Samples are from three ODP holes recovered during Leg 130 in the WEP (Fig. 1, Table 1): Hole 806A (0∘19.140′ N, 159∘21.660′ E, 2520.7 m water depth), Hole 806B (0∘19.110′ N, 159∘21.660′ E, 2519.9 m water depth) and Hole 807A (3∘ 36.420′ N, 156∘ 37.500′ E, 2803.8 m water depth) (Shipboard Scientific Party, 1991). Sites 806 and 807 are not likely to have experienced major tectonic changes over the last 20 million years.

Microfossils in sediments at these sites, as with any sedimentary sequences, have the potential to be influenced by diagenesis. Despite evidence of authigenic carbonate formation, recent modeling work concluded that the influence of dissolution and reprecipitation at Sites 806 and 807 was relatively minor (Mitnick et al., 2018​​​​​​​). Prior work has also found minimal impacts on the B/Ca ratio of Pliocene foraminifera from Site 806 (White and Ravelo, 2020) and on the Mg/Ca ratio of Miocene Dentoglobigerina altispera shells at Site 806 (Sosdian et al., 2020). The weight-to-shell ratio is commonly used to monitor dissolution, and the only published record at Site 806 for the Pliocene does not show a trend consistent with dissolution of T. sacculifer (Wara et al., 2005). We do note that while the “coccolith size-free dissolution” index reported in Si and Rosenthal (2019) indicates higher dissolution rates in the Miocene, their records were thought to be biased from changes in foraminifera assemblages as discussed in White and Ravelo (2020).

To further assess the potential impact of dissolution in our geochemical data, the weight-to-shell ratio was examined in our samples. The weight-to-shell data used to monitor dissolution does not exhibit any trend within the interval studied consistent with dissolution. Absolute weight-to-shell is increasing in the Miocene, which is not consistent with dissolution influencing the record (Fig. 2e). Additionally, reconstructed pH and pCO2 values also exhibit reasonable correspondence with the ice core data. Downcore δ11B values from Sites 806 and 807 are similar, despite evidence for higher authigenic carbonate at Site 807 relative to Site 806 (Mitnick et al., 2018). Further, despite different sedimentation rates, our δ11B and Mg/Ca results are consistent between Sites 806 and 807, and with data from Site 872 (Sosdian et al., 2018), which implies that diagenesis is not a primary driver of the reconstructed trends. A comparison of raw data as well as derived parameters is shown in Figs. 2 and 7.

The age model for Site 806 from 0–1.35 Ma is based on Medina-Elizalde and Lea (2005); calculated ages correspond well with ages from the Lisiecki and Raymo LR04 stack (Fig. 2a). The fourth polynomial regression-based biostratigraphy from Lear et al. (2015) was used for the rest of the record, following other work (Sosdian et al., 2020). Ages for Site 807 are based on published biostratigraphy (Berger et al., 1993) with additional constraints placed by Zhang et al. (2007) for the interval from 0–0.55 Ma. Benthic δ18O values from Sites 806 and 807 show good correspondence for the last 0.55 Myr, and the low-resolution benthic δ18O record for Site 806 (Lear et al., 2003, 2015) is consistent with the stack from Lisiecki and Raymo (2005) for the period studied (Fig. 3).

Samples were picked and cleaned to remove clays at UCLA (Los Angeles, CA) and the University of Western Brittany (Plouzané, France). A total of 50–100 foraminifera shells were picked from the 300–400 µm fraction size for T. sacculifer (without sacc) and from the 250–300 µm for G. ruber (white sensu stricto). Picked foraminifera were gently crushed, clays were removed, and they were checked for coarse-grained silicates. Samples were then cleaned using a full reductive and oxidative cleaning protocol following Barker et al. (2003). A final leach step with 0.001N HCl was done prior to dissolution in 1N HCl. Boron purification used a published microdistillation protocol (see Misra et al., 2014b; Guillermic et al., 2020, for more detailed methods).

