Most recent attempts to address the glacial-interglacial CO2 problem have been focused on nutrient utilization and biogeochemistry of the Southern Ocean [e.g., Martin, 1990; Watson et al., 2000; Anderson et al., 2002; Bopp et al., 2003]. The model here reproduces most of the CO2 differences between the modern ocean and the glacial ocean through a simple switch in the circulation. The biogeochemistry in the model continues to do what it does in the modern state. The circulation switch operates in response to a change in the wind forcing around Antarctica. The Discussion below revisits past discussions about the glacial-interglacial CO2 problem and shows how they are reframed by this role for the circulation and the winds.
5.1. Whither the Biological Pump?
 Broecker  showed that the low pCO2 of the glacial atmosphere was associated with a larger δ13C gradient between the surface ocean and the deep ocean. He drew on the δ13C evidence to suggest that the glacial ocean retained more CO2 because it had more nutrients and more biological productivity, i.e., a stronger “biological pump.” Evidence to support a massive increase in biological productivity never materialized, however, and the search for a cause shifted to the Southern Ocean after the publication of the “Harvardton Bears” papers in 1984 [Sarmiento and Toggweiler, 1984; Siegenthaler and Wenk, 1984; Knox and McElroy, 1984].
 The current debate about the role of the Southern Ocean is framed around two perspectives on the Harvardton Bears result. One builds on the biological pump idea. The other is built on circulation and gas exchange. A brief synopsis of the two perspectives is given below.
 All of the density classes in the deep ocean outcrop at the surface around Antarctica, and the surface waters in these outcrop areas have high concentrations of nitrate and phosphate. One interpretation is that this nitrate and phosphate is “unused” or “underutilized” by the phytoplankton. The ocean can retain more respired CO2 if the organisms in the Southern Ocean are more efficient in their use of this underutilized resource. The high-nutrient surface waters of the modern Southern Ocean have a high pCO2 that ultimately sets the pCO2 of the atmosphere. More efficient use of the nutrients by organisms would lower the pCO2, lower the surface nutrients, and cause respired CO2 to build up in the deep ocean.
 The other view does not see the nutrients in the Southern Ocean as being “underutilized.” High southern nutrient concentrations are instead a manifestation of the upwelling of deep water that takes place along the southern flank of the ACC [Toggweiler, 1999; Stephens and Keeling, 2000; Gildor et al., 2002]. The atmospheric pCO2 is high during interglacial periods because the circulation brings deep water that is rich in respired CO2 right up to the surface. In this way, most of the respired CO2 in the upwelled deep water is given up to the atmosphere before the upwelled water sinks again [Toggweiler et al., 2003]. Organisms do not have it within their power to utilize more than a small portion of the upwelled nutrients.
 The critical distinction between the two perspectives can be narrowed down to the region in the Southern Ocean where this ventilation process occurs. The box models in the Harvardton Bears papers assume that this ventilation process occurs over the entire Southern Ocean. Organisms all over the Southern Ocean get a crack at the nutrients in the upwelled deep water before it can sink again. This is not very realistic.
 The other perspective breaks up the Southern Ocean into two domains, one close to Antarctica, and one further out, the subantarctic zone beyond the Polar Front. Biology and nutrient utilization are important only in the subantarctic zone. The switch operates in the upwelling/sinking zone close to Antarctica. Control of atmospheric CO2 belongs to the processes next to Antarctica that govern the amount of ventilated southern water that sinks back into the deep ocean [Archer et al., 2003; Toggweiler et al., 2003].
5.2. Whither the Westerlies?
 Early attempts to locate the glacial westerlies with paleo proxies were equivocal; one school argued for equatorward shifted westerlies [Heusser, 1989] and another supported poleward shifted westerlies [Markgraf, 1989]. No one in the early 1990s identified the westerlies as a particularly important factor for the global climate.
 Observers of the modern atmosphere, meanwhile, began to notice that the mean position of the Southern Hemisphere westerlies has been shifting poleward over the last 40 years [Hurrell and van Loon, 1994]. This change in position is reflected in a pattern of month-to-month and year-to-year variability called the Southern Annular Mode (SAM). Model simulations from the mid 1990s hinted that there might be a small intensification and a poleward shift of the westerlies in our greenhouse future [Kushner et al., 2001; Intergovernmental Panel on Climate Change, 2001]. Current coupled models produce much more substantial poleward shifts in their greenhouse warming scenario runs [Yin, 2005; Stouffer et al., 2006].
