Gold Open Access: This paper is published under the terms of the CC-BY license.


Eocene paleoclimate reconstructions are rarely accompanied by parallel estimates of CO2 from the same locality, complicating assessment of the equilibrium climate response to elevated CO2. We reconstruct temperature, precipitation, and CO2 from latest middle Eocene (ca. 38 Ma) terrestrial sediments in the posteruptive sediment fill of the Giraffe kimberlite in subarctic Canada. Mutual climatic range and oxygen isotope analyses of botanical fossils reveal a humid-temperate forest ecosystem with mean annual temperatures (MATs) more than 17 °C warmer than present and mean annual precipitation ∼4× present. Metasequoia stomatal indices and gas-exchange modeling produce median CO2 concentrations of ∼630 and ∼430 ppm, respectively, with a combined median estimate of ∼490 ppm. Reconstructed MATs are more than 6 °C warmer than those produced by Eocene climate models forced at 560 ppm CO2. Estimates of regional climate sensitivity, expressed as ∆MAT per CO2 doubling above preindustrial levels, converge on a value of ∼13 °C, underscoring the capacity for exceptional polar amplification of warming and hydrological intensification under modest CO2 concentrations once both fast and slow feedbacks become expressed.


Efforts to understand climate response to sustained greenhouse gas forcing commonly focus on periods of peak Cenozoic warmth during the Paleocene–Eocene thermal maximum and early Eocene (e.g., Zachos et al., 2008; Lunt et al., 2012a). The subsequent cooling trend of the middle and late Eocene (Pagani et al., 2005) is also relevant because atmospheric CO2 concentrations dovetail the range projected for the coming century (Maxbauer et al., 2014; Jagniecki et al., 2015; Anagnostou et al., 2016; Steinthorsdottir et al., 2016), ultimately crossing the threshold necessary to maintain continental ice sheets (∼500 ppm; Royer, 2006). Observations from the Arctic Ocean suggest that ice rafting may have been initiated by the middle Eocene (e.g., Tripati et al., 2008), in apparent conflict with the warmth implied by the terrestrial biota (e.g., Eberle and Greenwood, 2012). Climate models struggle with these critical early Cenozoic intervals because unrealistically high CO2 forcing is required to produce the temperature responses implied by proxies, particularly for the sparse network of terrestrial high-latitude sites (Lunt et al., 2012a). Furthermore, paleoclimate and CO2 reconstructions are not commonly derived from the same sedimentary archive; this complicates assessment of proxy-model mismatch and frustrates efforts to understand the sensitivity of past equilibrium climate response to greenhouse gas forcing.

Our objective is to assess climate and greenhouse-gas forcing for Northern Hemisphere subarctic latitudes during the latest middle Eocene by exploiting a remarkable terrestrial sedimentary archive. The Giraffe kimberlite locality (paleolatitude ∼63°N) comprises the posteruptive sedimentary fill of a maar formed when kimberlite intruded Precambrian cratonic rocks of the Slave Province at 47.8 ± 1.4 Ma (Creaser et al., 2004). Pollen, δ18O from wood cellulose, and foliar stomata from this locality provide a comprehensive reconstruction of late middle Eocene climate and CO2 for the northern subarctic latitudes.


Exploration drill core BHP 99-01 (see Appendix DR1 in the GSA Data Repository1) captures ≥50 vertical-equivalent meters of lacustrine sediment overlain by 32 m of peat (Fig. 1), together representing the progressive infilling of the maar basin. Both facies have remarkable preservation of aquatic and terrestrial plant fossils (Wolfe et al., 2006; Doria et al., 2011). We analyzed a 21 m section (vertical equivalent depth) of peat in core BHP 99-01, representing ∼20 k.y. assuming reasonable accumulation rates and only moderate compaction. The common sampling interval over which we estimate the mean climate state and CO2 concentration includes multiple samples from 7 m of vertical equivalent thickness (Fig. 2), or ∼7 k.y. of continuous deposition. Two rhyolitic tephra beds are present in the core directly below the lacustrine-to-peat transition (Fig. 1C). Glass fission track dating (Westgate et al., 2013) of both tephra beds gives a weighted mean age (±1σ) of 37.84 ± 1.99 Ma (Table DR1 and Appendix DR1).

