This research used samples from Ocean Drilling Program (ODP) Site 1092, located at 46.41∘ S and 7.08∘ E (Fig. 1) at a water depth of 1973 m. The interval studied (∼ 178–184 mcd; see Table S1 in the Supplement) consists of nannofossil ooze with excellent carbonate preservation, as shown by undissolved coccoliths, which are susceptible to corrosion (Fig. 2). The shown example is from ∼ 13.8 Ma, where pH was low, but shell preservation is similarly good throughout the entire record. From the 250–315 µm size fraction, approximately 200 specimens (∼ 2.5 mg) of the cold-water-dwelling planktonic foraminifer G. bulloides were picked for δ11B analysis from 35 processed 10 cm3 sediment samples. In addition, about five to eight specimens of the benthic foraminifer Cibicidoides wuellerstorfi were picked from 14 intervals to reconstruct deep-sea pH (Table S1). Today, surface-water pCO2 at Site 1092 is close to equilibrium with the atmosphere (ΔpCO2sea-air<-30 µatm; Takahashi et al., 1993, 2009). Plankton tow data revealed that G. bulloides in the Atlantic sector of the Southern Ocean is predominantly found within the upper 100 m of the water column (Mortyn and Charles, 2003). Based on DIC (dissolved inorganic carbon) and TA (total alkalinity) from seasonal TCO2+TALK (Goyet et al., 2000) and temperature and salinity from WOA13 data sets, modern gradients in pH and pCO2 are small at this site within the upper 200 m of the water column. During the austral spring season, when the shell flux of G. bulloides is highest (Jonkers and Kučera, 2015; Raitzsch et al., 2018), the differences in pH and pCO2 within this depth range are -0.03 and 17 µatm, respectively. Hence, we expect G. bulloides to be a reliable recorder of subsurface pH and pCO2.

The age model used in this study is adopted from Kuhnert et al. (2009) but was revised to bring it in line with the astronomically tuned stable isotope record of the reference IODP Site U1338 (Holbourn et al., 2014) (Fig. 3, Table S2). For direct comparison of our pCO2 record with the boron-based reconstructions available from the literature, the age models of ODP 761 (Holbourn et al., 2004; Foster et al., 2012; Sosdian et al., 2018) and the Blue Clay Formation of Ras-il-Pellegrin on Malta (Abels et al., 2005; Badger et al., 2013) were also retuned to match the reference curve U1338 (Table S2). Although the global stratotype section and point (GSSP) of the Langhian–Serravallian boundary (Mi-3b) in the Malta section was not explicitly placed at 13.82 Ma as suggested by Abels et al. (2005), the boundary in the revised age model (13.81 Ma) is in very close agreement, supporting the validity of the retuned record. Based on these revised age models, the δ18O and δ13C profiles of Sites 1092, ODP 791, and the Blue Clay Formation generally show a good agreement with those of Site U1338 (Fig. 3).

Foraminifer shells were cleaned following the protocol of Barker et al. (2003). Boron (B) isotope ratios were measured following Raitzsch et al. (2018). Briefly, the cleaned samples were dissolved in 60 µL of 1 N HNO3 and micro-distilled on a hotplate to separate boron from the carbonate matrix. The micro-distillation method has been proven to yield a B recovery of ∼ 100 %, a low procedural blank, and accurate results, even at low B concentrations (Gaillardet et al., 2001; Wang et al., 2010; Misra et al., 2014; Raitzsch et al., 2018). The procedural blank contribution was 10–50 pg B, which equates to ∼ 0.2 %–0.8 % of the total [B] in the micro-distillation vial. The distillate containing only boron was diluted with 2 % HNO3 and analyzed for isotopes in triplicate using a Nu Plasma II multi-collector inductively coupled plasma mass spectrometry (ICPMS) at AWI (Bremerhaven, Germany) that is equipped with a customized detector array of 16 Faraday cups and 6 secondary electron multipliers (SEM), also termed ion counters (IC). The 11B and 10B samples were collected in IC5 and IC0, respectively, at a boron concentration of ∼ 3 ppb. As three of the SEMs were later replaced by Daly detectors, 11B and 10B of C. wuellerstorfi samples were collected in D5 and D0, respectively, at a boron concentration of ∼ 2 ppb. 11B/10B was standardized against concentration-matched NBS 951 using the standard–sample–standard bracketing technique, and frequent analysis of control standard AE121 with an isotopic composition similar to that of foraminifers was monitored to ensure accuracy of measurement. Measurement uncertainties are reported as 2 standard deviations (2σ) derived from triplicate measurements or as ±0.30 ‰ (for SEM-analyzed samples) and ±0.25 ‰ (for Daly-analyzed samples) determined from the long-term reproducibility (2σ of per-session δ11B averages) of the control standard, whichever is larger (see Table S1).