 The SAM and its northern cousin, the Northern Annular Mode (NAM), are characterized by a seesaw in atmospheric mass between the polar cap regions poleward of 60° latitude and the surrounding zonal rings centered near 45° latitude [Thompson and Wallace, 1998, 2001; Thompson and Solomon, 2002]. The atmospheric pressure inside the polar caps is low with respect to the pressure at 45° in accord with the strong westerly flow between 45° and 60°. When the pressure difference between the polar cap and the surrounding ring is larger than normal, i.e., when the annular modes are in their “high-index” polarity, the belt of westerly winds is displaced poleward from its climatological mean position. When the pressure difference is smaller than normal, i.e., in the “low-index” state, the westerly belt is displaced toward the equator.
 The poleward shifts in the westerlies over the last 40 years and the evidence tying these shifts to greenhouse warming would suggest that the westerlies also shifted poleward as CO2 increased and the climate warmed at the end of the last ice age. This would imply that the westerlies at the LGM were equatorward of their Holocene position. Indeed, the recent work with paleo proxies supports the idea of a very large equatorward shift at the LGM.
 Moreno et al.  put the glacial westerlies in South America at around 41°S compared with 50°S today. This result is based on pollen studies that track the migrations of vegetation types found near the tip of South America. The most distinctive is the bog assemblage of the Magellanic Moorlands, which thrives today under the cold and wet conditions within the westerly storm track (48°–56°S). This vegetation type is found at lower latitudes only at high elevations. At the LGM, this vegetation type is found near sea level at 41°S. Moreno et al. conclude that the northern edge of the westerly storm track must have been north of 41°S.
 Lamy et al.  and Stuut and Lamy  have examined the sediments recovered from sites off the coasts of Chile and Namibia. They identify the sediments as wind-blown dust or as fluvial sediment eroded from the continents by rain. Today, these sites receive mainly wind-blown dust from arid environments because they are so far north of the westerly storm track. These same sites received mainly fluvial sediment at the LGM. A Chilean site at 33°S that is located near a coastal desert today received fluvial input from low-elevation coastal range source rocks. This suggests that the westerly storm track at the LGM was close by.
 This pattern of wet and dry extends into the Holocene. Lamy et al.  and Lamy et al.  show that the warm middle Holocene (5000–8000 kyrs B.P.) and the Medieval Warm Period are dry periods in central Chile and times of relative warmth with less sea ice around Antarctica. The cool late Holocene and Little Ice Age are wet periods in central Chile and colder periods with more sea ice around Antarctica. This is further evidence that the westerly storm track moves systematically with the mean climate.
 The westerly shifts over the last 40 years can be linked to changes in the upper troposphere that extend up into the stratosphere. Steinbrecht et al. , Thuburn and Craig  and Santer et al.  have shown that the troposphere/stratosphere boundary has been displaced upward as the climate has warmed. This is consistent with the fact that there has been more warming in the upper troposphere than at the surface. The stratosphere, meanwhile, has been cooling in response to higher CO2.
 These changes in the upper troposphere indicate that the pocket of warm air near the Earth's surface is not only getting warmer, it is also expanding upward at low latitudes, and the thermal contrast between the pocket of warm air and the surrounding envelope of cold air in the stratosphere is increasing. This means that there is more baroclinicity and a stronger circulation aloft than there was 40 years ago. The poleward shifted, high-index westerlies at the surface are an expression of the stronger circulation aloft [Thompson and Solomon, 2002; Shindell and Schmidt, 2004; Yin, 2005].
 Broecker and Denton  showed that the snow line on mountain peaks was uniformly lower at the LGM; that is, individual isotherms in the middle of the troposphere were about 1 km lower at the LGM in relation to their positions today. This observation suggests that the pocket of warm air in the troposphere was cool and thin relative to today. It would also suggest that there was less baroclinicity and a weaker circulation aloft, a state that would be consistent with equatorward shifted westerlies at the surface.
5.3. Southern Ocean Stratification and the Role of Sea Ice
 Antarctica is surrounded by a layer of relatively fresh surface water that has the potential to isolate the deep water below from the atmosphere. The stratification associated with the fresh layer is often described as a barrier for CO2 [Francois et al., 1997; Sigman and Boyle, 2000; Gildor et al., 2002]. The same can be said of the sea ice at the surface that can limit the exchange of CO2 across the air-sea interface [Stephens and Keeling, 2000]. The hypothesis in this paper is that glacial-interglacial CO2 changes take place along with a shift in the westerlies but the stratification and sea ice barriers are also playing a role.
