Equilibrium state and sensitivity of the simulated middle-to-late Eocene climate

While the early Eocene has been considered in many modelling studies, detailed simulations of the middle and late Eocene climate are currently scarce. To understand Antarctic glaciation at the Eocene-Oligocene Transition (∼34Ma) as well as middle Eocene warmth, it is vital to have an adequate reconstruction of the middle-to-late Eocene climate. Here, we present a set of high resolution coupled climate simulations using the Community Earth System Model (CESM) version 1. 5 Two middle-to-late Eocene cases are considered with new detailed 38Ma geographical boundary conditions with a different radiative forcing. With 4x pre-industrial concentrations of CO2 (i.e. 1120 ppm) and CH4 (∼2700 ppb), the equilibrium sea surface temperatures correspond well to available late middle Eocene (42-38 Ma) proxies. Being generally cooler, the simulated climate with 2x pre-industrial values is a good analog for that of the late Eocene (38-34 Ma). Deep water forma10 tion occurs in the South Pacific Ocean, while the North Atlantic is strongly stratified and virtually stagnant. A shallow and weak circumpolar current is present in the Southern Ocean with only minor effects on southward oceanic heat transport within wind-driven gyres. Terrestrial temperature proxies, although limited in coverage, also indicate that the results presented here are realistic. The reconstructed 38Ma climate has a reduced equator-to-pole temperature gradient and a more sym15 metric meridional heat distribution compared to the pre-industrial reference. Climate sensitivity is similar (∼0.7◦C/Wm2) to that of the present-day climate (∼0.8◦C/Wm2; 3◦C per CO2 doubling), with significant polar amplification despite very limited sea ice and snow cover. High latitudes are mainly kept warm by albedo and cloud feedbacks in combination with global changes in geography and the absence of polar ice sheets. The integrated effect of geography, vegetation and ice accounts 20 for a 6-7◦C offset between pre-industrial and 38Ma Eocene boundary conditions. These 38Ma sim1 Clim. Past Discuss., https://doi.org/10.5194/cp-2018-43 Manuscript under review for journal Clim. Past Discussion started: 12 April 2018 c © Author(s) 2018. CC BY 4.0 License.