Chemical separation was performed in a boron-free clean lab at the University of Cambridge (Cambridge, UK). Calcium concentrations were measured on an ICP-AES^® Ultima 2 HORIBA at the Pôle Spectrometrie Océan (PSO), UMR6538 (Plouzané, France). Elemental ratios (e.g., X/Ca ratios) were analyzed on a Thermo Scientific^® Element XR HR-ICP-MS at the PSO, Ifremer (Plouzané, France). Boron isotopic measurements were carried out on a Thermo Scientific^® Neptune+ MC-ICP-MS equipped with 1013 Ohm resistor amplifiers (Lloyd et al., 2018) at the University of Cambridge (Cambridge, UK).

Variations in B isotope ratios are expressed in conventional delta (δ) notation with δ11B values reported against the reference standard NIST SRM 951 (NIST, Gaithersburg, MD, USA): 1δ11B(‰)=1000×11B/10BSample11B/10BNIST SRM 951-1.

Multiple analyses of external standards were performed to ensure data quality. For boron isotopic measurements, JCp-1 ​​​​​​​(Geological Survey of Japan, Tsukuba, Japan, Gutjahr et al., 2020) was used as a carbonate standard, and NEP, a Porites sp. coral from University of Western Australia and Australian National University was also used (McCulloch et al., 2014). A boron isotope liquid standard, ERM© AE121 (certified δ11B = 19.9 ± 0.6 ‰, SD), was used to monitor reproducibility and drift during each session (Vogl and Rosner, 2012; Foster et al., 2013; Misra et al., 2014b). For trace elements, external reproducibility was determined using the consistency standard Cam-Wuellerstorfi (University of Cambridge) (Misra et al., 2014a).

Detailed calculations can be found in the Supplement. Briefly, Mg/Ca was used to reconstruct sea surface temperature (SST) using the framework from Gray and Evans (2019) correcting for influences of pH, salinity and secular variation in seawater Mg/Ca. δ11Bcarbonate was corrected using an empirical δ11Bcarbonate weight-to-shell ratio relationship. δ11Bborate was determined using species-dependent sensitivities of δ11Bcarbonate to δ11Bborate (Guillermic et al., 2020). pH was calculated using δ11Bborate with different scenarios of secular seawater δ11B changes (Lemarchand et al., 2002; Raitzsch and Hönisch, 2013; Greenop et al., 2017). pCO2 was reconstructed using pH-based δ11Bcarbonate and different scenarios of alkalinity (Tyrrell and Zeebe, 2004; Ridgwell and Zeebe, 2005; Caves et al., 2016 and Rae et al., 2021). Further details including equations are provided in the Supplement.

Geochemical data used in this study are presented in Fig. 2. Mg/Ca data (Fig. 2c) are consistent with previously published Mg/Ca values for Site 806 on T. sacculifer (Wara et al., 2005; Tripati et al., 2009; Nathan and Leckie, 2009). Although the record we generated does not overlap with Site 872, they are 1 Myr apart (15.7 and 16.7 Ma); there is a good correspondence between our Mg/Ca data and the published Mg/Ca record from T. trilobus at Site 872 (Sosdian et al., 2018). Mg/Ca from a different species, D. altispera (Sosdian et al., 2020), is also plotted with an offset, for comparison.

Comparison with Site 872 data that are part of the compilation from Sosdian et al. (2018) shows that their δ11B data are in line with our dataset (Fig. 2b), and all sites examined in the WEP (Sites 806, 807, and 872) are above the lysocline (Shipboard Scientific Party, 1991). The δ11B data for T. sacculifer exhibit a significant increase (4.2 ‰) from the Miocene to the present. Figure 2b also compares the δ11B data used in this study with published data from other sites and shows that raw δ11B data for the WEP can be lower than values for other regions.