To reconstruct pH and pCO2 for the middle Miocene from the boron isotopic composition of foraminiferal shells, more constraints and assumptions of boundary conditions need to be made, as the ocean chemistry was different from today. This includes the boron isotopic composition of seawater (δ11Bsw), which has a direct effect on that of foraminifer shells (δ11Bforam) but also on the δ11Bforam/δ11Bborate relationship, which in turn may lead to different pH yields. Further, we need to consider the seawater concentrations of calcium [Ca] and magnesium ions [Mg], which affect both the ocean's buffering capacity and the Mg / Ca ratios in foraminifer shells used to reconstruct seawater temperatures. Ultimately, we need to constrain a second carbonate system parameter (besides pH) to determine absolute atmospheric pCO2 values as accurately as possible.

Our boron-based record indicates a steady pCO2 increase from ∼ 380 to 520 ± 100 µatm between 14.25 and 14.08 Ma and a subsequent decrease to ∼ 320 µatm until 13.87 Ma, culminating after the onset of the major ice-sheet expansion (Fig. 4e). It should be noted that this ∼ 200 µatm drop in pCO2 occurred at a time of northward-shifting Southern Ocean fronts, which is only related to a step-like drop in SST and salinity around 14 Ma at this site (Fig. 4c) (Kuhnert et al., 2009). At 13.82 Ma, when SST had already reached a lower stable level and after the inception of the ∼ 1 ‰ rise in benthic δ18O (Fig. 4a), pCO2 increased rapidly by ∼ 100 µatm and displayed high-amplitude variations of more than 50 µatm until 13.57 Ma. This transient rise in pCO2 ended with a decrease to ∼ 320 µatm and much reduced variability after 13.53 Ma (Fig. 4e). Despite the high 2σ uncertainty of ∼ 100 µatm for the overall pCO2 record, the interpretation of comparatively small changes of ∼ 50 µatm is reasonable, due to the smaller uncertainty of relative variations. The latter is estimated to be approximately 50 µatm (2σ), i.e., when uncertainties in the δ11Bforam/δ11Bborate calibration and δ11Bsw are neglected. Sea-surface pH values within the entire record range from 7.94 to 8.11 ± 0.09, while deep-sea pH varies between 7.77 and 7.87 ± 0.08 over the same interval (Figs. 4d and S4). Interestingly, the pH offset between the surface and deep ocean is nearly identical to the modern gradient and apparently does not change substantially within our Miocene record (Fig. S4), but the high uncertainties associated with the pH estimates hamper a reasonable statistical analysis of the temporal change in gradient.

The pCO2 variations within the studied time interval show a remarkable agreement with variations in δ13C that correspond to CM5b and CM6 (Fig. 4). Accordingly, the maxima at ∼ 14.1 and 13.7 Ma as well as the minima at ∼ 13.9 and 13.5 Ma, indicate that atmospheric CO2 levels were paced by 400 kyr cycles, as also demonstrated by evolutive harmonic and power spectral analyses (Fig. 5), but due to the phase lag it is difficult to substantiate a direct link between long eccentricity cycles and atmospheric CO2.

