Revised alkenone d 13 C based CO 2 estimates during the Plio-Pleistocene

Here we revisit reported alkenone d13C (d13CC37) based CO2 records during the Plio-Pleistocene and apply a refined approach to better constrain the dynamic range of CO2 during the time. Specifically, we consider ways to correct for regional differences in physical oceanographic factors. As a result of our relatively simple approach we find that offsets of ~150 ppm between reported d13CC37 CO2 records from different sites can be significantly reduced. This confirms that better constraints 10 on environmental variables, including physical oceanographic controls on depth and season of production are key aspects for improving d13CC37 based CO2 estimates. The revised d13CC37 CO2 datasets suggest that Plio-Pleistocene CO2 levels are 180350 ppm, which is consistent with the most of reported CO2 reconstructions, though their upper end of Pliocene CO2 levels are lower than that of some CO2 estimates.

as discussed above. dorg is derived from the d 13 C values of di-unsaturated alkenones and corrected for the 4.2‰ depletion of 90 alkenones relative to biomass (Bidigare et al., 1997).

Calculation of CO2
The ep values of marine algae are a function of not only CO2(aq) but also growth rate (Bidgare et al., 1997) and cell geometry (Popp et al., 1998). Therefore, correction of those effects on ep values is required. Field and experimental work both indicate that the relationship between ep and related factors is expressed as the following: where ef is the isotope fractionation associated with carbon fixation by the ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO) enzyme (25 ‰;Bidigare et al., 1997). The b-term is an expression of physiological factors -including growth rate, membrane permeability, and cell geometry -that affect carbon discrimination by regulating the flux of CO2 into and out of the cell. For d 13 CC37-derived ep values, and thus for the limited range of organisms synthesizing such compounds and for which 100 cell geometry effects are thought to be minimal, field data reveal a significant correlation between the b-term and [ The modern range of [PO4 3-] in the surface mixed layer is used to calculate b-term. Because a reliable [PO4 3-] proxy is not available so far, we assume that [PO4 3-] did not differ for the open marine settings in Fig.1 over the last 5 Myr. Finally, 105 atmospheric CO2 levels are calculated using these b-term values and assuming air-sea equilibrium (using salinity, SSTs, and Henry's Law to convert CO2(aq) to gas CO2; Weiss, 1974).

Result and discussion
3.1 d 13 CC37 derived CO2 records based on the standard diffusive model approach Figure 4a shows CO2 records derived from the standard approach used in most previous publications: We have calculated the 110 b-term using the global regression , which assumes alkenone production depths at the surface of the mixed layer, and which assumes air-sea equilibrium at all sites. Although the trends are generally similar, indicating decreasing CO2 levels over the past 5 Myr, the ranges are large (>150 ppm) in both recent and older sediments. In particular, CO2 estimates are relatively low in the extra-tropics (Sites 982 and 1208) and higher in the tropics (Sites 806, 925, and 1241) except for the Caribbean Sea (Site 999). 115 The first application of the alkenone d 13 CC37 CO2 proxy suggested it could be used to accurately reconstruct [CO2(aq)] when SSTs and the b-term are well constrained (Pagani et al., 2002). If constraints on those parameters are the key to reconstruct CO2 from the proxy, the observed large gaps among the sites could be attributed to insufficient constraints on such key parameters. It should be noted that the ep value of marine algae is a function of not only [CO2(aq)], but also the b-term. Modern ranges of [PO4 3-] in the euphotic zone are typically used to determine the b-term in paleo-investigations (Seki et al., 2010;120 Pagani et al., 2010). However, the relationship between the b-term and [PO4 3-] depends on the oceanographic setting, resulting in significant variability in the global regression equation (Bidigare et al., 1997;Pagani et al., 2005) (Fig. 2). Therefore, we applied regional rather than global regressions to better constrain the b-term at specific sites (Site 806 and 1241; Fig. 4c; see Section 3.2. for the detail). Another factor in CO2 determination is regional variation in air-sea disequilibrium (∆CO2). Previous studies have reconstructed CO2 levels based on the assumption of equilibrium between CO2 in seawater and the atmosphere. 125 However, air-sea ∆CO2 spatially varies, generally with excess [CO2(aq)] in the tropics and deficiencies in the extratropics (Takahashi et al., 2009). Hence, we have reconsidered these data by first correcting for disequilibrium between ocean and atmosphere ( Fig. 4b; See Section 3.3. for details). Also, we discuss other factors that potentially influence CO2 reconstruction in Sections 3.4. and 3.5.

