Glacial abrupt climate changes and Dansgaard-Oeschger oscillations in a coupled climate model

Transcription

1 PALEOCEANOGRAPHY, VOL. 21,, doi: /2005pa001238, 2006 Glacial abrupt climate changes and Dansgaard-Oeschger oscillations in a coupled climate model Zhaomin Wang 1,2 and Lawrence A. Mysak 1 Received 3 November 2005; accepted 14 December 2005; published 7 April [1] There are three fundamental features which characterize large glacial millennial (Dansgaard-Oeschger) oscillations: (1) the climatic transitions were abrupt and large; (2) the lengths of both interstadials and stadials and the period of Dansgaard-Oeschger oscillations were not uniform; and (3) there were no large millennial oscillations during an early stage of a glacial period and a peak glacial period. In this modeling study we offer a consistent explanation for these three features by employing an Earth system Model of Intermediate Complexity. We demonstrate that a moderate global cooling forces the Atlantic meridional overturning circulation (MOC) into an unstable state and hence causes the flip-flop of the Atlantic MOC between a strong mode and a weak mode. The durations of both interstadials and stadials associated with these millennial oscillations are modulated by the changing background climate in qualitative agreement with the observations. In a warm climate the Atlantic MOC is strong and stable, with the deep water formed mainly by intense heat loss to the atmosphere. In a cold climate the Atlantic MOC is weak and stable, and this mode is largely maintained by the process of sea ice brine rejection. Since the Dansgaard-Oeschger oscillations result from an alternation between these two MOC states during an intermediate phase climate, we conclude that brine rejection plays a necessary role in the oscillations, confirming a hypothesis suggested in some proxy data studies. Citation: Wang, Z., and L. A. Mysak (2006), Glacial abrupt climate changes and Dansgaard-Oeschger oscillations in a coupled climate model, Paleoceanography, 21,, doi: /2005pa Introduction [2] Greenland ice core records reveal that much of the last glacial period was punctuated by millennial-scale climatic fluctuations, which are termed Dansgaard-Oeschger (D-O) oscillations [Dansgaard et al., 1993; North Greenland Ice Core Project Members, 2004]. A much longer North Atlantic sediment core record shows that D-O oscillations also occurred during previous glacial periods [McManus et al., 1999]. There are three key features of D-O oscillations. First, there are abrupt and large climatic transitions in the oscillations. During one cycle of the oscillations, there is a sudden warming over several decades or less, a gradual cooling phase followed by a rapid cooling, and finally a gradual warming phase leading up to the next sudden warming. Shifts of 9 to 16 C were observed over Greenland during a cycle [Schwander et al., 1997; Lang et al., 1999]. Second, the millennial oscillations are quasiperiodic, with the period being modulated by the nature of the glacial background climate: long interstadials (warm states) occurred during late (oxygen isotope) stage 5 and early stage 3 when the background glacial climates were relatively warm, while long stadials (cold states) and short interstadials occurred during stages 4 and 2, when the background glacial climates were relatively cold [Bond et al., 1999; 1 Earth System Modelling Group, Department of Atmospheric and Oceanic Sciences, McGill University, Montreal, Quebec, Canada. 2 Now at British Antarctic Survey, Cambridge, UK. Copyright 2006 by the American Geophysical Union /06/2005PA001238$12.00 Schulz, 2002]. The D-O periods were thus relatively long when the background climate was either relatively warm or cold; when the background climate state was at an intermediate phase, the periods were relatively short. Third, the extensive North Atlantic Ocean sediment record reveals that the millennial oscillations were significantly suppressed when the background climate was either warm (interglacial and early stage of a glacial) or very cold (peak or maximum glacial) [McManus et al., 1999]. [3] Numerous modeling studies have investigated the possible causes of millennial oscillations, with the model complexity ranging from box [Birchfield et al., 1994] to comprehensive general circulation models [Tziperman, 2000], and from stand-alone ocean circulation models [Marotzke, 1989] to coupled multicomponent climate models [Ganopolski and Rahmstorf, 2001, 2002; Knutti et al., 2004]. There have also been many proxy data studies of D-O oscillations [Broecker and Denton, 1990; Bond et al., 1999; Dokken and Jansen, 1999; van Kreveld et al., 2000], and a variety of mechanisms have been proposed for these oscillations. Early studies suggested that D-O oscillations would be generated by internal oceanic processes [Broecker and Denton, 1990; Weaver, 1999]. Recently, in coupled climate models under the forcing of glacial boundary conditions, an external weak periodic millennial-scale freshwater forcing with noise excitation [Ganopolski and Rahmstorf, 2002] or strong freshwater forcing [Sakai and Peltier, 1997; Timmermann et al., 2003; Knutti et al., 2004] triggered D-O oscillations. Ice sheet-thermohaline circulation interactions have also been proposed as a cause for D-O oscillations [Birchfield et al., 1994; van Kreveld et al., 2000; Wang and 1of9

