Tropical seaways played a more important role than high latitude seaways in Cenozoic cooling

Abstract. Following the Early Eocene climatic optimum (EECO, ~55–50 Ma), climate deteriorated and gradually changed the earth from a greenhouse into an icehouse, with major cooling events at the Eocene-Oligocene boundary (∼34 Ma) and the Middle Miocene (∼15 Ma). It is believed that the opening of the Drake Passage had a marked impact on the cooling at the Eocene-Oligocene boundary. Based on an Early Eocene simulation, we study the sensitivity of climate and ocean circulation to tectonic events such as the closing of the West Siberian Seaway, the deepening of the Arctic-Atlantic Seaway, the opening of the Drake Passage, and the constriction of the Tethys and Central American seaways. The opening of the Drake Passage, together with the closing of the West Siberian Seaway and the deepening of the Arctic-Atlantic Seaway, weakened the Southern Ocean Deep Water (SODW) dominated ocean circulation and led to a weak cooling at high latitudes, thus contributing to the observed Early Cenozoic cooling. However, the later constriction of the Tethys and Central American Seaways is shown to give a strong cooling at southern high latitudes. This cooling was related to the transition of ocean circulation from a SODW-dominated mode to the modern-like ocean circulation dominated by North Atlantic Deep Water (NADW).


Introduction
The Cenozoic Era was a period of long-term cooling, with major cooling events at the Eocene-Oligocene boundary and the Middle Miocene (e.g., Miller, 1992;Zachos et al., 2001;Liu et al., 2009).There are several hypotheses trying to explain the mechanisms behind the Cenozoic cooling (e.g., Kennett, 1977;Raymo and Ruddiman, 1992;DeConto and Pollard, 2003;Coxall et al., 2005).One hypothesis is that past reorganizations of ocean circulation could have been an important factor.Among the tectonic events that had an impact on ocean circulation, the opening of the Drake Passage is thought to have contributed to cooling at the Eocene-Oligocene boundary.The opening of the Drake Passage enabled the development of the Antarctic Circumpolar Current (ACC), which thermally insulated Antarctica through reduced southward heat transport (e.g., Toggweiler and Bjornsson, 2000;Nong et al., 2000;Sijp and England, 2004), and also influenced the formation of deep water in the North Atlantic (NADW) (Toggweiler and Bjornsson, 2000;Scher and Martin, 2008).
This thermal isolation hypothesis is, however, in conflict with recent studies (Livermore et al., 2005;Scher and Martin, 2006;Lagabrielle et al., 2009) indicating that the opening of the Drake Passage might have been much earlier than the Eocene-Oligocene boundary.Further, modelling studies have demonstrated that the change of ocean heat transport with the opening of the high latitude seaways was weak (Huber and Nof, 2006), and the cooling caused by the changes in ocean heat transport is insignificant compared to the cooing caused by the drop of atmospheric CO 2 levels (DeConto and Pollard, 2003).
Published by Copernicus Publications on behalf of the European Geosciences Union.
Z. Zhang et al.: Tropical seaways in Cenozoic cooling The warm greenhouse climate of the Early Eocene marked the beginning of the long term Cenozoic cooling trend, and has been the focus of several recent modelling studies (Huber et al., 2004;Heinemann et al., 2009;Roberts et al., 2009;Shellito et al., 2009;Tindall et al., 2010;Winguth et al., 2010;Speelman et al., 2010;Lunt et al., 2010).These studies have tested the sensitivity of climate to a large increase in atmospheric greenhouse gases (Shellito et al., 2009;Winguth et al., 2010;Lunt et al., 2010), explored the mechanism behind the Arctic warming (Heinemann et al., 2009;Shellito et al., 2009), simulated the isotope distribution of ancient seawater and precipitation in the Early Eocene (Tindall et al., 2010, Speelman et al., 2010) , and also considered the possible climatic effects caused by the changes to the high latitude seaways, including the West Siberian Seaway, Atlantic-Arctic Seaway (Roberts et al., 2009), Drake Passage and Tasman Seaway (Huber et al., 2004).However, the impact of tropical seaways on the warm Early Eocene climate has not been addressed in detail.More importantly, under the Early Eocene land-sea configuration, with a narrow and shallow Drake Passage and a narrow Tasman Seaway, the ACC was relatively weak (Huber and Sloan, 2001;Zhang et al., 2010), challenging the thermal isolation hypothesis which requires the development of a strong ACC once the Drake Passage opens.
Here, starting from a simulation of Early Eocene climate, we use FOAM (Fast Ocean Atmosphere Model) (Jacob et al., 2001) to examine the climate sensitivity to changes in the seaways, including the closing of the West Siberian Seaway, the deepening of the Arctic-Atlantic Seaway, the opening/deepening of Drake Passage and in particular the constriction of the tropical Atlantic seaways: the Tethys Seaway and the Central American Seaway.
In this paper, Sect. 2 introduces the climate model FOAM and the experimental design.Sections 3 and 4 present the Early Eocene control experiment and four sensitivity experiments with changes in the most important seaways.Section 5 compares the simulations with geological evidence and previous modelling studies, and discusses the implications of this study for understanding Cenozoic cooling.A summary follows in Sect.6.