Geographical and temporal variations in the b-term 130
As suggested by the empirical relationship between the b-term and [PO4 3-] in modern surface ocean waters (Bidigare et al., 1997), the growth rate is a primary factor controlling the b-term in diverse phytoplankton, including haptophytes (Bidigare et al., 1997(Bidigare et al., , 1999Popp et al., 1999). Although that is factored into our calibration at each site, significant past changes in growth rate (which we infer to be broadly correlated to a regime's primary productivity history) will bias CO2 trends and estimates (Mix et al., 2003). For these reasons, we have precluded Sites 882 and 1012, for which there is evidence for strong past 135 oceanographic and productivity variations which are hard to constrain (Pagani et al., 2010).
All of the remaining CO2 data used in this study exhibit similar long-term decreasing trends (Fig. 4a), suggesting that growth rate variations are subordinate controls on long-term trends. However, there are likely local variations in the controls on the bterm. Although there is a strong positive relationship (r 2 = 0.78) between phosphate concentration and the b-term in the global dataset (Bidigare et al., 1997(Bidigare et al., , 1999Pagani et al., 2005), there are systematic variations in the relationship among different 140 oceanographic regimes. For example, b-term values are lower than the global regression trend in the Equatorial Pacific, Southern Ocean, and sub-Antarctic and higher in Santa Monica, the Northeast Pacific, and the Peru upwelling margin (Fig. 3).
Such regional differences are not surprising because phosphate is unlikely to limit directly haptophyte growth, which is instead is probably governed by micronutrients (such as iron, zinc, and cobalt) (Bidigare et al., 1997). The relationships between concentrations of phosphate and the actual growth-limiting micronutrients might differ among regions. 145 Thus, we explore here the application of regional regressions to better constrain the b-term. Currently We apply the regional regression (4) to these sites (Table 1).

Geographical and temporal variations in air-sea disequilibrium 150
As described earlier, the conversion from [CO2(aq)] to atmospheric [CO2] is based on Henry's Law (Weiss, 1974) and using SSS (Table 1) and SST (described above), while assuming air-sea equilibrium. However, this assumption is likely invalid in some of the settings investigated. There is excess CO2(aq) in the equatorial ocean due to active upwelling and deficiency of CO2(aq) in extratropical regions (Takahashi et al., 2009). Therefore, we also show CO2 reconstructions, in which we have corrected for air-sea ∆CO2 at each site ( Fig. 4b; correction shown in Table 1). 155

Temporal variations in cell geometry
Previous studies showed that variations in haptophyte cell geometry, if significant enough, can potentially affect ep values (Henderiks et al., 2007;Bolton et al., 2015;Stoll et al., 2019;Zhang et al., 2020). In a previous study (Seki et al., 2010), we explored the potential effects of such changes based on the approach of Henderiks et al. (2007), who proposed a geometrycorrected b-term using the coccolith size record of Reticulofenestra, the postulated alkenone-producing haptophytes in the 160 Caribbean Sea prior to the Pliocene (Kameo and Bralower, 2000). However, applying this approach to Site 999 results in significantly lower CO2 levels (<150 ppm) for the Pliocene and Pleistocene compared to the other CO2 estimates (Kürschner et al., 1996;Siegenthaler et al., 2005;Badger et al., 2013;Pagani et al., 2010;Martínez-Botí et al., 2015;Cui et al., 2020;de la Vega et al., 2020). This underestimation potentially illustrates the limitations for applying the relationship between Reticulofenestra coccolith and cell sizes obtained from early Miocene species to the smaller species of the Pliocene. 165 Moreover, it has been suggested that Cyclicargolithus was the most important alkenone producer during the Miocene (Plancq et al., 2012). Similarly, the alkenone concentration record in Caribbean Sea sediments (Seki et al., 2010) does not resemble that of Reticulofenestra abundances (Sites 998, 999, and 1000) (Kameo and Sato, 2000). This decoupling suggests that Reticulofenestra may not be a major alkenone producer. Thus, we suggest that a cell geometry correction based on Reticulofenestrais is not appropriate for the Caribbean Sea. Therefore, the effect of changes in growth rate and cell geometry 170 of alkenone producers remains uncertain for CO2 reconstruction. However, the effect of geometry change on b-term is probably not significant on Plio-Pleistocene timescales, since cell size does not significantly change in the tropical ocean during the