2 Mysak, 2001]. Northern North Atlantic sea ice brine rejection has been found to play an important role during the oscillations in some proxy data studies [Dokken and Jansen, 1999; van Kreveld et al., 2000]. Routing switch of river discharge has also been recently hypothesized as a potential mechanism for the D-O oscillations [Clark et al., 2001]. In the work of Gildor and Tziperman [2003] and Kaspi et al. [2004], sea ice switch-like behavior, driven by variability in a weak ocean circulation, causes large temperature changes during a glacial period. In their box model studies, Sima et al. [2004] and Olsen et al. [2005] have simulated that the oscillations are modulated by slowly changing background forcing. [4] In this study, we invoke the stability of the glacial Atlantic meridional overturning circulation (MOC) in a slowly changing background climate in an Earth system Model of Intermediate Complexity (EMIC) to answer the following three major questions: (1) Why were the glacial climatic transitions abrupt and large? (2) Why were the lengths of both interstadials and stadials and the period of D-O oscillations not uniform? (3) Why were there no large millennial oscillations during an early stage of a glacial period and a peak glacial period? [5] While early modeling studies [Ganopolski and Rahmstorf, 2001; Schmittner et al., 2002] have shown that the stability of the glacial Atlantic MOC was reduced, the thermal modulation effects on the stability of this circulation by the slowly changing glacial background climate have not been thoroughly studied in state-of-the-art coupled climate system models. We will show that the Atlantic MOC is not stable in a background climate which is undergoing a period of moderate cooling. This Atlantic MOC instability results in D-O oscillations, which are characterized by flip-flops between strong and weak modes. The Atlantic MOC is stable when the background climate is either relatively warm or relatively cold, with the circulation being strong for the relatively warm climate and weak for the relatively cold climate. We will also show that the durations of the interstadials and stadials vary with the background climate, and also that the period of the oscillations is not uniform throughout the glacial period. Thus in this modeling study, we offer a consistent explanation for the above three features of D-O oscillations using an EMIC. For the first time, we will also show theoretically that sea ice brine rejection plays a necessary role in maintaining the weak Atlantic MOC mode associated with these oscillations, which corroborates a hypothesis suggested in some proxy data analysis studies [Dokken and Jansen, 1999; van Kreveld et al., 2000; Clark et al., 2002]. 2. Model and Experimental Design [6] In this study, the McGill Paleoclimate Model-2 (MPM-2) [Wang, 2005] is employed. The MPM-2 is an extension of Wang and Mysak [2000, 2002], and Wang et al. [2005], with several major improvements. The variables in the MPM-2 are sectorially or zonally averaged across each continent and ocean basin, with variables in or over North America and Eurasia downscaled to a 5 by 5 resolution in order for us to couple a two-dimensional dynamic ice sheet model [Marshall and Clarke, 1997] to the model [Wang and Mysak, 2002]. A new parameterization for the solar energy disposition, which concerns the amount of solar energy absorbed by the atmosphere and by the surface and the amount of solar energy reflected to space, has been employed [Wang et al., 2004], and a global dynamic vegetation model (VECODE (Vegetation Continuous Description)) [Brovkin et al., 2002] has been coupled to the model [Wang et al., 2005]. In the MPM-2, the model domain is extended from (75 S, 75 N) to (90 S, 90 N), surface winds are parameterized [Petoukhov et al., 2000], and there are no oceanic heat and freshwater flux adjustments [Wang, 2005]. The MPM-2 is now a global climate model which consists of atmosphere, ocean, sea ice, land surface, continental ice and vegetation components. [7] In the control experiment of this study, the standard LGM boundary conditions used by the Paleoclimate Modeling Intercomparison Project (PMIP) [Joussaume and Taylor, 2000] are employed to obtain a glacial background climate; that is, the orbital forcing is for 21 kyr B.P., the ice sheets are prescribed as at the LGM [Peltier, 2004], the atmospheric CO 2 level is at 200 ppm, and the global salinity is increased one ppt to take into account the global sea level drop. The sea ice meridional advection velocity is prescribed: its annual mean value increases from 0.15 cm/s in the Arctic Ocean to 3.0 cm/s at 40 N. In the Southern Ocean, the prescribed sea ice velocity increases from 1 cm/s at 70 S to 5 cm/s at 40 S. A white noise freshwater forcing is also applied over the N latitude band of the North Atlantic. The standard deviation of the integrated white noise freshwater flux is 0.03 Sv and the decorrelation time step is 1 year. 3. Abruptness, Periodicity and Atlantic MOC Modes [8] Under the forcing of the glacial boundary conditions and the imposed white noise freshwater forcing over N of the North Atlantic, the model was run for 20 thousand years in the control experiment. After the spin-up phase, a periodic oscillation persists throughout the whole model run, with a period of approximately 1400 years. It is argued by Winton and Sarachik [1993], Ganopolski and Rahmstorf [2002] and Timmermann et al. [2003] that ocean dynamics mainly accounts for this timescale. [9] Figure 1 shows the time series of the Atlantic MOC (we use the maximum stream function below the Ekman layer in the Atlantic to represent the intensity of the Atlantic MOC), Atlantic sea surface salinity at 57.5 N, surface air temperature (SAT) at 62.5 N over the North Atlantic and SAT at 62.5 S for the time interval from 10 to 15 thousand model years. During one cycle, the Atlantic MOC jumps abruptly from a relatively weak state to a peak one and then slowly weakens; however it stays at a strong state for several hundred years. When the Atlantic MOC weakens to a critical value, it abruptly switches to a weak state and then it takes almost 1000 years to recover until the next abrupt jump (Figure 1a). Similar behavior of the Atlantic MOC was also simulated by Ganopolski and Rahmstorf [2001] by prescribing a periodic and weak freshwater 2of9

