The role of tectonic Southern Ocean gateway changes in driving Antarctic climate change at the Eocene–Oligocene boundary remains a topic of debate. One approach taken in previous idealized modeling studies of gateway effects has been to alter modern boundary conditions, whereby the Drake Passage becomes closed. Here, the authors follow this approach but vary atmospheric pCO2 over a range of values when comparing gateway configurations. They find a significantly greater sensitivity of Antarctic temperatures to Southern Ocean gateway changes when atmospheric pCO2 is high than when concentrations are low and the ambient climate is cool. In particular, the closure of the Drake Passage (DP) gap is a necessary condition for the existence of ice-free Antarctic conditions at high CO2 concentrations in this coupled climate model. The absence of the Antarctic Circumpolar Current (ACC) is particularly conducive to warm Antarctic conditions at higher CO2 concentrations, which is markedly different from previous simulations conducted under present-day CO2 conditions. The reason for this is the reduction of sea ice associated with higher CO2. Antarctic sea surface temperature and surface air temperature warming due to a closed DP gap reach values around ∼5° and ∼7°C, respectively, for high concentrations of CO2 (above 1250 ppm). In other words, the authors find a significantly greater sensitivity of Antarctic temperatures to atmospheric CO2 concentration when the DP is closed compared to when it is open. The presence of a DP gap inhibits a return to warmer and more Eocene-like Antarctic and deep ocean conditions, even under enhanced atmospheric greenhouse gas concentrations.
One of the most profound climatic reorganizations in the geological record occurred at the Eocene–Oligocene (E–O) boundary (∼34 million years ago), where rapid cooling and glaciation of Antarctica (inferred from a benthic δ18O shift) represented an important step in Cenozoic climate cooling (Zachos et al. 2001). The apparently close temporal proximity in the geological records between this turning point and the opening of the Tasman Seaway led Kennett (1977) and Berggren and Hollister (1977) to hypothesize a causal relationship. In this view, the oceanic thermal isolation arising from the development of the Antarctic Circumpolar Current (ACC) leads to cool Antarctic conditions favorable for the emergence of permanent large-scale terrestrial glaciation. However, Stickley et al. (2004) in later work suggest that the opening of the Tasman Seaway occurred about 2 million years before the onset of Antarctic glaciation. The length of this time lag would preclude a direct causal relationship. Furthermore, Huber et al. (2004) find no enhanced poleward heat transport (PHT) in their fully coupled model of the Eocene where the Tasman Seaway is closed and conclude that the opening of any Southern Ocean (SO) gateway is unlikely to have caused Antarctic glaciation. This view is further reinforced by the seminal modeling work of DeConto and Pollard (2003), who examine the development of terrestrial Antarctic ice in a range of Cenozoic scenarios and also conclude that changes in atmospheric carbon dioxide may have played a more significant role than tectonically driven SO gateway changes. The timing of the opening of Drake Passage (DP), the seaway between Antarctica and South America, is less well constrained to between 20 and 40 million years ago (Barker and Burrell 1977; Lawver and Gahagan 1998; Scher and Martin 2006), but recent timing estimates by Livermore et al. (2005) led these authors to argue that the opening of DP may have constituted a trigger for Antarctic glaciation. The effect of ocean gateways on Antarctic glaciation remains unclear.
The benthic δ18O shift at the E–O boundary may reflect both a cooling and an increase in terrestrial ice. There has been debate about the relative contribution of ice and temperature to the δ18O shift. For instance, increases in ice volume could be accommodated by Antarctica alone (e.g., DeConto et al. 2008), or additional Northern Hemisphere (NH) glaciation may have been necessary (Tripati et al. 2008). Katz et al. (2008) integrate Mg/Ca ratios of benthic foraminifera, oxygen isotopes, and sea level reconstructions to show that stepwise cooling and glaciation occurred. Liu et al. (2009) find a significant benthic cooling between 3° and 5°C with concurrent Antarctic glaciation and argue that additional NH glaciation need not be invoked to explain the δ18O shift at the E–O boundary. Indeed, DeConto et al. (2008) use their thermomechanical ice-sheet model and atmospheric model (DeConto and Pollard 2003) to suggest that the threshold below which NH glaciation occurs (∼280 ppm) is much lower than for Antarctica (∼750 ppm in their model).