Other factors affecting ep37:2
As outlined above, [CO2(aq)] estimates are highly sensitive to the b-term, which represents an integration of various 175 physiological variables but is especially dependent on haptophyte growth rate (Bidigare et al., 1997). Equation (2) implicitly assumes that CO2 diffuses into the haptophyte cell, even though there is much evidence that haptophytes, including those that produce alkenones, employ carbon concentrating mechanisms (CCMs) (Raven et al., 2011;Reinfelder, 2011). Nor does the model account for the competing requirements for dissolved inorganic carbon for both cell growth and coccolith production and the resultant internal partitioning of the DIC pool. These factors are implicitly considered because Equation (2), as well as the relationships between the b-term and nutrient concentrations, are empirical (Bidigare et al., 1997). However, the recently developed ACTI-CO model of Bolton and Stoll (2013) reveals the complexity of the controls on haptophyte ep values; in particular, it illustrates how isotopic partitioning between organic and inorganic components will be governed by a complex range of environmental and physiological factors. As such, we consider our CO2 estimates to be semi-quantitative and include them primarily to allow comparison to previous investigations. Nonetheless, the ACTI-CO model still indicates that higher 185 [CO2(aq)] will yield higher ep values; as such, we consider the trends in atmospheric CO2 levels generated from d 13 CC37-derived ep values to be robust.

Revised d 13 CC37 CO2 over the past 5 Myrs
With taking the factors described above into account, we revised published CO2 data from the Pacific Ocean and the Atlantic Ocean. Figure 4d shows revised CO2 records based on the refined approach (e.g. correcting offsets estimated for both ∆CO2 190 and regional [PO4 3-]). The large difference (>150 ppm) in CO2 estimates among sites based on the standard approach (Fig. 4a) has been markedly reduced by ~100 ppm when we apply this refined method (Fig. 4d). Indeed, a combination of corrections, using both regional b-term regressions and accounting for air-sea ∆CO2, yields similar Pliocene to Pleistocene CO2 records ( Fig. 4d). Importantly, the late Pleistocene CO2 estimates fall within the range of ice core CO2 records (Siegenthaler et al., 2005;Lüthi et al., 2008) Despite the importance of oceanographic conditions -export production depth of alkenones, [PO4 3-] depth profiles and their relationship to b-term, air-sea ∆CO2 -we have assumed that they did not differ markedly in the past. If these parameters have changed significantly over time, this would impact absolute reconstructions of CO2 levels as well as the temporal trends. As 200 records from several sites all exhibit similar long-term trends, despite representing diverse oceanographic settings, we suggest that these parameters have been relatively stable, or at least a relatively minor factor, over the past 5 Myrs for the open ocean environment. Indeed, a relatively stable b-term on Plio-Pleistocene timescales has been suggested in some regions (Bolton et al., 2015;Mejia et al., 2017;Stoll et al., 2019).

