atmospheric component was T42 (about 2.8°× 2.8°) with 20 vertical levels and that of the ocean component was about 1.4°× 1°with 43 vertical levels. The MIROC 4m produced vigorous AMOC under the LGM when submitted to PMIP2 (Otto-Bliesener et al., 2007). The coefficient of the horizontal isopycnal layer thickness diffusivity of the ocean model was changed to 7.0 × 10 6 cm 2 /s from 3.0 × 10 6 cm 2 /s, and the present model produces weak AMOC under the LGM because of enhanced dense Antarctic Bottom Water (AABW) formation (Kawamura et al., 2017). The MIROC has been used to investigate the climate of the LGM with radiative forcing and climate feedback (Yoshimori et al., 2009), the effect of ice sheets on the climate and the AMOC (Abe- Kawamura et al., 2017;Sherriff-Tadano et al., 2018), ocean biogeochemical cycles (Kobayashi et al., 2015;Kobayashi & Oka, 2018;Yamamoto et al., 2019), and mass balance of the Antarctic ice shelves (Kusahara et al., 2015;Obase et al., 2017).
We conducted transient simulation from the LGM (21 ka) to the end of the BA transition (13 ka) by changing insolation, atmospheric greenhouse gas (GHG) concentrations, and meltwater input, following the protocol of PMIP4 (Ivanovic, Gregoire, Kageyama, et al., 2016). The model was spun-up with the condition of the LGM lasting for more than 30,000 years, and the orbital parameters and atmospheric GHG concentrations changed over time based on reconstructions (Figures 1a and 1b, Berger & Loutre, 1991;Bereiter et al., 2015;Loulergue et al., 2008;Schilt et al., 2010). The coastlines, bathymetry, and ice sheet height and extent were fixed to those of the LGM throughout the experiment to remove the complex influence of ice sheets on the AMOC. The time series of glacial meltwater flux was uniformly applied to 50−70°N in the North Atlantic following a conventional method (Kageyama et al., 2013). The time series of meltwater was from an ice sheet reconstruction (Ice-6G) following PMIP4 (Figure 1c), and the cumulative meltwater was applied not to exceed the sea level rise since the LGM. As the meltwater flux was applied uniformly to the fixed area of the North Atlantic within 50-70°N, and meltwater input into the other ocean  Kawamura et al., 2017); (h) global mean ocean temperature. The reconstructed global mean ocean temperature is relative to present day ("Mix" of Bereiter et al., 2018), and the simulated value is the raw value. The 15-year running mean is displayed for the simulated 2-m air temperature, and the vertical bars are scaled to make the amplitude of reconstructions and model results closer. The blue and red shaded areas correspond to the periods of H1 and BA, following Ivanovic, Gregoire, Kageyama, et al. (2016). basins was not considered, the cumulative meltwater input into the North Atlantic was less than the actual sea level rise ( Figure 1d). There was no large reduction in meltwater input, as indicated by a threedimensional dynamic ice sheet model . The amount of seawater applied to the North Atlantic was collected from the rest of the global ocean to maintain the global salinity and freshwater content, as well as bathymetry. The global salinity content was assumed to be the same as the present day. We first conducted a number of experiments that differed in the time series of meltwater fluxes and found that the timing of abrupt increase in the AMOC depended on the meltwater flux (Supporting Information, Figure S1). We selected one experiment to serve as a reference experiment (REF), as the timing of BA warming was close to reconstructions (red lines of Figure 1). In the REF experiment, the time series of the meltwater input followed the reconstructed volume of the Northern Hemisphere ice sheet (ICE-6G, Peltier et al., 2015) during 21 to 16 ka, and the meltwater flux was kept to 0.06 Sv thereafter (Figure 1d).