Process studies using ocean general circulation models have attempted to demonstrate the oceanic thermal isolation of Antarctica by applying a small land bridge between Antarctica and South America in present-day (Mikolajewicz et al. 1993; Nong et al. 2000; Sijp and England 2004, 2005) or idealized (Toggweiler and Bjornsson 2000) model topography, shutting off the DP gap. Closure of the DP gap enables meridional oceanic flow with a western land boundary, allowing the development of a zonal pressure gradient and a geostrophic meridional current across the DP latitudes. This leads to a large Southern Hemisphere (SH) overturning cell when DP is closed and enhanced PHT across the Southern Ocean. When surface salinity is not restored to present-day values NH overturning is also absent when DP is closed (Sijp and England 2005). Toggweiler and Bjornsson (2000) show that the existence of a gap between South America and Antarctica cools the Southern Ocean by around 3°C in their model and warms the NH by a similar magnitude. Sijp and England (2004) find increased southward oceanic heat transport when DP is closed, but sea surface temperature (SST) remains remarkably similar poleward of 60°S. In particular, in their study a closed DP gap is associated with significant localized SH SST warming (e.g., up to 10°C around the Brazil–Malvinas confluence) but much smaller changes close to Antarctica. Opening of the DP gap leads to the initiation of NH overturning in their model, leading to SH cooling and NH warming. Indeed, NH overturning is also found in a model by von der Heydt and Dijkstra (2006) and von der Heydt and Dijkstra (2008) for the late Oligocene. Interestingly, because of the presence of a low-latitude ocean gateway connecting the Pacific to the Atlantic at the location of present-day Panama, sinking occurs in the North Pacific as well as the Atlantic. Omta and Dijkstra (2003) examine changes in the horizontal ocean flow during the Tertiary, but their shallow-water model precludes an analysis of the meridional overturning circulation (MOC). Most notably they find and explain a plausible flow reversal through the Panama straits in response to the widening of DP and closure of the Tethys Seaway. In conclusion, model results where DP is closed lend some support to Kennett’s (1977) hypothesis that the development of the ACC led to Antarctic glaciation but fall short of explaining the dramatic changes in Antarctic climate across the Eocene–Oligocene boundary.
Also, the lack of a global cooling response to opening Southern Ocean gateways in previous model results emphasizes the need for including additional factors in explanations of Eocene warmth, and subsequent studies have stressed atmospheric CO2 and the associated atmospheric reorganization as an important factor in driving higher polar temperatures during the Eocene (DeConto and Pollard 2003; Huber et al. 2004; Huber and Nof 2006). This is in agreement with proxy-based estimates of atmospheric CO2 of at least twice modern levels (Pagani et al. 2005).
Here, we further examine the interplay between ocean gateway changes and atmospheric CO2 concentration (pCO2) in determining Antarctic and deep ocean warmth. In particular, we examine the global climate response to DP geometry in experiments exploring atmospheric CO2 over a range of concentrations from 280–2000 ppm. We find that deep ocean temperature and Antarctic SST warm because of closing DP in the model and that the magnitude of this warming due to tectonic gateway changes alone increases with increasing pCO2. This relates to the limitation on temperature changes arising from the freezing point of seawater and is mainly due to the reduction in Antarctic sea ice at high levels of pCO2. Also, unlike experiments carried out at preindustrial CO2 levels (280 ppm), closing DP leads to deep ocean warming at higher pCO2. Deep ocean temperature is also more sensitive to global CO2-induced surface warming when DP is closed, and significant deep ocean warming occurs as a result of closing DP under fixed high pCO2 in the range of Eocene values.
Much has been inferred from the previous gateway model studies despite their simplicity with regard to forcing and configuration. For instance, DeConto and Pollard (2003) base their parameterized changes in oceanic PHT to Antarctica arising from the opening of DP on the studies of Nong et al. (2000) and Toggweiler and Bjornsson (2000). We will show that the results of these previous model studies depend rather strongly on atmospheric pCO2.
2. The model and experimental design
We use the intermediate complexity coupled model described in detail in Weaver et al. (2001), the so-called University of Victoria (UVic) model. This model comprises an ocean general circulation model [Geophysical Fluid Dynamics Laboratory (GFDL) Modular Ocean Model (MOM) Version 2.2; Pacanowski 1995] coupled to a simplified one-layer energy–moisture balance model for the atmosphere and a dynamic–thermodynamic sea ice model of equal global domain and horizontal resolution of 3.6° longitude by 1.8° latitude. Although air–sea heat and freshwater fluxes evolve freely in the model, a noninteractive wind field is employed. The wind stress forcing is taken from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis fields (Kalnay et al. 1996) and averaged over the period 1958–97 to form a seasonal cycle from the monthly fields. No flux corrections are used, allowing surface temperature and salinity to vary freely in the model. This allows the full operation of the thermal and salinity feedback (Bryan 1987; Sijp and England 2004). Vertical mixing is modeled similarly to Bryan and Lewis (1979), using a diffusivity that increases with depth, taking a value of 0.3 cm2 s−1 at the surface and increasing to 1.3 cm2 s−1 at the bottom. Convective overturning of the water column is modeled using the convective adjustment algorithm described by Rahmstorf (1993; see also Pacanowski 1995). The effect of subgrid-scale eddies on tracer transport is modeled after Gent and McWilliams (1990) and implemented as an advection velocity acting on the tracer fields. A rigid-lid approximation is used and surface freshwater fluxes are represented by way of an equivalent salt flux calculated using a fixed reference salinity of 34.9 psu. The surface ocean layer has thickness of 50 m, and no mixed layer scheme is included.
