Different Pathways to an Early Eocene Climate

12 The early Eocene was characterised by much higher temperatures and a smaller equator-to13 pole surface temperature gradient than today. Comprehensive climate models have been rea14 sonably successful in simulating many features of that climate in the annual average. How15 ever, good simulations of the seasonal variations, and in particular the much reduced Arctic 16 land temperature seasonality and associated much warmer winters, have proven more difficult. 17 Further, aside from an increased level of greenhouse gases, it remains unclear what the key 18 processes are that give rise to an Eocene climate, and whether there is a unique combination 19 of factors that leads to agreement with available proxies. Here we use a very flexible General 20 Circulation Model to examine the sensitivity of the modelled climate to differences in CO2 con21 centration, land surface properties, ocean heat transport, and cloud extent and thickness. Even 22 in the absence of ice or changes in cloudiness, increasing the CO2 concentration leads to a polar23 amplified surface temperature change because of increased water vapour and the lack of con24 vection at high latitudes. Additional low clouds over Arctic land generally decreases summer 25 temperatures and, except at very high CO2 levels, increases winter temperatures, thus helping 26 achieve an Eocene climate. An increase in the land surface heat capacity, plausible given large 27 changes in vegetation and landscape, also decreases the Arctic land seasonality. In general, var28 ious different combinations of factors – high CO2 levels, changes in low-level clouds, and an 29 increase in land surface heat capacity – can lead to a simulation consistent with current proxy 30 data. 31 Plain Language Summary 32 During the early Eocene, some 50 million years ago, the Earth was approximately 13 33 degrees warmer and the equator-to-pole surface temperature difference was much smaller than 34 that of today. We now have proxy data on the surface temperature at different latitudes and 35 the seasonality of the surface temperature (for land at high-latitudes), the amount of carbon 36 dioxide in the air, the nature of the vegetation, and the land configuration. However, much of 37 this data is quite uncertain. Modern climate models have been used to estimate what the Eocene 38 climate was like, but they are complicated to use, hard to understand, and in some ways are 39 tuned to the present climate. Here we use a simpler, more flexible climate model to simulate 40 the Eocene climate and examine how differences in the CO2 concentration, land surface prop41 erties, ocean heat transport, and cloud extent and thickness affect the simulated climate. We 42 find that different combinations of CO2 concentration, surface albedo, cloudiness and surface 43 heat capacity of land can lead to simulations that are within estimated values from the data, 44 suggesting there are multiple pathways to simulating a climate consistent with what is currently 45 known about the Eocene. 46