Model introduction
FOAM version 1.5 is a fully coupled General Circulation Model (GCM), run without flux corrections (Jacob et al., 2001).FOAM consists of a parallel version of the Community Climate Model (CCM2) (Hack et al., 1993), with atmospheric physics upgraded following CCM3.6 (Kiehl et al., 1996).The atmospheric model is run with a horizontal resolution of R15 (4.5 × 7.5 degrees) and 18 vertical levels.The Ocean Model (OM3) is a finite-difference, z-coordinate ocean model with a horizontal resolution of 1.4 × 2.8 degrees, 24 vertical levels, and a free surface.The basic equations solved in OM3 are the same as those for the Modular Ocean Model (MOM) (Bryan, 1969;Cox, 1984), and it uses Richardson-number-based vertical mixing.The basic land model in FOAM is a simplified version of the default land model of CCM2 (Hack et al., 1993).The sea ice model uses the thermodynamics of the NCAR CSM Sea Ice Model (Bettge et al., 1996).
FOAM is intended for long century-scale integrations.It provides a good simulation of the mean and variability of the modern climate (e.g., Liu and Wu, 2004).It has also been used in paleoclimate studies of the Holocene (e.g., Liu et al., 2000) and Pleistocene (e.g., Lee and Poulsen, 2006), as well as periods from the deep geological time, such as the Neoproterozoic snowball Earth (Poulsen et al., 2001), Mesozoic paleoclimate and paleoceanography (Poulsen and Huynh, 2006;Donnadieu et al., 2007).Details on the FOAM model can be found here: http://www.mcs.anl.gov/research/projects/foam/.

Boundary conditions
In the Paleocene/Earliest Eocene, the Arctic connected with the Atlantic by the shallow Atlantic-Arctic Seaway, and with the Tethys Sea via the West Siberian Seaway (e.g., Scotese, 2001).At the same time the Drake Passage closed.However, there was a direct passage between the Pacific and Atlantic through the Central America Seaway.Reorganizations of these seaways occurred gradually.In the Middle Eocene, the West Siberian Seaway closed (Akhmet'ev and Beniamovski, 2006), and the Arctic-Atlantic Seaway deepened by the extension of the Mid-Atlantic Ridge (e.g., Scotese, 2001).Although age estimates for the opening of the Drake Passage are still under debate (Barker, 2001;Lagabrielle et al., 2009), there is evidence indicating an early shallow opening in the Early Eocene (Livermore et al., 2005) and subsequent deepening at ∼ 41 Ma in the Middle Eocene (Scher and Martin, 2006).If this is the case, the opening/deepening of the Drake Passage is simultaneous with the closing of the West Siberian Seaway and the deepening of the Arctic-Atlantic Seaway.
Changes to the tropical seaways occurred later.The narrowing of the Tethys Seaway (Barrier and Vrielynck, 2008) and the Central American seaway (e.g., Droxler et al., 1998) commenced in the Late Eocene.Further gradual shoaling of the Central American seaway began at ∼16 Ma (e.g., Droxler et al., 1998), and the Tethys seaway was permanently closed at about 15 Ma (Rögl, 1999).
The Early Eocene boundary conditions (Fig. 1) are set following the paleogeography of Scotese (2001) and the paleobathymetry is estimated using the method outlined by Bice et al. (1998).The paleogeography contains the reconstructions of mountains, coastlines and shallow ocean basins, as well as magnetic lines on the ocean floor.The coastlines outlined by Comparison of deep water temperature among the simulation (bold green line), paleotemperature estimation (green shaded bar) (Lear et al., 2000) and benthic foraminiferal δ 18 O (small green markers).(f) Latitudinal gradients of benthic foraminiferal δ 13 C in the Early Eocene.All geological data used here are listed in the supplementary materials.LMA means leaf-margin analysis.(e) Comparison of deep water temperature among the simulation (bold green line), paleotemperature estimation (green shaded bar) (Lear et al., 2000) and benthic foraminiferal δ 18 O (small green markers).(f) Latitudinal gradients of benthic foraminiferal δ 13 C in the Early Eocene.All geological data used here are listed in the supplementary materials.Scotese (2001) are set at 0 m, and the shallow ocean basins are given a depth of 200 m.The age-depth relationships established by Bice et al. (1998) are used to calculate the depth of each magnetic line.Finally, this depth data is interpolated into a global areal configuration.
Starting from the reconstructed Early Eocene bathymetry, the main seaways discussed above are changed in the sensitivity experiments (Table 1, Supplement).The West Siberian Seaway is closed in experiment WSSC.The Arctic Atlantic Seaway is deepened in experiment AASD, giving a maximum depth of 1900 m, and an Arctic with a water depth between 2000 m to 3000 m.The Drake Passage is deepened to about 2500 m in experiment DPGD.The Tethys Seaway is fully closed and the Central American Seaway is narrowed in experiment TSCN, as the tectonic changes to these two seaways happened close in time.The changes of these two seaways are simplified; however paleogeographical reconstructions (Droxler et al., 1998;Rögl, 1999) illustrate the complicated evolution of the Tethys and Central American Seaway.
In order to focus on the climatic effects caused by changes to the seaways, other boundary conditions are kept fixed.In all experiments, the solar constant and orbital parameters are set to present conditions; vegetation on land is prescribed as shrubland/grassland; and the concentration of atmospheric   CO 2 is set to 8 times the preindustrial level, following estimates for the Early Eocene (Royer, 2006).This is done to better isolate the roles of the seaways.

