Climate variability, heat distribution, and polar amplification in the warm unipolar “icehouse” of the Oligocene

The Oligocene (33.9–23.03 Ma) had warm climates with flattened meridional temperature gradients, while Antarctica retained a significant cryosphere. These may pose imperfect analogues to distant future climate states with unipolar icehouse conditions. Although local and regional climate and environmental reconstructions of Oligocene conditions are available, the community lacks synthesis of regional reconstructions. To provide a comprehensive overview of marine and terrestrial climate and environmental conditions in the Oligocene, and a reconstruction of trends through time, we review marine and terrestrial proxy records and compare these to numerical climate model simulations of the Oligocene. Results, based on the present relatively sparse data, suggest temperatures around the Equator that are similar to modern temperatures. Sea surface temperatures (SSTs) show patterns similar to land temperatures, with warm conditions at mid- and high latitudes (∼60–90°), especially in the Southern Hemisphere (SH). Vegetation-based precipitation reconstructions of the Oligocene suggest regionally drier conditions compared to modern times around the Equator. When compared to proxy data, climate model simulations overestimate Oligocene precipitation in most areas, particularly the tropics. Temperatures around the mid- to high latitudes are generally underestimated in models compared to proxy data and tend to overestimate the warming in the tropics. In line with previous proxy-to-model comparisons, we find that models underestimate polar amplification and overestimate the Equator-to-pole temperature gradient suggested from the available proxy data. This further stresses the urgency of solving this widely recorded problem for past warm climates, such as the Oligocene.

1 Introduction

Simulations of future climate change, by current-generation fully coupled climate models, indicate that global average surface warming will continue over the coming centuries depending on future CO₂ emissions and sequestration (IPCC, 2022). The models and available temperature, CO₂, and sea-level reconstructions of past Mesozoic and Cenozoic warm climates suggest that Earth's climate may ultimately move towards unipolar conditions, with ice only remaining on Antarctica (Burke et al., 2018; Clark et al., 2016). Climate models additionally predict a global equilibrium surface warming between 1.5–4.5 °C per doubling of atmospheric CO₂ concentrations relative to pre-industrial values, most likely with a value around 3 °C (IPCC, 2022). This warming will be amplified at higher latitudes, notably the Arctic, by a factor of 2–3 relative to the global average (Fischer et al., 2018; Holland and Bitz, 2003; IPCC, 2022). However, model projections, particularly for such distant future non-analogue states, still include large uncertainties and are ideally independently constrained by data. Proxy-based reconstructions of past climates provide useful insights into the Earth's natural response to CO₂ changes and are therefore an independent opportunity to quantify the sensitivity of various climate parameters to greenhouse forcing, including sea-level and polar amplification (e.g., Burke et al., 2018; Lunt et al., 2016; Palaeosens Project Members, 2012). This way, climate models which are simulating past climate conditions can be compared against proxy data; hence the performance of the models can be evaluated.

It is likely that important climate parameters such as equilibrium climate sensitivity and polar amplification depend on the state of the climate (e.g., Farnsworth et al., 2019; Gaskell et al., 2022; Hutchinson et al., 2021; Köhler et al., 2015; Masson-Delmotte et al., 2013). Therefore, these parameters have been investigated for several past climate states. Traditional targets include the Pleistocene, Pliocene, and Eocene (e.g., Burke et al., 2018) and, more recently, the Miocene (Steinthorsdottir et al., 2021). These time intervals encompass a wide range of climate states, including those with ice sheets in both the Southern Hemisphere (SH) and Northern Hemisphere (NH), the Southern Hemisphere only, and ice-free states, in addition to a wide range of atmospheric greenhouse gas concentrations (e.g., Rae et al., 2021).

Recent work (e.g., O'Brien et al., 2020) has highlighted the Oligocene as a potentially useful climate state which allows us to assess the dynamics of global climate with only an Antarctic ice sheet present. Though geographical boundary conditions during the Oligocene (33.9–23.03 Ma – million years ago) were different to today, along with the Miocene, the Oligocene is a useful and relatively recent analogue to future unipolar icehouse climate states (e.g., O'Brien et al., 2020; Liebrand et al., 2017; Miller et al., 1988). Sparse, glacially deposited sediments suggest the presence of NH glaciers as young as the late Eocene (Eldrett et al., 2007; St. John, 2008), but there is no evidence for late Eocene large-scale continental glaciation. Instead, the cryosphere potentially comprised localised glaciers and restricted sea ice in the Arctic Ocean (DeConto et al., 2008; Stickley et al., 2009). Reconstructions of atmospheric CO₂ range from over 1000 ppm (parts per million) to as low as ∼300 ppm for the Oligocene (Foster et al., 2017; Rae et al., 2021), similar to the range projected for the future based on various emission scenarios (IPCC, 2022). Despite potentially low atmospheric CO₂ conditions, the few available sea surface temperature (SST) reconstructions indicate temperatures warmer than modern climates throughout the Oligocene, with remarkable polar amplification (O'Brien et al., 2020).

Across the Eocene–Oligocene transition (EOT), atmospheric CO₂ concentrations dropped from >1000 ppm during the Eocene (56.0–33.9 Ma) to ∼750 ppm or lower at the beginning of the Oligocene (Heureux and Rickaby, 2015; Pagani et al., 2005; Pearson et al., 2009). This drop coincides with a large increase (∼1 ‰–1.5 ‰) in deep-ocean benthic foraminifer oxygen isotope ratios (δ¹⁸O), which includes the effects of both the formation of ice sheets and a drop in deep-sea temperatures (e.g., Coxall and Wilson, 2011). The forcings underlying the onset of the Oligocene so-called “icehouse” climate (i.e., with polar ice) are still highly debated. As of now, the leading hypothesis invokes a strongly non-linear response to orbital forcing superimposed on a long-term drop in atmospheric CO₂ levels across a critical threshold (DeConto et al., 2008; DeConto and Pollard, 2003; Galeotti et al., 2016). However, the question that remains is if, or to what extent, tectonic changes and associated oceanographic reorganisations in the Southern Ocean (SO) played a role (e.g., Hill et al., 2013; Houben et al., 2019; Huber et al., 2004; Ladant et al., 2014; Sauermilch et al., 2021). Changes associated with the onset of polar glaciation include a drop in the global average temperature (Eldrett et al., 2009; Kotthoff et al., 2014; Liu et al., 2009; Meckler et al., 2022; Sluiter et al., 2022; Thompson et al., 2022; Zanazzi et al., 2007) and a profound change in deep-water temperatures (Meckler et al., 2022). Across the EOT, surface cooling (Liu et al., 2009), the accumulation of land ice that reached the Antarctic coastlines (Salamy and Zachos, 1999), and the consequent appearance of sea ice (Houben et al., 2013) were associated with pronounced changes in SH atmospheric circulation and oceanographic conditions (Diester-Haass and Zahn, 1996; Houben et al., 2019; Liu et al., 2009; Tripati et al., 2005) and an increase in the poleward ocean heat transport (Goldner et al., 2014). Global change at the beginning of the Oligocene also influenced the global turnover of flora and fauna (e.g., Solé et al., 2020; Sun et al., 2014). Oceanographic changes, including upwelling and the formation of sea ice, rapidly transformed circum-Antarctic marine ecosystems to such an extent that they may have influenced the evolution of large animal groups, such as the diversification amongst odontocete and mysticete (baleen) whales (e.g., Fordyce, 1980; Salamy and Zachos, 1999; Houben et al., 2013). Thus, the EOT seems to mark a prominent change in the global climate system (Westerhold et al., 2020) as continental ice sheets expanded.