The energy-conserving thermodynamic sea ice model calculates ice thickness, areal fraction, and ice surface temperature. The atmospheric radiative balance is fed by solar insolation, where orbital values are fixed at the year 1850. Atmospheric moisture transport is achieved by way of advection and diffusion using constant moisture diffusivity. Precipitation occurs when relative humidity is greater than 85%. Precipitation that falls over land returns to the ocean instantaneously via prescribed river basins. Precipitation creates snow cover when the surface air temperature (SAT) drops below −5°C. Snow cover is limited to a maximum height of 10 m, which is more than can melt during summer. Both snow and sea ice influence planetary albedo. Model atmospheric temperatures decrease over uplifted land areas according to a fixed lapse rate.
No dynamic terrestrial ice-sheet model is included. Instead, elevation and prescribed surface albedo over Antarctica and Greenland are used in the UVic model to mimic the presence of modern-day glaciation in these regions. In the present study, we remove this parameterization by employing a fixed surface elevation of 350 m throughout Antarctica and Greenland and no prescribed albedo. This allows an idealized look at Antarctic climate in the absence of land ice and determines a first-order estimate of the ability of snow to survive the Antarctic summer. The uniform Antarctic elevation and the restriction on the height of snow in the model may lead us to underestimate Antarctica’s propensity to snow accumulation, as no glaciers descending from cool elevated plateaus are modeled. We do not attempt to reproduce Eocene orography and refer to the seminal study by DeConto and Pollard (2003) for an in-depth analysis of Antarctic ice-sheet formation at the Eocene–Oligocene boundary. Rather, the strength of the present study lies in the inclusion of the full 3D dynamic ocean model, allowing a detailed response in ocean temperature and ocean heat transport. We also advance previous studies by exploring the climate response to gateway changes under a suite of atmospheric CO2 levels, including a range of values suitable to the Oligocene–Eocene boundary.
This study aims to examine the interplay between elevated atmospheric pCO2 concentration and the absence of the ACC in determining Eocene Antarctic warmth. Therefore, we have run the model to equilibrium for a period of 6000 years using the present-day topography where either the DP is open (experiment DPopen) or DP is closed by a small land bridge (experiment DPclsd). Both configurations have been run to equilibrium for CO2 values ranging between 280 and 2000 ppm (see Table 1 for the range of CO2 values used). Experiments are denoted by their geometry (DPopen or DPclsd) and their CO2 value (280–2000 ppm). For instance, DPclsd 280 ppm denotes the experiment where DP is closed and atmospheric pCO2 is 280 ppm. In contrast to some previous studies (e.g., Sijp and England 2004), we will present temperature anomalies arising from tectonic changes as DPclsd − DPopen to indicate the contribution of a closed DP geometry to Antarctic warmth. This also allows a comparison with the contribution of CO2 to Antarctic warmth.
Figure 1 shows the steady-state annual-mean meridional overturning streamfunction for the 280-ppm experiments. The overturning is very similar to that found in Sijp and England (2004) and other studies (e.g., Mikolajewicz et al. 1993), whereby the DPopen case is characterized by North Atlantic Deep Water (NADW) formation and weak Antarctic Bottom Water (AABW) formation, and the DPclsd geometry is associated with a strong SH sinking cell. NADW outflow is somewhat weak in our model. Nonetheless, experiments with a later version of our model where NADW outflow is tuned to be significantly stronger show very similar results to those shown here. In contrast, the fully coupled Eocene model used by Huber and Sloan (2001), Huber et al. (2003), and Huber et al. (2004) shows no strong SH overturning in the global average when the Tasman Seaway between Antarctica and Australia is closed, although significant vertical mass exchange between the surface and the deep ocean occurs around Antarctica, whereby upwelling largely cancels sinking in the zonal mean (M. Huber 2008, personal communication). Unlike our study, they also find NH overturning. This difference may be due to enhanced southward atmospheric moisture transport in their model (see, e.g., Sijp and England 2009). The ∼34 Sv (1 Sv ≡ 106 m3 s−1) of Antarctic sinking is a result of cold Antarctic conditions and the possibility of geostrophic flow across the latitudes of the DP. Figure 2 shows the yearly averaged SST for DPclsd, the SST difference DPclsd − DPopen, and surface ocean heat fluxes for DPclsd where atmospheric pCO2 is 280 and 1250 ppm. Near-freezing conditions extend to 60°S in DPclsd 280 ppm, whereas ∼4°−8°C SST isotherms intersect with the Antarctic coast in the DPclsd 1250-ppm case (Figs. 2a,b). The effect of a closed DP gap (DPclsd − DPopen) for 280 ppm (Fig. 2c) is similar to that found by Sijp and England (2004), where significant SH warming occurs in the South Atlantic, but is restricted equatorward of 60°S. In contrast, these warmer conditions in DPclsd are of significantly greater spatial extent under the higher CO2 concentrations of 1250 ppm (Fig. 2d), where the warming in DPclsd − DPopen now encircles Antarctica and extends also beyond 60°S up to the continental shore.