Initial conditions and model spin-up
In order to reduce the spin up time we use geological evidence (e.g., Lear et al., 2000;Tripati et al., 2003;Sluijs et al., 2006;Pearson et al., 2007) to construct an initial temperature field for the ocean model.We prescribe the temperatures at four points, SST of 34 to calculate the temperature in each layer at the equator and poles.d here is the depth of each layer in the ocean model.Then, a trigonometric sine function is used to set the meridional temperature gradient for each layer, with the temperature difference between the equator and the pole being the amplitude of the sine function.Finally, the meridional ocean temperature profile is applied at all longitudes, giving a threedimensional temperature field to initialize the model.The EECO experiment is spun up for 1250 yr in a fully coupled configuration, in order to reach equilibrium not only at the ocean surface but also in the deep ocean.After the 1250 yr spin-up, we continue the EECO experiment to model year 2000.We use the 1250 of the spin-up to initialize the four sensitivity experiments.These sensitivity experiments run for another 750 yr to reach quasi-stationary ocean states.All results reported here are the averages of the last 100 yr of each experiment.

The Early Eocene simulation
The Early Eocene experiment simulates a warm climate and an ocean circulation dominated by Southern Ocean Deep Water (SODW).The global mean surface air temperature (SAT) is 25 • C, and the global mean sea surface temperature (SST) is 27 • C (Fig. 1).The average SST in the Arctic basin is about 6 • C and in the Southern Ocean about 15 • C. SODW divides into two branches: one flows north in the Atlantic and upwells in the tropics; the other branch flows through the Indian Ocean and upwells in the Pacific.
Reconstructions of surface temperature for the Early Eocene are based on oxygen isotope data (e.g., Zachos et al., 1994;Schmitz et al., 1996;Fricke and Wing, 2004;Eldrett et al., 2009), Ma/Ca ratios (e.g., Tripati et al., 2003) and the TEX 86 method (a paleothermometer based on the distribution of crenarchaeotal membrane lipids) (Sluijs et al., 2006;Brinkhuis et al., 2006;Bijl et al., 2009;Hollis et al., 2009).They give a rough estimate of the zonal mean surface temperature gradient during the Early Eocene.The simulated zonal mean SAT and SST fall within the range of the reconstructed temperature gradient (Fig. 1d).
It is important to note the uncertainties in these reconstructions of surface temperature.The reconstructions of SST have uncertainties due to the choice of oxygen isotopic composition of mean seawater, the estimated Mg/Ca of seawater, and the calibration of TEX 86 (Huber, 2008).The recent SST reconstructions based on oxygen isotopes (e.g., Pearson et al., 2007;Ivany et al., 2008) are warmer than the earlier results (Zachos et al., 1994).The SST estimates based on TEX 86 (Sluijs et al., 2006(Sluijs et al., , 2007;;Brinkhuis et al., 2006;Bijl et al., 2009;Hollis et al., 2009) appear to be warmer than those from oxygen isotopes and Mg/Ca.Most of the terrestrial temperature estimates, even though they are mostly from North America (e.g., Fricke and Wing, 2004;Greenwood et al., 2005), show a large range.
The simulated surface air temperature falls within the range of terrestrial temperature estimates.The simulated sea surface temperature agrees better with the most recent estimates based on Mg/Ca and oxygen isotopes (e.g., Tripati et al., 2003;Pearson et al., 2007;Ivany et al., 2008), but is significantly lower than the estimates based on TEX 86 , both at low and high latitudes.The model-data discrepancy appears in the northern high latitudes, particularly in the Arctic.The simulated surface temperature is about 5∼6 • C at the surface in the northern high latitudes, which agrees well with other simulations with the same atmospheric CO 2 level (Shellito et al., 2009).However, the simulated annual mean temperatures at northern high latitudes are still lower than most proxy based estimates which fall within the range 10 to 15 • C (Eberle et al., 2010).
In the deep ocean (Fig. 1e), the simulated temperature is consistent with the estimated range of ∼12-14 • C based on Mg/Ca ratios (Lear et al., 2000).Since the Early Eocene simulation presented here has reached equilibrium, the biases in model-data comparison of deep ocean temperature should be small.The simulated Early Eocene ocean circulation is supported by the reconstruction of circulation based on δ 13 C ratios from benthic foraminifera (Fig. 1f), which indicates active deep water formation in the Southern Hemisphere and its northward flow (aging) (Nunes and Norris, 2006).

