Clim. Past, 17, 1483–1506, 2021 https://doi.org/10.5194/cp-17-1483-2021 © Author(s) 2021. This work is distributed under the Creative Commons Attribution 4.0 License. Deep ocean temperatures through time Paul J. Valdes1,z, Christopher R. Scotese2, and Daniel J. Lunt1 1School of Geographical Sciences, University of Bristol, Bristol BS8 1SS, UK 2Northwestern University, Dept Earth & Planetary Sci, Evanston, IL, USA zInvited contribution by Paul J. Valdes, recipient of the EGU Milutin Milankovic Medal 2015. Correspondence: Paul J. Valdes ([email protected]) Received: 15 June 2020 – Discussion started: 6 July 2020 Revised: 8 May 2021 – Accepted: 10 May 2021 – Published: 19 July 2021 Abstract. Benthic oxygen isotope records are commonly ocean), and the extent of polar amplification (e.g. ice albedo used as a proxy for global mean surface temperatures dur- feedbacks). Deep ocean sediments prior to the Cretaceous are ing the Late Cretaceous and Cenozoic, and the resulting es- rare, so extending the benthic foraminifera proxy further into timates have been extensively used in characterizing ma- deeper time is problematic, but the model results presented jor trends and transitions in the climate system and for here would suggest that the deep ocean temperatures from analysing past climate sensitivity. However, some fundamen- such time periods would probably be an unreliable indicator tal assumptions governing this proxy have rarely been tested. of global mean surface conditions. Two key assumptions are (a) benthic foraminiferal tempera- tures are geographically well mixed and are linked to surface high-latitude temperatures, and (b) surface high-latitude tem- 1 Introduction peratures are well correlated with global mean temperatures. To investigate the robustness of these assumptions through One of the most widely used proxies for estimating global geological time, we performed a series of 109 climate model mean surface temperature through the last 100 million years simulations using a unique set of paleogeographical recon- is benthic δ18O measurements from deep-sea foraminifera structions covering the entire Phanerozoic at the stage level. (Zachos et al., 2001, 2008, Cramer et al., 2009, Friedrich et The simulations have been run for at least 5000 model years al., 2012, Westerhold et al., 2020). Two key underlying as- to ensure that the deep ocean is in dynamic equilibrium. We sumptions are that δ18O from benthic foraminifera represents find that the correlation between deep ocean temperatures deep ocean temperature (with a correction for ice volume and global mean surface temperatures is good for the Ceno- and any vital effects) and that the deep ocean water masses zoic, and thus the proxy data are reliable indicators for this originate from surface water in polar regions. By further as- time period, albeit with a standard error of 2 K. This uncer- suming that polar surface temperatures are well correlated tainty has not normally been assessed and needs to be com- with global mean surface temperatures, then deep ocean iso- bined with other sources of uncertainty when, for instance, 18 topes can be assumed to track global mean surface temper- estimating climate sensitivity based on using δ O measure- atures. More specifically, Hansen et al. (2008) and Hansen ments from benthic foraminifera. The correlation between and Sato (2012) argue that changes in high-latitude sea sur- deep and global mean surface temperature becomes weaker face temperatures (SSTs) are approximately proportional to for pre-Cenozoic time periods (when the paleogeography is global mean surface temperatures because changes are gen- significantly different from the present day). The reasons for erally amplified at high latitudes but that this is offset because the weaker correlation include variability in the source re- temperature change is amplified over land areas. They there- gion of the deep water (varying hemispheres but also varying fore directly equate changes in benthic ocean temperatures latitudes of sinking), the depth of ocean overturning (some with global mean surface temperature. extreme warm climates have relatively shallow and sluggish The resulting estimates of global mean surface air temper- circulations weakening the link between the surface and deep ature have been used to understand past climates (e.g. Zachos Published by Copernicus Publications on behalf of the European Geosciences Union. 