3 Figure 1. Simulated D-O oscillations in the control experiment with the standard glacial boundary conditions used by Paleoclimate Modeling Intercomparison Project [Joussaume and Taylor, 2000] and the imposed white noise freshwater forcing. (a) Atlantic meridional overturning circulation (MOC) (represented by the maximum stream function below the Ekman layer). (b) Sea surface salinity at 57.5 N in the North Atlantic. (c) Surface air temperature (SAT) at 62.5 N over the North Atlantic. (d) SAT at 62.5 S over the Southern Ocean. forcing. The evolution of the sea surface salinity in the North Atlantic has the same abrupt and periodic behavior, with an amplitude of 2 ppt (Figure 1b). This is similar to what is found by Ganopolski and Rahmstorf [2001] and it is in agreement with proxy data reconstructions [van Kreveld et al., 2000]. [10] The change of zonally averaged Atlantic SAT over N clearly shows rapid northern climate transitions, either from a cold state (stadial) to a warm state (interstadial) or from a warm state to a cold state on a multidecadal timescale (Figure 1c). After each rapid warming or cooling, there is a gradual cooling or warming, respectively. The maximum temperature change is up to 6 C. The zonally averaged SAT changes over North America and Eurasia at the same latitude band are only 2.5 C and 1.4 C respectively, and the zonally averaged Pacific SAT change over N is just 1.8 C, indicating that the North Atlantic is the origin of the warming and cooling events during D-O oscillations. The strong cooling and warming events are mainly caused by the Atlantic MOC mode switches, but these changes are significantly amplified by the concurrent large North Atlantic sea ice extent changes. The Northern Hemisphere sea ice area decreases from a peak stadial value of km 2 to a peak interstadial value of km 2. This change happens mainly in the North Atlantic. [11] The zonally averaged SAT in the southern high latitudes also shows a periodic, delayed out-of-phase behavior as compared with the North Atlantic SAT (Figure 1d). However, the amplitude of these oscillations is small (around 0.5 C) because the oceanic heat transport change is small (see the discussion in section 3). Note that the atmospheric CO 2 level is prescribed. If a variable atmospheric CO 2 were used in the model, these Southern Ocean oscillations would perhaps be larger. Observations show CO 2 rises of less than 10 ppm during the cool phase of less pronounced D-O oscillations and up to 20 ppm rises during long lasting cool phases of large oscillations [Stauffer et al., 1998; Stocker and Marchal, 2000]. However, the existence of the SAT changes over the Southern Ocean corresponding to less pronounced D-O oscillations in the proxy data is less evident [Blunier and Brook, 2001; Clark et al., 2002]. [12] The stream function in the Atlantic Ocean basin is shown in Figure 2a and 2b for both the strong and weak modes; these plots are snapshot outputs at model year 10,430 and 10,660, respectively. Deep water forms in the subpolar region of the North Atlantic in both the strong and weak modes, with a shallower and less vigorous overturning circulation for the weak mode. Although some modeling results suggest a southward shift of the deep water convection site [Ganopolski et al., 1998; Knorr and Lohmann, 2003], proxy data support the concept that the deep water formed more or less continuously in the same subpolar region of the North Atlantic during the last glacial period [Vidal et al., 1998; Dokken and Jansen, 1999; Sarnthein et al., 2000; Clark et al., 2002]. The UVic intermediate complexity climate model also simulates the deep water formation in the subpolar region under LGM boundary conditions [Weaver et al., 1998]. Our subsequent sensitivity experiments show that northern sea ice brine rejection is responsible for the continuous maintenance of the subpolar deep water formation in this sea-ice-covered region (see the related discussion in section 5), which supports the hypothesis based on proxy data reconstructions [Vidal et al., 1998; Dokken and Jansen, 1999]. [13] The annual mean oceanic heat transport contrast between the strong and the weak mode is much larger in the Atlantic (Figure 3a) than in the Indo-Pacific (Figure 3b). At 40 N in the Atlantic the northward heat transport for the strong mode is 0.4 PW larger than that for the weak mode, while in the Indo-Pacific, the heat transport difference is small in the northern middle and high latitudes. This demonstrates again that the large northern climate change 3of9

4 Figure 2a. This is because the oceanic heat transport at the peak state of the Atlantic MOC is just slightly larger than that at the state as shown in Figure 2a. Figure 3a shows that the maximum oceanic heat transport change between the two modes is 0.4 PW at 40 N, which is close to the maximum heat transport change of 0.3 PW given by Kaspi et al. [2004]. 4. Modulation by a Changing Background Climate [15] Under the prescribed standard PMIP LGM boundary conditions [Joussaume and Taylor, 2000], the global annual mean SAT drop relative to the present-day SAT is 4.1 C averaged over one cycle in our model. This drop is at the low end of the range found by most PMIP1 investigators (i.e., 4 6 C) [Joussaume and Taylor, 2000] and smaller than that obtained by Ganopolski et al. [1998] (6.2 C). Note, however, that some radiative forcings are missing in the MPM-2, such as CH 4 and dust, and that the increased continental area because of sea level drop [Broccoli, 2000], which affects the surface albedo, is neglected. At this stage, it is difficult to include all radiative forcings with enough reliability and accuracy. In order to investigate how the millennial oscillations are modulated by the changing glacial background climate state, in the next experiment the atmospheric CO 2 level is gradually decreased from 240 ppm at a rate of 1 ppm per thousand years; at the same time, we Figure 2. Atlantic MOC modes in the control experiment. (a) Stream function in the Atlantic at model year 10,430, defined as the strong mode. (b) Stream function in the Atlantic at model year 10,660, defined as the weak mode. during the oscillations is mainly induced by the Atlantic MOC mode switch. The cooling and warming that occur over the Pacific are principally induced by atmospheric heat transports. In the Southern Hemisphere, the change of the heat transport in the Atlantic is opposite to and larger than the change in the