Results
In our simulations, we found that the abrupt increase in the AMOC could be caused without reduction in meltwater. In the REF experiment, a relatively weak AMOC (7 to 8 Sv) compared with the present-day simulation (16 Sv) continued from the LGM to H1 (Figure 1e). The AMOC did not weaken during H1, as the meltwater had no peak at this time in this experiment. At the beginning of the BA transition, the AMOC abruptly strengthened to about 30 Sv. The abrupt increase in AMOC strength across the BA transition was consistent with reconstructions based on kinematic proxies of Pa/Th (Figure 1e, Ng et al., 2018). The Greenland surface air temperature abruptly rose as the AMOC strengthened ( Figure 1f). Although the simulated temperature changes were smaller than those of reconstructions, it is notable that the speed of the warming event was consistent with surface temperature reconstructions (Buizert et al., 2014). Antarctic warming stopped soon after the recovery of the AMOC (Figure 1g), and a cooling trend continued for~2,000 years, as recorded in Antarctic ice cores (Kawamura et al., 2017), as the AMOC maintained a vigorous mode. The global mean ocean temperature turned into a cooling trend after the BA transition (Figure 1h), similar to the Antarctic temperature, which was consistent with reconstructions based on atmospheric noble gas concentrations (Bereiter et al., 2018).
The abrupt climate change accompanied large changes in sea ice and deep water formation in the North Atlantic. At 21 ka (LGM), the winter sea ice edge in the North Atlantic was close to~30°N, and the deep water formation in the North Atlantic was weak (Figure 2). The winter sea ice extent at the LGM was consistent with reconstructions (de Vernal et al., 2005). The sea ice extent and the AMOC at 15 ka were similar to those of the LGM, probably because the state of the AMOC was weak throughout this period. The winter sea ice retreated to about 60°N at 14.7 ka, and this retreat of sea ice initiated the deep water formation and associated convective mixing (Figure 2). This change triggered a drastic release of heat stored in the subsurface of the North Atlantic by wintertime sea surface cooling and increased heat transport from the South Atlantic (Figure 3), which promoted further retreat of sea ice. The winter sea ice expanded slightly at 13 ka, and a strong AMOC persisted. These changes in the North Atlantic sea ice and ocean associated with abrupt resumption of the AMOC are consistent with previous studies (Buizert et al., 2014;Zhang et al., 2017;Zhu et al., 2014). Although there was little change in sea ice extent and meridional overturning circulation in the North Atlantic between 21 to 15 ka, there was gradual but steady change in the deep Atlantic Ocean, characterized by warming and freshening at the bottom of the Atlantic Ocean ( Figure 3). This gradual change was likely induced by warming and sea ice melting in the Southern Ocean, which promoted warming and freshening of bottom water of the North Atlantic by the AABW cell (Liu et al., 2015). The gradual warming and freshening of the bottom of the North Atlantic contributed to weakened stratification in the North Atlantic ( Figure S2). From 15 to 14.7 ka, the tendencies of the temperature and salinity changed significantly as a result of the overshoot in the AMOC. The salinity and density trends during 14.7 to 13 ka were similar to the reverse of 21 to 15 ka. Therefore, gradual warming of the deglaciation caused gradual weakening of the stratification of the North Atlantic and finally caused the drastic retreat of sea ice and initiation of deep convection. We calculated Mov, a stability index of the AMOC defined by freshwater transport across the Southern and Northern boundaries of the Atlantic (Liu et al., 2015). Mov was negative from the LGM to H1 and had decreasing trend, and it changed to positive as soon as the AMOC recovered at the BA transition ( Figure S3). A decrease in the salinity of South Atlantic contributed to the increase in Mov ( Figure S4) as Figure 4 of Liu et al. (2015). This change of Mov during the deglaciation has large similarity between our study and Liu et al. (2015), despite the relation between AMOC change and meltwater was very different.