Although the Oligocene has been the subject of numerous studies, the documentation of global Oligocene climate conditions and their variability, including the hemispheric distribution of heat, meridional temperature gradients, and biotic change, is sparse and greatly relies on benthic foraminifer isotope data. With the Earth's high-latitude cryosphere and climate directly responding to astronomical insolation changes, astronomical forcing studies of high-resolution benthic foraminifer δ¹⁸O records are of great importance. Those studies (e.g., De Vleeschouwer et al., 2017; Galeotti et al., 2016; Levy et al., 2019; Liebrand et al., 2017; Naish et al., 2001; Pälike et al., 2006b) suggest significant variability in continental ice volume, paced by eccentricity and obliquity. Multiple-proxy SST data, albeit of much lower resolution than the deep-ocean δ¹⁸O records, revealed that the Oligocene was characterised by generally warm climates, with flattened meridional temperature gradients (Gaskell et al., 2022; O'Brien et al., 2020). Still, Antarctica retained a significant cryosphere (e.g., Hoem et al., 2021). The recorded trends, cycles, and events provide ample opportunity to study the dynamics of climate and the carbon cycle in what has been called a “doubthouse” or “intermediate” climate state (O'Brien et al., 2020).

In this paper, we aim to review the current state of knowledge regarding the Oligocene climate to provide a baseline for focused future research. To this end, we first provide basic constraints regarding important climatic boundary conditions, such as palaeogeography and atmospheric CO₂ levels. Additionally, a review of marine and terrestrial climate proxy records is presented, building on the compilation of marine records by O'Brien et al. (2020), to assess long-term trends and variability in the Oligocene climate. The temperature and precipitation data were subsequently compared to the results of two sets of Oligocene climate model simulations to evaluate how well we understand the data from a climate physics point of view. Lastly, we identify specific points of interest for follow-up research.

2 Oligocene stratigraphical and chronological framework

2.1 Oligocene chronostratigraphy

The Oligocene represents the epoch between two formal Global Stratotype Sections and Points (GSSPs), the Eocene–Oligocene boundary (EOB) and the Oligocene–Miocene boundary (OMB), at 33.9 and 23.04 Ma following GTS 2020 (Gradstein, 2020). The EOB GSSP was set in 1992 at the Massignano Quarry (Italy) and is defined by the extinction of the two planktic foraminifer genera Hantkenina and Cribrohantkenina at 33.9 Ma (Silva and Jenkins, 1993). The GSSP for the OMB was defined by Steininger et al. (1997) in the Tertiary Piedmont Basin in Italy on the magnetic reversal from polarity chron C6Cn.2r–C6Cn.2n between two subunits of the Rigoroso Formation. Later, Beddow et al. (2018) dated the base of C6Cn.2n at 23.040 Ma. Within the Oligocene, Hardenbol and Berggren (1978) were the first to distinguish the Rupelian (33.9–27.29 Ma) from the Chattian (27.29–23.040 Ma) of northwestern Europe. They separated the two periods based on lithostratigraphy in Belgium into an open marine clayey unit which overlies a shallower marine sandy unit. The Rupelian (chron C13r–C9n) was introduced by Dumont (1849), describing the Boom Clay formation along the Rupel and Scheldt rivers in Belgium. The Chattian (chron C9n–C6Cn) was first officially mentioned by Fuchs (1894), who studied the “Kasseler Meeressande” (marine sands) in Hessen and Bünde, Germany (De Man et al., 2010; Van Simaeys, 2004; Van Simaeys et al., 2004). The GSSP for the Rupelian–Chattian boundary was set in 2016 by Coccioni et al. (2018) at the Monte Cagnero section near Urbania (Italy) and was bound by the (highest) last common occurrence (LCO) of the planktonic foraminifer Chiloguembelina cubensis at the base of the planktonic foraminifer zone O5. Currently, the official GSSP age for the Rupelian–Chattian boundary is 27.29 Ma, after Coccioni et al. (2018).

2.2 Oligocene isotope stratigraphy

Several informal definitions are used to describe the various stratigraphic isotope events associated with the Oligocene (Fig. 1). The Eocene–Oligocene transition (EOT) refers to the numerous climatic and environmental events broadly associated with the epoch boundary (e.g., Coxall and Pearson, 2007; Eldrett et al., 2009; Houben et al., 2012; Zanazzi et al., 2007). However, Hutchinson et al. (2021) recently defined it as the ∼790 kyr interval between the extinction of the coccolithophore species Discoaster saipanensis (∼34.46 Ma) and the Earliest Oligocene oxygen Isotope Step (EOIS). Hutchinson et al. (2021) also defined several other globally recognisable Oligocene isotope events, together replacing the classic Oligocene oxygen isotope zones of Miller et al. (1991). The EOIS, previously known as the onset of Oligocene oxygen isotope zone 1 (Oi-1; Miller et al., 1991), is an ∼40 kyr long ≥0.7 ‰ δ¹⁸O increase (Coxall and Wilson, 2011; Zachos et al., 1996) which peaks at ∼33.71 Ma. The EOT and the EOIS are followed by the Early Oligocene Glacial Maximum (EOGM), which lasted ∼490 kyr from ∼33.71 to ∼33.22 Ma and chronostratigraphically correlates to most of the geomagnetic polarity timescale (GPTS) magnetochron C13n (33.726–33.214 Ma) (Hutchinson et al., 2021). The EOGM was first introduced by Liu et al. (2004) and spans the separate δ¹⁸O maxima which were defined by Zachos et al. (1996) as Oi-1a (∼33.66 Ma) and Oi-1b (∼33.26 Ma), a consecutive period of colder climate and/or glaciation. The Mid-Oligocene Glacial Interval (MOGI) represents an ∼1 Myr long phase of strong variability in δ¹⁸O with marked maxima representing profound cooling and/or glacial expansion between 27.3 and 26.3 Ma (Liebrand et al., 2017), previously referred to as Oi-2b. The MOGI was followed by three warming phases (∼26.3, ∼25.5, and ∼24.22 Ma), after which cooling led up to the Oligocene–Miocene transition (OMT; 23.88–23.04 Ma). The beginning of the Miocene (23.040–5.33 Ma) is marked by an ∼1 ‰ rise in deep-ocean benthic foraminifer δ¹⁸O values, traditionally referred to as the Mi-1 event (Billups et al., 2002; Flower et al., 1997; Miller et al., 1991).