of Eocene climates suggested that temperatures at low latitudes increased far less than tem-78 peratures at high latitudes, so much so that climate models struggled to represent the appar-79 ent much reduced equator-to-pole temperature gradient (Huber et al., 2003, for example). How-80 ever, more recent estimates of tropical temperatures seem to indicate low-latitude temperatures 81 were higher than was previously estimated (Pearson et al., 2007), albeit with large error bars, 82 and recent climate models show a better proxy-model match in surface temperature gradient 83 (D. Lunt et al., 2020). Thus, at least on the annual average, it seems there may in fact no longer 84 be a large discrepancy between climate models and Eocene proxies. The generally-accepted 85 reason for the high overall temperature in the Eocene is high CO 2 levels, and climate mod-86 els give fair agreement with proxies (Huber & Caballero, 2011), albeit often with higher lev-87 els of CO 2 than are now thought to have existed (Anagnostou et al., 2020). The required level 88 of CO 2 needed for such high temperatures could be reduced if there were an increase in ab-89 sorbed solar radiation (i.e., a reduced planetary albedo). This might be achieved, for exam-90 ple, through a decrease in aerosol production leading to a decrease in cloud condensation nu-91 clei and a reduction in cloud cover (Kiehl & Shields, 2013;Carlson & Caballero, 2017). The 92 warming from CO 2 could also potentially lead to a reduction in cloud cover which reduces the 93 planetary albedo (Zhu et al., 2019). 94 Although the annual average Eocene temperature can arguably be reproduced by climate 95 models, much more difficulty arises when trying to understand the seasonality of Arctic tem- Our goal in this paper is to clarify the conditions required to reproduce an Eocene cli-111 mate, with particular attention to the seasonal cycle and the maintenance of relatively warm 112 winters over Arctic land. To this end we use a very flexible GCM, configured with Eocene land 113 and topography, that enables us to independently vary CO 2 levels, cloud distributions, ocean 114 -3-manuscript submitted to Paleoceanography and Paleoclimatology heat transport, and various land-surface parameters. We thereby seek to understand how these 115 processes, separately and together, affect the global-mean temperature, the equator-to-pole sur-116 face temperature gradient, and the seasonality in Arctic land temperature. We begin with a de-117 scription of the model itself (Section 2), and follow this with a description of experiments in 118 which we change the surface boundary conditions (Section 3), the clouds (Section 4), the land 119 surface heat capacity (Section 5), and ocean heat transport (Section 6). 120 2 Model and Reference Simulations 121 We construct our models using the Isca climate modeling framework (Vallis et al., 2018) 122 configured with no sea ice, a slab mixed-layer ocean boundary condition, and a simple rep-123 resentation of land and topography following Eocene-like continental outlines taken from com-124 prehensive climate model simulations of the Eocene (D. J. Lunt et al., 2021). Meridional ocean 125 heat transport is represented by imposing a q-flux, as described further in Section 6, although 126 in many simulations this is set to zero. The cloud scheme diagnoses large scale clouds from 127 the relative humidity, with adjustments for marine low stratus clouds and polar clouds (Liu et 128 al., 2020). The effective radius of liquid and ice cloud droplets is set to 14 and 25 microns 129 respectively, and the in-cloud liquid water mixing ratio is set to 0.18 g/kg. These parameters  Lunt et al., 2021). Land also differs from oceans by its heat ca-136 pacity, which we set to 0.2 meters equivalent water depth for continents (Merlis et al., 2013) 137 and 20 meters for oceans, by the roughness constant, which is set to be 10 times higher over 138 land than ocean, and by the land evaporative resistance which is set to 0.5 (parameter in equa-139 tion 10 of Vallis et al. (2018)). We use the Eocene's land distribution (the contour is visible 140 in fig. 1), and notice that most modern day continents are recognizable, though the continen-141 tal configuration may have an impact on ocean circulation. Simulations are run at spectral T42 142 resolution, which corresponds to approximately 2.8 degrees resolution at the equator. Convec-143 tion is calculated using a simplified Betts-Miller convection scheme (Frierson, 2007). Large 144 scale condensation is parameterized such that relative humidity does not exceed one and con-145 densed water immediately returns to the surface, and the cloud distribution is not directly cou-146 pled to the precipitation. 147 We first describe five reference simulations with a fixed set of control parameters in which 148 CO 2 concentrations are set to 300 ppm, 900 ppm (3 × 300 ppm), 1800 ppm (6×300 ppm), 2700 ppm 149 (9×300 ppm), and 3600 ppm (12 × 300 ppm). Following that we discuss a set of experiments 150 where the surface albedo and land evaporative resistance are modified, a set where we prescribe 151 various high-latitude cloud distributions, and a set where we reduce the land's surface heat ca-152 pacity. Finally, we test the importance of ocean heat transport by prescribing a meridional heat  Table 1.
155 Figure 1 shows the annual-mean and winter (December, January, and February mean (DJF)) 156 surface temperature for the 300 ppm and 3600 ppm simulations. At 300 ppm, the winter tem-157 peratures reach below −30°C in parts of the Arctic land whereas at 3600 ppm, the winter tem-158 peratures are above zero almost over the whole Arctic land surface. At 2700 ppm the temper-159 atures fall below zero for periods in winter, as seen in fig. 2, although given the uncertainties 160 in the proxies it is difficult to be definitive as to whether this falls outside of bounds of the ob-    As noted in the introduction, atmospheric models produce polar amplification -mean-175 ing an enhanced warming at and near the the surface at high latitudes -when CO 2 is increased, 176 even without changes in ice cover. To understand this, suppose first that the vertically aver- is absent at high latitudes, leading to an effective low-level polar amplification. In addition, 181 the overall increase in water vapor due to a higher temperature and increase in latent heat trans-182 port leads to bottom-heavy atmospheric temperature change at high latitudes (Henry et al., 2021).

183
In addition to polar amplification, the increased temperatures that results from the ad-184 ditional CO 2 forcing alone reduces the seasonality of Arctic land temperature due to the small 185 heat capacity of land (Henry & Vallis, 2021b). This effect arises from the nonlinearity of the 186 temperature dependence of surface longwave emission , which is proportional to 4 , where 187 is the surface temperature. Surfaces at low temperature need to warm more than those at 188 high temperature in order to achieve the same increase in emission, leading to a larger increase 189 in surface temperature in winter than in summer. The seasonality is naturally larger over land  were hypothesized to be important in maintaining warm Arctic winters (Sloan & Pollard, 1998).

253
In order to test these various hypotheses as to how clouds affect Arctic warming, we prescribe  . 5c).

297
The additional radiative effect of high clouds is, at least in these simulations, relatively weak     The characteristics of the land surface were likely quite different in the Eocene, espe-319 cially at high latitudes where frozen soil and ice is replaced by abundant vegetation and pos-320 sibly swamps and lakes. We therefore explore the sensitivity of our results to an increase in 321 land surface heat capacity. Specifically, we set the mixed layer depth over land to 2m instead 322 of 0.2m and to see how this affects the seasonal cycle at high CO 2 levels. The increase in the 323 'mixed layer depth' of land to 2m does not substantially change the zonal-mean annual-mean 324 surface temperature ( fig. 7a and b compared to fig. 2a and b). However, the seasonal cycle of 325 Arctic land temperature is almost consistent with proxies (dark grey box) at 2700 ppm and fully 326 consistent with proxies at 3600 ppm ( fig. 7c).