The sensitivity experiments
In the sensitivity experiments starting from the Early Eocene control climate (EECO), two types of overturning circulation can be discriminated: one dominated by the formation of SODW and the other by NADW (Fig. 2).As in the control simulation, SODW continues dominating ocean circulation in the experiments with a closed West Siberian Seaway (WSSC), a deep Arctic-Atlantic Seaway (AASD), and a deep Drake Passage (DPGD).However, the intensity of SODW formation is weakened, with a maximum overturning at about 40 • S reduced from 31 Sv in the control run (EECO) to 20 Sv in experiment DPGD.Once the tropical seaways (the Tethys and Central American Seaway) are closed/narrowed in experiment TSCN, the deep ocean circulation changes drastically with the commencement of deepwater formation in the North Atlantic and the dominance of NADW (Figs. 2f and 3b).
With the transition to a NADW-dominated ocean circulation in experiment TSCN, global mean SST is reduced by about 0.4 • C compared to the Early Eocene control climate (EECO).In contrast to the nearly symmetric cooling in the high latitudes of both hemispheres in experiment DPGD, the www.clim-past.net/7/801/2011/Z. Zhang et al.: Tropical seaways in Cenozoic cooling  cooling in experiment TSCN is focused to the East Antarctic section of the Southern Ocean (Fig. 3d).A significant decrease of surface air temperature is also observed over East Antarctica (not shown).Note that the initiation of Antarctic glaciation at the Eocene-Oligocene boundary and the expansion at the Middle Miocene were focused to East Antarctica (Zachos et al., 2001).No cooling can be observed over West Antarctica, on which the glaciation occurred later in the Late Miocene (Zachos et al., 2001).
The pattern of the SST changes is similar to the familiar sea-saw pattern (Crowley, 1992), with a strong cooling in the South Atlantic and a warming in the North Atlantic.The sea-saw pattern is best seen when comparing the TSCN and DPGD experiments (Fig. 4a).
A similar pattern of changes in sea surface salinity (SSS) is observed in response to the closed/narrowed tropical seaways.SSS increases by 2 psu in the North Atlantic, and decreases ∼2 to 4 psu in the South Atlantic (Fig. 4b).At the same time, the western boundary currents along the east coast of Southern American are strengthened (Fig. 4c).
The above changes of SSS and currents are caused by the reduction in transport through the Tethys and Central Amer-ica Seaways.Due to the salinity of inflowing Indian Ocean water feeding the Tethys Sea, the salinity in the Tethys Sea is lower than in the tropical and subtropical North Atlantic in the EECO experiment.After the tropical seaways are narrowed/closed in the sensitivity experiments, the flow of water through these two seaways is reduced, particularly in experiment TSCN (Table 2).The closing of the Tethys Seaway shuts down the relatively fresh surface current that flows into the North Atlantic, increasing SSS in the subtropical North Atlantic.Similarly, the narrowing of the Central American Seaway reduces the inflow of relatively fresh Pacific water and supports the development of the western boundary current in the Atlantic.The change of the surface salinity and currents in the Atlantic, caused by the reduced inflow from the Pacific and Tethys Sea, is beneficial for the commencement of NADW formation.
In the sensitivity experiments dominated by SODW formation, there are only small changes in meridional heat transport.Compared with the control run (EECO), experiments WSSC, AASD and DPGD show very small changes in ocean heat transport (Fig. 5).c) surface currents at 50m (cm/s) between the TSCN and the DPGD experiment.SST and SSS changes with confidence levels greater than 95% are illustrated in (a) and (b).The two red rectangles in (a) highlight the Central American Seaway (Left) and the section of the Tethys Seaway (Right) for calculating water mass in Table 2.  2. However, once a NADW-dominated ocean circulation is established in experiment TSCN, northward heat transport is intensified, particularly in the Atlantic (Fig. 5c).As a result, there is a cooling in the Southern Hemisphere (Fig. 6).Although there is a strong cooling over parts of the Southern Ocean, simulated surface temperatures remain above freezing, and the area of sea ice does not increase.Due to the production of deep water in the North Atlantic, more heat is transported from the surface to the deep ocean (Fig. 6c and d).