Figure 1 (a) In blue: SST of ODP 1168A (west of Tasmania, TEX${}_{\mathrm{86}}^{\mathrm{H}}$; Guitián and Stoll, 2021; Hoem et al., 2021). In brown: SST of U1404 (northwestern Atlantic, Uk37; Liu et al., 2018). (b) Published pCO₂ records of the Oligocene. Dark-blue line and shading represent the median and 95 % credible interval. Grey squares: phytoplankton data. Brown squares: leaf gas exchange reconstructions. Black dots: boron isotopic data. Green triangles: land plant δ¹³C data. Brown crosses: palaeosol data. Green stars: stomatal frequency data (Greenop et al., 2019; Moraweck et al., 2019; Pagani et al., 2005, 2011; Roth-Nebelsick et al., 2014; Witkowski et al., 2018; Zhang et al., 2013; The CenCO2PIPm, 2023). (c, d) Deep-ocean benthic foraminifera stable carbon isotope and oxygen isotope records, respectively (Westerhold et al., 2020, notably representing the record of Pälike et al., 2006a). Red colour block: Eocene–Oligocene transition (EOT). Grey: Eocene–Oligocene Glacial Maximum (EOGM). Blue: Mid-Oligocene Glacial Interval (MOGI). Green: Oligocene–Miocene transition (OMT).

3 Boundary conditions for Oligocene climate

3.1 Geographical boundary conditions

Oligocene plate tectonic geography differed from the modern configuration regarding several regions that are relevant to climate (Fig. 2). Specifically, plate tectonic movements may have been important for oceanographical change, influencing regional climate through changes in meridional and zonal heat transport. Here, we discuss the most prominent tectonic differences.

Figure 2 Palaeogeographic reconstruction of the Oligocene (∼28 Ma). Yellow dots: Tex₈₆-based data. Purple hexagons: nearest living relative (NLR) data. Green squares: U${}_{\mathrm{37}}^{\mathrm{K}}$ data. Red diamonds: δ¹⁸O data. Map created using GPlates, using the Scotese and Wright (2018) plate rotation.

One of the most discussed tectonic changes since the Oligocene is the uplift of the Tibetan Plateau, the Himalayas, and the Hengduan Mountains. Although the collision between India and the Eurasian Plate predates the Oligocene (60–50 Ma; van Hinsbergen, 2022; Wang et al., 2014), continued collision also created further uplift of the Himalayas and the Tibetan Plateau in the Oligocene. The climatic consequences of the uplift are investigated both regionally (e.g., SE Asia; Ding et al., 2017; Su et al., 2019) and globally as a source of chemical weathering during the Cenozoic (e.g., Raymo and Ruddiman, 1992). Today, Tibet has an average elevation of over 4.5 km. During the Oligocene the elevation of Tibet was between 2.3 and 3 km, with central Tibet most likely at similar altitudes to today (Spicer et al., 2020, 2021a, b; Su et al., 2019).

Another important tectonic event during the Oligocene was the convergence of the European and Adriatic plates which led to the formation of the Alpine system. While the collisional stage of the Alps began in the earliest Paleocene (65 Ma), during the Oligocene a slab from the subducted oceanic European lithosphere broke off, which resulted in the rapid and continued uplift of the Alps, which lasted until today but became stable in the mid-Miocene (Meschede and Warr, 2019). This resulted in an Oligocene uplift of ∼1 km of the Alpine area, from an average elevation of <1 to 2 km (Dielforder, 2017; Winterberg et al., 2020). Due to the strong tectonic changes around the eastern Alps and Tibet during the Oligocene, the so-called “Paratethys” (Laskarev, 1924), which reached from the western Molasse basin in Switzerland to the Aral Sea between Kazakhstan and Uzbekistan, became a semi-isolated inland sea at the beginning of the Oligocene (Schulz et al., 2005; Steininger and Wessely, 1999). The Paratethys consisted of a series of adjacent sedimentary basins, of which the interconnections were rather unstable, resulting in the separation of the Paratethys into three main parts: the western (Alpine), central (Balkan), and eastern (Caucasian) Paratethys (Kováč, 2017; Palcu and Krijgsman, 2023; Rögl, 1998). A connection to the Mediterranean might have been established in the late Oligocene, connecting the Paratethys to the main oceans (Kováč, 2017).

Oceanic gateways in the Southern Ocean, including the Drake Passage (DP) and the Tasmanian Gateway, underwent tectonic changes in the Paleogene that were originally hypothesised to have strongly affected regional ocean circulation and associated heat and salt transport, as inferred from sedimentary data (e.g., Kennett, 1977; Murphy and Kennett, 1986). However, based on model simulations and biogeography, the extent to which the opening of these gateways affected Cenozoic cooling, either globally or regionally, remains the subject of debate (Houben et al., 2019; Huber et al., 2004; Kennett, 1977; Sauermilch et al., 2021; Scher and Martin, 2006; Toumoulin et al., 2020). While exact age estimates for the first opening of the DP lie most likely in the middle to late Eocene (∼50 Ma; Eagles and Jokat, 2014), it remains likely that the DP opened once or experienced intermittent closures that resulted in staggered throughflow over several tens of millions of years (van de Lagemaat et al., 2021). The DP opened due to tectonic processes between the South American, Antarctic, and central Scotia plates (Eagles, 2016), and it involved a complex opening of isolated ocean basins which became oceanographically connected into one deep throughflow during the Oligocene (∼26 Ma; van de Lagemaat et al., 2021). During the early to middle Oligocene (34–26 Ma), subduction initiated between South America and the Scotia Plate (Crameri et al., 2020), which led to the opening and deepening of several ocean basins in the area between 29.5 and 21.2 Ma (van de Lagemaat et al., 2021). This tectonic development facilitated the formation of a deeper DP gateway (Maldonado et al., 2014).

Although spreading between Australia and Antarctica initiated in the Cretaceous, the South Tasman Rise connected the continents until the latest Eocene (see overview in Bijl et al., 2021). Dinoflagellate cyst biogeographical evidence suggests shallow-water connections between the Australo-Antarctic Gulf and the southwestern Pacific initiated close to the early–middle Eocene transition (Bijl et al., 2013). Lithological evidence for rapid deepening of the South Tasman Rise at ∼35.7 Ma (Stickley et al., 2004) was later found to be a Southern Ocean-wide phenomenon related to the initiation of the throughflow of a vigorous Antarctic Counter Current and a proto-Antarctic Circumpolar Current (ACC; Houben et al., 2019). It is generally accepted that the Tasmanian Gateway was open to deep waters by Oligocene times (Stickley et al., 2004), but the Australian continent obstructed the optimal flow of strong circumpolar ocean currents until the late Neogene (Evangelinos et al., 2022; Hill et al., 2013; Sauermilch et al., 2021).