Introduction
The Eocene-Oligocene Transition (EOT) is one of the most dramatic climate transitions of the Cenozoic, thought to be associated with the formation of a continental-scale ice sheet on Antarctica (Zachos et al., 1994;Coxall et al., 2005;Lear et al., 2008).A possible cause for the inception of ice is a long-term decline of greenhouse gas concentrations through the middle Eocene, eventually 30 crossing a threshold for glaciation (DeConto and Pollard, 2003;DeConto et al., 2008;Gasson et al., 2014).Following the early Eocene (∼50Ma), a general cooling levelled off to a plateau in the middle Eocene  and eventually reversed into a warming until ∼40Ma (Zachos et al., 2001(Zachos et al., , 2008;;Bijl et al., 2009).At the Middle Eocene Climatic Optimum (MECO; Bohaty and Zachos 2003; Sluijs et al. 2013), conditions returned to values close to those seen in the early Eocene (Zachos 35 et al., 2008;Cramer et al., 2009;Bijl et al., 2010Bijl et al., , 2013) ) and quickly cooled down again into the late Eocene (∼38Ma).This was followed by a cooling event at ∼37.3Ma characterised by the Priabonian Oxygen isotope Maximum (PrOM, Scher et al. 2014).Significant swings in global temperature thus occurred prior to the EOT.Yet, conditions only allowed the growth of a continental-scale Antarctic ice sheet after 34Ma, although indications for significant ice volume in the late Eocene have been 40 found (Scher et al., 2014;Passchier et al., 2017;Carter et al., 2017).It remains a question to what extent continental geometry (e.g.opening of Southern Ocean Gateways) next to gradual shifts in both the atmospheric and oceanic circulation, was a driver to both regional and global climate change during the Eocene (Bijl et al., 2013;Bosboom et al., 2014;Sijp et al., 2014Sijp et al., , 2016)).Both the timing and effects of Southern Ocean Gateways opening during the Eocene and Oligocene remain uncertain 45 and are not necessarily related to the EOT (Stickley et al., 2004;Lagabrielle et al., 2009).
Prior to the EOT, not only the exceptionally warm early Eocene (Greenwood and Wing, 1995;Bijl et al., 2009;Huber and Caballero, 2011), but also the later part of the Eocene were characterised by a low equator-to-pole temperature gradient (Bijl et al., 2009;Hollis et al., 2012;Douglas et al., 50 2014; Evans et al., 2018).In general, the Eocene greenhouse climate has proven challenging to simulate adequately with climate models (Huber and Sloan, 2001;Huber and Caballero, 2011).Previous model simulations needed very high radiative forcing in order to reproduce high-latitude warmth but at the expense of equatorial temperatures being significantly higher than indicated by proxy data (Huber and Caballero, 2011;Lunt et al., 2012).In more recent studies, consensus between models 55 and proxy data is growing as equatorial temperature estimates have risen (Pearson et al., 2007; In-construction from Baatsen et al. (2016).The POP ocean component runs on a ∼1 • (384 x 320) curvilinear grid with its northern pole located in central Greenland, and 60 vertical layers.The thickness of these layers depends on depth, being 10m in the upper 20 layers and gradually increasing to 500m at a maximum depth of 5500m.The ocean is coupled to the CAM4 atmosphere component that uses a ∼2 • (90 x 144) finite volume grid and 26 vertical layers extending upwards to 2hPa.Veg-110 etation is fixed; generally tropical or subtropical in low-to-middle latitude regions (savannah/shrubs in sub-tropical continental interiors) and mixed forests further poleward.Aerosol concentrations are fixed and determined by running the atmospheric component of the model for 50 years as a bulk aerosol model (BAM; Kiehl et al. 2000).The 38Ma simulations have 2x and 4x pre-industrial levels (280ppm; 671ppb) of both CO 2 and CH 4 , referred to as 2x and 4x PIC.Using the estimated radia-115 tive forcing from Etminan et al. (2016), our simulations are comparable to 2.15× (∼600ppm) and 4.69× (∼1300ppm) pre-industrial CO 2 -equivalent for 2x and 4x PIC, respectively.For reference, a pre-industrial simulation is performed using the same model version with 1x PIC and present-day boundary conditions for geography and vegetation.A short overview of the main characteristics of each simulation is given in Table 1.120 Both 38Ma model simulations start from the same initial conditions: a stagnant ocean with a horizontally homogeneous temperature distribution.The initial ocean temperature decreases linearly with depth, from 15 • C at the surface to 9 • C at the bottom.The pre-industrial reference is initialised 125 using temperature and salinity fields from the PHC2 dataset (Steele et al., 2001).A long spin-up (see Table 1) is performed to allow the deep ocean to equilibrate sufficiently.An overview of absolute (∆T , ∆S) and normalised (∆T /T , ∆S/S) drifts over the last 200 model years is given in  Spin-up 3000 years 3000 years 4000 years Table 1.Overview of characteristics for all CESM 1.0.5 simulations that were performed (BAM: bulk aerosol model).
Table 2 for each spin-up.Drifts are generally ∼ 10 −4 K/year for global mean, volume weighted average ocean temperature.Similarly, for globally averaged salinity drifts at the end of the spin-up are   stratified North Atlantic Ocean (Coxall et al., 2018).Deep water formation occurs only in the South various Eocene simulations (Thomas et al., 2014;Baatsen et al., 2018;Hutchinson et al., 2018) and also suggested by proxy data (Hague et al., 2012).
Most of the global circulation is dominated by sub-tropical and sub-polar gyres (Figure 3c), with the strongest cell in the South Indo-Pacific.A frontal zone is present in the Southern Ocean, separating cyclonic (sub-tropical) and anticylonic (sub-polar) gyres.This front coincides with an Antarctic Cir-180 cumpolar Current (ACC) that is, however, still strongly restricted in its path and depth (200-500m).
Strikingly, the front is located at 55-60  (Bijl et al., 2009;Douglas et al., 2014).On the other hand, a strong eastward flow Eocene corresponding to a benthic δ 18 O difference of 0.5-1‰ (Zachos et al., 2001(Zachos et al., , 2008 in position of the upwelling cells between Eocene and pre-industrial circulation is smaller than those seen for the surface polar front in Figure 3. An otherwise predominantly wind-driven gyre circulation is reflected by symmetric meridional oceanic heat fluxes into both hemispheres (Figures 4d).This is in contrast to the pre-industrial situation, where approximately a 1PW maximum difference between hemispheres is observed making the 220 Northern Hemisphere relatively warm (Trenberth and Caron, 2001).A major part of the heat flux is generated in the Pacific Ocean, with the exception of the southern low latitudes.A large contribution arises from both the South Indian and South Atlantic sub-tropical gyres.A subtle but important difference between hemispheres is seen at high latitudes, where about 0.5PW is transported southward at 45 • S while the heat transport is close to zero at 45 • N.This difference can be partly explained by The ocean heat transport patterns in our 38Ma Eocene simulations generally agree with those seen in previous model studies.In situations with no (or a restricted) ACC, increased southward heat transport is found in the Southern Hemisphere and attributed to both changes in the horizontal gyre circulation (in particular sub-polar gyres; Huber et al. 2004;Huber and Nof 2006;Sijp et al. 2011) 235 and more directly to the meridional overturning circulation (Toggweiler and Bjornsson, 2000;Sijp and England, 2004;Sijp et al., 2009).These differences with respect to the present (pre-ndustrial) ocean circulation have been referred to as the Drake Passage effect Toggweiler and Bjornsson (2000), which has been shown to also exist (with different strength) in model simulations resolving meso- While it seems unlikely that these regions would be colder than they are today, the model possibly also underestimates strong seasonal cooling in shallow coastal waters and other local effects near 290 the coastline.The complex palaeogeography at both locations is not sufficiently represented even in our relatively high-resolution simulations.SST estimates from other locations at similar latitude, such as ODP 1052 on Blake Nose (east of Florida) are much higher bringing the model and proxies into good agreement.A better match is generally found at middle and high latitude regions (apart from the Arctic) when summertime temperatures are considered.Since SST proxies are based on 295 past living organisms, it is possible that at higher latitudes these proxies have a bias towards the warm season as their activity and sedimentation are directly or indirectly affected by the amount of sunlight (Sluijs et al., 2006(Sluijs et al., , 2008;;Bijl et al., 2009;Hollis et al., 2012;Schouten et al., 2013).
A comparison of 2x PIC model SST with 38-34 Ma proxy estimates shows a similar result (Appendix: Table A2 and Figure A1).Good agreement between proxy and model results exists in equa-300 torial regions, with again a large spread in proxies and generally a better match with summer temperatures at higher latitudes.Assuming that temperature proxies represent the annual mean in equatorial waters but are skewed towards the warm season at middle and high latitudes, the model does seem to reconstruct the middle-to-late Eocene (42-34 Ma) temperature distribution quite well.In addition to the annual mean fields shown in Figure 3, winter and summer averaged SST, wind stress, barotropic stream function and mixed layer depth are presented in Figure 7. Sea surface salinity and upper 200m potential density are not considered here as they show little seasonality.At southern high latitude regions, SST seasonality is strong with winter temperatures of mostly 6-12 • C 310 rising to 16-24 • C in summer.A combination of solar heating and gyre circulations causes the coldest summer-time temperatures in the Southern Hemisphere to occur at the centre of sub-polar gyres, rather than near the Antarctic coastline (zonal mean temperatures are also seen to increase slightly beyond 65 • S in Figure 6).Not surprisingly, the South Pacific holds the warmest high latitude waters as it exhibits additional southward heat transport in the meridional overturning circulation.