Indo-Pacific. Consequently, the zonal integral of the oceanic southward heat transport is slightly larger for the weak mode than for the strong mode. The southern high latitude SAT thus has a cooling (warming) corresponding to the warming (cooling) in the Northern Hemisphere. The opposite changes of the oceanic heat transport in the Atlantic and Indo-Pacific lead to a small difference in the global southward heat transport between the two Atlantic MOC modes, which is one of the reasons for a small simulated zonally averaged SAT change at 62.5 S (Figure 1d). This also indicates a complicated spatial pattern for the global effect of the Atlantic MOC mode switch. [14] In the work of Gildor and Tziperman [2003] and Kaspi et al. [2004], sea ice switch-like behavior driven by a small Atlantic MOC variability is responsible for large temperature changes during a glacial period. Although the Atlantic MOC jumps to about 40 Sv in this study, the sea ice extent (image not shown) and temperature (see Figure 1c) corresponding to this state are just slightly different from those corresponding to the Atlantic MOC state shown in Figure 3. Oceanic heat transports in the control experiment. Oceanic heat transports in the (a) Atlantic, (b) Indo- Pacific, and (c) global ocean. The red curve is for the strong mode, and blue curve is for the weak mode. 4of9

5 Figure 4. Ice core data and the simulated modulation of D-O oscillations by the background climate in the experiment with a decreasing atmospheric CO 2 level (from 240 to 160 ppm) and other conditions fixed as in the control experiment. (a) Time series of d 18 O from the Northern Greenland Ice Core Project ice core [North Greenland Ice Core Project Members, 2004]. (b) Simulated Atlantic MOC. (c) Global annual mean SAT. (d) SAT at 62.5 N over the North Atlantic. Oxygen isotope stages (OIS) 5, 4, 3, and 2 are shown in Figure 4a with red representing a relatively low ice volume and a warm climate and blue representing a relatively large ice volume and a cold climate [Lambeck and Chapell, 2001]. The red and blue lines in Figure 4d mark the lengthy interstadials and the lengthy stadials, respectively; the oscillations associated with these warm and cold events evidently have longer periods than the oscillations in the middle of the time series. Note that the time interval is from 10 to 60 kyr for Figure 4d. maintain the other conditions and the white noise freshwater forcing as in the control experiment. We note that the modulation effects of a changing glacial background climate on millennial oscillations can also be obtained by (1) prescribing the evolution of ice sheets between 60 and 30 kyr B.P. [Claussen et al., 2003], (2) imposing a North Atlantic freshwater forcing which increases with ice volume [Sima et al., 2004], and (3) introducing a variable orbital forcing [Olsen et al., 2005]. The model is integrated for 80 thousand years after the spin-up phase. The final CO 2 level of 160 ppm is thus reached at the end of the model run. By changing the CO 2 level from 240 to 160 ppm, a gradual but fairly large cooling of the background climate is realized. The global annual mean SAT changes from 10.9 C (a 3.2 C cooling relative to the present day) to 8.3 C (a 5.8 C cooling). [16] The proxy data from the North Greenland Ice Core Project (NGRIP) ice core [North Greenland Ice Core Project Members, 2004] (Figure 4a) suggest that millennial oscillations are suppressed during the early part of the glacial and at the LGM. A much longer ocean sediment core record [McManus et al., 1999] confirms that this type of behavior was also the case for the past 0.5 million years, which covers the past five glacial-interglacial cycles. These observations thus strongly suggest that a moderate global cooling favors the millennial oscillations. Also, the glacial millennial oscillations are not of uniform structure: they have a much longer period during late stage 5 and early stage 3 (warmer background climate) and during stages 4 and 2 (colder background climate) (see Figure 4a). When the background climate is warmer, the interstadial phases are much longer, whereas when the background climate is colder, the stadial phases are much longer [Bond et al., 1999; Schulz, 2002]. During a cooling Bond cycle, which has smaller ice sheets and a higher atmospheric CO 2 level [Stauffer et al., 1998; Stocker and Marchal, 2000] at the beginning, the interstadial phase is longer for the first millennial cycle; the stadial phase is longer and the interstadial phase is much shorter for the last millennial cycle [Bond et al., 1993]. We also note that the long stadial for the last cycle is usually accompanied by a massive iceberg discharge event (Heinrich event) several hundred years after the rapid cooling [Bond and Lotti, 1995]. [17] Figure 4b shows that the Atlantic MOC gradually slows down when the global annual mean SAT drops (see Figure 4c). Once a threshold value T 1 of the SAT (T 1 = 10.5 C when CO 2 is ppm) is reached, millennial oscillations appear. When another SAT threshold value T 2 (T 2 = 9.7 C when CO 2 is ppm) is reached, the millennial oscillations disappear. Thus in our simulation the millennial oscillations exist for the SAT in the range of 10.5 to 9.7 C and for CO 2 in the range of to ppm under prescribed LGM ice sheets. The period is longer when the SAT drops just beyond the first threshold value T 1 and each oscillation has a relatively long interstadial phase (marked by red bar in Figure 4d); the period also becomes longer when the SAT is approaching the second threshold value T 2 and each oscillation has a relatively long stadial phase and short interstadial phase (marked by blue bar in Figure 4d). This thermal modulation effect on the durations 5of9

