Palaeoclimate reconstructions of periods with warm climates and high atmospheric CO2 concentrations are crucial for developing better projections of future climate change. Deep-ocean1,2 and high-latitude3 palaeotemperature proxies demonstrate that the Eocene epoch (56 to 34 million years ago) encompasses the warmest interval of the past 66 million years, followed by cooling towards the eventual establishment of ice caps on Antarctica. Eocene polar warmth is well established, so the main obstacle in quantifying the evolution of key climate parameters, such as global average temperature change and its polar amplification, is the lack of continuous high-quality tropical temperature reconstructions. Here we present a continuous Eocene equatorial sea surface temperature record, based on biomarker palaeothermometry applied on Atlantic Ocean sediments. We combine this record with the sparse existing data4,5,6 to construct a 26-million-year multi-proxy, multi-site stack of Eocene tropical climate evolution. We find that tropical and deep-ocean temperatures changed in parallel, under the influence of both long-term climate trends and short-lived events. This is consistent with the hypothesis that greenhouse gas forcing7,8, rather than changes in ocean circulation9,10, was the main driver of Eocene climate. Moreover, we observe a strong linear relationship between tropical and deep-ocean temperatures, which implies a constant polar amplification factor throughout the generally ice-free Eocene. Quantitative comparison with fully coupled climate model simulations indicates that global average temperatures were about 29, 26, 23 and 19 degrees Celsius in the early, early middle, late middle and late Eocene, respectively, compared to the preindustrial temperature of 14.4 degrees Celsius. Finally, combining proxy- and model-based temperature estimates with available CO2 reconstructions8 yields estimates of an Eocene Earth system sensitivity of 0.9 to 2.3 kelvin per watt per square metre at 68 per cent probability, consistent with the high end of previous estimates11.
Access optionsAccess options
Rent or Buy article
Get time limited or full article access on ReadCube.
All prices are NET prices.
Subscribe to Journal
Get full journal access for 1 year
only $3.90 per issue
All prices are NET prices.
VAT will be added later in the checkout.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
In this research, we used samples and data provided by the International Ocean Discovery Program (IODP) and its predecessor, the Ocean Drilling Program. This work was carried out under the programme of the Netherlands Earth System Science Centre (NESSC), financially supported by the Dutch Ministry of Education, Culture and Science. A.S. thanks the European Research Council (ERC) for ERC Starting Grant number 259627 under the European Union Seventh Framework Program. This study was made possible by the Netherlands Organisation for Scientific Research (NWO) grant number 834.11.006, which enabled the purchase of the UHPLC-MS system used for GDGT analyses. P.K.B. and F.P. acknowledge NWO-ALW Veni grants number 863.13.002 and number 863.13.016, respectively. M.H. was funded by the US National Science Foundation (NSF) grant OCE-0902882; the CESM model is also supported by the NSF. C.A. acknowledges the University of Padova grant number BIRD161002. We thank B. Wade (University College London) for converting the Eocene TDP data to GTS2012, P. Sexton (The Open University) for converting the middle Eocene Site 1258 data to GTS2012, J. Kiehl and C. Shields for providing their simulation data, G.-J. Reichart (Royal NIOZ and Utrecht University) for discussions, L. van der Heijden (Utrecht University, now at University of La Rochelle, France), M. Nicolai (Utrecht University), A. Mets (NIOZ), N. Welters, A. van Dijk and D. Kasjaniuk (Utrecht Geolab) for analytical support.Reviewer information
Nature thanks G. Inglis, C. Lear, K. Littler and S. Robinson for their contribution to the peer review of this work.
Extended data figures and tables
Age–depth plot showing calcareous nannofossil and chemostratigraphic tie-points (diamonds; vertical error bars are indicate the minimum and maximum depth of the tie-point), as presented in Extended Data Table 1b. B, BC and T stand for base, base common and top, respectively. Blue-shaded regions represent depth intervals for which sedimentation rates (blue lines) were calculated. The hiatus of ~1.5 Myr in Core 35 is indicated as a red curly line. Epochs and ages are expressed in Myr ago (Ma), following GTS2012.
