Characterizing past climate states is crucial for understanding the future consequences of ongoing greenhouse gas emissions. Here, we revisit the benchmark time series for deep ocean temperature across the past 65 million years using clumped isotope thermometry. Our temperature estimates from the deep Atlantic Ocean are overall much warmer compared with oxygen isotope–based reconstructions, highlighting the likely influence of changes in deep ocean pH and/or seawater oxygen isotope composition on classical oxygen isotope records of the Cenozoic. In addition, our data reveal previously unrecognized large swings in deep ocean temperature during early Eocene acute greenhouse warmth. Our results call for a reassessment of the Cenozoic history of ocean temperatures to achieve a more accurate understanding of the nature of climatic responses to tectonic events and variable greenhouse forcing.

The early Eocene hothouse experienced highly elevated atmospheric CO2 levels and multiple transient global warming events, so-called hyperthermals. The deep ocean constitutes an assumed setting to estimate past global mean temperatures. However, available deep-sea temperature reconstructions from conventional benthic foraminiferal oxygen isotopes and magnesium/calcium ratios rely on uncertain assumptions of non-thermal influences, associated with seawater chemistry and species-specific physiological effects. Here we apply the carbonate clumped isotope thermometer, a proxy not governed by these uncertainties, to evaluate South Atlantic deep-sea temperatures across two hyperthermal events in the early Eocene (Eocene Thermal Maximum 2/H1 and H2; ~54 Myr ago). Our independent reconstructions indicate deep-sea temperatures of 13.5 ± 1.9 °C (95% CI) for the background conditions and average hyperthermal peak temperatures of 16.9 ± 2.3 °C (95% CI). On average, these absolute temperatures are three degrees warmer than estimates from benthic oxygen isotopes. This finding implies a necessary reassessment of (1) the Eocene seawater isotope composition and (2) pH changes in the deep ocean and its potential influence on benthic foraminiferal oxygen isotope records.

Abstract. The aim of paleoclimate studies to resolve climate variability from noisy proxy records can in essence be reduced to a statistical problem. The challenge is to isolate meaningful information on climate events from these records by reducing measurement uncertainty through a combination of proxy data while retaining the temporal resolution needed to assess the timing and duration of the event. In this study, we explore the limits of this compromise by testing different methods for combining proxy data (smoothing, binning and sample size optimization) on a particularly challenging paleoclimate problem: resolving seasonal variability in stable isotope records. We test and evaluate the effects of changes in the seasonal temperature and hydrology cycle as well as changes in accretion rate of the archive and parameters such as sampling resolution and age model uncertainty on the reliability of seasonality reconstructions based on clumped and oxygen isotope analyses in 33 real and virtual datasets. Our results show that strategic combinations of clumped isotope analyses can significantly improve the accuracy of seasonality reconstructions if compared with conventional stable oxygen isotope analyses, especially in settings where the isotopic composition of the water is poorly constrained. Smoothing data using a moving average often leads to a dampening of the seasonal cycle, significantly reducing the accuracy of reconstructions. A statistical sample size optimization protocol yields more precise results than smoothing. However, the most accurate results are obtained through monthly binning of proxy data, especially in cases where growth rate or water composition cycles dampen the seasonal temperature cycle. Our analysis of a wide range of natural situations reveals that the effect of temperature seasonality on isotope records almost invariably exceeds that of changes in water composition. Thus, in most cases, isotope records allow reliable identification of growth seasonality as a basis for age modelling and seasonality reconstructions in absence of independent chronological markers in the record. These specific findings allow us to formulate general recommendations for sampling and combining data in paleoclimate research and have implications beyond the reconstruction of seasonality. We discuss the implications of our results for solving common problems in paleoclimatology and stratigraphy, including cyclostratigraphy, strontium isotope dating and event stratigraphy.