327
Since the increased prescribed low clouds over land seemed a promising way to get a 328 climate consistent with proxies ( fig. 4), we also test a higher land surface heat capacity with 329 increased prescribed low clouds over land. This does not substantially change the winter Arc-  Passage could reduce the temperature difference between high and low latitudes by between 337 5°C and 9°C. However, other studies that use dynamical, three-dimensional atmospheric mod-338 els have tended to find that changes in ocean heat transport are largely compensated by changes 339 in atmospheric energy transport and the surface temperature is then largely unaltered, even over 340 the ocean (Farneti & Vallis, 2013;Rencurrel & Rose, 2020). 341 We explore the importance of ocean heat transport by imposing a meridional heat flux 342 (a 'q-flux') to the slab ocean that mimics equator-to-pole energy transport by the ocean, as in  . 8b). This is not to say that the ocean 351 heat flux has no effect; rather, if the atmosphere is responding by changing its meridional en-352 ergy flux then the intensity of its circulation (and hence such things as the mid-latitude storm 353 tracks) will change correspondingly; however, we do not explore that here. the Eocene period, we test how a 33% reduction in surface albedo affects the global climate 377 and find that it has a similar gross effect to increasing the CO 2 levels (as also noted by Carlson 378 and Caballero (2017) for example). In our simulations, the simulation with a 33% reduction 379 in surface albedo has roughly the same temperature at 2700 ppm as the reference 3600 ppm 380 simulation.

381
More difficulty arises in simulating the seasonal cycle, and in particular in obtaining win-382 ter temperatures that are more-or-less consistent with the proxies without going to CO 2 lev-383 els higher than observations suggest and which in turn leads to summer temperatures that are 384 too high. Varying the cloud amounts is one way that better agreement can be achieved, and Arctic land increases winter Arctic land temperatures for low CO 2 , but has little effect at high 394 CO 2 since the additional greenhouse effect is then relatively small. However, the increased low 395 cloud reduces summer Arctic land temperatures for all CO 2 levels, bringing Arctic land sea-396 sonality closer to the proxies.

397
The physical mechanisms whereby cloud cover could change in an Eocene climate are 398 less clear. We found that the land evaporative resistance (essentially a measure of the wetness 399 of the surface) had a large impact on low cloud formation over land, with increasing wetness 400 leading to more low cloud. This is a plausibly important effect, given that the high latitude 401 land surface in the Eocene may have been dotted with lakes and rainforest-like vegetation. Nev-402 ertheless, even with this effect, the only way to make the Arctic land above freezing year-round 403 is to increase the land surface heat capacity over its present value by a factor of 10. This, too, 404 is a plausible effect given the difference in land-surface properties in the Eocene compared to 405 those of today. If we additionally prescribe increased low land clouds, the winter Arctic land 406 temperature is not affected (at high CO 2 levels), but the summer Arctic land temperature is 407 reduced (for all CO 2 levels). Finally, we note that, perhaps surprisingly, even large changes 408 in ocean heat transport have very little impact on the zonal-mean surface temperature and Arc-409 tic land temperature seasonality ( fig. 8). This is largely consistent with previous studies (Farneti  There are, evidently, various pathways to get an Eocene climate simulation that is con-412 sistent with proxies:

416
• By reducing the surface albedo by about one third, the temperature is within proxy bounds 417 ( fig. 3) for 2700 ppm instead of 3600 ppm.

418
• Adding low clouds over land reduces summer Arctic land temperatures for all CO 2 lev-419 els and increases winter Arctic land temperatures only al low CO 2 levels. Thus, at the 420 higher levels of CO 2 appropriate for an Eocene climate, low clouds reduce the season-421 ality and help to bring the climate closer to proxies ( fig. 4).

422
• Increasing the surface heat capacity of land has little effect on the meridional gradient 423 in temperature, but reduces the Arctic land seasonality such that at 3600 ppm, the land 424 surface temperature is above freezing year-round ( fig. 7). 425 Given the relatively limited measurements, and the potentially similar effects that some 426 of the changes have (e.g., reduced albedo vs increased CO 2 , increased low clouds and increased 427 surface heat capacity), it is difficult to say what the 'correct' set of parameters is that can re-428 produce an Eocene climate. Undoubtedly, an increased level of CO 2 is needed, likely to val-  Finally, we draw some more general conclusions. The reduced equator-to-pole temper-443 ature gradient and much warmer winters over land of warm past climates can, to a first ap-444 proximation, be explained by robust, known processes (e.g., changes in lapse rate in warmer 445 climates, Planck feedbacks) and those effects can be captured by modern climate models, as 446 both our results and those from the DeepMIP ensemble suggest. The proxies are not exactly 447 matched, but the difference is not wholly unreasonable and do not suggest truly 'unknown physics'. 448 Further, the reduced temperature gradient is likely not the result of a wholesale change in the 449 general circulation of the atmosphere -the mid-troposphere temperature gradient need be lit-450 tle altered, for example. But having said that, care should be taken in using the Eocene to con-451 strain the equilibrium climate sensitivity (to a doubling of CO 2 levels) of today's climate, for 452 even if proxy temperature measurements were exact, effects not present in today's climate come 453 into play. Purely radiative effects imply that the ECS will increase somewhat as temperature