Due to the profound effect of pH on Mg / Ca in shells of G. bulloides, which might bias reconstructed SSTs, we applied the modified MgCaRB tool of Gray and Evans (2019) to iteratively solve Mg / Ca temperatures and δ11B-derived pH. In addition, we applied a correction factor to account for the effect of [Mg / Ca]sw on shell Mg / Ca using the method of Evans and Müller (2012). However, to evaluate the difference between iteratively adjusted and non-adjusted pH and temperature values, we also performed the conventional calculations using the G. bulloides δ11B calibration from Raitzsch et al. (2018) and Mg / Ca-to-temperature component from Gray and Evans (2019), respectively. It should be noted that the Mg / Ca sensitivity of G. bulloides to temperature determined by Gray and Evans (2019) is much larger than that by Mashiotta et al. (1999). Therefore, the calculated unadjusted temperature decrease between 14.05 and 13.85 Ma is on the order of 13 ∘C, compared to ∼ 6 ∘C determined with the equation of Mashiotta et al. (1999) and reported in Kuhnert et al. (2009). After the iterative correction of Mg / Ca and pH, the results show that before the EAIS expansion the adjusted temperature and pCO2 values are up to ∼ 3 ∘C and ∼ 60 µatm lower, respectively (Fig. S3). On the contrary, after 13.9 Ma the adjusted temperature and pCO2 reconstructions are very similar, suggesting that approximately one-fourth of the observed pre-glacial Mg / Ca decrease was induced by a contemporaneous pH rise.

Given that we use TA as the second variable to calculate the inorganic carbon chemistry, which might affect calculated pCO2 discernibly, we carried out a number of sensitivity tests. The first tests demonstrate that either constant salinity, a ∼ 1 unit change in salinity, or TA co-varying with salinity using a modern TA–S relationship (∼ 42 µ mol kg-1 per S unit), has no discernible effect on pCO2 estimates (Fig. S5a). As ODP Site 1092 was apparently influenced by a northward migration of the Southern Ocean fronts, we tested the potential impact on calculated pCO2 by coupling TA to temperature. Even at an unrealistically high total change of 400 µmol kg-1, calculated relative pCO2 changes are similar after the EAIS expansion at ∼ 13.9 Ma (Fig. S5b). Given the modern sea-surface TA gradient across the polar frontal system of ∼ 100 µmol kg-1 (higher in the south), we conclude that the increasing influence of southern-sourced surface waters on pCO2 estimates at our study site is comparatively small. When in our simulations TA was coupled to pH (Fig. S5c, d) or eccentricity (Fig. S5e–h) by ±100 µmol kg-1, pCO2 reconstructions are almost identical among each other. On the other hand, at TA variations of ±400 µmol kg-1, the pCO2 estimates may differ substantially between the different scenarios, but the general shape of the record remains more or less the same. Given that total variations in TA of 800 µmol kg-1 are very unlikely, we suggest that potential variations in TA do not affect our pCO2 estimates significantly.

At face value, the reconstructed pCO2 curve displays a remarkable resemblance to the global benthic δ13C record (Fig. 4), suggesting that there is a first-order link between the two. Our record, however, is the first that shows an increase of pCO2 after the onset of Antarctic glaciation, which is counterintuitive at first glance, given that most hypotheses refer to enhanced Corg burial as the main mechanism having caused the rise in δ13C and presumably an atmospheric CO2 drawdown. Hence, this raises the question as to whether Site 1092 truly records a global pCO2 signal or a local one instead. It is difficult to tackle this question because we neither have a complete, independent pCO2 reconstruction for this time interval at similar or higher resolution for comparison, nor can we ascertain that any site millions of years ago was not influenced by local processes affecting the geochemical signal. Therefore, we cannot completely rule out a local or regional imprint on our pCO2 record, particularly due to the proximity of Site 1092 to the polar front (Fig. 1).