Extended Data Fig. 2 Comparison between different TEX86-to-SST calibrations and different GDGT ratios.
a, TEX86–SST calibration lines (trend lines for BAYSPAR) for one logarithmic and several linear calibrations. Plotted symbols are the Site 959 TEX86 record, to illustrate which part of the calibration is relevant for this study. Compared calibrations are: BAYSPAR19,41 with default settings (search tolerance for 2 TEX86 standard deviations, 0.13; dark-grey line, dark-grey diamonds), BAYSPAR with increased search tolerance (0.2) (dashed line, light-grey diamonds), Kim et al.18 logarithmic core-top calibration (red line, red diamonds), linear core-top calibration18 (light-blue line) and linear subset core-top calibration without Red Sea and polar ocean data18 (dark-blue line). It is of note that the logarithmic starts strongly diverging from the linear BAYSPAR and subset calibrations from TEX86 values of >0.8. b, Site 959 SST record using different TEX86 calibrations. Calibrations and line colours and types are as in a. c, Ratio of crenarchaeol to GDGT-0 against TEX86. Data are from a core-top compilation41 (black circles; Red Sea subset, purple circles) and our Site 959 record (red squares). d, Ring index sensu Zhang et al.50 against TEX86. Data are from a core-top compilation (black circles; Red Sea subset, purple circles) and our Site 959 record (red squares). The exponential regression line of Zhang et al. through the core-top data is plotted as a black line.
a, Top, tropical SST compilation. Proxy data are compiled as described in Methods (red symbols), with all TEX86-based records converted to SST using the BAYSPAR calibration (default settings, search tolerance as described in Methods). The fitted LOESS model is plotted as a black line, with the 95% confidence interval as grey shading. Bottom, ice-free deep-ocean temperature compilation. δ18O-based proxy data are compiled as described in Methods. The fitted LOESS model is plotted as a black line and the 95% confidence interval as dark blue shading. b, Calculated MTG based on LOESS fits of proxy data (lines, propagated 95% confidence intervals shown as shading). The black line with the grey silhouette shows the tropical compilation with calibration and the blue line with the blue silhouette shows the BAYSPAR calibration. c, Proxy (blue diamonds, tropical compilation; red diamonds, Site 959) deep-ocean temperature against tropical SST using the BAYSPAR calibration for TEX86-based records. Lines represent Deming regression analysis through proxy data. The slope (polar amplification factor) is 1.42 ± 0.14 (±1 standard error) for the tropical compilation and 0.75 ± 0.04 for Site 959. Proxy data grouped into 1-Myr bins from 34–58 Myr ago, with error bars representing one standard deviation due to binning. This sensitivity analysis shows that calculated MTGs and the constant polar amplification factor are relatively robust to the specific TEX86 calibration used, although MTGs are less reduced in the early Eocene when using the BAYSPAR calibration. Polar amplification factors are lower, but reflect a linear relationship.
Extended Data Fig. 4 Regression analysis between reconstructed SST and abundance of upwelling indicators.
a, -based SST (red diamonds, upper left vertical axis), protoperidinioid abundance (percentage of total dinocyst assemblage; brown dots, right vertical axis) and TOC (percentage of sediment; black dots, lower left vertical axis) records of ODP Site 959. Dashed lines represent a hiatus in Site 959. Age is in GTS2012. b, Regression analysis between SST and percentage of protoperidinioid dinocysts of total dinocyst assemblage, showing a non-significant relationship with a very low fit (brown line, 90% confidence interval shown as brown shading; R2 = 0.00, P = 0.75) and a better fit (R2 = 0.35) that is significant (P < 0.01) when only the late Eocene (post-MECO) part of the record is considered (blue–grey line; 90% confidence interval is shown as blue–grey shading). c, Regression analysis between SST and percentage of TOC in sediment, showing a significant negative correlation for the whole record (R2 = 0.39, P < 0.001; dark-grey line, with the 90% confidence interval shown as dark-grey shading) and the late Eocene subset (R2 = 0.37, P < 0.01; blue–grey line, with the 90% confidence interval shown as blue–grey shading).