The role of tropical seaways
The simulated surface temperature gradient and ocean circulation agrees well with previous modelling studies (Shelltio et al., 2009;Winguth et al., 2010).Further, the comparison to proxy data illustrates that the experiment EECO simulates the Early Eocene climate reasonably well, although our simulation can not reproduce the full magnitude of the reconstructed Early Eocene warming in the Arctic region.
Here, we focus on the important role of tropical seaways in the evolution of Cenozoic ocean circulation and climate.Starting from a realistic Early Eocene simulation, the sensitivity experiments indicate that the transition in ocean circulation has a cooling effect on southern high latitudes.The constriction of the main tropical Atlantic seaways appears to play a leading role in this transition.Earlier modelling studies with a shallow Central American Seaway and a closed Tethys Seaway show the existence of NADW in the South Atlantic (Nisancioglu et al., 2003), a strong cooling in the Southern Hemisphere (von der Heydt and Dijkstra, 2006) and development of a western boundary current in the Atlantic (Omta and Dijkstra, 2003).All these studies show that the NADW dominated ocean circulation exists with the Central American Seaway fully opened, suggesting that the Tethys Seaway is more important for the transition of ocean circulation.
The role of the Tethys Seaway is supported by the timing of key geological events (Fig. 7).Detailed reconstructions of the paleogeography (Rögl, 1999) illustrate that the Tethys Seaway was open in the early Miocene (∼24-19 Ma), closed in the late early Miocene (∼19-17 Ma), opened again in the middle Miocene (∼17-15 Ma), before it permanently closed at about 15 Ma.The reduced carbon isotope gradient between the Atlantic and Pacific indicates that NADW formation was active during a period in the late early Miocene and dominated ocean circulation again after about 15 Ma (Wright et al., 1992), though this may be complicated by lower oceanic nutrient levels.Oxygen isotope data reveal that climate was warm in the earliest Miocene, followed by a relatively cold period in the late early Miocene, then came the warm middle Miocene Climate Optimum  the establishment of a ice-sheet on East Antarctica (∼15-14 Ma BP) (Zachos et al., 2001(Zachos et al., , 2008)).These events show that stages with a closed (open) Tethys Seaway correspond well to active (inactive) NADW formation and a cold (warm) climate, supporting the results simulated here.In addition to the Tethys Seaway, the gradual shoaling of the Central American Seaway beginning ∼16 Ma contributed to the transition of ocean modes and the consequent cooling in the Miocene.Note also that the Tethys Seaway is important for the devel-opment of the monsoon climate in East Asia (Ramstein et al., 1997;Zhang et al., 2007a, b).
The earliest evidence of NADW formation has been found in deep sea cores from the Early Oligocene (Davies et al., 2001;Via and Thomas, 2006;Scher and Martin, 2008).This is approximately the time when the Arabia-Eurasia collision happened (Allen and Armstrong, 2008).The collision might have caused notable narrowing of the Tethys Sea, though the sea was still open in the Oligocene (Rögl, 1999).Simulations of Oligocene climate (von der Heydt and Dijkstra, 2008) suggest that NADW formation could have existed before the Tethys Seaway permanently closed.Therefore, the timing for the onset of NADW is still under debate, and the two candidates are the Early Oligocene and the Middle Miocene.However, both of these two time periods start with a cooling.