There is indication for development of the subantarctic ocean frontal system and a better stratification of the upper ocean at Site 1092 following Antarctic ice-sheet expansion, suggested by a divergence in δ18O between shallow- and deep-dwelling planktonic foraminifera (Paulsen, 2005). Although the occurrence of ocean fronts should have involved changes in biological productivity that may result in altered surface-water [CO2], the opal and Corg contents in the sediment remain steadily low throughout the middle Miocene (Diekmann et al., 2003), suggesting that Site 1092 was in an oligotrophic setting within the time interval studied. Moreover, if the δ11B signal was mainly driven by a regional change in stratification after 13.85 Ma, this would not explain the low pH earlier during CM5, as shown in our record (Fig. 4).

Further, a better stratification should have resulted in a reduced vertical mixing of the water column and thus to reduced sea–air gas exchange, leading to decreased atmospheric CO2, such as in the Southern Ocean during the Last Glacial Maximum (François et al., 1997). On the contrary, our data do not imply such a situation, as the pH gradient between deep and surface ocean appears to have remained constant (Fig. S4), suggesting that DIC accumulated in both shallow and deep waters at this site, which resulted in an overall pH decrease. Indeed, Leutert et al. (2020) suggested that the expansion of the Antarctic Circumpolar Current (ACC) following ice-sheet expansion could have moved the frontal systems and westerly winds further north, which reduced Southern Ocean upwelling and hence provided a mechanism to increase carbon storage in the deep ocean, drawing down pCO2. If the deep ocean in the south was indeed less well ventilated, we would expect benthic δ13C to decrease after the onset of the EAIS expansion because the ocean had more time to accumulate 13C-depleted CO2 from decomposed Corg, which is the opposite (increasing δ13C, Fig. 4). All these complications call for more proxy records at sufficient resolution from spatially distinct locations to better constrain pCO2 evolution across the MMCT.

Our pCO2 data broadly agree with reconstructed long-term trends based on paleosol δ13C (Ji et al., 2018) and planktonic foraminiferal δ11B (Foster et al., 2012), except for the higher value at ∼ 14.2 Ma (Fig. 6). The boron isotope data from the Mediterranean Sea by Badger et al. (2013) also show pCO2 values that match our data well after 13.8 Ma, but our data do not exhibit a generally higher pCO2 level immediately before 13.8 Ma. On the one hand, the relative pH increase of ∼ 0.1 accompanying Antarctic glaciation was backed up by comparing a pH-dependent (Mg / Ca) with independent temperature proxies (clumped isotopes and TEX86) obtained from the same set of samples (Leutert et al., 2020). On the other hand, the strong SST decrease of ∼ 4–5 ∘C after the onset of Antarctic glaciation, as suggested by the Mediterranean data, is more than 2 ∘C larger than those recorded in other low-latitude sites (Foster et al., 2012; Leutert et al., 2020; Sosdian and Lear, 2020), yet the reason for this discrepancy needs to be examined. In contrast, the pCO2 reconstruction based on alkenone δ13C data from Badger et al. (2013) is similar to the estimates of Super et al. (2018), which are generally lower than the boron-based data and do not exhibit significant changes during CM6. Recent studies suggest that phytoplankton may enhance cellular carbon supply for photosynthesis via carbon concentrating mechanisms under limited aqueous [CO2], making this proxy less sensitive at low to moderate pCO2 levels (Badger et al., 2019; Stoll et al., 2019).