The high-latitude warming poleward of 60°S in DPclsd 1250 ppm stands in stark to contrast to the previous modeling study of Sijp and England (2004), where SH warming in the absence of an ACC is restricted equatorward of ∼60°S, and more significant high-latitude SAT warming associated with a closed DP gap is limited to the (austral) winter. The high-latitude warming in DPclsd − DPopen 1280 ppm occurs in areas where SST is close to freezing in the 280-ppm case, but between 4° and 8°C in the 1250-ppm case. Unlike the 1250-ppm case, there is little high-latitude SST change between DPopen and DPclsd in the 280-ppm case because temperatures remain near their lower bound dictated by the freezing point of seawater in both experiments and no significant temperature difference can develop. Furthermore, the presence of sea ice in DPclsd 280 ppm inhibits oceanic PHT to high latitudes. Sea ice persists poleward of 60°S in DPclsd at 280 ppm, whereas this area is largely ice free at 1250 ppm (Figs. 2e,f). Furthermore, less of the extra oceanic heat transport (HT) in DPclsd (compared to DPopen) is consumed in the latent heat budget of sea ice formation in the warmer ambient climate at CO2 levels of 1250 ppm, and therefore more energy is available to raise SST. The sea ice found poleward of 60°S in DPclsd 280 ppm (Fig. 2e) also insulates the ocean from heat loss in winter. This insulating effect is largely absent in DPclsd 1250 ppm, resulting in a large increase in oceanic heat loss compared to DPclsd 280 ppm, particularly between the Ross Sea and the Antarctic Peninsula (Fig. 2f).
Figure 3 shows the oceanic and the atmospheric northward heat transport for DPclsd 280 ppm, DPclsd 1250 ppm, DPopen 280 ppm, and DPopen 1250 ppm. Although there is greater southward oceanic PHT in DPclsd 280 ppm than in DPopen 280 ppm because of greater AABW formation and the absence of NADW in DPclsd, this difference is modest and restricted to the NH and the midlatitudes of the SH (Figs. 3a,b). In contrast to the 280-ppm case, oceanic PHT is significantly greater in DPclsd 1250 ppm compared to DPopen 1250 ppm, and the difference penetrates to higher southern latitudes, delivering heat close to the Antarctic continent (Figs. 3a,b). Unlike the 280-ppm case, the opening of DP in the 1250-ppm case leads to a significant oceanic thermal isolation of Antarctica right up to high southern latitudes. This is in agreement with the warmer high-latitude Antarctic SST conditions in DPclsd compared to DPopen in the 1250-ppm case (see Fig. 2). The greater oceanic PHT at latitudes poleward of 60°S in DPclsd 1250 ppm compared to DPclsd 280 ppm is a result of the increased oceanic heat loss to the cold overlying atmosphere in the ice-free high southern latitudes (see Fig. 2). We will also see that there is a modest increase in AABW formation, contributing to enhanced oceanic PHT. Atmospheric PHT values respond so as to partially compensate the oceanic changes and are thus reduced in DPclsd 1250 ppm compared to DPclsd 280 ppm (Fig. 3b) in response to enhanced oceanic PHT.
Figure 4 shows the deep ocean and Antarctic climate response to atmospheric pCO2 changes for DPclsd and DPopen. All temperature changes shown in Fig. 4 are calculated relative to the DPopen 280-ppm experiment over a range of CO2 concentrations. We interpret the difference between the red and black curves as the effect of closing DP alone on Antarctic temperature at each level of atmospheric pCO2. Globally averaged SAT change is very similar for DPclsd and DPopen (Fig. 4a), whereas DPclsd exhibits a markedly stronger Antarctic SAT and SST sensitivity to pCO2 than DPopen for atmospheric CO2 values up to 1250 ppm (Figs. 4b,c). At pCO2 greater than 1250 ppm, the effect of closing DP no longer changes with increasing pCO2, yielding a similar Antarctic SAT and SST sensitivity for DPclsd and DPopen (see the dashed curve at pCO2 > 1250 ppm). In summary, the magnitude of the effect of closing DP is dependent on the warmth of the ambient climate, whereby higher levels of CO2 lead to a stronger gateway impact. Antarctic SAT warms by ∼3°C because of a closed DP at 280 ppm, but at higher levels of CO2 the SAT anomaly of DPclsd − DPopen can be ∼7°C. Antarctic SST exhibits no significant response to closing DP at 280 ppm, whereas SST is ∼5°C warmer in DPclsd than DPopen for high values of pCO2. Similarly, the deep ocean warms by ∼5°C in DPclsd compared to DPopen at 1500 ppm, whereas this warming is less than ∼2°C for DPopen. Deep ocean warming due to closing DP at 2000 ppm is ∼4°C. Interestingly, the Antarctic SST and deep ocean temperature difference DPclsd − DPopen (dashed; Figs. 4c,d) are larger than the CO2-induced warming values in DPopen at each CO2 value, indicating that the high-latitude warming due to a closed DP gap alone is significantly larger than the warming achieved by CO2 increases. As a result, deep ocean temperatures are also more sensitive to the closure of the DP gateway than to CO2 increases. This is a remarkable result given some previous studies (e.g., Sijp and England 2004; Mikolajewicz et al. 1993; Toggweiler and Bjornsson 2000). In contrast to these studies, Najjar et al. (2002) do find deep ocean warming in a simulation using an Eocene topography with a simplified energy (heat) balance atmosphere and restoring sea surface salinity (SSS) to zonally averaged present-day observations. The salinity restoration in their model could be the cause of their bipolar overturning whereby AABW formation is enhanced while NADW formation remains present in their Eocene topography. This precludes the compensating effect of a shift in MOC polarity (as found by Sijp and England 2004) on deep ocean temperatures in their Eocene climate. Salinity restoration to present conditions precludes a realistic MOC response in models when departing significantly from present-day conditions [see Sijp and England (2004, 2005) for a discussion and references] and prevents the operation of a positive salt feedback crucial for the shutdown of MOC cells.