We sought to assess whether there is evidence for air–sea equilibrium or disequilibrium in the WEP during the large amplitude late Pleistocene glacial–interglacial cycles, in order to validate our approach. We reconstructed pCO2 for the last 800 kyr (n=16, Fig. 3). For the last 800 kyr, reconstructed pCO2 values for Sites 806 and 807 are in the range of ice cores (Fig. 3, Petit et al., 1999; Siegenthaler et al., 2005; Lüthi et al., 2008; compilation from Bereiter et al., 2015). The two critical diagnostics we used for method validation are (1) that the δ11B-based reconstruction of pCO2 is consistent with ice core atmospheric CO2 and (2) that the boron-based reconstruction empirically reproduces interglacial–glacial amplitudes from ice cores. Figure 3b shows that both of these criteria are met despite large scatter. We also created a cross plot comparing these two independent constraints on pCO2 (Fig. 3c). Two regressions between ice core pCO2 and boron-based pCO2 are shown, a simple linear regression (grey line) and a Deming regression that takes into account error in variables (blue line). Bootstrapping was used to calculate uncertainties in the regression models (n=1000, Fig. 3c, Table S6). While slopes and intercepts are not statistically different from a 1:1 line, the regressions do not reach a high significance level (p=0.25); boosting the resolution of the record could help provide better constraints for this type of comparison. No significant difference in variability was observed at either site. The age models for the sites are based on comparisons of the benthic δ18O records for both Sites 806 and 807 (Fig. 3a, Zhang et al., 2007; Lear et al., 2003, 2015) to the published isotopic stack (Lisiecki and Raymo, 2005).

We also note that reconstructed pCO2 uncertainties (both accuracy and precision) could potentially arise from Mg/Ca-derived estimates of temperature; these uncertainties could be reduced using independent temperature proxies for the WEP such as clumped isotope thermometry (Tripati et al., 2010​​​​​​​, 2014), a technique that is not sensitive to the same sources of error as Mg/Ca thermometry and therefore is an area planned for future work. Other sources of uncertainty that have a larger effect on pCO2 calculations are the weight–shell correction, while the TA and seawater boron isotope composition have a minor effect over this time interval.

Between MIS 7 and 6, our reconstructions exhibit a decrease in temperature (ΔT) of 1.2 ∘C, an increase in pH (ΔpH) of 0.08 and a decrease in pCO2 (ΔpCO2) of 58 ppm. Between stage 3 and 1, we observed an increase in temperature of 2.0 ∘C, a decrease in pH of 0.13 and an increase in pCO2 of 76 ppm. We also compare results with recent reconstructions in Figs. S1 and S2 (Sosdian et al., 2018; Rae et al., 2021). These results highlight that we are able to reproduce the range of atmospheric pCO2 in the ice core record and reproduce the amplitude of changes between transitions, with uncertainties typical for this type of work (Hönisch et al., 2019).

Mg/Ca data are consistent at Site 806 (Wara et al., 2005; Tripati et al., 2009, 2011; Nathan and Leckie, 2009) and Site 872 (Sosdian et al., 2018) in the WEP. The Mg/Ca in T. sacculifer has to date not shown a pH dependency (Gray and Evans, 2019), but Mg/Ca of G. ruber does and was therefore corrected from this effect (see Supplement). Data for both species were corrected from salinity and seawater Mg/Ca changes. Mg/Ca temperatures for Site 872 were reconstructed using published data and the same framework we use here and are presented in Fig. 4. Recalculated values for Site 872 are from D. altispera, with an offset applied relative to T. sacculifer, and show similar variations to our record for the Miocene Climate Optimum and Middle Miocene Climate Transition (MCO–MMCT) periods (Sosdian et al., 2020). Temperatures from Tex86 and U37K′ are plotted for comparison but those records are limited to the last 12 and 5 Myr, respectively (Zhang et al., 2014).