315
Considering the barotropic stream function, warm waters in the Southwest Pacific are carried southward through the East Australia Current (EAC) down to 55 • S where the EAC meets the proto-Leewin current from the west, and northward in the Ross Gyre further poleward.This is mostly consistent with the picture for the middle Eocene given by Bijl et al. (2009), but shifted southward (with a much smaller Ross Gyre) compared to Huber et al. (2004).In contrast, seasonality outside near-equatorial To put the results presented here into perspective, the 4x PIC case is compared to previous model simulations from Goldner et al. (2014) (hereafter: GH14) in Figure 8.The latter use a 45Ma hot spot referenced reconstruction for their geographical boundary conditions and 4x pre-industrial levels of CO 2 (i.e.1120ppm).The main difference between both studies is the horizontal resolution; ∼1 • and 335 ∼2 • in our 38Ma simulations versus ∼3 • and ∼4 • in GH14 for the ocean and atmosphere grids, respectively.While the 38Ma 4x PIC case also has a quadrupling of CH 4 and the 45Ma one does not, the difference in global temperature is much larger than would be expected based on radiative forcing alone.The grid and resolution used here for the atmospheric component (CAM4) are shown to both increase the sensitivity to a doubling of CO 2 by Bitz et al. (2012).Aditionally, using the 340 finite volume instead of a spectral grid greatly influences cloud radiative forcing and causes a warming compared to GH14.Indeed, similar global temperatures are found under comparable radiative forcing by Hutchinson et al. (2018), using a lower resolution but comparable finite difference dynamical core in the Atmospheric component.Finally, a warm initialisation of the deep ocean, general circulation changes and aerosol changes may add further to temperatures differences.

345
The most prominent differences in geography are the representation of Antarctica, the Tasmanian Gateway opening and the position of India.Especially the Southern Ocean shows large differences in circulation and the resulting temperatures, mainly related to the different geography.The formation of a proto-ACC and extent of the East Australia current act to shift the polar front in the South Pacific southward for the 38Ma case, while the opposite happens for 45Ma.These changes can be 350 linked directly to the continental configuration and associated shifts in zonal wind stress (maximum at 55 • S versus 45 • S).The more southerly position of the polar front can explain the heterogeneity in the Southern Ocean; a difference of 6-8 • C is seen between Tasmania and the tip of the Antarctic Peninsula in agreement with proxy indications from Douglas et al. (2014).Generally, western boundary currents (e.g.Kuroshio, Agulhas, East Australia Current) and the effects of bottom topog-355 raphy are more pronounced in the 38Ma results.Finally, an issue in the 45Ma results with very low (negative) salinities in the Arctic, although having seemingly little impact on the general circulation, is mostly resolved (lowest salinities down to ∼10 psu) in the 38Ma case by having several shallow passages.

360
A comparison of zonal mean SST's in the 38Ma 2x PIC and 4x PIC cases, 45Ma 4x CO 2 from GH14 and pre-industrial reference simulation along with proxy estimates is shown in the Appendix (Figure A3).Apart from the 38Ma cases being overall warmer, the zonal mean temperature profile  (Song and Zhang, 2009;Bellucci et al., 2010).Effects of orographic lift are evident on westward facing coastlines and ridges across the middle and high latitudes.
The temperature at 700hPa (Figure 9c) highlights warm mid-level air masses in persistent continen-385 tal high pressure regions in the sub-tropics.Antarctica is substantially warmer than the Arctic at this pressure level because of its elevation and continental climate.Similar to the Tibetan Plateau today, it therefore acts as an elevated heat island (Hoskins and Karoly, 1981;Ye and Wu, 1998).Wind speeds at 200hPa show maxima linked to the positions of both sub-tropical and polar jet streams.Several stationary Rossby Waves are present, but a persistent upper level trough is evident around 90 • E, 390 where both jets coincide.Such a pattern can also be seen at present day but only in the Northern Hemisphere (located at 120 • E in the pre-industrial reference), caused by the influence of the Tibetan Plateau.
Finally, the 500hPa geopotential height field (Figure 9d) shows a nearly meridionally symmetric pattern, with the sharpest gradient in middle-latitude regions.The 500hPa surface is generally about 395 200m higher than in the pre-industrial reference, mainly due to warming of the air column.The current ∼200m asymmetry between southern (∼5km) and northern (∼5.2km) high latitudes does not occur in the 38Ma Eocene climate (∼5.4km for both).The corresponding tropopause height (using the WMO definition of 2 • C/km) is at 17-18km in the tropics and does not fall below 11-12km at high latitudes, being 1 and 2km higher than it is in the pre-industrial climate, respectively.