Abstract. The aim of paleoclimate studies resolving climate variability from noisy proxy records can in essence be reduced to a statistical problem. The challenge is to extract meaningful information about climate variability from these records by reducing measurement uncertainty through combining measurements for proxies while retaining the temporal resolution needed to assess the timing and duration of variations in climate parameters. In this study, we explore the limits of this compromise by testing different methods for combining proxy data (smoothing, binning, and sample size optimization) on a particularly challenging paleoclimate problem: resolving seasonal variability in stable isotope records. We test and evaluate the effects of changes in the seasonal temperature and the hydrological cycle as well as changes in the accretion rate of the archive and parameters such as sampling resolution and age model uncertainty in the reliability of seasonality reconstructions based on clumped and oxygen isotope analyses in 33 real and virtual datasets. Our results show that strategic combinations of clumped isotope analyses can significantly improve the accuracy of seasonality reconstructions compared to conventional stable oxygen isotope analyses, especially in settings in which the isotopic composition of the water is poorly constrained. Smoothing data using a moving average often leads to an apparent dampening of the seasonal cycle, significantly reducing the accuracy of reconstructions. A statistical sample size optimization protocol yields more precise results than smoothing. However, the most accurate results are obtained through monthly binning of proxy data, especially in cases in which growth rate or water composition cycles obscure the seasonal temperature cycle. Our analysis of a wide range of natural situations reveals that the effect of temperature seasonality on oxygen isotope records almost invariably exceeds that of changes in water composition. Thus, in most cases, oxygen isotope records allow reliable identification of growth seasonality as a basis for age modeling in the absence of independent chronological markers in the record. These specific findings allow us to formulate general recommendations for sampling and combining data in paleoclimate research and have implications beyond the reconstruction of seasonality. We briefly discuss the implications of our results for solving common problems in paleoclimatology and stratigraphy.

<p>Reconstructing deep ocean temperature is important to infer deep water mass structure and hence ocean circulation patterns in the past. The late Paleocene-early Eocene experienced the warmest climates of the Cenozoic, with highly elevated CO<sub>2</sub> levels and no ice sheets on the continents [1,2]. Benthic foraminiferal δ<sup>18</sup>O records suggest relatively stable deep ocean conditions on long time scales (>100 kyr) in this hothouse [2–4]. However, interpretations from benthic δ<sup>18</sup>O records are complicated by influences of factors other than temperature, such as the isotope composition of the seawater (δ<sup>18</sup>O<sub>sw</sub>), pH, and species-specific physiological effects [5,6]. Carbonate clumped isotope thermometry (Δ<sub>47</sub>) has the major advantage that it is independent of the isotope composition of the fluid source, and is not measurably affected by other non-thermal influences [7–10]. Early Cenozoic clumped isotope reconstructions from the North Atlantic have revealed surprisingly large deep-sea temperature swings under hothouse conditions [11]. Extreme warming is recorded at the onset of the Early Eocene Climatic Optimum (EECO) [11]. To explore the spatial extent of these deep-sea temperature changes, we reconstructed early Eocene Δ<sub>47</sub>-based deep-sea temperatures from the South Atlantic Ocean, a location that is considered to capture a global signal [2–4]. We find similar deep-sea temperatures as those from the North Atlantic. Cooler temperatures of ~12 °C stand out in the interval (54–52 Ma) before the peak warmth of the EECO (52–50 Ma) of ~20 °C. This result overthrows the classic view of a gradual early Eocene warming trend based on benthic δ<sup>18</sup>O records, at least for the deep Atlantic Ocean. Our findings raise new questions on the regions of deep water formation, changes in deep ocean circulation, and the driving mechanisms in the early Cenozoic hothouse.<br><br><strong>References</strong><br>[1] Anagnostou, E. <em>et al</em>. (2016). <em>Nature</em>, <em>533</em>(7603), 380-384.<br>[2] Zachos, J. <em>et al</em>. (2001). <em>Science</em>, <em>292</em>(5517), 686-693.<br>[3] Lauretano, V. <em>et al</em>. (2018). <em>Paleoceanography and Paleoclimatology</em>, <em>33</em>(10), 1050-1065.<br>[4] Westerhold, T. <em>et al</em>. (2020). <em>Science</em>, <em>369</em>(6509), 1383-1387.<br>[5] Ravelo, A. C., & Hillaire-Marcel, C. (2007). <em>Developments in marine geology</em>, <em>1</em>, 735-764.<br>[6] Pearson, P. N. (2012). <em>The Paleontological Society Papers</em>, <em>18</em>, 1-38.<br>[7] Ghosh, P. <em>et al</em>. (2006). <em>Geochimica et Cosmochimica Acta</em>, <em>70</em>(6), 1439-1456.<br>[8] Tripati, A. K. <em>et al</em>. (2015). <em>Geochimica et Cosmochimica Acta</em>, <em>166</em>, 344-371.<br>[9] Guo, W. (2020). <em>Geochimica et Cosmochimica Acta</em>, <em>268</em>, 230-257.<br>[10] Meinicke, N. <em>et al</em>. (2020). <em>Geochimica et Cosmochimica Acta</em>, <em>270</em>, 160-183.<br>[11]<strong> </strong>Meckler, A. N.<em> et al</em>. (in revision).</p>