6 Figure 5. Schematic diagram of the glacial Atlantic MOC intensity and stability versus the background climate. T 1 = 10.5 C, T 2 = 9.7 C, S = 22 Sv, and W = 16 Sv in this modeling study. of interstadial and stadial phases by the background climate was also simulated in a simple atmosphere-ocean model [Winton, 1997]. These simulated features of the glacial millennial oscillations are consistent with paleoclimate data (see Figure 4a, Bond et al. [1999] and McManus et al. [1999]). [18] Figure 4 demonstrates that a warm climate favors a strong and stable Atlantic MOC and a cold climate favors a weak and stable Atlantic MOC. An intermediate cold climate forces the Atlantic MOC to oscillate on a millennial timescale. A relatively long interstadial is produced by a relatively warm climate. The opposite is true for a relatively cold climate. Although a massive iceberg discharge event could also produce a long stadial, the modulation by the background climate should be taken into account. [19] Sensitivity studies have been also carried out to investigate the role of the imposed stochastic freshwater forcing in the above runs. In the experiment without the stochastic freshwater forcing, millennial oscillations still occur. However, the background climate range in which millennial oscillations occur becomes narrower. By doubling the stochastic forcing intensity, the background climate range widens. Also the imposed stochastic freshwater forcing slightly shortens the periods. [20] On the basis of the modeling results presented above, we highlight, in Figure 5, an important new mechanisms for the large glacial millennial oscillations. After the early stage of a glacial cooling period, further global cooling could weaken the Atlantic MOC by decreasing the equator-to-pole oceanic temperature contrast (and hence the horizontal oceanic density gradient) in the upper layer [Prange et al., 1997; Wang et al., 2002] and increasing the vertical stratification in the North Atlantic deep water formation region [Winton, 1997; Wang et al., 2002]. When the cooling reaches a critical state as indicated by T 1 in Figure 5, the Atlantic MOC is forced into an unstable state (the green dashed line) between S and W. This means that this thermalforcing-induced Hopf bifurcation point is passed and a limit cycle can be obtained under constant external forcing, with the Atlantic MOC oscillating between the two solid green curves. Thus a periodic oscillation occurs in the control experiment with constant glacial boundary conditions. When the background climate is further cooled down, another critical state as indicated by T 2 is reached. The Atlantic MOC evolves into a weak but stable state after this point. The oscillations hence disappear. In other words, large millennial oscillations can only occur in an intermediate cold glacial state. This mechanism has been hypothesized by Alley et al. [1999], Ganopolski and Rahmstorf [2002], and Shaffer et al. [2004]. [21] A strong salinity perturbation or freshwater forcing can also push the Atlantic MOC into an unstable state after passing a Hopf bifurcation point and hence cause the Atlantic MOC oscillations [Tziperman et al., 1994; Sakai and Peltier, 1997; Tziperman, 1997, 2000; Timmermann et al., 2003]. However, a strong freshwater forcing at the warm (interstadial) phase is at odds with the evidence that abrupt cooling occurred before the large freshwater forcing event (Heinrich event) [Bond and Lotti, 1995]. 5. Role of Sea Ice Brine Rejection [22] As mentioned in section 4, the change of sea ice extent significantly amplifies the warming and cooling induced by the rapid Atlantic MOC mode switch. This is due to two thermal features of sea ice: high surface albedo and low heat conductivity (insulation effect). We next show that sea ice brine rejection also plays a necessary role in the oscillations. [23] When sea ice forms, salt is rejected; when sea ice melts, freshwater is released. If sea ice forms and melts in the same region over a certain period, the integrated freshwater forcing is canceled out over this period. However, sea ice is generally transported away from a region of formation to a region of melt. In addition, we note that the area-integrated freshwater forcing is zero over sea-ice-covered regions, except for the case in which the total sea ice volume changes because of climate changes. In the MPM-2, the rate of the total sea ice volume change is small in terms of freshwater forcing (on the order of Sv or less). [24] Figure 6a shows the time series of the integrated freshwater flux over two latitude bands in the northern North Atlantic. In Figure 6b, we show the time series of integrated brine rejection (freshwater equivalent) to the north of 60 N (red curve) and freshwater release to the south of 60 N (blue curve). Figure 6b clearly indicates that when the Atlantic MOC is in a strong mode, both salt rejection to the north of 60 N and freshwater release to the south of 60 N are very small, consistent with a very small sea ice extent and small sea ice mass in the North Atlantic; when the Atlantic MOC is in a weak mode, both strong brine rejection and freshwater release occur, which are consistent with a large sea ice extent and large sea ice mass. The large changes of freshwater fluxes in Figure 6a are mainly due to the sea ice brine rejection/freshwater release. The modulated freshwater flux change can be as large as 0.1 6of9

7 Figure 6. (a) Integrated North Atlantic freshwater flux over 45 to 60 N (blue curve) and 60 to 75 N (red curve) for the control experiment. ( positive means that the ocean obtains freshwater.) (b) Integrated North Atlantic sea ice brine rejection (freshwater equivalent) and freshwater release to the south of 60 N (blue curve) and to the north of 60 N (red curve) for the control experiment. (c) Simulated Atlantic MOC in the experiment with the brine rejection/freshwater release after model year 11,000 fixed at the value at model year 10,430 (blue curve) and in the experiment with the brine rejection/freshwater release after model year 11,830 fixed at the value at model year 11,000 (red curve). The black curve in Figure 5c is the one from the control experiment. Sv regionally. Note that the salt rejection to the north of 60 N is nearly balanced by the freshwater release to the south of 60 N (Figure 6b). Thus the sum of freshwater fluxes over N and N in Figure 