In an earlier study, with an idealized model investigating the role of the tropical seaways, Hotinsky and Toggweiler (2003) found that the closing of a Tethyan circumglobal passage caused a cooling of the Northern Hemisphere.This is in contradiction to our results showing a warming of the Northern Hemisphere when the tropical seaways are constricted.This difference is most likely due to the very different model configurations.The previous study used a highly idealized bathymetry, or "water planet" with strong wind-driven upwelling in the tropics, and did not realistically simulate salinity changes caused by fresh water input associated with changes to the land-sea distribution.As our simulations demonstrate, the increase of sea surface salinity is one important reason for the transition of ocean circulation to the NADW mode associated with warming in the Northern Hemisphere.
The simulations indicate that the role of the Drake Passage opening/deepening on ocean circulation and climate is less important compared to the impact of the Tethys and Central Cenozic cooling based on deep-sea oxygen isotopes and ice-sheet development (Zachos et al., 2001(Zachos et al., , 2008)), 3) the evolution of ocean circulation.The Tethys evolution is summarized according to Akhmet'ev and Beniamovski (2006), Rögl (1996), Barrier and Vrielynck (2008).The changes of other seaways are summarized according to Scotese (2001), Livermore et al. (2005) and Scher and Martin (2006).The Early Oligocene evidence of NADW formation is summarized according to Davies et al. (2001), Via and Thomas (2006) and Scher and Martin (2008).The red line show the NADW index based on benthic carbon isotopes (Wright et al., 1992).The high value of Δδ 13 C indicates active NADW, and the low value indicates inactive NADW.(Zachos et al., 2001(Zachos et al., , 2008)), (3) the evolution of ocean circulation.The Tethys evolution is summarized according to Akhmet'ev and Beniamovski (2006), Rögl (1996), Barrier and Vrielynck (2008).The changes of other seaways are summarized according to Scotese (2001), Livermore et al. (2005) and Scher and Martin (2006).The Early Oligocene evidence of NADW formation is summarized according to Davies et al. (2001), Via and Thomas (2006) and Scher and Martin (2008).The red line shows the NADW index based on benthic carbon isotopes (Wright et al., 1992).The high value of δ 13 C indicates active NADW, and the low value indicates inactive NADW.American Seaways.These results challenge the prevailing hypothesis that the Drake Passage is the key player in major cooling events during the Cenozoic and the glaciation of Antarctica.
The cooling effect of the Drake Passage and its impact on Cenozoic glaciations might have been overestimated in the earlier sensitivity experiments (e.g., Nong et al., 2000;Sijp and England, 2004), as they were carried out with present topography and bathymetry.In this case, the intensity of the Antarctic Circumpolar Current (ACC) is above 100 Sv, once the Drake Passage is opened.However, as shown in the present study, the Drake Passage was relatively narrow and shallow in the early Cenozoic, and ocean circulation was dominated by the formation of SODW.As a consequence, the ACC during the early Cenozoic is thought to have been significantly weaker than present (Huber and Sloan, 2001;this study).With this in mind, the effect of the Drake Passage on ocean circulation and the climate of Antarctica during the early Cenozoic should be reconsidered.