There is no other direct evidence for increasing pH immediately following the onset of EAIS expansion than the δ11B data by Badger et al. (2013), which is supported by Sosdian et al. (2020) who used planktonic foraminiferal B / Ca ratios from ODP Site 761 (Fig. 1) to infer relative changes in DIC across the middle Miocene. As B / Ca in planktonic foraminifera is thought to be governed by the ratio between borate and DIC concentrations [B(OH)4- / DIC], an increase in B / Ca can be interpreted as a decrease in DIC and/or an increase in [B(OH)4-] (e.g., Allen et al., 2012). The high-resolution B / Ca record of Sosdian et al. (2020) reveals that the long-lasting Monterey excursion is accompanied by generally low B / Ca ratios, whereas CM intervals are characterized by higher B / Ca, suggesting periodic DIC decreases within a “high-DIC” world. The authors concluded that during the CM events, ice-sheet expansion and global cooling enhanced marine productivity and Corg burial, which resulted in a reduction of surface-water DIC and pCO2, and a rise in δ13C.

Within the overlapping period, which encompasses two full 400 kyr cycles, our record exhibits some striking similarities with the B / Ca record of Sosdian et al. (2020), including the long-term decrease in pCO2, accompanied by declining surface-water DIC, and the higher pCO2 during CM5b (Fig. 6) that is accompanied by increased DIC. Intriguingly, the correlation between the two records collapses at the onset of CM6, when the B / Ca record indicates a decline in DIC, which is corroborated by a lowering in pCO2 (Badger et al., 2013). This is in contrast to our record showing that pCO2 increased during CM6 (Fig. 4), although the records are once again in agreement at the end of CM6 (∼ 13.5 Ma). This discrepancy could be attributed to non-identified factors influencing either of the two proxies or regional effects on either of the two records. The potential influence of the polar front on Site 1092 is addressed in Sect. 4.1, leading to the question of whether the data by Sosdian et al. (2020) could be alternatively explained by increased deep-water production. When the density of surface waters in the Southern Ocean increased due to cooling in a generally warm middle Miocene, the invigorated deep-ocean venting could have brought respired CO2 to the atmosphere, leaving the overall ocean depleted in DIC and enriched in CO32- and possibly also in δ13C. The higher CO32- in turn could have favored the burial of carbonate in the deep sea, producing even more CO2 that is vented to the atmosphere, as suggested for the Pleistocene glacial–interglacial cycles (Toggweiler, 2008). This way, the loss of CO2 in shallow waters, caused by enhanced marine productivity and Corg burial (Sosdian et al., 2020), could have been outweighed by CO2 added to the atmosphere by deep-ocean ventilation. However, this is speculative and needs to be validated in future studies.

It is worth noting that our pCO2 record before Antarctic glaciation reveals a striking match with the box model output from Ma et al. (2011), while after EAIS expansion our values are some 50 µatm lower, although they still exhibit similar relative changes (Fig. 6). The divergence after ∼ 13.9 Ma may be related to the model by Ma et al. (2011), which was designed to simulate orbitally paced carbon reservoir changes for the Miocene Climate Optimum (MCO) and hence does not include ice sheets. Consequently, the model does not consider the effects of falling sea level on the shelf-to-basin fractionation of carbonate burial (McKay et al., 2016; Ma et al., 2018), or of ice-sheet expansion on ocean circulation and bottom-water ventilation (e.g., Shevenell et al., 2004; Holbourn et al., 2007, 2013), all of which should have influenced δ13C and pCO2. However, the similarity between our record and the model suggests that the ocean carbon reservoir varied in response to weathering-induced nutrient input, which in turn was controlled by varying orbital configuration. This model was sufficient to explain the observed δ13C pattern of the MCO (Ma et al., 2011) but could also be capable of explaining a portion of the observed carbon cycle changes in a world with an ice-covered Antarctica.