3.2 Ocean circulation

Using marine magnetic data, Barker and Burrell (1977) found that the opening of the Southern Ocean gateways (SOGs) (i.e., the Drake Passage and the Tasmanian Gateway) preconditioned the formation of the Antarctic Circumpolar Current (ACC) between Antarctica, South America, and Australia. As Earth's strongest ocean current, the modern ACC is not only responsible for the regulation of heat and carbon exchange from and to the atmosphere, but it also influences deep-water formation and nutrient distribution (Cox, 1989; Scher et al., 2015). The ACC encircles Antarctica and, in doing so, connects the deep waters of the Pacific, Indian, and Atlantic oceans. Models using middle Eocene to early Oligocene geographies and CO₂ values have sought to understand the influence of the SOG openings on the global ocean (e.g., Goldner et al., 2014; Hutchinson et al., 2018; Kennedy et al., 2015; Kennedy-Asser et al., 2019; Sauermilch et al., 2021; Baatsen et al., 2016). Model simulations suggest that, as soon as the DP opened, a weak current (the proto-ACC) from the Pacific to the Atlantic would have been established (Ladant et al., 2014). Additionally, the coupled model of Toggweiler and Bjornsson (2000) shows that winds around Antarctica raise cold, dense water, cooling the region. This upwelled water then becomes fresher and warmer as it moves northwards due to Ekman transport. North of the ACC, this lighter and warmer water is transported downwards into the thermocline. This thickens the lower thermocline and creates a bigger density contrast across the Icelandic sills, ultimately enhancing the formation of North Atlantic Deep Water (NADW). Subsequently, NADW cools the water, facilitating its southward transport. The model of Toggweiler and Bjornsson (2000) shows that the DP opening thus might have led to a cooling of the air and oceans around Antarctica of around 3 °C. The water in the SH takes up the solar heat, which is then transported northwards where it is released, consequently warming up the NH by the equivalent amount the SH was cooled (Toggweiler and Bjornsson, 2000).

Lagabrielle et al. (2009) also discuss the influence of the proto-ACC on the formation and strength of Northern Component Water (NCW; later turns into NADW), which brings water from the NH to the Southern Ocean. Exactly when the formation of NCW began is unclear, but around 34 Ma the North Atlantic deepened rapidly due to its separation from Greenland in response to the Iceland mantle plume collapse, and significant deep-water production started in the NH (Lagabrielle et al., 2009; Via and Thomas, 2006). Hence, along with tectonic changes in the North Atlantic region, the proto-ACC contributed to the inception of NCW/NADW and modulated its strength.

In the Eocene the SOGs were not open to deep-ocean circulation. Rather, shallow-ocean connections south of 60° S allowed a westward Antarctic Counter Current (e.g., Bijl et al., 2013; Houben et al., 2019). Although SOGs progressively opened in the Oligocene (Stickley et al., 2004), oceanographic changes associated with that were restricted to the Southern Ocean (Houben et al., 2019; Scher and Martin, 2008), and there was little effect on the Southern Ocean oceanography for the remainder of the Oligocene (Evangelinos et al., 2020; Hill et al., 2013; Hoem et al., 2021; Wright et al., 2018). Only in the late Oligocene did Southern Ocean latitudinal SST gradients increase, and perhaps the ACC strengthened, due to deep opening of Drake Passage (Hoem et al., 2022), although the ACC weakened again during the Miocene Climatic Optimum (Evangelinos et al., 2022; Sangiorgi et al., 2018).

3.3 Atmospheric pCO₂

Only a few records of atmospheric pCO₂ cover the Oligocene entirely. Most are focused on the EOT and the OMT or are of low resolution. Trends are quite inconsistent between records and proxies (CenCO2PIP, 2023). The available records for the Oligocene are based on higher plant leaf gas exchange, phytoplankton ¹³C-fractionation, and foraminifer boron isotope ratios (Fig. 1). Pagani et al. (2005) were the first to produce a pCO₂ record for the Oligocene using ¹³C-fractionation of di-unsaturated alkenones extracted from various Deep Sea Drilling Project (DSDP) and Ocean Drilling Program (ODP) sediments. They recorded decreasing pCO₂ throughout the Oligocene from ∼1500 ppm at the EOT to modern levels by the late Oligocene. Pagani et al. (2011) evaluated regional differences in pCO₂ and pCO₂ trends over the EOT by contrasting alkenone carbon isotope values from six DSDP and ODP sites. The estimated pCO₂ values yielded highly variable results among the different sites, showing a general atmospheric pCO₂ decline from around 1200 to around 600 ppm throughout the Oligocene, with pCO₂ decreasing around 40 % at the EOT. Zhang et al. (2013) critically evaluated confounding factors of the alkenone pCO₂ proxy and excluded data from several locations, arriving at a continuous CO₂ record covering the past 40 Ma based on di-unsaturated alkenone ¹³C-fractionation at ODP Site 925 in the western tropical Atlantic Ocean. The general findings of Zhang et al. (2013) agreed with the pCO₂ trends reported in Pagani et al. (2005, 2011) but showed pCO₂ values to decrease from ∼1000 ppm at the EOT to ∼400 ppm in the late Oligocene. Lastly, Witkowski et al. (2018) compiled the longest consecutive pCO₂ record of the past ∼100 Ma, solely based on phytane ¹³C-fractionation from marine sediment and oil samples. Their results concur with the findings of Zhang et al. (2013), showing pCO₂ ranges from ∼600–1000 ppm throughout the Oligocene with a decreasing trend from the EOT towards the OMT (Fig. 1). Both Roth-Nebelsick et al. (2014) and Moraweck et al. (2019) used fossil leaf stomata to reconstruct Oligocene atmospheric pCO₂ levels. Oligocene fossil leaves of Platanus neptuni from various sites in Saxony (Germany) suggest lower pCO₂ levels than the alkenone-based results, with a modelled range of ∼400–600 ppm for the Oligocene (Roth-Nebelsick et al., 2014). Moraweck et al. (2019) reconstructed pCO₂ from the middle Eocene to the Oligocene using P. neptuni and Rhodomyrtophyllum reticulosum leaves from seven central European sites. They found a similar pCO₂ range to Roth-Nebelsick et al. (2014), with values also varying between ∼400–600 ppm in the Oligocene. Greenop et al. (2019) created the only available boron-isotope-based pCO₂ record; however they only focus on the OMT. While Greenop et al. (2019) did not find a strong decreasing trend over the OMT, they generally found low, stable values ranging from around 220 to 350 ppm, which increased to around 400 ppm after the OMT.

While there is a lot of variability between Oligocene pCO₂ records, with plant and boron data showing relatively stable pCO₂ levels, most other records suggest a steady decline in atmospheric pCO₂ towards the Miocene (CenCO2PIP, 2023).

4 Climate proxy data

We compiled marine and terrestrial climate proxy data to assess long-term trends and variability in climate across the Oligocene. For sea surface temperatures, we have added recently published records to the compilation of O'Brien et al. (2020). To assess terrestrial climate, we compiled published records of fossil plant remains, notably pollen, spores, and macro-remains (Table A1). Where the fossil plant remains had been assigned taxonomic affinities, the nearest living relatives (NLRs) were determined and used as input for NLR-based probability density modelling, following the methodology of Willard et al. (2019) and Reichgelt et al. (2023), to assess terrestrial palaeoclimate. We adopt the age determination from the original sources (Table A1), corrected for the GTS 2020 stage boundaries (Gradstein, 2020), where, if absolute age determination is unavailable, an average age was taken. Based on the distribution of the NLR for each fossil species, the probability of plant co-existence in an assemblage is calculated for 60 000 combinations of mean annual temperature (MAT), winter mean temperature (WinT), mean annual precipitation (MAP), and driest-month precipitation (DMP), as plant species distributions are sensitive to these variables and significant differences exist in the analysed plant groups for these variables (see Supplement). Up to 20 different plant taxa were compared at a time, and, where there were more than 20 taxa, sets of 10 were randomly chosen to maximise data variability. We report the highest-probability climate combination, and the uncertainty range is based on those climatic combinations with a probability of ≥2.5 % of the maximum probability combination.