 Stratification in the polar oceans is enhanced by colder temperatures [Sigman et al., 2004]. Thus a cool climate perturbation that lowers atmospheric CO2 by increasing CO2 solubility would also strengthen the stratification around Antarctica and cause the ice pack to expand and thicken. The cooling effect on the barriers could then cause respired CO2 to build up in the deep ocean, which would lead to an even greater reduction in CO2. Perhaps the barriers are capable of lowering atmospheric CO2 through a feedback of their own. Is a wind shift really necessary?
 A windless feedback may not be big enough to get the ball rolling. The modern Ekman transport draws up warm salty deep water from below to replace the fresh surface waters that are continuously carried away to the north [Gordon, 1971]. The warm upwelled water also melts the ice at the surface and keeps the ice pack fairly thin. A cool perturbation that does not shift the westerlies has to overcome these effects. Hence a more likely scenario is that the barriers work together with the westerlies to produce a stronger feedback. A cool perturbation that shifts the westerlies equatorward reduces the upwelling from below and makes it easier for the stratification and a thicker ice pack to develop.
 Adkins et al.  recovered ancient pore water from Southern Ocean sediments, which shows that the bottom water of the glacial Southern Ocean was the saltiest water in the deep ocean at that time. They suggested that the salty deep water was produced from the brine rejected by an expanded glacial ice pack. These findings support the idea that thick ice was a factor in the low CO2 at the LGM [Stephens and Keeling, 2000]. They do not tell us how and when the ice buildup came about.
 Stouffer and Manabe  reproduced the Adkins et al.  result in a coupled model that was run with half the preindustrial level of atmospheric CO2. The low CO2 imposed in the model leads to the production of thick sea ice, which drives a strong circulation in the salty bottom waters. The abyssal circulation in the Stouffer and Manabe model develops after the low CO2 is imposed and the model climate becomes very cold.
 The Adkins et al. and Stouffer and Manabe results are not part of the model solutions generated here, but they are not inconsistent. Adkins et al.'s salty bottom water was apparently very depleted in δ13C (Figures 1 and 2). This means that the salty bottom water had no contact with the atmosphere. A vigorous deep circulation that is initiated under thick sea ice and is not in contact with the atmosphere is the same as a dead blue circulation in the present context. Unless it can be shown that the salty bottom water developed before the atmospheric pCO2 came down, we will assume that Adkins et al.'s salty bottom water is a feature that developed during the cold state.
5.4. Temporal Setting for the Feedback
 The δ13C observations from Hodell et al.  in Figure 2 show that the deep ocean accumulates respired CO2 over a short period of time at the beginning of each glacial period and then abruptly loses all of this CO2 at the beginning of the next interglacial. It tends to stay in the same state from one transition to the next. The intervals between these transitions seem to be fairly regular. Respired CO2 accumulated at 70,000, 180,000, 370,000, and 460,000 years ago, i.e., about every 100,000 years. It was vented away at 15,000, 120,000, 320,000, and 410,000 years ago, about 50,000 years after the preceding accumulation event.
 This pattern is similar to the pattern in the CO2 time series from the Vostok ice core [Petit et al., 1999] in the sense that the atmosphere is also spending most of its time with a pCO2 that is close to one of its extremes. The atmospheric pCO2 resides in an average state only briefly while it is transitioning from one extreme to the other.
 The CO2 time series from the Vostok ice core extends back 420,000 years. This is long enough that the mean pCO2 of the time series, roughly 230 ppm, should reflect the long-term steady state pCO2 that is maintained by volcanoes and weathering. This makes the behavior of atmospheric CO2 over the last half million years seem particularly odd: despite the fact that volcanoes and weathering are pulling the atmospheric pCO2 back to a long-term mean state between the glacial and interglacial extremes, the system seems to be going out of its way to avoid ever being in the mean state.
 Why does the CO2 system exhibit this behavior? Why are the deep ocean and atmosphere avoiding their mean states, and why do they flip to the opposite state with the particular timescale seen in Figure 2? We interpret these transitions as evidence for a threshold in the climate system.