The pCO2 increase preceding and following the EAIS expansion closely tracks the global δ13C history of CM5b and CM6 (Fig. 4), emphasizing a fundamental link between changes in the ocean carbon cycle and atmospheric carbon dioxide. The CM peaks correspond to minima in the 400 kyr eccentricity cycle, suggesting that δ13C variations were related to changes in monsoon and weathering intensity (Holbourn et al., 2007; Ma et al., 2011). Eccentricity maxima lead δ13C minima at the 400 kyr band by on average more than 50 kyr, which is comparable to the phase lag of 47 ± 18 kyr observed in other Miocene records, which can possibly be attributed to the slow response of weathering to orbital forcing (Holbourn et al., 2007). Monsoon intensity is paced by solar radiation variations caused by precessional cycles but their amplitude variations are modified by eccentricity. More precisely, higher eccentricity results in larger precession amplitudes and hence in larger wet–dry variations in the tropics (e.g., Wang, 2009), stronger physical and chemical weathering, and ultimately in an increased input of particulate organic material, dissolved inorganic carbon, alkalinity and nutrients to the oceans (e.g., Clift and Plumb, 2008; Ma et al., 2011; Wan et al., 2009). The potential influence of alkalinity variations on our pCO2 record is shown in Fig. S5 and evaluated in Sect. 3.3.

The increased nutrient supply enhanced primary production and organic carbon burial, which in turn lowered pCO2. At the same time, the burial of shallow and total calcium carbonate (CaCO3) increased, while CaCO3 burial in the deep ocean decreased (Holbourn et al., 2007). However, as the net burial of CaCO3 (enriched in 13C) in relation to Corg (depleted in 13C) increased at high eccentricity, δ13CDIC values decreased, which is in line with the global isotope signature (Holbourn et al., 2007; Ma et al., 2011). Estimates of Miocene shallow-water carbonate burial and shelf–basin fractionation of carbonate accumulation are only available in rather low temporal resolution and do not capture orbital variability that would allow assessment of global trends (Boudreau and Luo, 2017; Boudreau et al., 2019; van der Ploeg et al., 2019). Another limitation is that attempts to estimate shallow water carbonate burial on an orbital timescale are limited to regional studies in very different environments (Kroon et al., 2000; Reuning et al., 2002; Williams et al., 2002; Aziz et al., 2003; Auer et al., 2015; Betzler et al., 2018; Ohneiser and Wilson, 2018; Reolid et al., 2019) that cannot be easily synthesized to provide global estimates.

An increase in carbonate burial in shallow seas would have removed alkalinity from seawater and thus lowered pH and released CO2 to the atmosphere (e.g., Zeebe and Wolf-Gladrow, 2001) during 100 kyr eccentricity maxima; however, the resolution of our record is too low to detect this. On longer timescales (400 kyr cycles), increased alkalinity input from rivers, dissolution of deep-sea carbonates, and the enhanced burial of Corg in tropical regions during eccentricity maxima might have contributed to the long-term decrease in pCO2. This agrees well with the box model output from Ma et al. (2011), suggesting that during the long-eccentricity maxima pCO2 decreased, when the monsoon was most intense and the organic carbon burial and river fluxes were high. Conversely, weathering and nutrient supply to the ocean in low latitudes decreased when eccentricity was low, resulting in a net decrease in the CaCO3-to-Corg burial ratio and hence in CO2 release and higher δ13C.

The model of Ma et al. (2011) was designed to explain the long eccentricity signal in the δ13C record throughout the warm MCO and hence does not consider processes caused by expanding Antarctic ice sheets and their impacts on the climate system. Indeed, the modeled pCO2 reveals smaller amplitudes after ∼ 13.9 Ma than our reconstruction (Fig. 6) and a less pronounced CM6 in relation to the previous CM events. Changes in shelf–basin partitioning of carbonate burial and enhanced global deep-water ventilation, as well as increased weathering, resulting from a major sea level fall at 13.8 Ma (Ma et al., 2018) are possible mechanisms to explain the higher amplitude of CM6. Additional mechanisms may have contributed to the amplification of CM6, including (1) increased Corg burial by enhanced carbon sequestration through the biological pump linked to intensified upwelling in low latitude areas, as well as (2) substantial outgassing of CO2 from the deep ocean (δ13C of approx. -1 ‰ to -2 ‰) into the atmosphere (δ13C of approx. -6 ‰ to -6.5 ‰), which would not only result in increased atmospheric pCO2 but also in a positive excursion in atmospheric δ13C. The size of the relevant reservoirs (38 000 Gt in the deep ocean vs. 750 Gt in the atmosphere, if the proportions were similar to today) would suggest that this process can be sustained over considerable time intervals. (3) The amplitude of the 405 kyr cycle-paced δ13C excursions depends on the size of the global marine DIC pool (e.g., Paillard and Donnadieu, 2014). This size may have decreased following the loss of shelf seas caused by the major regression associated with the onset of CM6, resulting in a higher amplitude δ13C response to eccentricity forcing.