Implications for Cenozoic cooling
The mechanisms behind the long term Cenozoic cooling trend and major cooling events are still under debate.There are four main hypotheses/theories, including the thermal iso-lation of Antarctica (Kennett, 1977;Toggweiler and Bjornsson, 2000), declining levels of atmospheric CO 2 (DeConto and Pollard, 2003;Huber and Nof, 2006), uplift of large mountains (Raymo and Ruddiman, 1992), and modified insolation patterns and seasonality (Coxall et al., 2005).The thermal isolation hypothesis for understanding the cooling at the Eocene-Oligocene boundary has been challenged by geological evidence (Moran et al., 2006;Tripati et al., 2008), early modelling studies (DeConto and Pollard, 2003;Huber and Nof, 2006) and the present study.The CO 2 theory suggests that the declining levels of atmospheric CO 2 is more important in understanding the cooling at the Eocene-Oligocene boundary.This theory is strongly supported by reconstructions of Cenozoic atmospheric CO 2 (e.g., Pagani et al., 2005;Zachos et al., 2008, Pearson et al., 2009) and modelling studies (DeConto and Pollard, 2003;Huber and Nof, 2006).However, this does not answer the question why atmospheric CO 2 levels decreased greatly during the Cenozoic, and in particular at the Eocene-Oligocene boundary.The uplift hypothesis links mountain uplift with the drop of atmospheric CO 2 levels in the Cenozoic (Raymo and Ruddiman, 1992).The uplift of the Himalayas-Tibetan Plateau, Rocky Mountains and Andes Mountains would increase chemical weathering, thereby reducing atmospheric CO 2 levels and cause a cooling.This hypothesis is still important in understanding the long-term cooling trend in the Cenozoic.However, there are uncertainties in the history of the uplift, and it is difficult to estimate how much atmospheric CO 2 was removed in the process.The weak insolation hypothesis suggests that weak insolation caused by the Earth's orbital configuration was the ultimate trigger for the cooling events at the Eocene-Oligocene boundary (Coxall et al., 2005).However, this hypothesis is based on a tuned age model, with its inherent uncertainties (Kuiper et al., 2008).In summary, the key remaining problem in explaining the Cenozoic cooling is why atmospheric CO 2 levels were greatly reduced, in particular at the Eocene-Oligocene boundary.
In the present study, our experiments illustrate the linkage between tropical seaways and ocean circulations associated with the Cenozoic cooling.The constriction of the main tropical Atlantic seaways appears to play an important role in the transition of ocean circulation from a SODW to a NADW ocean circulation mode.The transition of ocean circulations causes a significant cooling in the Southern Hemisphere.We suggest that this linkage is potentially important for understanding the Cenozoic cooling, in particular for the cooling at the Eocene-Oligocene boundary.
However, there are still some limitations in our present study.In order to better isolate the role of the seaways, we carry out the experiments with the atmospheric CO 2 level fixed at eight times the preindustrial level.Due to the high atmospheric CO 2 level, we are limited to address the possible impact of the decrease in atmospheric CO 2 levels on ocean circulations.As earlier studies illustrated, the changes of atmospheric CO 2 level may also influence ocean circulation modes (Lunt et al., 2010), and climate sensitivities to seaways associated with different atmosphere CO 2 levels can be different (Sijp et al., 2009).Furthermore, we do not distinguish the respective roles of the Central American Seaway and the Tethys Seaway.Based on the Early Eocene land-sea distribution conditions, we are not able to consider the possible influence of the widening of the Atlantic on the meridional overturning circulation (von der Heydt and Dijkstra, 2008).With these limitations in mind, the linkage between the tropical seaways and the ocean circulations still needs to be tested by more experiments with reduced atmospheric CO 2 levels.
In our experiments, the local cooling in the Southern Hemisphere caused by the transition from a SODW to a NADW dominated ocean circulation is significant (Figs. 3d and 6a,b).However, the global effect of the transition is minimal, with a global mean cooling of 0.4 • C in SST.The global cooling is weaker than what is estimated from available geological evidence (e.g., Liu et al., 2009;Eldrett et al., 2009), indicating that changes in ocean circulation alone can not fully explain the cooling at the Eocene-Oligocene boundary.Additional factors must be involved to account for the full scale of the cooling, such as the drop of atmospheric CO 2 levels (DeConto and Pollard, 2003;Huber and Nof, 2006).
Here, we suggest the potential link between the transition of ocean circulation from a SODW to a NADW dominated mode and the large observed drop in atmospheric CO 2 levels at the Eocene-Oligocene boundary.Since the ocean plays a vital role in the Earth's carbon cycle, it is likely that the transition of ocean circulation to a NADW dominated mode might have played an important role in causing the drop in atmospheric CO 2 levels in the Cenozoic.The response of the carbon cycle can be very fast (e.g., Dunkley et al., 2010).Further improving our knowledge of the link between ocean circulation and the carbon cycle is important for understanding the cooling pattern at the surface (Liu et al., 2009), the debate of bipolar versus polar glaciation (DeConto et al., 2008;Lear et al., 2008), and extinction and environmental events (Pearson et al., 2008) at the Eocene-Oligocene boundary.However, how the switch from a SODW to a NADW dominated mode of circulation impacted the carbon cycle remains an open question, which must be addressed by additional proxy data and modelling studies.
Although the simulations presented here and the studies of DeConto and Pollard (2003), Huber et al. (2004) and Huber and Nof (2006) indicate that the Drake Passage plays a minor role in cooling the Antarctic continent, the impact of the Drake Passage opening/deepening on the long-term Cenozoic cooling trend should not be neglected.With a gradual expansion and deepening of the Drake Passage, the strength of the ACC is expected to have increased, possibly, playing a role in the observed long-term global cooling in the Cenozoic.