In contrast to studies examining the effect of gateways alone, studies examining the effect of atmospheric pCO2 do find significant deep cooling in response to a decline in greenhouse gases. Huber and Sloan (2001) were the first to present results from a fully coupled general circulation model simulation of the Eocene and find a warming of bottom water temperatures of 7°C with respect to the present. More recently, Liu et al. (2009) find a benthic cooling of 4°C (from which they infer a 3°–5°C cooling in the real ocean) in response to a reduction in atmospheric pCO2 from 2240 to 560 ppm in their coupled model.
Figure 5 shows the dependence of various factors affecting Antarctic climate on atmospheric pCO2 for DPclsd and DPopen. Antarctic permanent snow cover decreases sharply with increasing pCO2 beyond ∼500 ppm in DPclsd (Fig. 5a), culminating in ice-free conditions beyond ∼1250 ppm. In contrast, terrestrial snow accumulation is remarkably robust with respect to CO2 up to this value in DPopen and remains above ∼75% for all CO2 values. In agreement with Sijp and England (2004), the annually averaged Antarctic sea ice area is less in DPclsd than in DPopen for 280 ppm (Fig. 4b). Ice-free conditions are achieved at ∼1250 ppm for DPclsd, whereas some sea ice remains present, albeit reduced, for all CO2 values in DPopen. Poleward oceanic heat transport (Fig. 4c) is significantly greater in DPclsd 280 ppm than in DPopen 280 ppm because of the large SH overturning cell (refer to Fig. 1). This discrepancy increases with increasing CO2, whereby PHT sensitivity to CO2 is greater for DPclsd than for DPopen for pCO2 less than ∼1250 ppm. The increasing oceanic HT in DPclsd results from decreased sea ice cover under warmer climatic conditions (see also Fig. 2). Finally, the overturning in the SH cell modestly increases with increasing CO2 in DPclsd (Fig. 4d), taking values between ∼35 and ∼46 Sv, also contributing to greater oceanic PHT to Antarctica.
Figure 6 shows the seasonal Antarctic climate response to atmospheric pCO2 for DPopen and DPclsd for the same range of CO2 as in Figs. 4 and 5. In agreement with the annual averages, SST and SAT exhibit stronger sensitivity to pCO2 in DPclsd than in DPopen for both December–February (DJF) (austral summer; Fig. 6a) and June–August (JJA) (austral winter; Fig. 6b). Whereas SST is close to the freezing point of seawater for DPclsd 280 ppm and DPopen 280 ppm, open and closed DP temperatures diverge increasingly with increasing values of CO2. Antarctic summer SST values of up to ∼8°C are reached in DPclsd under CO2-induced warming, and yet a warming of only ∼3°C is achieved in DPopen (Fig. 6). The discrepancy in CO2 sensitivity between DPclsd and DPopen is even greater in winter, where temperatures of ∼5°–6°C are reached for DPclsd at high levels of atmospheric CO2, and yet only ∼0°C is reached in DPopen, despite atmospheric CO2 approaching 2000 ppm.
Antarctic summer SAT exerts a strong control over the development of a terrestrial ice sheet, as warm summers may prevent snow cover from surviving into the subsequent winter. In DPclsd, Antarctic summer SAT poleward of 70°S exceeds ∼0°C for pCO2 beyond ∼1000 ppm (Fig. 6c), whereas above-freezing terrestrial conditions are not even reached in any of our DPopen experiments (Fig. 6c, red curve). As a consequence, Antarctic minimum yearly snow cover is highly sensitive to atmospheric CO2 beyond ∼500 ppm in DPclsd and yet remarkably robust in DPopen (Fig. 6d). In agreement with the permanent snow cover shown in Fig. 5, no snow survives summer in DPclsd 1500 ppm (Fig. 6d). Antarctic summer sea ice vanishes at relatively low pCO2 (around 500 ppm) in DPclsd (Fig. 6e), and year-round ice-free conditions prevail beyond ∼1250 ppm in DPclsd (Fig. 6f). In contrast, some summertime sea ice remains in DPopen even when pCO2 ∼1250 ppm (Fig. 6e), with substantial volumes returning in winter despite high CO2 concentrations (Fig. 6f).