400
As was done for sea surface temperatures in the previous section (Figure 6), an overview of zonally averaged 2m air temperatures is given in Figure 10.Considering annual mean temperatures, a meridionally symmetric pattern with a low equator-pole gradient stands out again.This pattern looks 405 similar for both 38Ma simulations, being about 3 • C colder in the 2x PIC case with a slight amplification towards the poles.Regardless of greenhouse gas concentrations, the pre-industrial reference temperatures are in sharp contrast with those seen in the Eocene cases.In addition to being considerably colder (global mean 2m temperature of 13.7 • C), the pre-industrial climate features a stronger equator to pole temperature gradient with high latitudes up to 40 • C colder than those seen in the 410 38Ma 2x PIC case.This shows that the absence of ice sheets and changes in continental geometry have a significant impact on global mean temperature as well as its distribution.Despite changes in the general circulation, total meridional heat fluxes are not substantially different in the Eocene climate compared to the pre-industrial reference (Figure 4c).The reduced equator to pole temperature gradient is thus a result of reduced heat loss at high latitudes rather than enhanced heat fluxes.seasonality is weak at southern middle latitudes it is stronger at northern middle and high latitudes, 420 again caused by differences in land coverage.Remarkably strong seasonality is seen on the Antarctic continent, where average winter (DJF) temperatures drop below -10 • C while easily reaching 30 • C in summer (JJA).The continental configuration thus causes the coldest place in winter on the Northern Hemisphere to be at ∼70 • N (rather than the North Pole), while the warmest place in summer south of ∼45 • S is on Antarctica.The near meridional symmetry in temperatures is reflected by both 425 oceanic and atmospheric heat fluxes, shown in Figure 4c-d.
A comparison between middle-late Eocene terrestrial proxies (Appendix: Table A3, data from Greenwood and Wing 1995;Gregory-Wodzicki 1997;Smith et al. 1998;Wolfe et al. 1998; Green-  Traiser et al. (2005); Kowalski and Dilcher (2003)) and MBT-CBT (Peterse et al., 2012).In contrast to Figure 10, only temperatures over land are considered in the zonal mean and pre-industrial values are not shown.The latitudinal distribution of the available terrestrial proxies is rather limited and covers mostly middle latitude regions.
A good agreement is seen with proxy data from the North Atlantic (Greenland), South America 440 (Chile), Europe, Australia and the Antarctic margin.While most higher latitude (>40 • N) proxies agree fairly well with model temperatures over China and North America, there is a large discrepancy at lower latitudes.As discussed before, continental interiors are seemingly too warm in the model especially in summer.Meanwhile, many of the proxy estimates from China likely underestimate summer temperatures (possibly explaining the apparent lack of a latitudinal temperature gradi-445 ent) as there is no present-day equivalent of vegetation that can withstand such high heat stresses.In addition, proxy temperatures are representative of a specific location while the horizontal resolution of the simulated atmosphere is limited (∼ 2 • ).Therefore, the model will underestimate local effects by having a strongly smoothed topography.A4 and Figure A2), with again a good agreement in middle latitude regions but larger differences over interior China.
The model results presented here are again compared to those from Goldner et al. (2014) (GH14) and shown in the Appendix (Figure A4).Similar to the ocean results, all of the considered Eocene 460 simulations have a comparable zonal mean temperature distribution with the 38Ma cases being generally warmer.High summertime temperatures over low-middle latitude continental regions are more pronounced in the 38Ma 4x PIC case, but also present for 45Ma 4x CO 2 from GH14.The large seasonality over Antarctica is also enhanced in the 38Ma simulations compared to GH14, with summer temperatures well over 30 • C in the 4x PIC results.Because of these extreme summertime temper-  tionary Rossby Waves, are most prominent in winter.Smaller stationary waves are seen in summer and linked to sub-tropical ridges where high surface temperatures are mixed upwards to 700hPa.At this level the effect of solar heating at high altitudes (∼2km) on East-Antarctica stands out, where 490 temperatures are about 10 • C higher than those seen over the Southern Ocean.
Finally, some of the peculiar features in the atmospheric circulation of the 4x PIC case are considered by looking at the zonal mean zonal wind and temperatures for June-August (Figure 13a) 495 and December-February (Figure 13b).The zonal mean temperature shows a typical pattern for the Northern Hemisphere summer.In the troposphere, a steady cooling towards the poles can be seen, but in the stratosphere further cooling only takes place on the winter pole because of solar heating at the summer pole.Persistent cold temperatures are present around the top of the tropical tropopause, at about 70hPa (compared to 90hPa at present-day).Associated with the meridional temperature gra-500 dients in the troposphere are the sub-tropical and polar jets, where the former is more pronounced in the zonal mean and both jets are stronger in the winter hemisphere.Thermal wind balance demands an increase of the zonal wind with height as long as temperatures decrease poleward.As a result, the polar jet extends into the stratosphere in the winter hemisphere, while decreasing again above the tropopause in summer.The polar regions are thus surrounded by a cyclonic polar vortex in winter 505 that extends from the surface to the top of the stratosphere.
An extension of the subtropical jet towards the equator can be seen, resulting in positive (westerly) zonal winds at the equator and thus superrotation at the tropopause even in the annual mean.This is caused by the momentum transfer of atmospheric Rossby waves (causing the 'V' shape in the tropical tropopause in Figure 9d) and the strong winter cell of the Hadley circulation.As it was shown 510 by Caballero and Huber (2010), superrotation does occur in very warm climate simulations and is therefore no surprise to emerge in the 4x PIC case where equatorial SSTs reach 35 • C.Although weaker, zonal mean equatorial westerly winds are still present in the 2x PIC climate.The highest (annual mean, zonal mean) equatorial SSTs in this case are nearly 33 • C and thus close to the threshold found in Caballero and Huber (2010).In the lower tropical troposphere, perennial easterlies are reverses in summer across southern high latitudes.The Antarctic continent thus acts as a (slightly 520 elevated) heat island, reversing the meridional gradients in both pressure and temperatures and thus also zonal winds through the thermal wind balance.While the sub-tropical jet remains unaffected (driven by the middle latitude gradients), the polar jet becomes anticyclonic in both the troposphere and stratosphere with a maximum near the tropopause.This makes the Antarctic summer climate resemble that of the sub-tropics rather than a typical polar summer.).Note that both simulations were integrated to equilibrium separately, without ever imposing a strong perturbation of greenhouse gas doubling.Even in this virtually ice-free world significant polar amplification can be 540 observed, with most of the warming occurring in high latitude continental regions.As in the present day climate the polar amplification is in part due to albedo effects from vegetation, snow cover and (limited) seasonal sea ice.Low latitude regions, on the other hand, warm less as they are more strongly governed by moist processes and tied to SST's.
Averaging the response (to ∆RF 4× 2× ) globally yields an equilibrium climate sensitivity of S EO = 545 0.69 • C/Wm −2 for these simulations.The Eocene ECS is smaller than the reported 0.82 for the present-day version of the CESM model, using ∆RF CO2 and ∆T = 3.13  14a).The zonal variation in the warming signal is highest in the middle latitude regions because of land-ocean contrasts, while differences in seasonal response are largest in polar regions.
555 El Niño-like conditions in the warmer 38Ma climate.The equatorial westerlies seen in Figure 13 may play an important role in the regional climate through a feature resembling the Madden-Julian Oscillation, similar to the one found in Caballero and Huber (2010).A general increase in deep convection and precipitation can be seen in most sub-tropical monsoon regions and the middle-latitudes in the 38Ma Eocene climate.A prominent decrease of low level clouds in the Arctic is also seen 580 due to the Arctic Ocean warming up and decreasing the strength of a summertime marine inversion.
Over Antarctica, on the other hand, a large increase in low-level cloudiness is related to increased moisture helping to keep up winter temperatures at the surface (Figures 14e-f).

585
The Earth System Sensitivity (ESS) for this model set-up can be estimated by comparing the 38Ma Eocene simulations to the pre-industrial reference.In the middle-late Eocene climate not only greenhouse gases are different, but also the land-sea distribution, orography, vegetation and land ice cover (Lunt et al., 2010a).The global mean air temperature warming from the pre-industrial ref- Using the radiative forcing for a CO 2 doubling (∆RF CO2 = 3.8 W/m 2 ), the values of ESS determined here correspond to a 5.68 − 8.97 W/m 2 per CO 2 doubling.The lower of these values (considering the 4x PIC climate) lies close to the 6 • C per CO 2 doubling ESS estimate from geological data (Royer et al., 2012).