As stated in the previous section, the δ13C of CM6 might be related to global deep-water ventilation, due to cooling in the Southern Ocean that invigorated deep-water formation. The MMCT bears similarities to the Eocene–Oligocene transition (EOT) ∼ 34 Ma ago, when Antarctic ice expansion was also accompanied by a large positive δ13C excursion, in terms of both magnitude and duration (Coxall et al., 2005). The δ13C excursion following Antarctic glaciation was accompanied by a transient pCO2 rise of more than 300 µatm (Pearson et al., 2009; Heureux and Rickaby, 2015), which bears similarity to the pCO2 rise during the mid-Miocene EAIS expansion, although the amplitude was much larger. A modeling study suggested that a sea-level-fall-induced shelf-basin carbonate burial fractionation and a temporal enhancement of deep-water formation (by 150 %) in the Southern Ocean were sufficient to produce both the positive δ13C excursion and a transient pCO2 rise by ∼ 80 µatm in the EOT (McKay et al., 2016).

Similarly, ocean circulation may have intensified during the Miocene EAIS expansion, as the latitudinal temperature gradient increased and atmospheric circulation strengthened. This is supported by benthic foraminifers, Mn / Ca and XRF data from Southeast Pacific sites, which indicate improved deep-water ventilation and carbonate preservation after 13.9 Ma, particularly during colder climate phases (Holbourn et al., 2013). Improved deep-water ventilation may also have led to enhanced advection of silica-rich waters toward low latitudes, culminating in an increased upwelling and diatom productivity between 14.04 and 13.96 Ma and between 13.84 and 13.76 Ma in the equatorial East Pacific, as indicated by massive opal accumulation found at IODP Site U1338 (Holbourn et al., 2014). We speculate that increased upwelling may have resulted in CO2 outgassing, if the supply exceeded consumption by primary productivity.

Such a scenario involving Southern Ocean cooling that led to enhanced deep-water formation is also in line with the carbon-cycle box model by Toggweiler (2008), showing that the associated ocean venting could have transported respired CO2 from deep waters to the atmosphere. Consequently, as the ocean was left depleted in DIC and enriched in CO32- (and possibly also in δ13C), this could have favored the burial of carbonate in the deep sea, removing TA and releasing even more CO2 that is vented to the atmosphere (Toggweiler, 2008). The assumed rise in deep-ocean CO32- following EAIS expansion and associated deepening of the carbonate compensation depth (CCD) is not contradictory to the benthic foraminiferal B / Ca ratios from the deep South China Sea that indicate an increase in the calcite saturation state spanning CM6, i.e., a temporary deepening of the CCD (Ma et al., 2018).

It is noteworthy that CM6 exhibits a more pronounced δ13C change than all other CM events of the Monterey excursion, which is not reflected by our pCO2 record but rather shows a similar amplitude of change to CM5b. One possible explanation for this phenomenon is that enhanced deep-water formation during CM6 could have placed Site 1092 into a frontal system in the Southern Ocean, thus recording increasing aqueous [CO2] in a climate of decreasing atmospheric pCO2, which might have also contributed to the divergence in the records of B / Ca (Sosdian et al., 2020) and pCO2 from this study. However, based on our data we can only speculate on the unique nature of CM6 in comparison to other CM events, which must be further explored in future studies.