Compilation presented in Fig. 2, here plotted per site and proxy, with data sources in the key. The abbreviations Moro., Aca., and Turbo. stand for foraminifera genera Morozovella, Acarinina and Turborotalia, respectively. The dashed line in the Site 959 record represents a hiatus. Conservative estimates of propagated calibration and analytical errors (1 s.d.) are ±2.6 °C for , ±1.6 °C for δ18O and ±1.3 °C for Mg/Ca. Δ47 uncertainties are as reported in the original study6, with the minimum and maximum per-sample uncertainty. Uncertainties are plotted on the same relative vertical temperature scale as the data to facilitate comparison. The age is in Myr ago, following GTS2012.
a, Top, tropical SST compilation; proxy data are compiled as described in Methods (red symbols). The fitted LOESS model is plotted as a black line and the 95% confidence interval as grey shading. Bottom, deep-ocean temperature compilation; δ18O-based proxy data are compiled as described in Methods. Ice-free deep-ocean temperatures and fitted LOESS model are shown as grey dots and line, respectively, and the deep-ocean temperature compilation and fitted LOESS model including late Eocene ice volume effect (Methods) as blue dots and line, correspondingly. 95% LOESS confidence intervals are shown as shading. b, Calculated MTG based on LOESS fits of proxy data (lines; propagated 95% confidence intervals are shown as silhouettes). The black line with grey silhouette shows results obtained using ice-free deep-ocean temperatures, and the blue line with blue silhouette includes the late Eocene ice volume effect on the deep-ocean temperature. c, Proxy (blue diamonds, tropical compilation; red diamonds, Site 959) deep-ocean temperature, including the late Eocene ice volume effect, against tropical SST. Lines represent Deming regression analysis through proxy data. The slope (polar amplification factor) is 2.07 ± 0.25 (±1 standard error) for the tropical compilation and 1.19 ± 0.06 for Site 959). Proxy data grouped into 1-Myr bins from 34–58 Myr ago, with error bars representing one standard deviation due to binning. d, Sensitivity of δ18O of Eocene seawater (‰ VSMOW) to the build-up of 0–107 km3 of ice with varying isotopic composition.
Site 1172 TEX86-based SST (record plotted in Fig. 2) against Site 959 TEX86-based SST. Lines represent Deming regression analysis through proxy data (polar amplification factor, 1.66 ± 0.57). Proxy data are grouped into 1-Myr bins from 34–58 Myr ago, with error bars representing one standard deviation due to binning. Peak PETM and peak MECO SSTs are plotted as separate points, which fall within the uncertainty of the regression line.
a, Top, tropical SST compilation; proxy data are compiled as described in Methods (red symbols). The fitted LOESS model is plotted as a red line, with the 95% confidence interval as pink shading. Bottom, ice-free deep-ocean temperature compilation; δ18O-based proxy data are compiled as described in Methods. Generalized cross-validation (GCV)-optimized fitted LOESS model (as in Fig. 3) plotted as grey line, with the 95% confidence interval as grey shading. An alternative LOESS model with small bandwidth (0.25 times the GCV-optimized span) that tracks deep-ocean PETM temperature more closely is shown as the blue line, with the 95% confidence interval as blue shading. Data are plotted together with the PETM simulation of Kiehl and Shields28 (black open squares, seasonal range error bars) with altered cloud parameters (CP_PETM). b, Calculated MTG based on LOESS fits of proxy data and of the model simulation CP_PETM. The grey line is obtained using a GCV-optimized fitted LOESS model and the blue line using a smaller-bandwidth deep-ocean LOESS model (propagated 95% confidence intervals are shown as shading). We note that the PETM MTG pattern is complex owing to its sensitivity to the specific records, age models and filtering used, and might have evolved over the course of the event. Nevertheless, peak PETM MTG matches the simulation CP_PETM poorly. The age is in Myr ago, following GTS2012.
a, b, ESS estimates using proxy (a) and model (b) temperatures in combination with proxy-based CO2 concentrations, derived as described in Methods. Eocene ESS is separated into the late Eocene relative to the EECO (red), the middle Eocene relative to the EECO (purple) and the middle Eocene relative to late Eocene (blue). ESS estimates of the EECO relative to preindustrial temperature (black) have lower error owing to the high precision of preindustrial CO2 concentration and temperature, but include additional long-term non-CO2 effects.
This file contains Supplementary Data Tables 1–5. Supplementary Table 1 shows calcareous nannofossil counts, Supplementary Table 2 shows fractional abundances of isoprenoid GDGTs, Supplementary Table 3 shows relative and absolute abundance of upwelling-indicative dinocysts, Supplementary Table 4 shows selected temperature output of CESM simulations EO1–EO4 and EO_CP, and Supplementary Table 5 shows Eocene tropical sea surface temperature compilation.