6a is almost a constant. [25] In order to see the effect of sea ice brine/freshwater release, we designed two other experiments in which the same glacial boundary conditions and stochastic freshwater forcing are employed as in the control experiment. In one experiment, starting at model year 11,000, the brine rejection/freshwater release is fixed at its value at year 10,430. At year 11,000, the Atlantic MOC is weak (this year is marked by the red vertical dash-dotted line in Figure 6c) and both the salt rejection to the north of 60 N and freshwater release to the south of 60 N are large (Figure 6b), while at year 10,430, the Atlantic MOC is strong (see the blue vertical dash-dotted line in Figure 6c) and both the salt rejection and freshwater release are very small (Figure 6b). After a large brine rejection at year 11,000 is replaced by a small one, the weak mode Atlantic MOC goes to an even weaker and finally a collapsed state (see the blue curve in Figure 6c). In another experiment, starting at model year 11,830, the brine rejection/freshwater release is fixed at its value at year 11,000. In this experiment, a very large brine rejection/freshwater release value corresponding to a weak mode of the Atlantic MOC is used to replace a small one corresponding to a strong mode of the Atlantic MOC. The strong mode is then maintained throughout the end of the model run, without any switch to a weak mode. The results of these two experiments demonstrate the important and necessary role of sea ice brine rejection processes in the millennial oscillations of the Atlantic MOC. [26] Sensitivity experiments with no brine rejection taken into account or with very small sea ice meridional advection velocity show that only a very weak or collapsed Atlantic MOC can be obtained under the forcing of the glacial boundary conditions. By increasing (decreasing) the sea ice advection velocity, the oscillation period is shortened (lengthened), further demonstrating the important role of sea ice brine rejection. [27] In addition to the forcing by a moderately cold background climate, internal oceanic feedbacks [see, e.g., Winton, 1997; Paul and Schulz, 2002] and the imposed stochastic forcing, we have demonstrated that the process of sea ice brine rejection during sea ice formation also plays a necessary role in the D-O oscillations. Figure 7 illustrates a negative feedback loop between the Atlantic MOC, North Atlantic climate and northern sea ice brine rejection under an appropriate background climate forcing. If we have a strong (weak) Atlantic MOC initially, we will have a high Figure 7. Negative feedback loop involving the Atlantic MOC, North Atlantic climate, and sea ice brine rejection. 7of9

8 (low) SST and SAT in the North Atlantic and hence a warm (cold) North Atlantic. A warm (cold) North Atlantic leads to reduced (enhanced) sea ice extent and thickness, i.e., sea ice mass. The southward sea ice mass transport must be reduced (increased). The sea ice brine rejection in the subpolar North Atlantic is consequently small (large). The small (large) brine rejection does not favor (favors) a strong Atlantic MOC and thus a weak (strong) Atlantic MOC may appear. 6. Discussion and Conclusions [28] An extensive parameter sensitivity study of the D-O oscillations simulated here is not presented in this paper, although some sensitivity experiments have been done. It is well known that model results are sensitive to the descriptions of mixing processes in the ocean. Uncertainties in the simulations of global energy and hydrological cycles may lead to uncertain surface forcing on the ocean circulation. However, the occurrence of a weakened Atlantic MOC in a cold climate like the LGM appears to be a robust feature that has been simulated by many climate models [see, e.g., Winton, 1997; Ganopolski et al., 1998; Kim et al., 2003; Knorr and Lohmann, 2003; Shin et al., 2003]. Our early version of this model [Wang et al., 2002] also showed a weakened Atlantic MOC for a very cold climate, and proxy data support this conclusion [Wang et al., 2002]. A recent proxy data study shows that the LGM Atlantic MOC mode is the same as the Younger Dryas mode which is a weak mode [Keigwin, 2004]. Therefore there must be a transition between the strong mode (corresponding to an interglacial and an early stage of a glacial) and the weak mode (corresponding to a peak glacial) during a glacial period. We note that whether or not the large millennial oscillations occur during the transition phase is model-dependent. In this study, we demonstrate that the simulated large glacial millennial oscillations are sensitive to the process of sea ice salt rejection. [29] The role of active continental ice sheets is not addressed in this study. There appears to be little agreement about the response of ice sheet mass balance to millennial oscillations in current climate models. For example, in one modeling study [Gildor and Tziperman, 2003], cooling leads to the ice mass loss, while in another modeling study [Schmittner et al., 2002], cooling reduces the ice mass loss. Future work is needed in this aspect. In another sensitivity experiment with our model, we found that an active vegetation component plays a minor role in the oscillations. [30] By employing an EMIC, in this study we have demonstrated that an intermediate cold glacial background climate could force the Atlantic MOC into an unstable state and that the Atlantic MOC oscillates on a millennial scale with rapid decadal-scale switches between the strong mode and the weak mode. During these oscillations, deep water forms continuously in the subpolar North Atlantic. For the first time, we have also demonstrated that sea ice brine rejection/freshwater release and sea ice transport play an important role in the maintenance of the weak mode. Our model results capture many important features of the glacial millennial oscillations. The Atlantic MOC mode switch, along with the amplification effect by sea ice, causes the large and abrupt northern climate changes. The southern climate changes are very