600
When comparing global mean temperature increase between the pre-industrial reference and both Eocene simulations, clearly not all of the warming is a result higher greenhouse gas concentrations.
To separate the effect of palaeogeography (including albedo and vegetation changes, i.e. also land-ice distribution) from the greenhouse gas driven warming we assume that the 38Ma Eocene equilibrium 605 climate sensitivity S EO can be calculated using the combined radiative forcing due to greenhouse gas changes and palaeogeography changes: where ∆T is the temperature difference between two climate states, ∆R the radiative forcing change from greenhouse gases and G the radiative forcing due to the (integral) effect of palaeogeography.

610
Note that the latter also includes changes in the distribution of e.g.land ice and vegetation.In order to be compatible with estimates of ECS and ESS given above, this formulation gives S in normalised Royer et al. (2012).
Since both our 38Ma Eocene simulations use the same (except CO 2 and CH 4 ) boundary conditions,

615
it is reasonable to assume that S and G estimated from the comparison of each Eocene simulation with the pre-industrial case should be similar for both simulations.By comparing both Eocene runs to the pre-industrial reference, we can then estimate G from:  4.Not surprisingly, the estimate of G decreases when using only ocean temperatures as the 640 direct effect of cooler temperatures over land ice is removed.The results for S EO show less variation between different methods and are tied to the respective temperature changes between the 2x PIC and 4x PIC cases.
While considering global MAT changes will likely overestimate the effect of global geography changes, the opposite is true for equatorial SST's.Still we find a possible range of G = 5.61 − 645 10.23 W/m 2 , equivalent to ∼1.5-2.7 CO 2 doublings with a most likely value around 2 doublings or a ∼6 • C warming.This is considerably higher than the estimated 1.8 • C warming related to geography (using HadCM) found by Lunt et al. (2012) and more than the ∼2-4W/m 2 (i.e.0.6-1.1 CO 2 doublings) suggested by Royer et al. (2012).A considerably higher warming of ∼5 • C from integral geography effects in the Eocene has been found by Caballero and Huber (2013) that matches up with 650 our estimates.Finally, higher estimates of 2-6 • C can also be deduced using the different models used in Lunt et al. (2012).
Clim.Past Discuss., https://doi.org/10.When also taking the pre-industrial reference simulation into consideration, we find a fixed forcing G = 8.65W/m 2 from (integral) geography changes.Previous studies have noted this effect in terms of an offset in global mean temperature between pre-industrial and palaeo simulations (Ca-700 ballero and Huber, 2013;Lunt et al., 2012).Similar to what has been found in Caballero and Huber (2013) the direct effect of ice sheet distribution is limited, leaving a considerable warming due to other geography related changes.When using oceanic instead of atmospheric temperatures, the influence of topography and land surface changes is reduced mostly by the direct effect of the ice sheets and vegetation changes.Although smaller, the estimate for G is still larger than suggested in most 705 previous studies.This indicates a major contribution to G from changes in continental geometry and the related circulation patterns, that these simulations are better able to resolve.

Table 2 .Figure 1 .
Figure 1.Time series of a) upper 1000m and b) below 2000m volume-weighted average temperature for Pre-Industrial (dotted), 38Ma 2x (dashed) and 4x (solid) PIC spin-up simulations.Averages are shown globally (black) and for both the Pacific (red) and Atlantic (blue) ocean basins separately.Full depth global average temperature is also shown for each run with thick purple lines

3
Results: middle-to-late Eocene equilibrium climate 3.1 Ocean For each simulation, averages are made using over last 50 model years to represent the equilibrium 155 climate.The resulting annual mean ocean state for the 4x PIC 38Ma climatology is visualised by the set of fields shown in Figure 3. First of all, sea surface temperatures are quite warm at low latitudes (23 • S-23 • N average of 33.9 • C, locally >36 • C) but also mild in the high-latitude regions 6 Clim.Past Discuss., https://doi.org/10.5194/cp-2018-43Manuscript under review for journal Clim.Past Discussion started: 12 April 2018 c Author(s) 2018.CC BY 4.0 License.

Figure 2 .
Figure 2. Time series of a) upper 1000m and b) below 2000m volume-weighted average ideal age tracers for Pre-Industrial (dotted), 38Ma 2x (dashed) and 4x (solid) PIC spin-up simulations.Averages are shown globally (black) and for both the Pacific (red) and Atlantic (blue) ocean basins separately.c) 38Ma and d) pre-industrial horizontal distributions of ideal age tracers at 2500m depth for the end of each model simulation.Color shading in c) shows the ideal age of 4x PIC, while contours do so for the 2x PIC case (black line every 100 years up to 500 and every 250 years above, coloured contours use the color bar shown to the right).

Figure 3 .
Figure 3. Annual mean a) sea surface temperature and b) salinity, c) barotropic stream function (positive for clockwise flow) and zonal wind stress (contours every 2.5•10 −2 Pa, thick lines every 1•10 −2 Pa; solid positive and dashed negative, thick white line at 0Pa), and d) upper 200m mean potential density and mixed layer depth (contours every 50m, thick lines every 200m, thick white line at 500m) for the 4x PIC simulation.

Figure 4 .
Figure 4. Global oceanic meridional overturning stream function, averaged for the last 50 model years of the a) 4x (solid) and 2x (dashed) PIC Eocene simulation, and b) pre-industrial reference.Red contours show positive and blue negative values, drawn every 5Sv (thin lines show 1 and 2Sv in b) with the absolute maximum of the strongest cell (with sign) below 1000m specified.Top of model required total meridional heat flux is shown in c) for 4x PIC (solid black), 2x PIC (dashed dark grey) and pre-industrial (dotted grey) with the corresponding atmospheric fluxes in red (4x PIC) and dotted orange (pre-industrial).d) Total integrated meridional heat fluxes in the ocean, globally (blue) and Pacific-only (red).Note that all horizontal (latitude) scales cover the range [75 • S, 75 • N] with latitude increasing from left to right, for better comparison with oceanic fields.Despite differences in spin-up time, patterns of the equilibrium ocean circulation state are generally similar for the 2x and 4x PIC simulations.The annual mean, global mean sea surface temperature (SST) is 28.4 • C in the 4x PIC case versus 25.8 • C in the 2x PIC one (the pre-industrial model value is 18.4 • C).Temperature differences between both 38Ma cases of 2.5-3 • C are also seen in the upper and deep ocean in Figure 1, and are close to the global mean SST change of 2.6 • C. Without signifi-200

225
the presence of a deep meridional overturning cell (with negative sign) in the South Pacific, pulling warm waters into the southern high latitudes.In addition, land masses obstructing zonal flow at middle-high latitudes (more strongly than at present day) allow for stronger meridional contributions in gyres.Top of model net shortwave and longwave fluxes are integrated to obtain the required compensating meridional heat flux (Figure4c), which does not differ much between both Eocene 230 and the pre-industrial cases.

