Model simulations of past climates are increasingly found to compare well with proxy data at a global scale, but regional discrepancies remain. A persistent issue in modeling past greenhouse climates has been the temperature difference between equatorial and (sub-)polar regions, which is typically much larger in simulations than proxy data suggest. Particularly in the Eocene, multiple temperature proxies suggest extreme warmth in the southwest Pacific Ocean, where model simulations consistently suggest temperate conditions. Here we present new global ocean model simulations at 0.1° horizontal resolution for the middle-late Eocene. The eddies in the high-resolution model affect poleward heat transport and local time-mean flow in critical regions compared to the non-eddying flow in the standard low-resolution simulations. As a result, the high-resolution simulations produce higher surface temperatures near Antarctica and lower surface temperatures near the equator compared to the low-resolution simulations, leading to better correspondence with proxy reconstructions. Crucially, the high-resolution simulations are also much more consistent with biogeographic patterns in endemic-Antarctic and low-latitude-derived plankton, and thus resolve the long-standing discrepancy of warm subpolar ocean temperatures and isolating polar gyre circulation. The results imply that strongly eddying model simulations are required to reconcile discrepancies between regional proxy data and models, and demonstrate the importance of accurate regional paleobathymetry for proxy-model comparisons.

Our current understanding of global mean near-surface (land and sea) air temperature (GMSAT) during the Cenozoic era relies on paleo-proxy estimates of deep-sea temperature combined with assumed relationships between global mean deep-sea temperature (GMDST), global mean sea-surface temperature (GMSST), and GMSAT. The validity of these assumptions is essential in our understanding of past climate states such as the Early Eocene Climate Optimum hothouse climate (EECO, 56–48 Ma). The EECO remains relevant today, because EECO-like CO2 levels are possible in the 22nd century under continued high CO2 emissions. We analyze the relationship between the three global temperature indicators for the EECO using 25 different millennia-long model simulations with varying CO2 levels from the Deep-Time Model Intercomparison Project (DeepMIP). The model simulations show limited spatial variability in deep-sea temperature, indicating that local temperature estimates can be regarded representative of GMDST. Linear regression analysis indicates that compared to GMSST, both GMDST and GMSAT respond more strongly to changes in atmospheric CO2 by factors of 1.18 and 1.17, respectively. Consequently, this model-based analysis validates the assumption that changes in GMDST can be used to estimate changes in GMSAT during the EECO. Paleo-proxies of GMDST, GMSST, and GMSAT during EECO show the best fit with model simulations having an atmospheric CO2 level of 1,680 ppm, which matches paleo-proxies of atmospheric CO2 during EECO. Similar analyses of other past climate states are needed to examine whether these results are robust throughout the Cenozoic, providing insight into the long-term future warming under various shared socioeconomic pathways.