The Mg/Ca data support high temperatures of 35.2 ± 1.3 ∘C (2 SD, n=11) for the early Miocene until the MMCT, with a relatively small (ca. 1 ∘C) change into the MCO, and larger changes out of the MCO. Similarly warm SSTs for the MCO were reconstructed in the North Atlantic at Site 608 from Tex86 (Super et al., 2018). Despite a gap in our compilation from 11.5 to 9.5 Ma, there is a SST decrease of ∼6 ∘C from the MCO to ∼7 Ma where temperatures similar to present-day values are observed. A decline in temperature during the MMCT is coincident with the timing of the constriction of the Indonesian Seaway, the pre-closure of the trans-equatorial circulation and subsequent formation of a proto-warm pool (Nathan and Leckie, 2009; Sosdian et al., 2020). From 12 to 7 Ma, the Mg/Ca SST record diverges from Tex86 and U37K′-based reconstructions, with higher temperatures. At the same time, a record for the North Atlantic showed a decrease of ∼10 ∘C from the MCO to ∼9 Ma (Super et al., 2018). From 7 Ma to the present, the record from multiple proxies – Mg/Ca, Tex86 and U37K′ – in the WEP agree.

Figures 5 and 6 show the different histories of seawater δ11B and alkalinity used in our calculations, respectively. Details of calculations are provided in the Supplement methods. Following the approach of Tripati et al. (2009, 2014) and recent literature (Sosdian et al., 2018; Rae et al., 2021), we explored multiple scenarios for the evolution of seawater boron geochemistry (Fig. 5) and alkalinity for calculations of pCO2 (Figs. 6, S1 and S2). During the interval overlapping with the ice core record, we observe that the choice of model used does not make a significant difference in reconstructed values. During earlier time intervals, we see there is a greater divergence, reflecting larger uncertainties in seawater δ11B and alkalinity further back in Earth history.

Prior to 10 Ma and during the early Pliocene (∼4.5 to 3.5 Ma), calculations of pCO2 diverge from published values largely because of the different assumptions each study has used for past seawater δ11B (Fig. 5). However, we find that when the uncertainty in reconstructed pH is fully propagated, the differences in reconstructed pH values calculated using each of the δ11Bseawater histories is not significantly different (Figs. 5 and 6; see also Hönisch et al., 2019). In contrast to the results from Greenop et al. (2017), the record from Raitzsch and Hönisch (2013) exhibits substantial variations on shorter timescales. Such variability is a challenge to reconcile with the Li isotope record of Misra and Froelich (2012), given that Li has a shorter residence time than boron while having similar sources and sinks. For the remainder of this study, we use the δ11Bseawater history from Greenop et al. (2017) because it is in good agreement with seawater δ7Li (Misra and Froelich, 2012). The recent calculations of seawater pH (Sosdian et al., 2018; Rae et al., 2021) agree with values from our study when uncertainties are taken into account (Fig. 5).

The four alkalinity models used in this study diverge prior to 9 Ma, with a maximum difference at ∼13 Ma that is also reflected in reconstructed pCO2 values (Fig. 6). However, all four models yield pCO2 estimates that are within error of each other when the full uncertainty is considered. Uncertainty in the evolution of seawater alkalinity and seawater δ11B leads to differences in the absolute values of reconstructed pCO2 (Fig. S2), and a divergence in reconstructed pCO2 values that is largest in the Miocene. The two scenarios that produce the highest divergence in values are those calculated using constant alkalinity relative to those calculated using values from Caves et al. (2016), with a maximum difference at 15.06 Ma of up to 250 ppm CO2, and with the latter model producing lower values (Fig. 6b and e). Thus, for the MCO, alkalinity is a critical parameter in calculations of absolute pCO2 values. For the Miocene and earlier intervals, improved constraints on past secular variations of seawater δ11B and alkalinity will yield more accurate reconstructions of pCO2.

For the remainder of this paper, we use the model of Caves et al. (2016) to estimate alkalinity and δ11Bseawater determined by Greenop et al. (2017) (e.g., Fig. 6e). We note that two recent syntheses of boron isotope data have been published and compare our results to these findings (Figs. 8 and S2). Sosdian et al. (2018) report values that are in line with our results in the Miocene, but their study does not replicate results from ice cores. Rae et al. (2021) presents reconstructed values that are higher in the Miocene, due to the utilization of different scenarios of seawater δ11B and alkalinity compared to this work.