Figure 5 .Figure 6 .
Figure 5.Time series of a) Southern Ocean Gateway volume transports and b) maximum global meridional overturning for both 2x (dashed), 4x (solid) PIC Eocene and pre-industrial (dotted) spin-up simulations.Transports through three north-south transects in the Southern Ocean are considered: Drake Passage (black, 65 • W), Tasmanian Gateway (red, 150 • E) and Agulhas (blue, 25 • E), with positive values indicating eastward flow.

Figure 7 .
Figure 7. 4x PIC simulation a) June-July-August and b) December-January-February averaged sea surface temperature and zonal wind stress (contours every 2•10 −2 N/m 2 and thick white lines every 5•10 −2 N/m 2 ).c) and d) similar to a) and b), but for the barotropic stream function and maximum mixed layer depth (contours every 100m and thick white lines every 500m).

Figure 8 .
Figure 8. Annual mean (shading) and summer (contours every 2 • C) sea surface temperatures for the a) 38Ma 4x PIC simulation presented here and b) results from Goldner et al. (2014) (GH14) using a 45Ma gegraphy and 4x pre-industrial CO2.A similar comparison is made for c-d) sea surface salinity (psu) and mixed layer depth (contours) as well as e-f) barotropic stream function (shading) and zonal wind stress (contours).All color scales and contour intervals use the same conventions as those in Figures 3 and 7.
Figure 9. Annual mean for the 4x PIC simulation with a) 2m above ground level air temperature (coloured) and average min/max temperature (contours; magenta Tmin < 0 • C and white Tmax > 40 • C), b) precipitation (coloured) and mean sea level pressure (black contours every 10hPa, >1000hPa in white, thick line at 1000hPa), c) 700hPa temperature (coloured) and 200hPa wind speed (white contours every 5m/s starting at 10m/s, thick black lines every 20m/s), and d) 500hPa geopotential height (coloured) and dynamic tropopause height (contours every 1km starting at 12km, thick black line at 16km).Similar to oceanic fields, a climatology for the atmosphere is made from the last 50 years of the 4x PIC simulation of which the annual mean is shown in Figure9.Again, results for the 2x PIC case 375

415
Tropical temperatures are closely tied to prevailing water masses and show little variability, both seasonally and zonally.Summer temperatures in the sub-tropics are on average comparable to those in the tropics, but with higher zonal variability.Especially around 40 • N and 30 • S, land masses are more abundant in the Eocene allowing for summer temperatures of up to 45 • C at 4x PIC.While

Figure 10 .
Figure 10.Annual mean, zonal mean 2m air temperature (black line) for the 38Ma 4x PIC (solid), 2x PIC (dashed) and Pre-industrial (dotted) equilibrium climate.Zonal mean temperatures of the 4x PIC climate, averaged for June-July-August and December-January-February are shown in red and blue, respectively.For both of the latter, the longitudinal range of temperatures is indicated by shaded areas.The same is shown for the pre-industrial seasonality, but only considering the zonal mean.

Figure 11 .
Figure 11.As in Figure 10 for land-only temperatures, proxy estimates (squares) and corresponding model mean annual temperature (circles).Proxy temperatures are colour coded for their region and error bars (black dots) are indicative for the spread at each site.Similar to model SST, error estimates (black lines) show the spacial variation within a 5 • × 4 • box surrounding the corresponding location in the model.The inset shows a scatter plot comparing proxy and model air temperatures, where the latter are corrected for differences in model and reconstructed topography (uncorrected: small white squares).
Using the 0.1 • reconstruction fromBaatsen et al. (2016), model temperatures are corrected for the 450 20 Clim.Past Discuss., https://doi.org/10.5194/cp-2018-43Manuscript under review for journal Clim.Past Discussion started: 12 April 2018 c Author(s) 2018.CC BY 4.0 License.difference with model topography assuming a free atmospheric lapse rate of -6.5K/km.Especially North American sites show a large improvement when correcting for model topography, as shown in the inset of Figure 11 (comparing small squares to coloured circles).The discrepancy remains for the lower latitude part of the Chinese sites, where mean annual air temperatures are probably underestimated by proxies as well as overestimated by the model.Comparing 2x PIC model results to late Eocene proxies shows a similar result (Appendix: Table

465
atures on Antarctica and cloud-albedo feedbacks in the Arctic, annual mean (zonal mean) temperatures are globally above 5 • C in our 38Ma 4x PIC simulation.Seasonality in the atmosphere is studied in more detail by comparing the months June-August to 470 December-February for the 4x PIC case as shown in Figure 12.The coldest regions in winter are north-east Siberia and central Antarctica, where average temperatures drop below -10 • C. Nearby ocean temperatures keep Antarctic coastal regions mild and at or above freezing in winter.The continental sub-tropics show the highest summer temperatures on Earth, with maxima on average exceeding 50 • C in many locations.Such temperatures seem unrealistically high and do not agree 475 with the vegetation imposed in the model.This issue is probably related to fixed vegetation types, leading to trees losing their foliage in summer and creating a rough surface with very efficient drying and subsequent heating.Despite reduced equator-pole differences compared to the pre-industrial reference, middle latitude regions still exhibit a sharp temperature gradient, especially in winter.This leads to strong zonal flow and orographic lift seen in pressure and precipitation patterns (Figure 12c-480 d).The Indo-Pacific ITCZ is always double, but most intense in summer and linked to monsoonal troughs over land.Heavy rains from summer monsoons are most prominent across South-East Asia, (Northern and Southern) Africa and South America.In winter, plumes of precipitation extend poleward and eastward from the western low latitude ocean basins.These are typical storm tracks caused by baroclinic instability in the middle latitudes, 485 which are associated with sharp mid-tropospheric temperature gradients and high upper level winds (through the thermal wind balance, Figure 12e-f).Persistent jet streaks around 90 • E, linked to sta-

Figure 12
Figure 12. a) June-July-August and b) December-January-February mean 2m air temperature with contours indicating minimum and maximum values (magenta: Tmin < 0 • C, cyan: Tmin < −10 • C, blue: Tmin < −20 • C, white: Tmax > 40 • C and yellow: Tmax > 50 • C), c) and d) similar for total precipitation and mean sea level pressure, and e) and f) for 700hPa temperature and 200hPa wind speed.Color scales and contour values are all similar to those used in Figure 9.