We found 28 vegetation reconstructions of sufficient quality to assess palaeoclimate using the NLR method (see Table A1). The results can be assigned several potential quality “flags” based on diversity, depositional environment, and taxonomical assignments. Firstly, low convergence of multiple simulations of the same flora may suggest that the climate niche of one or multiple taxa has changed since the Oligocene (Reichgelt et al., 2023). Additionally, some floras had fewer than 20 taxa recorded, for which convergence could not be tested. Secondly, microfloras (pollen and spores) likely include upland or even extra-basinal input and are therefore less indicative of local climatic conditions than macrofloras (leaves, fruits, and flowers) (Reichgelt et al., 2023). Thirdly and finally, some palaeobotanical studies assign fossils to parataxa based on limited anatomical evidence or by using literature that is inappropriate for the study region. The majority of the data derives from NH mid-latitudes, a handful derives from SH mid-latitudes, and two datasets derive from the tropical realm. The absence of high-latitude data may be partly due to the lack of vegetation owing to cool conditions. However, there are pollen assemblages in sediments from the Antarctic margin (e.g., Askin and Raine, 2000; Prebble et al., 2005; Raine and Askin, 2001), but to our knowledge no quantitative data suitable for our NLR method have been generated for any high-latitude site.

4.1 Temperature

4.1.1 Continental mean annual temperature (MAT)

The data produced by the NLR allow a first assessment of the general Oligocene meridional temperature gradient on land. Unfortunately, the data are too sparse to assess trends or variability at any location, on any timescale (Fig. 3). Although the sparsity of data from low- and mid-latitude datasets limits our view on global gradients, the mid-latitude data can be compared to the modern and model simulations of the Oligocene and to reconstructed Eocene and Miocene gradients.

Figure 3 (a) Mean annual temperature (MAT; °C) plot over palaeolatitudes. In grey: pre-industrial (1900) MAT from Matsuura and Willmott (2018). (b) Winter temperature (WinT; °C) plot over palaeolatitudes. In grey: pre-industrial (1900) WinT from Matsuura and Willmott (2018). Darker colours represent a higher analytical certainty of the used site, and data with low reliability were excluded (see Table A1).

The two low-latitude data points suggest MATs of around 25 °C (±1.5 °C). At first sight, Oligocene low-latitude MATs were, on average, similar to modern MATs for the same latitudes. However, the NLR method is based on modern distributions; therefore reconstructed temperatures cannot exceed the global temperature maximum. Palaeobotanical temperature reconstructions from the tropics are susceptible to this problem and should therefore be regarded as minimum estimates. Most mid-latitude MAT reconstructions in both the NH and SH are between 12 and 17 °C, with an average error of ±2.9 °C (Fig. 3). Mid-latitude MATs during the Oligocene, particularly in the SH, are generally higher (up to 16 °C) than modern temperatures. We did not encounter suitable data for high-latitude regions.

The reconstructed winter temperatures (WinTs) range from ∼23 °C in the lower latitudes to ∼3 °C in the highest-latitude samples (∼52° N). Temperatures around the Equator reveal limited change in winter cooling (∼1.8 °C), while at higher latitudes the difference in temperature between WinT and MAT can be up to nearly 8 °C. Compared to modern values, Oligocene WinTs show the same trend as MATs, with similar values (possibly underestimations) around the Equator and higher temperatures in the mid-latitudes (Fig. 3). With the present dataset, it seems that Oligocene seasonality was similar to modern seasonality, with WinTs of the higher latitudes showing a difference of up to ∼12 °C compared to the MATs, thus reflecting a high temperature change between MATs and WinTs. Just as in the modern period, Oligocene WinTs had a large range at higher latitudes, with temperatures varying between ∼5–15 °C, whereas WinTs of lower latitudes barely show any temperature difference (Fig. 3).

4.1.2 Sea surface temperature

SSTs from three different proxies (U${}_{\mathrm{37}}^{{\mathrm{k}}^{\prime }}$, TEX₈₆, and biogenic calcite δ¹⁸O) were compiled for the low-, mid-, and high latitudes of the Oligocene (Fig. 4). The alkenone-based SST reconstruction (U${}_{\mathrm{37}}^{{\mathrm{k}}^{\prime }}$) relies on the temperature dependence of di- and tri-C₃₇ ketones (Prahl and Wakeham, 1987). At U${}_{\mathrm{37}}^{{\mathrm{k}}^{\prime }}$ values of >0.9 (at SSTs > 27 °C) the proportion of the tri-unsaturated C₃₇ alkenone becomes very low, virtually absent and/or undetectable, setting an upper limit for the application of this proxy (Tierney and Tingley, 2018). In addition, the low proportion of this alkenone introduces analytical uncertainties that cause noise. Consequently, we regard all U${}_{\mathrm{37}}^{{\mathrm{k}}^{\prime }}$ values > 0.9 as representing SSTs at or above 27 °C. Following the recommendations by Hollis et al. (2019), we use the calibration of Müller et al. (1998; see Table A2).

Figure 4 SST (°C) compilation for the Oligocene (CenCO2PIP, 2023). A linear interpolation was used between data points. Blue sites: SH high-latitude sites. Yellow sites: low-latitude sites. Green sites: NH high-latitude sites. See Fig. 2 for site locations and Table A3 for references.

The TEX₈₆ palaeothermometer is based on the temperature sensitivity of marine Thaumarchaeota membrane lipid (isoprenoid glycerol dialkyl glycerol tetraether (isoGDGT)) distributions (Schouten et al., 2002). The proportion of GDGTs containing a greater number of cyclopentane rings increases with higher temperatures and can thus be used to calculate SSTs using a modern surface sediment calibration (Wuchter et al., 2004). Discussion remains on how TEX₈₆ should be calibrated to represent seawater temperatures. The surface sediment calibration dataset shows virtually no response to temperatures below 15 °C, and it is debated if the response at the high-temperature end of the modern ocean – analogous to warmer climates in the past – can be assumed to be linear (e.g., O'Brien et al., 2017; Tierney and Tingley, 2014) or decreases exponentially (e.g., Cramwinckel et al., 2018; Kim et al., 2010). Moreover, isoGDGTs are barely produced in the mixed layer – they peak at ∼50–200 m depth and sometimes somewhat deeper (e.g., Hurley et al., 2018; van der Weijst et al., 2022). Most calibrations include surface ocean temperatures in their calibration datasets, leading to an overestimation of the proxy slope (Ho and Laepple, 2016). As the Oligocene was most likely warmer than today, we therefore prefer a conservative approach to assess SST, using an exponential calibration that has a drop in proxy response at higher temperatures. Even though it is associated with regression dilution (Tierney and Tingley, 2014), we use the TEX${}_{\mathrm{86}}^{\mathrm{H}}$ of Kim et al. (2010; see Table A2) to assess SST, rather than a linear model, for reasons outlined in Fokkema et al. (2024). Linear models produce much higher SSTs in the Oligocene TEX₈₆ range (Hollis et al., 2019). Moreover, any SST calibration assumes a similar relationship between surface temperature and the isoGDGT export zone in both modern and ancient oceans. Given the above uncertainties, it should be noted that absolute TEX₈₆-derived SST estimates come with large uncertainties.