Paleoclimate of the Latest Middle Eocene Subarctic

Pollen assemblages from Giraffe sediments are well preserved, diverse, and include numerous extant North American taxa (Fig. DR1). The relative abundance of angiosperm pollen (53%–74%) is higher than that of gymnosperms throughout the section (Fig. DR2). The former is strongly dominated by fagalean types (Quercoidites, Castanea, and Corylus), with lesser contributions from Ulmipollenites, Ericalean taxa, and the Eocene indicators Platycarya swasticoides and Pistillopollenites mcgregorii. Conifer pollen is strongly dominated by Cupressaceae, for which Metasequoia is likely the dominant source given the presence of well-preserved foliage and wood of this taxon (Fig. DR1). Pollen of Pinus and Picea is distributed throughout the section, whereas that of Ginkgo, Sciadopitys, and Tsuga occurs in trace amounts.

Mutual climate range (MCR) analysis (e.g., Ballantyne et al., 2010; Thompson et al., 2012; see the methods description in Appendix DR1) of these pollen assemblages using nearest living relatives yields reconstructed mean annual temperatures (MATs) of 12.5–16.3 °C (mean ± 1σ = 14.5 ± 1.3 °C), coldest month mean temperatures (CMMTs) of 0.5–4.5 °C (3.4 ± 1.7 °C), and warmest month mean temperatures (WMMTs) of 23.5–25.0 °C (24.5 ± 0.8 °C) (Figs. 2A–2C; Table DR2). Reconstructed mean annual precipitation (MAP) ranges from 1257 to 1292 mm, with a mean uncertainty of 310 mm. The present-day climate for Yellowknife (62.45°N, 114.40°W), the nearest long-term climate station situated 300 km southwest of the Giraffe locality, has a MAT of −4.3 °C, CMMT of -25.6 °C, WMMT of 17.0 °C, and MAP of 289 mm (1981–2010 climate normals; htpp:// Thus, middle Eocene MATs were 17–20 °C warmer, CMMTs 27–31 °C warmer, WMMTs 7–8 °C warmer, and MAP 3–5 times higher than present. The North American location with a modern climatology that most closely matches the reconstructed paleoclimate for the Giraffe locality is 3500 km to the southeast at Nashville (Tennessee, USA), with MAT of 15.2 °C, CMMT of 3.2 °C, WMMT of 26.3 °C, and MAP of 1200 mm (1981–2010 climate normals;

Non-permineralized wood in the posteruptive Giraffe sequence has exceptional preservation and can be assigned unambiguously to Metasequoia on the basis of xylotomy (Fig. DR1). This wood yields pristine α-cellulose (Fig. DR3) amenable to measurements of stable oxygen isotope ratios (δ18Ocell) by pyrolysis and continuous-flow isotope ratio mass spectrometry, which in turn can provide independent support for palynological estimates of MAT (Appendix DR1). The values of δ18Ocell range from 23.4‰ to 24.9‰ VSMOW (Vienna standard mean ocean water) (Fig. 2D; Table DR3). Using a Monte Carlo implementation of the Anderson et al. (2002) leaf-water model, we estimated values of δ18O for environmental waters (δ18Owater) accessed by the trees, and then calculated MAT from these inferred δ18Owater values using an empirical relation between Eocene environmental waters and MAT that accounts for Eocene latitudinal temperature gradients (Fricke and Wing, 2004). The δ18Ocell results yield a MAT estimate of 15.6 ± 2.0 °C at 1σ (Fig. 2E), which overlaps the pollen-based MCR estimate of 14.5 ± 1.3 °C MAT (Fig. DR4).