Closing the DP gap in a cold climate at 280 ppm leads to an Antarctic SAT warming of ∼3°C, while at 1250-ppm atmospheric CO2 concentration a warming of ∼7°C results. As a result, the surface area of Antarctic snow surviving summer is highly sensitive to CO2 in DPclsd, culminating in snow-free conditions beyond 1250 ppm. In contrast, the damped response of Antarctic SAT to pCO2 in DPopen allows summer snow to survive even at higher levels of CO2, leaving Antarctic summer snow cover relatively robust at CO2 levels up to 2000 ppm. In contrast, DeConto and Pollard (2003) find that the timing of Antarctic glaciation is sensitive to the presence of a DP gap but only if CO2 is within a critical range (700–840 ppm). They employ a slab ocean model, and assume a fixed 20% change in oceanic PHT at all latitudes associated with the presence of a DP gap for all CO2 values. This estimate is based on the more modest estimate of a 3°C warming in response to closing DP, a value derived from previous modeling results by Toggweiler and Bjornsson (2000) and Nong et al. (2000). Here, we show that the change in oceanic PHT arising from the presence of a DP gap in fact depends on the warmth of the ambient climate, particularly at high southern latitudes.
In contrast to our results, Huber et al. (2004) use a realistic geometry coupled ocean–atmosphere model of the Eocene to suggest that the opening of the Southern Ocean had a minimal effect on southward oceanic heat transport and SST around the Antarctic continent. Unlike our high-pCO2 cases, their model exhibits no substantial change in oceanic PHT at high southern latitudes. They find two MOC states in their model, both include NH sinking. Similar to their results, we find a significant compensating effect of atmospheric PHT. This is in agreement with Stone (1978), who argued that total heat transport should be independent of the detailed dynamical processes responsible for that transport but are determined only by insolation, the size of the earth, and mean planetary albedo. Enderton and Marshall (2009) find that, in warm climates with little ice, Stone’s result is a useful guide as they see little change in total PHT for widely different model geometries that lead to different climates. However, the repartitioning of PHT between the ocean and the atmosphere associated with these different climates shows that the distribution of PHT between the ocean and the atmosphere matters in determining climate. This is also borne out by our results, whereby enhanced oceanic PHT at high southern latitudes leads to warmer Antarctic conditions despite PHT compensations in the atmosphere. We find an amelioration of the SH temperature gradient, but gateway effects alone fall significantly short of explaining the low equator-to-pole temperature gradients, mild continental interiors, and greater global warmth inferred for the Eocene. This suggests that additional feedbacks and changes in pCO2 need to be considered. Indeed, Huber and Nof (2006) point out that changes in ocean gateways tend to redistribute heat between hemispheres. In addition, they find that Antarctic sensitivity to plausible concentration changes in atmospheric pCO2 to be as great as the sensitivity to a threefold change in oceanic PHT. They go on to point out that even an infinite efficiency of oceanic heat transport alone does not appear to ameliorate winter temperatures in continental interiors. Nonetheless, our results indicate that previous gateway studies have significantly underestimated the effect of SO gateways on Antarctic glaciation. Furthermore, the closure of DP has a significant local effect on Antarctic climate in our model, especially under realistically high pCO2.
The strong Antarctic climate response to gateway changes found in our study depends on the existence of strong SH overturning in the DP closed case, with sensitivity increasing as atmospheric CO2 increases. Importantly, there is evidence to support the existence of Southern Ocean deep-water formation during the early Paleogene (e.g., Via and Thomas 2006), thus our results likely have direct implications for past climate states even though debate on the polarity of the meridional overturning circulation during the Eocene is ongoing.
The geometry used in our model is that of the present day with a small modification at the DP. Also, we apply monthly varying observed winds to force ocean circulation and advect atmospheric moisture and heat. A bias toward the present-day climate may result from these factors. However, the distribution of the continents at the late Eocene was relatively similar to that of today. Furthermore, the general circulation of the atmosphere is likely to have included easterly winds at the tropics and westerlies at midlatitudes, although the polar easterlies may have been significantly modified during summer in the absence of permanent large-scale glaciation (Hay et al. 2005). However, it is worth noting that the large-scale wind field used in our model is not in dynamic balance with the temperature changes arising from enhanced greenhouse gases and temperature changes due to the closing of DP (although we address part of this issue with a sensitivity analysis elsewhere in this paper). Nonetheless, we uncover important limitations on the interpretation of previous gateway studies and show a plausible oceanic MOC response to the opening of DP under enhanced greenhouse gas conditions. At equilibrium we find the model has a preference for deep sinking in the SH because, being situated at higher latitudes, the Antarctic sinking regions are significantly colder than their NH counterparts even when DP is closed (refer also to Sijp and England 2009, their Fig. 4). In other words, strong SH sinking arises from an interhemispheric asymmetry in the thermal conditions over the ocean at high latitudes. This thermal asymmetry may also be a feature of the Eocene climate, where Antarctic sinking would then occur at high latitudes in the ancient geometry of landmasses.