Planktic foraminifer oxygen isotope ratios were also used to estimate Oligocene SSTs. This method is based on the direct correlation between the temperature-dependent fractionation of the oxygen isotopes ¹⁶O and ¹⁸O into biogenic calcite of foraminifera (Shackleton, 1974). Here, the calibration of Kim and O'Neil (1997) is used because it is based on inorganic calcite precipitated at temperatures between 10 and 40 °C and produces reliable results for foraminifera (Hollis et al., 2019). Foraminiferal calcite production has been found to decrease with increasing pH levels (Zeebe et al., 1999; Spero et al., 1997). The calibration of Kim and O'Neil (1997), may overestimate temperatures by up to 1.5 °C due to algal photosymbionts which modify the pH of the calcifying microenvironment (Spero and Williams, 1988). Although applying a direct correction is not recommended (Hollis et al., 2019) due to large uncertainties between symbiont activity levels, the influence of changing pH on SST reconstructions has to be considered.

Most available Oligocene SST data are from the mid-palaeolatitudes; records for the low- and high latitudes, especially in the NH, are scarce. Moreover, most records have low temporal resolution or cover only specific segments of the Oligocene (notably the EOT and OMT). The high latitude SSTs vary from 9.8 to 25.1 °C. It is worth noting that these records are restricted to latitudes no higher than 67° N and 68° S. SST estimates from mid-latitude locations have the largest temperature range, 6.0–32.1 °C, while SST estimates from low latitude sites span a narrower temperature range, 23.7–34.4 °C. The mid latitude SSTs show a slight increase (1–2 °C) between 34 and ∼27 Ma, followed by a small decrease of 1–2 °C towards 23 Ma (Fig. 4). However, overall, there is a remarkable absence of long-term trends in these SST records.

To assess long-term changes in temperature and meridional temperature gradients, we analyse data from three time slices: 33.7–33.2, 27.3–26.3, and 23.9–23 Ma, corresponding to the EOGM, MOGI, and OMT, which are averaged for an age of 33.4, 26.8, and 23.4 Ma, respectively. When SSTs are corrected for palaeolatitude (see Table A3), Oligocene SSTs are closer to late Eocene (∼38 Ma) than to modern values (Fig. 5). This is especially apparent in the Southern Ocean, where Oligocene SSTs are up to 10 °C warmer than modern SSTs. The high latitudes of the NH are challenging to assess due to data scarcity. However, the data available indicate that Oligocene SSTs were ∼2 °C colder than Eocene SSTs but still ∼4 °C warmer than modern SSTs. In contrast, low-latitude SST reconstructions show minimal differences, yielding similar temperature estimates for both the Oligocene and the Eocene. This leads to a flattened temperature gradient during the Oligocene between around 40° S and 40° N, where SSTs seem to be nearly the same.

Figure 5 Sea surface temperatures (SSTs; °C) over palaeolatitudes for 33.4 Ma (EOGM, black dots), 26.8 Ma (MOGI, blue triangles), and 23.4 Ma (OMT, orange squares). Brown shaded area: Baatsen et al. (2020) SST record for the late Eocene (38 Ma). Grey area: pre-industrial (1900) SST over latitude (Huang et al., 2015). Thick vertical error bars show the SST standard deviation, and thin vertical error bars represent the calibration error for each proxy. Larger symbols represent a higher data resolution, with larger symbols representing more data used and smaller points representing where fewer data were available. See Table A3 for data referral and references used.

4.2 Precipitation

Mean annual precipitation (MAP) and driest-month precipitation (DMP) were derived using the NLR approach (Table A1). The reconstructed MAP shows a range from ∼850 to 1750 mm yr⁻¹. The SH mid-latitudes show a generally higher MAP (∼1200–1750 mm yr⁻¹) for the Oligocene compared to the NH (∼850–1650 mm yr⁻¹). This differs from modern MAP values, where there is not a big discrepancy between SH and NH MAP. Generally, the Oligocene MAP values are higher than modern values, especially at mid-latitudes. The few data points in the tropics suggest a MAP that is similar to modern MAP. The values for the driest month range from ∼10 to 85 mm yr⁻¹, with generally lower values around the Equator and the NH (∼10–45 mm yr⁻¹) and higher DMP in the SH (∼20–85 mm yr⁻¹). The DMP values around the Equator are generally lower compared to modern values. In contrast to the Oligocene, the modern SH DMP values are on average lower than NH DMP values.

Figure 6 (a) Mean annual precipitation (MAP) plot over palaeolatitudes in mm yr⁻¹. In grey: pre-industrial (1900) MAP from Matsuura and Willmott (2018). (b) Driest-month precipitation (DMP) plot over palaeolatitudes in mm yr⁻¹. In grey: pre-industrial (1900) DMP from Matsuura and Willmott (2018). Darker colours represent a higher analytical certainty of the used site, and data with low reliability were excluded (see Table A1).

4.3 Data–model comparisons

The compiled surface temperature and precipitation data were regionally compared to the results from palaeoclimate model simulations (Figs. 7 and 8). Following the methodology described by O'Brien et al. (2020), two sets of modelling experiments were used: one from the NCAR Community Earth System Model version 1.0 (CESM1.0) and the other from the UK Hadley Centre Coupled Model version 3 (HadCM3L). The early and middle Oligocene simulations were performed using a ×3° nominal ocean and the T31 atmospheric resolution with varying glaciation conditions and pCO₂ of 560 and 1120 ppm. The late Oligocene simulation used a ×1° nominal ocean and 2° atmospheric resolution with a pCO₂ of 400 ppm. We compare data and simulations for three time slices: (a) early Oligocene, 33.9–33.0 Ma; (b) mid-Oligocene, 33.0–26.5 Ma; and (c) late Oligocene, 25.0–23.5 Ma (Fig. 7). For each time slice, the model ensemble mean is used to compare with the data, and the modelling details of the ensemble members are found in the Methods and Supporting Information Table S1 of O'Brien et al. (2020). The model annual mean values are derived from the nearest grid point to the study site.

Figure 7 Oligocene temperature data–model comparisons. All proxy data were compared to ensemble means from HadCM3L (UK) and CESM. (a–i) Sea surface and land temperature data–model comparisons for three Oligocene time slices. Data–model temperature difference data are displayed both spatially (a, b, d, e, g, h) and as zonal means (c, f, i). Temperature difference data in panels (a), (b), (d), (e), (g), and (h) are calculated as pointwise differences between the proxy mean value and the model annual mean. The pink and blue ribbons in panels (c), (f), and (i) represent the maximum and minimum differences associated with the zonal means. All proxy data are shown in Table A4.

Figure 8 Oligocene precipitation data (mm d⁻¹)–model comparisons. All proxy data were compared to ensemble means from HadCM3L (UK) and CESM. (a–i) Precipitation data–model comparisons for three Oligocene time slices. The data are displayed both spatially (a, b, d, e, g, h) and as zonal means (c, f, i) of the different precipitations. Differences in panels (a), (b), (d), (e), (g), and (h) are calculated as pointwise differences between the proxy mean value and the model annual mean and then plotted as points of difference. The pink and blue ribbons in panels (c), (f), and (i) represent the maximum and minimum differences associated with the zonal means. All proxy data are shown in Table A5.