1484 P. J. Valdes et al.: Deep ocean temperatures through time et al., 2008; Westerhold et al., 2020). Combined with esti- 2:5◦ in longitude–latitude (roughly corresponding to an av- mates of atmospheric CO2, they have also been used to esti- erage grid box size of ∼ 300 km) in both the atmosphere and mate climate sensitivity (e.g. Hansen et al., 2013) and hence the ocean. The atmosphere has 19 unequally spaced vertical contribute to the important ongoing debate about the likely levels, and the ocean has 20 unequally spaced vertical lev- magnitude of future climate change. els. To avoid singularity at the poles, the ocean model always However, some of the underlying assumptions behind the has to have land at the poles (90◦ N and 90◦ S), but the atmo- method remain largely untested, even though we know that sphere model can represent the poles correctly (i.e. in the pre- there are major changes to paleogeography and consequent industrial geography, the atmosphere considers there is sea changes in ocean circulation and location of deep-water ice covered ocean at the N. Pole but the ocean model has land formation in the deep past (e.g. Lunt et al., 2010; Nunes and hence there is no ocean flow across the pole). Though and Norris, 2006); Donnadieu et al., 2016; Farnsworth et HadCM3 is a relatively low-resolution and low complexity al., 2019a; Ladant et al., 2020). Moreover, the magnitude model compared to the current CMIP5/CMIP6 state-of-the- of polar amplification is likely to vary depending on the ex- art model, its performance in simulating modern climate is tent of polar ice caps and changes in cloud cover (Sagoo et comparable to many CMIP5 models (Valdes et al., 2017). al., 2013; Zhu et al., 2019). These issues are likely to modify The performance of the dynamic vegetation model com- the correlation between deep ocean temperatures and global pared to modern observations is also described in Valdes et mean surface temperature or, at the very least, increase the al. (2017), but the modern deep ocean temperatures are not uncertainty in reconstructing past global mean surface tem- described in that paper. We therefore include a comparison to peratures. present-day observed deep ocean temperatures in Sect. 3.1. The aim of this paper is two-fold: (1) we wish to docu- To perform paleosimulations, several important modifica- ment the setup and initial results from a unique set of 109 tions to the standard model described in Valdes et al. (2017) climate model simulations of the whole Phanerozoic era (last must be incorporated: 540 million years) at the stage level (approximately every 5 a. The standard pre-industrial model uses a prescribed million years), and (2) we will use these simulations to in- climatological pre-industrial ozone concentration (i.e. vestigate the accuracy of the deep ocean temperature proxy prior to the development of the “ozone” hole) which is in representing global mean surface temperature. a function of latitude, atmospheric height, and month of The focus of the work is to examine the link between ben- the year. However, we do not know what the distribu- thic ocean temperatures and surface conditions. However, we tion of ozone should be in these past climates. Beerling evaluate the fidelity of the model by comparing the model- et al. (2011) modelled small changes in tropospheric predicted ocean temperatures to estimates of the isotopic ozone for the early Eocene and Cretaceous, but no com- temperature of the deep ocean during the past 110 million prehensive stratospheric estimates are available. Hence, years (Zachos et al., 2008; Cramer et al., 2009; Friedrich et most paleoclimate model simulations assume unchang- al., 2012) and model-predicted surface temperatures to the ing concentrations. However, there is a problem with us- sea surface temperature estimates of O’Brien et al. (2017) ing a prescribed ozone distribution for paleosimulations and Cramwinckel et al. (2018). This gives us confidence because it does not incorporate ozone feedbacks associ- that the model is behaving plausibly but we emphasize that ated with changes in tropospheric height. During warm the fidelity of the simulations is strongly influenced by the climates, the model predicts that the tropopause would accuracy of CO estimates through time. We then use the 2 rise. In the real world, ozone would track the tropopause complete suite of climate simulations to examine changes in rise. However, this rising ozone feedback is not included ocean circulation, ice formation, and the impact on ocean and in our standard model. This leads to substantial extra surface temperature. Our paper will not consider any issues warming and artificially increases the apparent climate associated with assumptions regarding the relationship be- sensitivity. Simulations of future climate change have tween deep-sea foraminifera δ18O and various temperature shown that ozone feedbacks can lead to an overesti- calibrations because our model does not simulate the δ18O mate of climate sensitivity by up to 20 % (Dietmuller et of sea water (or vital effects). al., 2014; Nowack et al., 2015; Hardiman et al., 2019). Therefore, to incorporate some aspects of this feedback, 2 Simulation methodology we have changed the ozone scheme in the model. Ozone is coupled to the model-predicted tropopause height ev- 2.1 Model description ery model time step in the following simple way: We use a variant of the Hadley Centre model, HadCM3 -2 :0 × 10−8 kg kg−1 in the troposphere, (Pope et al., 2000; Gordon et al., 2000), which is a cou- -2 :0 × 10−7 kg kg−1 at the tropopause, pled atmosphere–ocean–vegetation model.