To examine the effect of changes in SH winds on our model results, we have conducted five additional experiments with DP closed and atmospheric CO2 at 1250 ppm. Because the SH westerlies may weaken in response to a reduced meridional temperature gradient, we halve the strength of the SH westerlies in one experiment DPclsd 1250 ppm reduced winds (RW). Because the SH westerlies may also move latitudinally, we conduct an experiment DPclsd 1250 ppm poleward winds (PW), where we shift the SH westerlies poleward by 6° latitude and an experiment DPclsd 1250 ppm equatorward winds (EW), where we shift the winds equatorward. We also conduct experiments DPclsd 1250 ppm RW EW and DPclsd 1250 ppm RW PW, where the winds are both halved and shifted equatorward and poleward, respectively, by 6°. Figure 7 shows the steady-state global meridional overturning streamfunction for these experiments. Halving the SH westerlies leads to a reduction in the maximum absolute value of the MOC streamfunction from 43 Sv in DPclsd 1250 ppm (see Table 1) to 36 Sv in DPclsd 1250 ppm RW (see Fig. 7a). Shifting the winds equatorward or poleward has no further significant effects (Figs. 7b,c), although the poleward case leads to a slight reduction in maximum streamfunction to 34 Sv (Fig. 7b). The difference in streamfunction between the halved winds case and the full winds case (Fig. 7d) shows that the reduction in wind strength has an effect in the SO but that there is no systematic reduction in the large interhemispheric MOC that is characteristic of the southern sinking solution when DP is closed. This is borne out by the Antarctic temperature changes in the RW and shifted winds (EW, PW) experiments where temperature changes remain within 1°C of the original values found for the full winds DPclsd 1250-ppm experiment (figure not shown). Therefore, our results are robust with respect to these relevant changes in the SH winds. We note that the reduction in the SH westerlies in these experiments is significant and that Sen Gupta and England (2007) find a weaker response in wind stress changes in response to a significantly reduced temperature gradient.
In the model, moisture diffusivity represents the effect of storms on atmospheric water transport. We use a globally constant atmospheric diffusivity to preclude any bias arising from assumptions regarding the spatial distribution of storms in past or present climates. For instance, Sijp and England (2009) find that enhanced atmospheric moisture diffusivity over the SO leads to NH sinking when the DP is closed and atmospheric CO2 concentration is 280 ppm. Similar to other key model parameters, the magnitude of the globally constant moisture diffusivity does not influence MOC polarity. Apart from the winds and the input of potential energy by parameterized turbulent ocean mixing, the model is only forced by insolation incident from space, leaving the climate free to evolve, for example, in regard to global heat and freshwater cycles. In summary, the strong SH sinking in our DPclsd geometry is not the result of any prescribed bias toward the present-day climate in our model; it comes about simply by virtue of model landmass geometry.
Our results do not support the global cooling suggested by Exon et al. (2001) as a direct result of the opening of DP (see Fig. 4a). Indeed, the cooling at the Eocene–Oligocene boundary shown in the proxy records of Katz et al. (2008) appears short lived (see their Fig. 1c), and Antarctic terrestrial ice growth accounts for most of the increase in the oxygen isotopic values recorded in the tests of benthic foraminifera (δ18O) once global temperatures have recovered almost to their previous values. This is in agreement with our similar global SAT values for DPopen and DPclsd. Our results indicate that the opening of DP cooled Antarctica significantly, without strongly affecting global climate.
There is substantial uncertainty surrounding atmospheric pCO2 values during the Paleogene, but Pagani et al. (2005) suggest that pCO2 ranged between 1500 and 1000 ppm in the middle to late Eocene and reached modern values sometime during the Oligocene. Pearson and Palmer (2000) estimate high pCO2 of more than 2000 ppm for the late Paleocene, followed by an erratic decline and values below 500 ppm at the early Miocene (about 24 million years ago). The global cooling trend from the Eocene to the Oligocene and beyond is likely to be related to this decline in pCO2 rather than gateway changes. Also, a pCO2 threshold may have been reached at the closing of the Eocene whereby Antarctic glaciation could occur during intervals where the earth’s orbit of the sun would favor cool summers, as seen in the reconstructions of Coxall et al. (2005) and also suggested by the modeling studies of DeConto and Pollard (2003) and de DeConto et al. (2008). Our results show that previous gateway modeling studies have underestimated the preconditioning effect of SO gateways on this glaciation threshold.
We note that Toggweiler and Bjornsson (2000) do not include a freezing point of seawater in their model and allow ocean temperature to drop below freezing. They find that Antarctic SST warms to 0°C from −4°C as a result of a closed DP gap at high southern latitudes. Our results suggest that the high-latitude portion of this temperature change in their model is allowed by the absence of a limiting freezing point and the absence of sea ice in their model. Inclusion of a freezing point and sea ice in their model would likely lead to a reduction in high-latitude change.