Figure 13
Figure 13.a) Zonal average zonal wind (shading) and temperature (contours every 10 • C) for the months June-July-August, b) similar to a) for December-January-February. Tropopause height (zonal mean) is indicated by the dashed black line and pressure is used as the vertical coordinate.
515present with two maxima associated with the double ITCZ and strongest winds at the winter hemisphere (corresponding to the stronger Hadley Cell).A remarkable component of the atmospheric circulation develops in summer over Antarctica.As suggested by temperatures shown in the Figures12e-fand 13, the meridional temperature gradient Clim.Past Discuss., https://doi.org/10.5194/cp-2018-43Manuscript under review for journal Clim.Past Discussion started: 12 April 2018 c Author(s) 2018.CC BY 4.0 License.
By comparing the 4x to the 2x PIC equilibrium climate states, a measure for the equilibrium climate sensitivity (ECS) can be obtained.To exclude possible effects of decadal-scale variability (see Fig- , 200 years instead of 50 are considered in the climatologies used here.The radiative forcing associated with doublings of CO 2 and CH 4 is estimated from the study of Etminan et al. (2016).Doubling of only CO 2 from the pre-industrial value (280 ppm) corresponds to a radiative forcing of ∆RF CO2 = 3.8 Wm −2 .The first doubling of both CO 2 and CH 4 starting from pre-industrial values results in a combined radiative forcing of ∆RF 2× = 4.23 Wm −2 and the second doubling gives 535 ∆RF 4× 2× = 4.48 Wm −2 .In Figures 14a-b the equilibrium temperature difference between the Eocene 4x PIC and 2x PIC simulations is shown, normalised by the associated radiative forcing (∆RF 4× 2×

Figure 14
Figure 14.a) Annual mean temperature response, normalised per W/m 2 of the 4x PIC compared to the 2x PIC equilibrium climate with contours showing the winter season only (using the same color scale).b) Zonal mean normalised temperature response; annual (black), December-January-February (blue) and June-July-August (red).The grey shading indicates minimum and maximum values for each latitude for the annual mean values.The (area weighted) global average (i.e.climate sensitivity) is 0.69 • C/Wm −2 and is indicated by the black dashed line.c) Clear-sky component of the net longwave flux change at the top of model (TOM) for 4x PIC vs. 2x PIC, similar for the shortwave flux in contours (blue: 1W/m 2 , cyan: 2W/m 2 and white: 5W/m 2 ).d) Zonal mean TOM flux change for longwave (black) and shortwave (red) fluxes, corresponding clear-sky components are shown using dashed lines.Note that longwave fluxes are define positive upward while shortwave fluxes are positive downward.e)Change in the occurrence of ZM parameterised deep moist convection and difference in precipitation (contours at 100, 200, 500 and 1000 mm, dashed for negative changes and thick line indicating 500mm).f) Zonal mean changes in cloud cover (in %) for low (red), medium (green) and high (blue) clouds, the thick black line indicates changes in total coverage.
erence to the Eocene 2x PIC simulation together with the radiative forcing due to the first CO 2 590 and CH 4 doubling (∆RF 2× = 4.23 Wm −2 ) results in ESS 2× = 2.36 • C/Wm −2 .Comparing the Eocene 4x PIC simulation to the pre-industrial reference, on the other hand, gives a lower ESS 4× = 1.49• C/Wm −2 .Because the radiative forcing of both CO 2 and CH 4 depends nonlinearly on the greenhouse gas concentrations, the radiative forcing due to a quadrupling is not equal to the sum Clim.Past Discuss., https://doi.org/10.5194/cp-2018-43Manuscript under review for journal Clim.Past Discussion started: 12 April 2018 c Author(s) 2018.CC BY 4.0 License.
Clim.Past Discuss., https://doi.org/10.5194/cp-2018-43Manuscript under review for journal Clim.Past Discussion started: 12 April 2018 c Author(s) 2018.CC BY 4.0 License.where ∆T 4× and ∆T 2× denote the temperature difference with respect to the pre-industrial climate 620 for the 4x PIC and 2x PIC case, respectively.This leads to an estimate of G = 10.32W/m 2 , and using equation 1 with either (∆T 4× , RF 4× ) or (∆T 2× , RF 2× ) leads to S = 0.69 • C/Wm −2 .As expected, the value of S EO is the same as the one found previously when comparing both Eocene simulations to each other.625 Alternatively, G can be estimated by comparing a pre-industrial climate under 2x PIC forcing, with the results from the modelled 2x PIC Eocene case.Using the pre-industrial ECS determined for the model (S P I = 0.82 • C/Wm −2 ) and radiative forcing of the first doubling of greenhouse gases (∆RF 2× , a temperature change of 3.48 • C would be expected.As the change in global mean temperature between the pre-industrial reference and 2x PIC Eocene climate is 9.97 • C, this leaves an 630 additional warming of 6.47 • C owing to integrated global geography changes.In order to convert this warming into an estimate of G, we can use either the Eocene S EO or pre-industrial S P I giving a possible range of G = 7.86 − 9.30 W/m 2 .Finally, we can only consider equatorial (< 23 • N/S) SST's (SST eq ; with a 3/2 ratio between global 635 and equatorial temperature change to account for polar amplification, as discussed in Royer et al. (2012)) or deep sea temperatures (T deep ) to be more compatible with ECS estimates from proxy data.An overview of used temperature differences: SST eq and T deep in addition MAT glob to (global mean, annual mean 2m temperature, used above) and the resulting values for S EO and G is given in Table