4.3.1 Temperature proxy–model comparison

Despite utilising two very distinct models with different boundary conditions, the temperature discrepancy between the model and data is similar. The comparison of sea and land temperatures of all three time slices show that the mid- and high-latitude proxy data generally suggest warmer local conditions than simulations predict (Fig. 7). For all investigated time slices, that discrepancy is largest in the North Atlantic and southwestern Pacific. Additionally, for every time slice, the tropics seem to be warmer in model simulations than what local proxy data find, with the most extreme discrepancies in continental southeastern Asia. For the early and middle Oligocene, there is a difference between the modelled higher latitudes and measured data. In both the early and middle Oligocene, the higher latitudes seem to be a lot warmer (up to 20 °C) in the proxy data than what the models predict. The lower latitudes, on the other hand, for both the early and middle Oligocene, seem somewhat colder (around 5 °C) than what models predict. In the early Oligocene, North America generally shows a similar temperature range to the European sites, with proxy data indicating warmer conditions compared to the model. This shifts in the middle Oligocene, where most of the recorded North American sites are colder than what the models predict. The late Oligocene seems to have a similar offset in the SH high latitudes, with reconstructed temperatures being up to nearly 20 °C higher than the model results. Due to the lack of proxy data in the NH high latitudes and the tropics, temperature differences between models and records cannot be determined for the late Oligocene.

4.3.2 Precipitation proxy–model comparison

Comparison of precipitation proxy data to modelled simulations shows that, for all three simulated times, models mostly slightly underestimate the daily precipitation on a global scale (Fig. 8). In particular, the SH mid- to lower latitudes seem to be much drier in the models that what the proxy data suggest. For the early Oligocene, only NH tropical and NH mid-latitude data are available, indicating a slightly wetter (300–400 mm yr⁻¹) climate in Europe, whereas eastern Asia and western North America appear to be drier (300–900 mm yr⁻¹) than model predictions. Due to limited proxy data, we cannot make definitive statements about the early Oligocene in North America and eastern Asia. In the middle Oligocene, although more proxy data are available compared to the early Oligocene, the patterns are similar. Compared to the model results, the proxy data suggest wetter climates (300–400 mm yr⁻¹) in Europe and eastern Asia (600–900 mm yr⁻¹) and somewhat drier conditions (300–900 mm yr⁻¹) in North America. In both the middle and late Oligocene, model simulations appear to underestimate precipitation in SH mid- to high latitudes (300–900 mm yr⁻¹) (Fig. 8). Similarly to the early and mid-Oligocene, late Oligocene precipitation over central Europe and the region corresponding to today's Middle East also appears to be underestimated by models (300–900 mm yr⁻¹), although precise quantifications cannot be made due to the lower proxy resolution.

5 Discussion

5.1 Temperature trends and variability

Not only does our compilation of data show that the Oligocene oceans were warm (O'Brien et al., 2020), but the sparsely available floral and faunal information from terrestrial settings shows pronounced warmth (Fig. 3). On average, the data indicate that the Oligocene was slightly cooler than the Eocene but much warmer than modern times (Figs. 3 and 5). Notably, the high latitudes were much warmer than in modern times. The current dataset suggests no or only modest tropical warming relative to the present, but it should be noted that this is based on few records and that the expected ocean warm pools have not yet been sampled. Both sea surface and terrestrial temperature gradients between the tropics and the SH midlatitudes were particularly small, which was previously recognised for the Eocene and for the Miocene (e.g., Baatsen et al., 2020; Burls et al., 2021; Hollis et al., 2019; Lunt et al., 2021). For both Eocene and Oligocene SSTs, this is mainly the result of very high proxy values in the southwestern Pacific and the Australo-Antarctic Gulf (Baatsen et al., 2020; O'Brien et al., 2020). The gradient is very steep beyond 50° S (Fig. 5), while the Equator-to-pole temperature gradient in the NH is more gradual. This means that, despite the presence of ice in Antarctica, the mid-latitudes of the SH seem to be especially warm in the Oligocene.

Single-site high-resolution benthic foraminifer isotope data show a gradual decrease in δ¹⁸O values across the late Oligocene up to the onset of the OMT (Fig. 1; De Vleeschouwer et al., 2017; Liebrand et al., 2016; Pälike et al., 2006b). This suggests pronounced warming during the late Oligocene, termed late Oligocene warming. Regionally, this warming is supported by biogeographical information (De Man and Simaeys, 2004). However, the compiled temperature records show no consistent evidence for long-term warming throughout the mid- to late Oligocene. Rather, they show relatively stable values, with only few records (e.g., ODP 925, 744) indicating cooling from the early to mid-Oligocene, followed by long-term warming from the mid- to late Oligocene (Fig. 4). This suggests that the decrease in δ¹⁸O values may reflect regional warming at deep-water formation sites rather than global warming.

In the shorter term, there is strong variability within the temperature data, especially for the high and mid-latitudes (Fig. 4). Currently the resolution of all temperature records is insufficient to assess if any of this variability corresponds to orbital cyclicity. Similar variability apparent in deep-ocean benthic δ¹⁸O records primarily reflects eccentricity signals on timescales around 110 ka (Pälike et al., 2006b; Westerhold et al., 2020). If the temperature variability is global in nature, its increasing amplitude towards higher latitudes would likely reflect a combination of climatic polar amplification (i.e., ice and/or snow albedo and humidity feedbacks) and/or oceanographic variability (i.e., fronts and upwelling). However, due to the low resolution and lack of temperature records, especially in low latitudes and in the NH, the nature of and mechanism behind the temperature variability remain uncertain.

When compared to model simulations (Figs. 7 and 8), it is evident that the models show significantly less polar amplification of warming relative to the present day than in the proxy data. This is due to the extremely strong warming at higher latitudes and comparatively modest warming in the tropics in the proxy data compared to in the simulations. The dataset is limited, as floral data might underestimate temperatures in tropical regions that are warmer than in modern times (e.g., Huber and Caballero, 2011). It should also be noted that some of the SST data are based on TEX₈₆, which suffers from large uncertainties in absolute SST reconstructions (see Sect. 4.1.2). However, plant-based (NLR) and U${}_{\mathrm{37}}^{{\mathrm{k}}^{\prime }}$-derived temperatures show similar warming to those based on TEX₈₆ in the high latitudes, corroborating strong polar amplification. In other parts, climate models are not able to fully reconstruct regional climate variations as closely as proxy data can and thus probably underestimate regional variability, particularly on land (e.g., Laepple et al., 2023). Additional high-quality SST data are necessary to fully evaluate tropical temperatures for the Oligocene.

In particular, the shallow gradient between the Equator and ∼40° N and S is difficult to reconcile with the simulations. Floral data and carbonate geochemical records support exceptional mid- to high-latitude warmth during the Eocene (e.g., Creech et al., 2010; Douglas et al., 2014; Willard et al., 2019), which still seems to be the case during the Oligocene. Collectively, the nature of the data might underestimate the latitudinal heat transfer for the Oligocene. Regardless, our findings agree with those of O'Brien et al. (2020), showing that most of the Oligocene was similar to the late Eocene greenhouse world. Like in the late Eocene (Baatsen et al., 2024), the recorded temperatures are difficult to reconcile with the formation and persistence of a large ice sheet on Antarctica. The model–data comparison highlights the ongoing challenges of fully understanding the complex nature of the Oligocene. Questions remain regarding the formation of ice in a world with a flattened meridional temperature gradient, when poles were much warmer than today and atmospheric CO₂ levels were high (e.g., Baatsen et al., 2020, 2024).