Global SAT is very similar for the DPopen and the DPclsd configuration at every level of atmospheric pCO2. This is because SH warming arising from closing the DP gap leads to a concurrent cooling of the NH due to the absence of interhemispheric northward HT associated with NADW formation (Crowley 1992). This is explained in detail by Toggweiler and Bjornsson (2000), who also point out that the ancient Tethyan Seaway connecting the Indian and the Atlantic Oceans during the Eocene is effective at inducing oceanic HT toward high northern latitudes. Najjar et al. (2002) use an ocean model with an energy balance atmosphere and realistic Eocene geometry, and attribute a warming of 4°C at 70°N with respect to the present-day climate to a connection between the Tethys Sea and the high northern latitudes. Therefore, among other factors, amplified polar warmth in the NH may have been maintained through this mechanism in conjunction with enhanced Antarctic Eocene warmth.
Our treatment of snow accumulation is simplified, as no orography is applied over Antarctica. Also, snow height is limited to below 10 m, precluding height–mass balance feedbacks. We therefore examine the propensity for Antarctic ice-sheet formation rather than its development. In particular, our results illustrate the influence of CO2 and gateway changes on climate conditions relevant to this process. Precise glaciation thresholds may differ between models.
Stickley et al. (2004) find no significant cooling at site 1172 (east Tasman Plateau) across the Eocene–Oligocene boundary in their analysis of marine sedimentary records recovered during the Ocean Drilling Program (ODP) leg 189. However, unlike Stickley et al. (2004), Exon et al. (2001) place these locations, along with sites 1170 and 1171, north of the ACC core during the earliest Oligocene (see their Fig. F25). Our results indicate that SST changes are small equatorward of the ACC, explaining the lack of cooling inferred for site 1172 by Stickley et al. (2004). Indeed, the sediments of the Tasmanian region were never south of the Polar Front (Exon et al. 2001). The Antarctic SST cooling in response to the opening of Drake Passage inferred by our model results is supported by the sudden and profound cooling across the Eocene–Oligocene boundary found by Via et al. (2008) at ODP site 748, located on the Kerguelen Plateau and south of the ACC core during the Eocene–Oligocene transition.
5. Summary and conclusions
We have shown that Antarctic SST is significantly more sensitive to atmospheric pCO2 when DP is closed compared to when DP is open. For example, the freezing point of seawater constitutes a lower bound to SST, and Antarctic Ocean temperatures remain near this threshold during a significant period of the year for DPclsd 280 ppm and DPopen 280 ppm. Closing the DP gap at a CO2 concentration of 280 ppm leads to enhanced oceanic PHT to Antarctica, but cold atmospheric conditions still prevent high-latitude SST to rise much above freezing point. The presence of sea ice in DPclsd 280 ppm prevents oceanic PHT from reaching high southern latitudes. Increasing atmospheric CO2 concentration in DPclsd leads to a rapid loss of Antarctic sea ice, exposing large areas of the high-latitude Southern Ocean to the overlying cold air. As a result, when enhanced ocean heat transport in DPclsd penetrates to high southern latitudes, there is enhanced oceanic heat loss facilitated by ice-free conditions, and thus warmer Antarctic SAT ensues. Antarctic SST warming due to a closed DP gap levels off at a value of ∼5°C for high concentrations of CO2 (above 1250 ppm), and here the effect of closing DP becomes less sensitive to the additional warmth of the ambient climate. At these high CO2 concentrations, Antarctic SST remains above its lower bound of freezing for all or almost all of the year. This shows that studies of the effect of closing DP under present-day or preindustrial atmospheric CO2 concentrations likely underestimate the warming associated with a closed Southern Ocean gateway during warmer climate eras. This renders inferences to climatic change at the Eocene–Oligocene boundary inappropriate unless higher CO2 is also considered.
Our model does not permit interactive wind feedbacks. This is a significant limitation that we intend to address in a future study using a fully coupled model, although we also note that air–sea coupling is only modest at mid–high southern latitudes (see Sen Gupta and England 2007). This is also suggested for the DP closed case by our wind-sensitivity experiments elsewhere in this paper. The greater sensitivity of Antarctic temperatures to atmospheric pCO2 when the DP gap is closed indicates that the absence of the ACC is particularly conducive to warm Eocene Antarctic conditions at higher CO2 concentrations and that the oceanic thermal isolation of Antarctica upon the establishment of the ACC is significant in warmer climates. The closure of the DP gap is a necessary condition for the existence of ice-free Antarctic conditions at high pCO2 in our model. Our results suggest that the presence of higher atmospheric CO2 concentrations during the Eocene and Oligocene (Pearson and Palmer 2000; Pagani et al. 2005) enhanced the thermal effect of oceanic gateway changes during this period. In our results, elevated levels of CO2 remain an important component of a full explanation of mild and ice-free Antarctic conditions of the early Cenozoic. Regardless of whether the principal cause of Antarctic glaciation lies with the opening of SO gateways or a decrease in atmospheric CO2 concentration, or both, a return to warmer Antarctic conditions is likely to be significantly more difficult once the ACC is established.
We thank the University of Victoria staff for support in usage of their coupled climate model. This research was supported by the Australian Research Council and the Australian Antarctic Science Program. We also thank Matthew Huber and Andreas Schmittner for constructive reviews that have improved the quality of this paper.
Corresponding author address: Willem P. Sijp, CCRC, University of New South Wales, Sydney, NSW 2052, Australia. Email: email@example.com