Figure A2 .
Figure A2.Annual mean, zonal mean (land-only) 2m air temperature for the 38Ma 2x PIC equilibrium climate (black line).Zonal mean temperatures, averaged for June-July-August and December-January-February are shown in red and blue, respectively, with their longitudinal range indicated by shaded areas.Markers indicate late Eocene proxy estimates (squares) and corresponding model mean annual temperature (circles).Proxy temperatures are colour coded for their region and error bars (black dots) are indicative for the spread at each site.Similar to model SST, error estimates (black lines) show the spacial variation within a 5 • × 4 • box surrounding the corresponding location in the model.The inset shows a scatter plot comparing proxy and model air temperatures, where the latter are corrected for differences in model and reconstructed topography (uncorrected: small white squares).
• S, which is 10 • poleward of where it is today.The region of maximum zonal wind stress coincides with this front, showing that it is truly fixed by the latitudes where winds are the least obstructed by continents.Due to the more southerly position of Australia in the Eocene, the temperature front separating warm Pacific from colder Antarctic waters also moves 185further southward.Interestingly, this allows low-latitude derived waters to extend much further south helping to explain high temperatures in many Southern Ocean regions, especially the Southwest Pacific in summer ; Cramer Clim.Past Discuss., https://doi.org/10.5194/cp-2018-43Manuscript under review for journal Clim.Past Discussion started: 12 April 2018 c Author(s) 2018.CC BY 4.0 License.et al., 2009).In our Eocene simulations, the globally averaged deep-sea temperature (below 2000m) reaches ∼11.5 • C in the 4x PIC case (and ∼8.5 • C in the 2x PIC case), which is much warmer than for the pre-industrial ocean.In therms of an oxygen isotope (δ 18 O) signal, this corresponds to a

Table 3 .
Average equilibrium temperatures at the end of each simulation, showing MATglob: global mean air temperature (at 2m reference height), SSTglob: global mean sea surface temperature, SSTeq: equatorial (< 23 • N/S) average SST, and Tdeep: global mean deep ocean ocean temperature (below 2000m).These values are calculated using the last 200 years of each simulation.
of two doublings(Etminan et al., 2016).As we only compare model equilibrium states that do not 595 result from transient quadrupling of greenhouse gases, a linear combination is assumed here such that:

Table 4 .
(Goldner et al., 2014)ript under review for journal Clim.Past Discussion started: 12 April 2018 c Author(s) 2018.CC BY 4.0 License.∆T2× ( • C) SEO ( • C/Wm −2 ) G (W/m 2 ) Temperature differences comparing the 4x PIC (∆T 4× ) and 2x PIC (∆T 2× ) Eocene climate to the pre-Using version 1.0.5 of the Community Earth System Model (CESM), we presented results of simulated 38Ma Eocene climates at both high (4x PIC) and low (2x PIC) concentrations of CO 2 and CH 4 , using the most recent palaeogeographic constraints.These are among the first simulations with a fully-coupled and detailed climate model to study the middle-late Eocene climate, using a new 38Ma geography reconstruction at high resolution.∼14•C)and a low equator to pole temperature gradient.The global heat budget is approximately meridionally symmetric, which is reflected by the zonal mean temperature pattern.Deep water formation occurs in the South Pacific Ocean, while the North Atlantic is stably stratified 665 and stagnant due to the outflow of brackish Arctic waters.A shallow and rather weak precursor of an Antarctic Circumpolar Current is present, mainly driven by seasonally dependent wind stresses.Continental low-middle latitude regions are characterised by high seasonality on both hemispheres and strong summer monsoons.Middle and high latitudes mostly have mild winters, warm summers and pronounced storm tracks.The Arctic is rather cool due to its geographic isolation and the Antarc-Past Discuss., https://doi.org/10.5194/cp-2018-43Manuscriptunderreviewfor journal Clim.Past Discussion started: 12 April 2018 c Author(s) 2018.CC BY 4.0 License.Previous Eocene simulations (at 4x PI CO 2 ) with the same model but a different (45Ma) continental geography and lower resolution have resulted in overall similar sea surface temperature distributions(Goldner et al., 2014).However, the 38Ma 4x PIC case presented here is about 4-5 • C warmer globally in both SST's and land temperatures.Higher resolution and a more specific geography re-Comparing the 38Ma 2x and 4x PIC cases, an equilibrium climate sensitivity S EO = 0.69 • C/Wm −2 was found, which is slightly lower than the same model's present-day value.The difference in equilibrium climate sensitivity between the Eocene and the present day reflects a state-dependence of the fast feedback processes allowing the Eocene climate to respond differently to a greenhouse gas • C) 680 Clim.690 695 doubling than the present day climate.

Table A1 .
• C) summer temperatures occur in the sub-tropics and are probably related to fixed vegetation types.Strong 710 seasonality is also seen on Antarctica, where summer temperatures reach up to 35 • C in the 4x PIC case.The absence of an ice sheet, warm waters surrounding the continent and summertime insolation cause te Antarctic continent to become a heat island.Sea ice coverage is very limited and only occurs sporadically during the coldest months for both 38Ma Eocene cases.Even without sea ice or extensive snow cover, significant polar amplification is seen for a doubling in CO 2 and CH 4 . .Past Discuss., https://doi.org/10.5194/cp-2018-43Manuscript under review for journal Clim.Past Discussion started: 12 April 2018 c Author(s) 2018.CC BY 4.0 License.Clim.Past Discuss., https://doi.org/10.5194/cp-2018-43Manuscript under review for journal Clim.Past Discussion started: 12 April 2018 c Author(s) 2018.CC BY 4.0 License.Clim.Past Discuss., https://doi.org/10.5194/cp-2018-43Manuscript under review for journal Clim.Past Discussion started: 12 April 2018 c Author(s) 2018.CC BY 4.0 License.Overview of 42-38 Ma SST proxies; site location, 38Ma reconstructed coordinates, SST estimate, calibration error, reference and method used.
715With limited surface albedo feedbacks, changes in cloud coverage and radiative forcing are the main drivers behind the further reduction of the meridional temperature gradient.Clim.Past Discuss., https://doi.org/10.5194/cp-2018-43Manuscript under review for journal Clim.Past Discussion started: 12 April 2018 c Author(s) 2018.CC BY 4.0 License.As the simulated middle-late Eocene (4x PIC) climate is in reasonable agreement with estimates from proxy records, the simulation results may be used to try to interpret proxy records in more 720 detail.Clim41 Clim.Past Discuss., https://doi.org/10.5194/cp-2018-43Manuscript under review for journal Clim.Past Discussion started: 12 April 2018 c Author(s) 2018.CC BY 4.0 License.