The global map of surface air temperature changes indicates that the temperature changes differed between the stages of the deglaciation (Figure 4a). The surface temperature change exhibited warming in both polar regions, whereas the tropics exhibited slight cooling from 21 to 18 ka. This temperature change was likely induced by the increase of obliquity, as the CO 2 rise during this period was small. The surface temperature increased globally from 18 to 16 ka, as the rise of CO 2 in this period overwhelmed orbital forcing. The Antarctic region exhibited greater warming from 16 to 15 ka, before the BA transition. The period of 15 to 13 ka, across the BA transition, was characterized by drastic warming in the Northern Hemisphere and gradual cooling in the Southern Hemisphere, consistent with the pattern of AMOC resumption (Pedro et al., 2018). The surface temperature averaged for the both hemispheres also showed that bipolar climate change occurred after the BA transition and that the cooling trend of the Southern Hemisphere continued for more than 1,000 years (Figure 4b).
We also found that the timing of abrupt increase in the AMOC was affected by the flux of meltwater. If no meltwater was applied (green lines of Figure 5a), an abrupt recovery in the AMOC occurred at around 19 ka. If the meltwater was less than that of the REF experiment (blue lines of Figure 5a), an abrupt recovery of the AMOC occurred at~16 ka, earlier than in reconstructions. We compared simulated surface temperature change at NEEM, Greenland with nitrogen isotopes as an indicator of abrupt surface temperature change (Rosen et al., 2014). The simulated abrupt warming in Greenland temperature occurred within 100 years in both simulations, which were consistent with the reconstruction (Figure 5b). These results indicate that small changes in the meltwater flux may affect the timing in the abrupt increase in the AMOC but do not affect the abruptness of the increase in the AMOC and surface temperature changes in Greenland.

Discussion
As a summary of the results, from transient simulations of the last deglaciation, we found that an abrupt increase in the AMOC could be caused without stopping meltwater in the North Atlantic. The results of the present study are consistent with previous studies in that gradual background climate change can cause abrupt increase in the AMOC, with an ocean general circulation model (Knorr & Lohmann, 2003;Knorr & Lohmann, 2007) and Earth system model of intermediate complexity (Ganopolski & Roche, 2009), and an AOGCM under gradual change in atmospheric CO 2 (Zhang et al., 2017). The new finding and strength of the study is that abrupt BA-like climate change can be simulated by an atmosphere-ocean coupled GCM under reconstructed forcing of insolation and GHGs of the last deglaciation, and meltwater from ice sheets. As found in Antarctic Ocean temperature, salinity, and density changes before the BA transition (Figure 3), we speculate that gradual warming freshened and warmed the Southern Ocean through reduction in sea ice production and formation of AABW (Kawamura et al., 2017;Liu et al., 2005). The AABW affected the water mass of the bottom water of the North Atlantic, and the threshold of the resumption of the AMOC was exceeded by the resumption of the AMOC by overcoming the effect of meltwater, which acted to weaken the AMOC. Since a critical question is the stability of the AMOC, we analyzed a stability indicator of the AMOC. Mov in this study is very similar to the previous study of the last deglaciation (Liu et al., 2015), despite the AMOC change in the present study is caused by the gradual warming, not a reduction in meltwater. Abrupt increase in indicator of the stability seems to be a response to the changes in the AMOC, and therefore it may be difficult to forecast the timing of abrupt increase in the AMOC from the stability index. The relative contributions of the processes that affected the abrupt changes in the AMOC should be clarified in the future.
The sensitivity experiments demonstrated that the timing of abrupt increase in the AMOC depended on the meltwater. A notable point is that BA-like climate change occurred at 19 ka, in a very early stage of the last deglaciation, if meltwater flux was not applied at all, as shown in green lines in Figure 5a. In this experiment, the model was forced only by orbital parameters and atmospheric GHGs. The surface temperature trends in the early stage of the last deglaciation exhibited warming in the polar regions and cooling in the tropics (Figure 4a), indicating that increase in obliquity (Figure 1a) was the primary forcing of this period. Therefore, the results of the green line experiment suggest that gradual forcing of insolation, mainly by obliquity, could also induce abrupt climate changes. This result is consistent with a previous study in that under lower CO 2 , high obliquity enhances the AMOC by reducing the AABW (Galbraith & de Lavergne, 2019). In the REF experiment, continuous meltwater from the LGM contributed to preventing BA-like climate changes for about several thousand years, by freshening the sea surface and strengthening the stratification of the North Atlantic.