 A long-term steady state pCO2 of ∼230 ppm corresponds to an intermediate temperature for the Earth that is also between the glacial and interglacial extremes. This intermediate temperature should, in turn, correspond to an intermediate position or an intermediate intensity of the Southern Hemisphere westerlies. Figure 11 shows a schematic diagram of the South Pacific with the position of the ACC and the approximate positions of the glacial and interglacial westerlies sketched in. The axis of the modern westerlies is shown overlying the ACC, while the axis of the glacial westerlies is far to the north. The long-term mean position of the westerlies is presumed to be somewhere in the middle between the modern and glacial limits.
Figure 11. Schematic diagram showing the position of the strongest westerlies today and the strongest westerlies at the Last Glacial Maximum in relation to the position of the Antarctic Circumpolar Current (ACC). The westerlies are pushed to these limits by the positive feedback described in the text. The intermediate position between these limits is assumed to be unstable with respect to the feedback and is therefore a threshold in the CO2/climate system. The threshold and the north-south limits on the westerlies extend into the Indian and Atlantic.
Download figure to PowerPoint
 We imagine that the westerlies have been sweeping back and forth between these limits in accord with the CO2 and temperature variations of the last 500,000 years. The feedback pushes the westerlies to the southern limit in a warming climate and it pushes the westerlies to the northern limit in a cooling climate. This implies that the intermediate position in between is unstable with respect to the feedback. The intermediate position is therefore a threshold in the CO2/climate system.
 The threshold is depicted as the green hatched region in Figure 11, i.e., as a position on a map. It could also be a critical wind intensity, like the 1.5x wind forcing in the model. The important thing is that the threshold, whatever it is, occupies an intermediate position between the extremes that mirrors the intermediate position/intensity of the westerlies favored by the long-term mean pCO2.
 The overarching idea behind this paper and the one to follow is that the CO2 cycles and big ice ages of the last 500,000 years owe their existence to this overlap: The position of the westerlies favored by the long-term steady state pCO2 overlies the position where the feedback has its threshold. This kind of overlap could explain the peculiar cyclic behavior seen in Figure 2 and in the Vostok CO2 record: the feedback is always driving the westerlies and the pCO2 away from the intermediate position occupied by the threshold while the Earth's volcanoes and weathering are always trying to bring them back. Once they fall back to the intermediate position, the next feedback event sends them off to the opposite extreme.
 The feedback described here can only push the westerlies and the pCO2away from the threshold. It is also bounded by the fact that the ocean can hold and give up only so much respired CO2. Hence the feedback is expended once the westerlies reach the southern and northern limits in Figure 11. At this point the feedback is powerless to stop the westerlies and the pCO2 from falling back from their extreme positions. The deep ocean, meanwhile, is set up to flip the other way. When the westerlies and pCO2 reach the threshold again, the feedback goes off in the other direction.
 Astute readers will have noted that the 50,000-year period between the accumulation and venting events in Figure 2 is much shorter than the 400,000-year timescale for volcanism and weathering to bring the pCO2 back to the long-term mean [Walker et al., 1981; Marty and Tolstikhin, 1998]. We will show in the follow-up paper that 50,000 years is an intrinsic timescale for the ocean's chemistry to adjust to the accumulation or venting of respired CO2. The pCO2 of the atmosphere tends to be “held away” from the threshold during this adjustment period. As such, the glacial and interglacial climates are “protected” from short-term perturbations like those seen during the last glacial period [Indermuhle et al., 2000; Blunier and Brook, 2001].
 After the ocean's chemical adjustment has run its course, this protection is over and the next perturbation to come along (with the right sign) can bring the pCO2 back to the threshold and set off the next feedback event. Thus volcanoes and weathering do not actually bring the pCO2 back to the threshold. They keep the long-term steady state pCO2 centered over the threshold so that random perturbations (of the appropriate sign) can bring the pCO2 back after the adjustment period. Two of these adjustments and two fast transitions add up to 100,000 years, the dominant period of variability in the Vostok CO2 record [Petit et al., 1999].
 The overlap between the position of the westerlies favored by the long-term mean pCO2 and the threshold is a coincidence, we would argue, that developed about 500,000 years ago when a slow decline in the volcanic output of CO2 brought the Earth's mean temperature and the mean position of the westerlies into alignment with the threshold. Before this time, the Earth was too warm and the westerlies were too strong or too far poleward. The earliest cycle presumably began about a million years ago when a lower pCO2 first brought the westerlies and the Earth's temperatures within range of the threshold. The cycles then became larger and more regular as the mean position of the westerlies came to overlie the threshold.