5.2 Precipitation

Surprisingly, the few MAP data points at low latitudes are similar to today (Fig. 6), whereas mid-latitude MAP is higher compared to pre-industrial values. DMP for the Oligocene low latitudes is especially low compared to the pre-industrial period. The Southern Hemisphere mid- to high-latitude DMP (∼100 mm yr⁻¹) is higher than pre-industrial records (<100 mm yr⁻¹), whereas the DMP for the Northern Hemisphere is lower (∼50 mm yr⁻¹), which is more in line with the pre-industrial records (0–250 mm yr⁻¹). During periods that are globally warmer than today, the tropics and the mid-latitudinal zone of converging westerlies (30–60° latitude) would be expected to be wetter than today, while the subtropics would be expected to be drier (e.g., Pierrehumbert et al., 2002). This is consistent with MAP reconstructions, especially at >30° (Fig. 6a), whereas the data from lower latitudes are less clear, but that is largely due to a dearth of data points. Overall, Northern Hemisphere MAP is lower than that of the Southern Hemisphere. This may be due to the prevalence of Northern Hemisphere continental climate systems in the Oligocene (e.g., Sun et al., 2014). The Oligocene reconstructed MAP is somewhat lower than that of the Eocene (Cramwinckel et al., 2022). This drier climate compared to the Eocene is also seen in other terrestrial records (e.g., Couvreur et al., 2021; Dupont-Nivet et al., 2007; Jaramillo et al., 2006; Kohn et al., 2015; Ma et al., 2012; Salard-Cheboldaeff, 1979; Sun et al., 2014), which suggests the expansion of arid regions and the reduction of rainforests during the Oligocene.

In addition, comparisons to modelled data show that the models underestimate, particularly in higher latitudes, precipitation levels from 300 up to over 900 mm yr⁻¹. Palaeoclimate models homogenise meso- and microclimates due to the large grid size, which leads to an averaged topography and thus less spatial precipitation variability. Proxy data, on the other hand, may experience a bias to wetter environments, as there is more plant data available where wetter climates persisted, including those in meso- or microclimates that were unrepresentative of the macroclimate, such as a riparian environment. Similar to temperature (Sect. 6.1), our understanding of global Oligocene precipitation relies on a limited dataset, mainly sampling Europe (Fig. 2); therefore a first-order goal to a more comprehensive understanding of Oligocene palaeoclimates would be to generate more terrestrial data from other continents, particularly from high-latitude regions.

6 Conclusions and outlook for future work

While the palaeogeography of the Oligocene differs from today, it still poses a useful analogue to a projected future unipolar climate state. It is becoming more and more clear that the Oligocene icehouse was in fact warmer than previously thought, particularly in mid- and high latitudes. However, the present dataset suggests that the tropical band was not much warmer than today. This is contrary to model simulations, which predict that tropical regions for the Oligocene were warmer than in modern times and are thereby most likely to underestimate polar amplification and subsequent Equator-to-pole heat distribution. From this perspective, the Oligocene also contrasts with reconstructions of tropical temperatures for the Miocene and Eocene (e.g., Steinthorsdottir et al., 2021; Hollis et al., 2019). Consequently, tropical climates during the Oligocene require further investigation in addition to the present dataset, which is notably based on biomarkers, so proxies based on well-preserved biogenic calcite derived from surface oceans are crucial. In contrast to tropical regions, proxy records at mid- and high-latitude regions suggest extreme warmth, as also noted for the Miocene and Eocene epochs (e.g., Steinthorsdottir et al., 2021; Hollis et al., 2019). The low temperature gradients indicated by our data remain a model–data mismatch that requires solving due to its implications for the processes governing polar amplification of greenhouse-gas-driven warming and the magnitude thereof.

The limited vegetation data available support higher precipitation in mid-latitude regions during the Oligocene, as predicted by theory and model simulations. However, the continental dataset is limited, and more data are required particularly for the tropical band and the high latitudes to test for tropical hydrology and high-latitude winter temperatures, respectively.

Our temperature compilation does not show systematic long-term changes in temperature during the Oligocene. With the typically low resolution of the available long-term records, temperatures remained approximately stable. This is at odds with the long-term trends in the benthic foraminifera δ¹⁸O record, which shows signs of early Oligocene cooling and late Oligocene warming, albeit low in magnitude. Higher-resolution SST reconstructions at multiple locations are required to fully evaluate if those trends are absent on Earth's surface. Moreover, the apparent absence of long-term trends in SST and the nature of the minor trends in benthic foraminifera δ¹⁸O are both inconsistent with the recorded long-term drop in atmospheric pCO₂. This is a truly interesting conundrum and one that requires long-term high resolution pCO₂ reconstructions to fully evaluate. Although the benthic foraminifera δ¹⁸O records have identified the orbital-scale dynamics of deep-ocean temperature and/or continental ice volume in detail, orbital-scale climate variability of the Oligocene on the surface is poorly constrained. SST records resolving orbital-scale variability are required to ultimately characterise the nature of global mean surface temperature variability and the magnitude of polar amplification, its dependence on atmospheric pCO₂, and its relation to global continental ice volume.

Finally, an outstanding question remains on the relation of climate variability and subsequent biotic change. There is ample micropalaeontological evidence for a biotic response to orbital-scale temperature variability for the Oligocene (e.g., De Man and Simaeys, 2004; Śliwińska et al., 2010; Hoem et al., 2021; Fenero et al., 2013). The interplay of long-term climate stability and superimposed orbital-scale variability provides a very interesting opportunity to investigate the systematic relation between such climate variability and biotic resilience. This requires long-term, high-resolution micropalaeontological records but might ultimately result in much better-defined thresholds of massive regime shifts.

Appendix A

Table A1 Results of the nearest living relative (NLR) analysis, showing mean annual temperature (MAT), winter mean temperatures (WinT), mean annual precipitation (MAP), driest-month precipitation (DMP), and their respective minimum and maximum values.

Table A2 Summary of the calibrations and references thereof used for the respective sea surface temperature (SST) proxies.

Table A3 Summary of all compiled sea surface temperatures (SSTs) for all site locations including the analytical and calibration errors used for each proxy. Top: available SST data for the Oligocene Miocene Transition (OMT). Middle: available SST data for the Mid-Oligocene Glacial Interval (MOGI). Bottom: available SST data for the Eocene Oligocene Glacial Maximum (EOGM).

Table A4 All sea surface temperature (SST) data per site location that were added to the O'Brien et al. (2020) data–model comparison, including standard deviations and lower quartile (LQ) and upper quartile (UQ) errors. Top: all available SST data between 33.9–33 Ma. Middle: all available SST data between 22–26.5 Ma. Bottom: all available SST data between 25–23 Ma.

Table A5 All precipitation (MAP) data per site location that were used in the data–model comparison, including standard deviations and lower quartile (LQ) and upper quartile (UQ) errors. Top: all available MAP data between 33.9–33 Ma. Middle: all available MAP data between 22–26.5 Ma. Bottom: all available MAP data between 25–23 Ma.