Atmospheric CO2 Reconstruction

Stomatal indices derived from Giraffe Metasequoia leaves (Doria et al., 2011) yield a combined median reconstructed atmospheric CO2 concentration for all stratigraphic levels of ∼630 ppm (433–1124 ppm at 68% confidence) (Figs. 2F and 3; Table DR4). Combined CO2 estimates from the Franks et al. (2014) gas-exchange model, applied to the same foliage (Appendix DR1), are somewhat lower, ranging from 353 to 561 ppm at 68% confidence with a median of ∼430 ppm (Fig. 3A). Given overlap between the two methods of CO2 reconstruction, and because the stomatal index proxy is unbounded at high CO2 concentrations (Doria et al., 2011), we resampled randomly from the combined stomatal index and gas-exchange model reconstructions to yield a consensus median CO2 concentration of ∼490 ppm (378–778 ppm at 68% confidence). This approach reduces biases inherent to either technique.

This CO2 reconstruction is lower than inferences of ∼800–1000 ppm from alkenone δ13C between 39 and 37 Ma (Zhang et al., 2013) and 650 ± 110 ppm (at 68% confidence) at 40.3 Ma from δ11B of pristine foraminiferal calcite (Anagnostou et al., 2016), but in agreement with estimates of 385–467 ppm (at 68% confidence) from the stomatal distributions of Canadian High Arctic Metasequoia foliage dating broadly to between 47.9 and 37.8 Ma (Maxbauer et al., 2014) and ∼350–650 ppm from the stomatal density of extinct fagalean foliage (Steinthorsdottir et al., 2016). The results from the Giraffe locality thus support lower CO2 concentrations than previously envisaged for greenhouse climate intervals (Franks et al., 2014).


High Polar Amplification under Modest CO2 Forcing

These data provide an integrated estimate of the mean climate state for the continental subarctic Giraffe locality over the multimillennial interval common to all proxies. The MCR-inferred paleotemperature and reconstructed CO2 concentrations can be plotted along a range of estimates for the sensitivity of MAT with respect to atmospheric CO2 (Fig. 4). Present-day estimates of global Charney climate sensitivity (CS) include a subset of fast feedbacks only, while Earth-system sensitivity (ESS) includes most fast and slow feedbacks (Royer, 2016). CS and ESS are typically expressed as globally averaged approximations of the temperature response to incremental CO2 doublings, expressed as ∆MAT relative to preindustrial conditions. However, recent studies have demonstrated the utility of regional approximations for these parameters (Dyez and Ravelo, 2013; Eagle et al., 2013).

The latest middle Eocene MAT and stomatal index CO2 reconstruction from Giraffe, when compared to the present climate of Yellowknife using the approach of Royer et al. (2012), yield a mean regional climate sensitivity of 12.7 °C per CO2 doubling for the North American subarctic latitudes (8.3–21.2 °C at 68% confidence; Fig. 3; Appendix DR1), more than twice the estimated ESS of ∼6 °C per CO2 doubling for the Pleistocene climate system (Hansen et al., 2008; Fig. 4). Use of CO2 estimates from the gas-exchange model produces even higher regional climate sensitivity values (14.0–32.8 °C at 68% confidence, median = 20.1 °C; Fig. 3).

These estimates of regional climate sensitivity, based on paleoclimate and CO2 proxies analyzed in parallel from the same sediment archive, highlight the exceptional magnitude of polar amplification under relatively modest CO2 forcing. This contention is supported by temperature reconstructions from Ellesmere Island and Siberia during the Pliocene (Ballantyne et al., 2010; Brigham-Grette et al., 2013), for which independent proxies (e.g., Zhang et al., 2013) indicate CO2 concentrations of ∼400 ppm (Fig. 4). Even greater ∆MATs (∼32 °C) are estimated from middle Eocene fossil floras of Axel Heiberg Island (Eberle and Greenwood, 2012), also dominated by Metasequoia, when CO2 was possibly as low as ∼420 ppm (Maxbauer et al., 2014). Pronounced middle Eocene polar amplification is likewise expressed in the Southern Hemisphere high latitudes, where temperate rainforests dominated by Nothofagus and araucarian conifers existed along the Wilkes Land margin of East Antarctica, implying MATs >10 °C and MAPs severalfold higher than present (Pross et al., 2012).