small and gradual. The longer D-O interstadials occur in a warmer climate, while a colder climate favors a longer stadial. The net result is that the period of the oscillations is longer when the background climate is either warmer or colder and the period is shorter for an intermediate state. [31] Acknowledgments. This work was supported by an NSERC Discovery Grant and a CFCAS Project Grant awarded to L. A. Mysak. We thank David Holland and Yi Wang for their comments on this manuscript. The constructive comments of the reviewers (Eli Tziperman and the anonymous reviewer), which helped to improve this paper, are gratefully acknowledged. References Alley, R. B., P. U. Clark, L. D. Keigwin, and R. S. Webb (1999), Making sense of millennial-scale climate change, in Mechanisms of Global Climate Change at Millennial Time Scales, Geophys. Monogr. Ser., vol. 112, edited by P. U. Clark, R. S. Webb, and L. D. Keigwin, pp , AGU, Washington, D. C. Birchfield, G. E., H. Wang, and J. J. Rich (1994), Century/millennium internal climate oscillations in an ocean-atmosphere-continental ice sheet model, J. Geophys. Res., 99, 12,459 12,470. Blunier, T., and E. J. Brook (2001), Timing of millennial-scale climate change in Antarctic and Greenland during the last glacial period, Science, 291, Bond, G. C., and R. Lotti (1995), Iceberg discharges into the North Atlantic on millennial time scales during the last glaciation, Science, 267, Bond, G., W. S. Broecker, S. Johnson, J. McManus, L. Labeyrie, J. Jouzel, and G. Bonani(1993), Correlations between climate records from North Atlantic sediments and Greenland ice, Nature, 365, Bond, G. C., W. Showers, M. Elliot, M. Evans, R. Lotti, I. Hajdas, G. Bonani, and S. Johnson (1999), The North Atlantic s 1 2 kyr climate rhythm: Relation to Heinrich events, Dansgaard/Oeschger cycles and the Little Ice Age, in Mechanisms of Global Climate Change at Millennial Time Scales, Geophys. Monogr. Ser., vol. 112, edited by P. U. Clark, R. S. Webb, and L. D. Keigwin, pp , AGU, Washington, D. C. Broccocli, A. J. (2000), Tropical cooling at the Last Glacial Maximum: An atmosphere-mixed layer ocean model simulation, J. Clim., 13, Broecker, W. S., and G. H. Denton (1990), The role of ocean-atmosphere reorganization in glacial cycles, Quat. Sci. Rev., 9, Brovkin,V.,J.Bendtsen,M.Claussen,A.Ganopolski, C. Kubatzki, V. Petoukhov, and A. Andreev (2002), Carbon cycle, vegetation and climate dynamics in the Holocene: Experiments with the CLIMBER-2 model, Global Biogeochem. Cycles, 16(4), 1139, doi: /2001gb Clark, P. U., S. J. Marshall, G. K. C. Clarke, S. W. Hostetler, J. M. Licciardi, and J. T. Teller (2001), Freshwater forcing of abrupt climate change during the last glaciation, Science, 293, Clark, P. U., N. G. Pisias, T. F. Stocker, and A. J. Weaver (2002), The role of the thermohaline circulation in abrupt climate change, Nature, 415, Claussen, M., A. Ganopolski, V. Brovkin, F.-W. Gerstengarbe, and P. Werner (2003), Simulated global-scale response of the climate system to Dansgaard/Oeschger and Heinrich events, Clim. Dyn., 21, Dansgaard, W., S. J. Johnsen, H. B. Clausen, D. Dahl-Jensen, N. Gundestrup, C. U. Hammer, J. P. Steffensen, A. E. Sveinbjornsdottir, J. Jouzel, andg. Bond(1993), Evidencefor general instability of past climate from a 250-kyr ice-core record, Nature, 364, Dokken, T. M., and E. Jansen (1999), Rapid changes in the mechanism of ocean convection during the last glacial period, Nature, 401, Ganopolski, A., and S. Rahmstorf (2001), Rapid changes of glacial climate simulated in a coupled climate model, Nature, 409, of9

9 Ganopolski, A., and S. Rahmstorf (2002), Abrupt glacial climate changes due to stochasic resonance, Phys. Rev. Lett., 88(3), doi: /physrevlett Ganopolski, A., S. Rahmstorf, V. Petoukhov, and M. Claussen (1998), Simulation of modern and glacial climates with a coupled global model of intermediate complexity, Nature, 391, Gildor, H., and E. Tziperman (2003), Sea-ice switches and abrupt climate change, Philos. Trans. R. Soc. London, Ser. A, 361, Joussaume, S., and K. E. Taylor (2000), WRCP- 111 The paleoclimate modeling intercomparison project, Rep. WMO/TD 1007, World Meteorol. Organ., Geneva, Switzerland. Kaspi, Y., R. Sayag, and E. Tziperman (2004), A triple sea-ice state mechanism for the abrupt warming and synchronous ice sheet collapses during Heinrich events, Paleoceanography, 19, PA3004, doi: / 2004PA Keigwin, L. D. (2004), Radiocarbon and stable isotope constraints on Last Glacial Maximum and Younger Dryas ventilation in the western North Atlantic, Paleoceanography, 19, PA4012, doi: / 2004PA Kim, S.-J., G. M. Flato, and G. J. Boer (2003), A coupled climate model simulation of the Last Glacial Maximum. part 2: Approach to equilibrium, Clim. Dyn., 20, Knorr, G., and G. Lohmann (2003), Southern Ocean origin for the resumption of Atlantic thermohaline circulation during deglaciation, Nature, 424, Knutti, R., J. Flucker, T. F. Stocker, and A. Timmermann (2004), Strong hemispheric coupling of glacial climate through freshwater discharge and ocean circulation, Nature, 430, Lambeck, K., and J. Chappell (2001), Sea level change through the last glacial cycle, Science, 292, Lang, C., M. Leuenberger, and J. Schwander (1999), 16 C rapid temperature variation in central Greenland 70,000 years ago, Science, 286, Marotzke, J. (1989), Instabilities and multiple steady states of the thermohaline circulation, in Ocean Circulation Models: Combining Data and Dynamics, edited by D. L. T. Anderson and J. Willebrand, pp , Springer, New York. Marshall, S. J., and G. K. C. Clarke (1997), A continuum mixture model of ice stream thermomechanics in the Laurentide Ice Sheet: 1. Theory, J. Geophys. Res., 102, 20,599 20,613. McManus, J. F., D. W. Oppo, and J. L. Cullen (1999), A 0.5-million-year record of millennial-scale climate variability in the North Atlantic, Science, 283, North Greenland Ice Core Project Members (2004), High-resolution record of Northern Hemisphere climate extending into the last interglacial period, Nature, 431, Olsen, S. M., G. Shaffer, and C. J. Bjerrum (2005), Ocean oxygen isotope constraints on mechanisms for millennial-scale climate variability, Paleoceanography, 20, PA1014, doi: /2004pa Paul, A., and M. Schulz (2002), Holocene climate bariability on centennial-to-millennial time scales: 2. Internal and forced oscillations as possible causes, in Climate Development and History of the North Atlantic Realm, edited by G. Wefer et al., pp , Springer, New York. Peltier, W. R. (2004), Global glacial isostasy and the surface of the ice-age Earth: The ICE-5G (VM2) model and GRACE, Annu. Rev. Earth Planet. Sci., 32, Petoukhov, V., A. Ganopolski, V. Brovkin, M. Claussen, A. Eliseev, C. Kubatzki, and S. Rahmstorf (2000), Climber-2: A climate system model of intermediate complexity. part I: Model description and performance for present climate, Clim. Dyn., 16, Prange, M., G. Lohmann, and R. Gerdes (1997), Sensitivity of the thermohaline circulation for different climates Investigations with a simple atmosphere-ocean model, Paleoclimates, 2, Sakai, K., and W. R. Peltier (1997), Dansgaard- Oeschger oscillations in a coupled atmosphere-ocean climate model, J. Clim., 10, Sarnthein, M., et al. (2000), Fundamental modes and abrupt changes in North Atlantic circulation and climate over the last 60 ky Concepts, reconstruction, and numerical modelling, in The Northern North Atlantic: A Changing Environment, edited by P. Schafer et al., pp , Springer, New York. Schmittner, A., M. Yoshimori, and A. J. Weaver (2002), Instability of glacial climate in a model of the ocean-atmosphere-cryosphere system, Science, 295, Schulz, M. (2002), The tempo of climate change during Dansgaard-Oeschger interstadials and its potential to affect the manifestation of the 1470-year climate cycle, Geophys. Res. Lett., 29(1), 1002, doi: /2001gl Schwander, J., T. Sowers, J.-M. Barnola, T. Blunier, A. Fuchs, and B. Malaizé (1997), Age scale of the air in the summit ice: Implication for glacial-interglacial temperature change, J. Geophys. Res., 102, 19,483 19,494. Shaffer,G.,S.M.Olsen,andC.J.Bjerrum (2004), Ocean subsurface warming as a mechanism for coupling Dansgaard-Oeschger climate cycles and ice-rafting events, Geophys. Res. Lett., 31, L24202, doi: / 2004GL Shin,S.-I.,Z.Liu,B.L.Otto-Bliesner,J.E. Kutzbach, and S. J. Vavrus (2003), Southern Ocean sea-ice control of the glacial North Atlantic thermohaline circulation, Geophys. Res. Lett., 30(2), 1096, doi: / 2002GL Sima, A., A. Paul, and M. Schulz (2004), The Younger Dryas-an intrinsic feature of late Pleistocene climate change at millennial timescales, Earth Planet. Sci. Lett., 222, , doi: /j.epsl Stauffer, B., et al. (1998), Atmospheric CO 2 concentration and millennial-scale climate change during the last glacial period, Nature, 392, Stocker, T. F., and O. Marchal (2000), Abrupt climate change in the computer: Is it real?, Proc. Natl. Acad. Sci. U.S.A., 97, Timmermann, A., H. Gildor, M. Schulz, and E. Tziperman (2003), Coherent resonant millennial-scale climate oscillations triggered by massive meltwater pulses, J. Clim., 16, Tziperman, E. (1997), Inherently unstable climate behaviour due to weak thermohaline ocean circulation, Nature, 386, Tziperman, E. (2000), Proximity of the presentday thermohaline circulation to an instability threshold, J. Phys. Oceanogr., 30, Tziperman, E., J. R. Toggweiler, Y. Feliks, and K. Bryan (1994), Instability of the thermohaline circulation with respect to mixed-boundary-conditions: Is it really a problem for realistic models?, J. Phys. Oceanogr., 24, van Kreveld, S. A., M. Sarnthein, H. Erlenkeuser, P.Grootes,S.Jung,M.J.Nadeau,U.Pflaumann, and A. Voelker (2000), Potential links between surging ice sheets, circulation changes and the Dansgaard-Oeschger cycles in the Irminger Sea, kyr, Paleoceanography, 15, Vidal, L., L. Labeyrie, and T. C. E. van Weering (1998), Benthic d 18 O records in the North Atlantic over the last glacial period (60 10 kyr): Evidence for brine formation, Paleoceanography, 13(3), Wang, Y., L. A. Mysak, Z. Wang, and V. Brovkin (2005), The greening of the McGill Paleoclimate Model. part I: Improved land surface scheme with vegetation dynamics, Clim. Dyn., 24, Wang, Z. (2005), Two climatic states and feedbacks on thermohaline circulation in an Earth system model of intermediate complexity, Clim. Dyn., 25, , doi: / s Wang, Z., and L. A. Mysak (2000), A simple coupled atmosphere-ocean-sea ice-land surface model for climate and paleoclimate studies, J. Clim., 13, Wang, Z., and L. A. Mysak (2001), Ice sheetthermohaline circulation interactions in a climate model of intermediate complexity, J. Oceanogr., 57, Wang, Z., and L. A. Mysak (2002), Simulation of the last glacial inception and rapid ice sheet growth in the McGill Paleoclimate Model, Geophys. Res. Lett., 29(23), 2102, doi: /2002gl Wang, Z., L. A. Mysak, and J. F. McManus (2002), Response of the thermohaline circulation to cold climates, Paleoceanography, 17(1), 1006, doi: /2000pa Wang, Z., R. Hu, L. A. Mysak, J.-P. Blanchet, and J. Feng (2004), A parameterization of solar energy disposition in the climate system, Atmos. Ocean, 42(2), Weaver, A. J. (1999), Millennial timescale variability in ocean/climate models, in Mechanisms of Global Climate Change at Millennial Time Scales, Geophys. Monogr. Ser., vol. 112, edited by P. U. Clark, R. S. Webb, and L. D. Keigwin, pp , AGU, Washington, D. C. Weaver, A. J., M. Eby, A. F. Fanning, and E. C. Wiebe (1998), Simulated influence of carbon dioxide, orbital forcing and ice sheets on the climate of the Last Glacial Maximum, Nature, 394, Winton, M. (1997), The effect of cold climate upon North Atlantic deep water formation in a simple ocean-atmosphere model, J. Clim., 10, Winton, M., and E. S. Sarachik (1993), Thermohaline oscillations induced by strong steady salinity forcing of ocean general circulation models, J. Phys. Oceanogr., 23, L. A. Mysak, Earth System Modelling Group, Department of Atmospheric and Oceanic Sciences, McGill University, Montreal, Quebec, Canada H3A 2K6. Z. Wang, British Antarctic Survey, High Cross, Madingley Road Cambridge, CB3 0ET UK. (zwa@bas.ac.uk) 9of9