On million-year timescales, atmospheric CO2 is controlled by its input through mantle degassing in the form of sub-aerial and sub-aqueous volcanic activity and its removal by chemical weathering of continental silicate rocks. Over the last 16 Myr, two relative maxima in atmospheric pCO2 are observed in our record, one during the MCO (at 15.67 Ma) and a second around the late Miocene–early Pliocene (beginning at 4.7 and 4.5 Ma) (Fig. 11), though the timing for the latter is not precise. The strong pCO2 increase from the early Miocene to MCO occurs when there is increasing volcanic activity associated with the eruption of the Columbia River Flood Basalts (Hooper et al., 2002; Foster al., 2012; Kasbohm and Schoene, 2018), with recent geochronological evidence published supporting higher eruption activity between 16.7 and 15.9 Ma (Kasbohm and Schoene, 2018), reinforcing the idea of an episodic pCO2 increase during the MCO due to volcanic activity. Underestimation of net CO2 outgassing from specific continental flood basalt eruption is possible, as both sub-aqueous and sub-aerial flood basalts, under right climatic conditions, are prone to enhanced chemical weathering. For example, the 4 ‰–5 ‰ drop in δ7Li record at the Cretaceous–Paleogene (K-Pg) boundary (Misra and Froelich, 2012) is attributed to rapid quasi-congruent weathering of the Deccan Traps (Renne et al., 2015​​​​​​​) during their eruption. Courtillot and Renne (2003) estimate that about 50 % of emitted CO2, roughly equivalent to the amount emitted by the eruption of a million cubic kilometers of Deccan Traps, may be missing due to chemical and physical weathering. Additionally, the early Eocene (at ∼50 Ma) 3 ‰–4 ‰ rise in seawater δ7Li at a time where there is not significant uplift of the Himalayas (Misra and Froelich, 2012) is also attributed to incongruent weathering of previously erupted Deccan Trap basalts as the Indian subcontinent moved from arid mid-latitudes to the wet low latitudes (Kent and Muttoni, 2008). Thus, a significant part of the outgassed CO2 can be consumed by chemical weathering of freshly erupted hot basalts (Courtillot and Renne, 2003). However, the congruency of chemical weathering of basalts, depending on regional climatic conditions (warm–wet vs. cold–arid), will determine the shape and position of inflection points in the seawater δ7Li record. The possible quantification of increased rates of silicate weathering inferred from δ7Li (mentioned below) can be utilized to determine total eruptive volume (missing + existing) and volatile emissions from the Columbia River Flood Basalts. At the same time as continental flood basalt emissions, enhanced seafloor production could also be a second possible source of CO2; however, we note there is evidence that the rate of seafloor production has remained virtually invariant over the last 60 million years (Rowley, 2002; Müller et al., 2016).

The second CO2 peak can possibly be caused either by the observed increase in global volcanism during the early–middle Pliocene (Kennett and Thunell, 1977) and/or by a change in silicate weathering regime. Strontium and lithium isotopes (87/86Sr and δ7Li) have been used as proxies for silicate weathering flux and congruency. Although the strontium isotope record exhibits a monotonous increase, lithium isotope data (Misra and Froelich, 2012) are more variable, with a transition from a period of increasing seawater δ7Li (e.g., non-steady state weathering) to stable seawater δ7Li (e.g., steady state weathering) beginning at roughly 6.8 Ma (Fig. 11).