The simulated patterns of surface temperature changes reflect different forcing and climate responses of the last deglaciation. Gradual increase in obliquity warmed the polar regions of both hemispheres in the early stage of the last deglaciation (21 to 18 ka), consistent with reconstructions (WAIS Project Members, 2013). For the period of 18 to 16 ka, the reconstruction indicates that there was significant warming in the Southern Hemisphere ), but the model shows warming in both hemispheres with weak warming of the Southern Hemisphere (Figure 4b). We speculate that this was because sea ice change during H1 was too small and surface temperature changes were overwhelmed by GHGs. The simulated AMOC strength during H1 was not so weak compared with the LGM (Figure 1d), which may be inconsistent with proxies. Improved representation of H1 by improving the meltwater scheme should be investigated in the future. Reconstructions from North Pacific regions suggest that synchronous warming between the North Pacific and Greenland started earlier than the BA transition (Praetorius & Mix, 2014). The pattern of surface temperature change suggests that global warming induced by GHGs may produce synchronous warming between distant areas (Figure 4a). Several boundary conditions were simplified in the present study compared with reality. First, the area of meltwater input in the North Atlantic was constant over time. The meltwater should have reached various ocean basins outside the North Atlantic through river runoff (Clark et al., 2001;Tarasov & Peltier, 2005). Previous studies have indicated that different regions of meltwater flux produce different responses of the AMOC (Otto-Bliesner & Brady, 2010;Roche et al., 2009;Weaver et al., 2003). Second, the ice sheet configurations in the Northern Hemisphere that affect the AMOC were kept to those of the LGM. In particular, the winds over the Atlantic Ocean, critical to the AMOC (Oka et al., 2012), would be significantly influenced by the Laurentide Ice Sheet (Sherriff-Tadano et al., 2018). As shown by Zhu et al. (2014), the gradual retreat of the Northern Hemisphere ice sheet may have weakened the AMOC and affected abrupt climate changes. The impacts of improved representation of these ice sheet boundary conditions should be clarified in the future.
Reconstruction has indicated that the warm period of the BA persisted until the Younger Dryas, for about 1,700 years. At roughly the same time as the BA transition, there was a rapid sea level rise called meltwater pulse 1A (Deschamps et al., 2012). The sea level rise of this period corresponds to meltwater flux of~0.4 Sv, significantly higher than the applied meltwater in the present study. We did not take into account such drastic increase in meltwater flux, as the AMOC would have significantly decreased if such an amount of meltwater were applied to the North Atlantic (Bethke et al., 2012;Ivanovic et al., 2018). As the meltwater from ice sheets flows into different ocean basins by river runoff (Ivanovic, Gregoire, Wickert, et al., 2016), the relationship between meltwater pulse 1A and vigorous AMOC during BA is beyond the scope of this study.
The present study has implications for the chain of events during the glacial termination. During the early stage of the last deglaciation, rising summer insolation led to melting continental ice sheets in the Northern Hemisphere, which delivered meltwater to the North Atlantic and weakened the AMOC during the Heinrich events (Denton et al., 2010). The reduced AMOC warmed the Southern Ocean through the bipolar seesaw mechanism and raised atmospheric CO 2 (Menviel et al., 2014). As found in the present study, gradual increase in atmospheric CO 2 during H1 may have contributed to cause warming and freshening in the deep water and weakened stratification of the North Atlantic. This may have contributed to the abrupt increase in the AMOC during the BA transition, much earlier than the end of the last glacial termination. However, from comparison of the last four deglaciations, not all deglaciations accompanied BA-like climate changes as in the last deglaciation (Cheng et al., 2009). We expect further investigations on the critical processes, and model-data comparisons will improve understanding of the nature of the climate system during deglaciations in the past.