Early Eocene climate model simulations for the latitudes of subarctic North America (Lunt et al., 2012a; Carmichael et al., 2016) underestimate the multiple proxy constraints presented here. For example, at 560 ppm CO2, the ensemble mean of three models (Lunt et al., 2012a) for the Giraffe region underestimates reconstructed MAT by 15.5 °C, with a minimum underestimate of 6.4 °C (Fig. 4). At 1120 ppm CO2, more than twice the inferred CO2 from Metasequoia foliage, the ensemble mean MAT is 11.3 °C lower than proxy MATs, with a minimum underestimate of 4.5 °C (Fig. 4). The model results compiled by Lunt et al. (2012a) consistently estimated colder-than-present preindustrial Yellowknife MATs in 280 ppm CO2 control runs. However, even when this model-dependent artifact is taken into account by expressing ∆MAT relative to instrumental Yellowknife MAT (Fig. DR5), model ∆MATs remain substantially lower than the proxy-based ∆MATs presented here for the Giraffe region (Table DR5).

Many mechanisms have been explored to explain the amplified warmth of high latitudes during the Cretaceous and Paleogene, including state-dependent CS (Caballero and Huber, 2013), decreased atmospheric pressure (Poulsen et al., 2015), altered cloud physics (Kiehl and Shields, 2013), biogenic aerosols (Beerling et al., 2011), and teleconnection dynamics with tropical oceans (Korty et al., 2008). Changes in atmospheric circulation such that low pressure centers and associated cyclogenesis became quasi-permanent features over the polar regions in the absence of perennial sea ice cover were also probable. Such configurations appear necessary to increase MAP by the amounts mandated by the proxy record at Giraffe and elsewhere, whereas intensification of the hydrologic cycle also increases poleward heat transfer by water vapor (Pagani et al., 2014; Carmichael et al., 2016). Despite obvious differences in boundary conditions with respect to the state of the cryosphere and biosphere, these configurations provide some degree of analogy with contemporary warming of the Arctic, where dramatic losses of Northern Hemisphere sea ice over the last decade have contributed to deepening lows over the Arctic Ocean, coupled to enhanced cyclogenesis that in turn exerts a strong positive feedback on remaining sea ice (Screen et al., 2011; Simmonds and Rudeva, 2012).

Because future temperatures are unlikely to decline appreciably over the time scales required for most fast and slow feedbacks to become fully expressed (centuries to millennia; Royer, 2016), even if all anthropogenic greenhouse gas emissions are eliminated (Archer and Brovkin, 2008), the latest middle Eocene forest ecosystem preserved in the Giraffe kimberlite offers considerable insight toward understanding high-latitude climate states under elevated, but not extreme, atmospheric CO2 concentrations.

We thank BHP Billiton Inc. and the Geological Survey of Canada for access to cores; the Natural Sciences and Engineering Research Council (Canada) (Greenwood, Reyes, Westgate, and Wolfe); the U.S. National Science Foundation (Siver); J. Basinger for discussions on Metasequoia; G. Braybrook for scanning electron microscopy; the Climate Change Consortium of Wales, D. McCarroll, and N. Loader for assistance with stable isotopes; R. Zetter for advice on pollen morphology; and G. Ludvigson, M. Steinthorsdottir, B. Jacobs, and several other anonymous reviewers. We dedicate this paper to our departed colleagues Leo Hickey (1940–2013) and Mark Pagani (1960–2016), who trailblazed much of our thought concerning greenhouse worlds of the past, and Art Sweet (1942–2017), pillar of Canadian palynology and the first to analyze Giraffe pipe pollen.

1GSA Data Repository item 2017202, detailed methods (Appendix DR1), data tables (Tables DR1–DR5) and supplementary Figures DR1–DR5, is available online at or on request from