It is interesting to note that the rise in δ7Li (Fig. 11b) from the early Miocene to the MCO is synchronous with the rise in pCO2. Before 18.5 Ma, the pCO2 is relatively stable, and δ7Li is increasing, suggesting the non-steady-state and incongruent nature of continental chemical weathering. From 18.6 to 16.7 Ma, the δ7Li record decreases by ∼2 ‰, consistent with decreasing weathering rates and an associated increase in pCO2. Between 16.7 and 15.9 Ma, when the eruption of the Columbia River Flood Basalts is at a maximum, δ7Li increases, in line with higher weathering rates that could arise from higher atmospheric CO2 and the presence of fresh basalts. The δ7Li record then decreases again until the end of the MCO at ∼14.7 Ma, in line with a decrease in the eruption rate, sustaining high atmospheric CO2. A constant increase in δ7Li is then observed, until the early Pliocene, where there is evidence for a shift to a steady-state weathering regime. This increase in δ7Li is also consistent with the decrease in pCO2 observed until the early Pliocene.

We developed a reconstruction of atmospheric pCO2 based on δ11B of planktic foraminifera from ODP Sites 806 and 807 located in the Western Equatorial Pacific for the past 16 million years and extended the record to 22 Ma by reprocessing data from Site 872 (Sosdian et al., 2018). We build on past efforts to reconstruct atmospheric pCO2 using different proxies from this region, including from carbon isotopes in marine organic matter (Rayno and Horowitz, 1996) and alkenones (Pagani et al., 2010), as well as foraminiferal B/Ca ratios (Tripati et al., 2009, 2011), all of which have been shown to have a number of complexities and potential sources of systematic error (e.g., Tripati et al., 2011). It also builds on efforts to use boron isotopes in other regions using MC-ICP-MS (Seki et al., 2010; Foster et al., 2012, 2014; Greenop et al., 2014; Martínez-Botí et al., 2015b; Stap et al., 2016; Chalk et al., 2017; Dyez et al., 2018; de la Vega et al., 2020), and our recent work constraining fractionation factors and measuring small samples of foraminifera (Guillermic et al., 2020).

Our study contributes a new long-term reconstruction of atmospheric pCO2 for the Neogene derived from boron isotopes from the tropical Pacific Ocean. Although the record is not continuous, with variable resolution, it captures both long-term and short-term variability associated with several key transitions and demonstrates the utility of examining sites in the Western Equatorial Pacific for future higher-resolution studies. Results for Sites 806 and 807 in the Western Equatorial Pacific reproduce the amplitude of late Pleistocene glacial–interglacial cycles in pCO2. These observations are consistent with the sites being in equilibrium with the atmosphere, although further work would be useful to explore sources of uncertainty and differences relative to ice core pCO2.

pCO2 values increase from the early Miocene to the MCO with estimated MCO pCO2 values of 511 ± 201 ppm (2 SD, n=3). These elevated values are potentially linked to the eruption of the Columbia River Flood Basalts, with values declining into the early Pliocene, including during Pliocene glacial intensification. The changes in pCO2 we observed are in line with changes in δ7Li, a proxy of silicate weathering, and future modeling of multiple proxy records should be insightful. Early Pliocene data for ∼4.7–4.5 Ma support high pCO2 of 419 ± 119 ppm, and elevated values during the mid-Pliocene Warm Period of 530 ± 110 ppm for the time interval ∼3.3–3.0 Ma. These data are low in resolution, thereby not fully sampling orbital and millennial-scale variability. The higher-resolution record for the Pliocene glacial intensification supports a reduction in pCO2 during several steps, with values at 2.7 Ma of 350 ppm, 2.6 Ma of ∼ 280 ppm and 2.4 Ma of ∼210 ppm. We find support for a larger reduction in glacial pCO2 during the Mid-Pleistocene Transition compared to interglacial pCO2, and a minimum in pCO2 during glacial MIS 30. These findings confirm a role for CO2 in the transition from a 41 to a 100 kyr world.

Higher-resolution boron isotope records from the WEP would allow for further resolution of these changes. Additional constraints on temperature, such as from clumped isotopes (Tripati et al., 2010) in the WEP (Tripati et al., 2014), could allow for uncertainties in pCO2 estimates from boron isotopes to be reduced and for new constraints on Earth climate sensitivity. Future constraints on the vertical structure of the tropical Pacific (Shankle et al., 2021) during these transitions may also potentially be illuminating.