Summary
In summary, starting from an equilibrium simulation of the Early Eocene, the important role of tropical seaways is highlighted in the present study.The simulation shows a potential important linkage between tropical seaways and the Cenozoic cooling.The Cenozoic cooling trend was enhanced by the transition between two different ocean circulation modes: one corresponding to a SODW-dominated mode and the other to a NADW-dominated mode.The closing of the Tethys Seaway was likely to be key to the transition from the SODW to the NADW.The constriction of the Tethys Seaway and the establishment of modern-like ocean circulation dominated by NADW formation might have triggered the Eocene-Oligocene cooling and glaciation over Antarctica, once atmospheric CO 2 levels reached a critical level.Although the role of the Tethys Seaway is supported by some geological evidence, the role of tropical seaways still needs to be investigated by more modelling and data studies in future.

Fig. 1 .
Fig. 1.Warm climate and SODW-dominated ocean circulation simulated in the Early Eocene (EECO) experiment.(a) Sea surface temperature (SST, • C).(b) Zonal mean ocean temperature ( • C).(c) Deep ocean circulation (cm s −1 ) at 2500 m.Arrows show flow directions, and colour scales show current speed.(d) Comparison of latitudinal sea surface temperature between the model simulation and proxy records.LMA means leaf-margin analysis.(e) Comparison of deep water temperature among the simulation (bold green line), paleotemperature estimation (green shaded bar)(Lear et al., 2000) and benthic foraminiferal δ 18 O (small green markers).(f) Latitudinal gradients of benthic foraminiferal δ 13 C in the Early Eocene.All geological data used here are listed in the supplementary materials.

Figure 3 .
Figure 3. Transition of ocean circulation from a SODW-dominated to a NADW-dominated mode and it impact on SST.Deep ocean currents (cm/s) at 2500m for experiments (a) DPGD and (b) TSCN.Geographic distributions of SST changes ( o C) in the DPGD (c) and the TSCN (d) experiments, compared to the control experiment (EECO).Arrows show flow directions, and colour scales show current speed.Changes in seaways are highlighted by red rectangles:1), closing of the West Siberian Seaway (WSSC); 2), deepening of the Arctic/Arctic-Atlantic Seaway (AASD); 3), deepening of the Drake Passage Gateway (DPGD); 4), constriction of the tropical Atlantic seaways (TSCN), with a fully closed Tethys seaway and narrow Central American seaway.The statistical signifcance has been tested with the t-test, based on annual mean SST time series for each point.Only SST changes with confidence levels greater than 95% are illustrated here.

Fig. 3 .
Fig. 3. Transition of ocean circulation from a SODW-dominated to a NADW-dominated mode and its impact on SST.Deep ocean currents (cm s −1 ) at 2500 m for experiments (a) DPGD and (b) TSCN.Geographic distributions of SST changes ( • C) in the DPGD (c) and the TSCN (d) experiments, compared to the control experiment (EECO).Arrows show flow directions, and colour scales show current speed.Changes in seaways are highlighted by red rectangles: (1) closing of the West Siberian Seaway (WSSC); (2) deepening of the Arctic/Arctic-Atlantic Seaway (AASD); (3) deepening of the Drake Passage Gateway (DPGD); (4) constriction of the tropical Atlantic seaways (TSCN), with a fully closed Tethys seaway and narrow Central American seaway.The statistical signifcance has been tested with the t-test, based on annual mean SST time series for each point.Only SST changes with confidence levels greater than 95 % are illustrated here.

Figure 4 .
Figure 4. Geographic changes of (a) SST ( o C) , (b) SSS and (c) surface currents at 50m (cm/s) between the TSCN and the DPGD experiment.SST and SSS changes with confidence levels greater than 95% are illustrated in (a) and (b).The two red rectangles in (a) highlight the Central American Seaway (Left) and the section of the Tethys Seaway (Right) for calculating water mass in Table2.

Fig. 4 .
Fig. 4. Geographic changes of (a) SST ( • C) , (b) SSS and (c) surface currents at 50 m (cm s −1 ) between the TSCN and the DPGD experiment.SST and SSS changes with confidence levels greater than 95 % are illustrated in (a) and (b).The two red rectangles in (a) highlight the Central American Seaway (Left) and the section of the Tethys Seaway (Right) for calculating water mass in Table2.

Table 1 .
Summary of boundary conditions.
• C at the equator, deepwater temperature of 12 • C at the equator, SST of 15 • C at both poles and deep-water temperature of 10 • C at both poles.Based on a rough empirical relationship of present day ocean temperature with depth, we use the following equa- tions: t 0 (d) = 12 + 22 × e − d 1000 and t 90 (d) = 10 + 5 × e − d 1000

Table 2 .
Water mass (Sv) transport through the Tethys Seaway and the Central American Seaway.Positive values are to the east.