Comparison of d 13 CC37 CO2 records with those derived from the ice cores and other CO2 proxies 205
Comparison of ice core and d 13 CC37 CO2 shows that some d 13 CC37 CO2 records based on the global b-term regression are higher than ice core CO2 data (Fig. 4a). However, the offset is significantly reduced if regional b-term regressions and air-sea ∆CO2 are applied (Fig. 4d)  estimates in the Pliocene since upregulation of CCM results in overestimation of pCO2 when dissolved CO2 level is lower than 230 7 µM (Badger, 2021). Also, it should be noted that the CCM effect is unlikely upregulated in mid to high latitude oceans where the surface water temperature is enough cold and thus dissolved CO2 concentration in the surface mixed layer is enough high to allow algae to adopt diffusive carbon transport.
Another potential reason for the overestimation of glacial CO2 is an underestimation of G-IG variability of dissolved CO2 d 13 C.
d 13 C of dissolved CO2 in the past is estimated from d 13 C of foraminifera carbonate shells, which is used to calculate ep. A 235 previous study shows that carbon isotope fractionates into foraminiferal calcite as a function of seawater pH/CO3 -2 (Spero et al., 19997) which must vary associated with the late Pleistocene G-IG cycle given that ~100 ppm variation of CO2 as suggested by the ice core CO2 record. Not considering this effect results in an overestimation of glacial CO2 levels. However, according to the relationship between foraminiferal d 13 C and CO3 -2 (Spero et al., 1997), this effect only explains approximately 10 ppm overestimation during the glacial. Thus, this effect is not sufficient to explain the mute signal of G-IG d 13 CC37 CO2 variability. Pliocene CO2 estimates based on other independent proxies such as leaf stomata are below 360 ppm (Kürschner et al., 1996).
Also, the dynamic range of G. ruber d 11 B CO2 (300-480 ppm) in the Pliocene is approximately twice as much as that of the deep glacial-interglacial cycles (180-280 ppm) in the late Pleistocene (Siegenthaler et al., 2005;Lüthi et al., 2008), despite the latter exhibiting smaller orbital-scale climate variations compared to that of the late Pleistocene. Furthermore, the fluctuation range of the statistical model CO2 in the Pliocene (200-620 ppm) is much larger than that of G. ruber d 11 B CO2 (Stoll et al.,250 2019) (Fig. 2b). Such a large fluctuation is difficult to explain given that G-IG climate fluctuation in Pliocene is much smaller than that of the late Pleistocene.
(2018) and de la Vega et al. (2020). They argued that biological effects such as long-term evolution of either foraminifera species might alter the relationship between d 11 B and carbonate chemistry. On the other hand, de la Vega et al. (2020) argue that the offset may be attributed to an analytical issue with T. sacculifer d 11 B. However, the revised alkenone CO2 records 260 agree well with T. sacculifer d 11 B CO2 (Fig. 5b), suggesting alternation of G. ruber d 11 B CO2 due to the biological effect.
Alternatively, another possible explanation for the high estimates of G. ruber d 11 B CO2 is that those CO2 records overestimate variability ranges possibly due calibration for the conversion of G. ruber d 11 B into CO2 concentration not being well-optimized. should be noted that the number of the culture data, which is used to establish the calibration, is limited. Although G. ruber d 11 B CO2 in ODP 999 seems to well represent the CO2 variability during the glacial-interglacial cycles documented in the Antarctic ice core record (Foster et al., 2008;Chalk et al., 2017), a close look reveals that the variability range of G. ruber d 11 B CO2 (150 to 340 ppm) is significantly larger than that of the ice-cores (180 to 280 ppm) (Fig. 2c). This suggests that G. 270 ruber d 11 B CO2 overestimates the variability of CO2 in the G-IG cycles. On the other hand, the dynamic range of T. sacculifer d 11 B CO2 (175-300 ppm) (Dyez et al., 2018), which shows relatively low CO2 in the Pliocene, agrees well with ice core values (180-280 ppm) (Fig. 2c) during the past 800 kyrs, supporting the reliability of CO2 estimates. The variation range (150-350 ppm) of the statistical model d 13 CC37 CO2 in the late Pleistocene also significantly exceeds that of ice core (180-280 ppm) (Fig.   2c), suggesting overcorrection of environmental factors. These considerations depict that the CO2 records, which overestimate Further studies for developing the calibrations and constraints of environmental and physiological factors will be needed to generate robust CO2 estimates. As for the alkenone d 13 C CO2 proxy, these are ongoing efforts within the community (Zhang et al., 2020;Phelps et al., 2020;Badger, 2021) and include refinements to the alkenone CO2 model (Zhang et al., 2020), culture based constrains (Phelps et al., 2020) and consideration of the lower limits of the proxy due to CCMs (Badger, 2021). Here 280 we highlight how relatively simple refinements for export production depth of alkenones (based on U !" #$ temperature) and using regional [PO4 3-] depth profiles can lead to major reconciliation of CO2 estimates with other lines of evidence.

Conclusion
In this study, previously reported d 13 CC37 based CO2 records from the Plio-Pleistocene were reexamined to better constrain CO2 evolution. We refined the conventional approach, which assumes constant physiological factors and a carbon diffusive 285 model over the time by including considerations of regional differences in physical oceanographic factors, specifically the mixed layer depth, production season and air-ocean disequilibrium of CO2 for each site (Fig. 4). This simple approach refines estimates of CO2 significantly, reducing large offsets previously observed in the reported d 13 CC37 CO2 records from different oceanographic settings. However, some of our data still do not obtain sufficiently low glacial CO2 values during the Pleistocene (Fig. 4d). This suggests varying physiological factors associated with G-IG cycles or upregulation of carbon concentration 290 mechanisms in glacial low CO2 conditions at some sites. Nevertheless, this study shows that better constraints on past environmental variables can be obtained using literature data and this this approach can improve d 13 CC37 CO2 estimates. Our revised d 13 CC37 CO2 datasets are consistent well with the previously reported CO2 estimates in the early to mid Pleistocene (Da et al., 2019;Dyez et al., 2018;Cui et al., 2020) with CO2 levels below 300 ppm. On the other hand, the revised d 13 CC37 CO2 CO2 levels in the Pliocene are lower than 350 ppm. Our low CO2 values in the Pliocene are consistent with other CO2 estimates 295 such as a T-inverse model (van de Wal et al., 2011;Berends et al., 2019), T. sacculifer d 11 B (Bartoli et al., 2011Dyez et al., 2018) and leaf stomata (Kürschner et al., 1996) based CO2 reconstructions (Fig. 5b), albeit the upper end of the Pliocene CO2 are lower than that of G. ruber d 11 B (Seki et al., 2010;Martínez-Botí et al., 2015;de la Vega et al., 2020) and statistical model d 13 CC37 (Stoll et al., 2019) CO2 estimates. Further efforts are needed to constrain the dynamic range of CO2 for time periods pre-dating the ice core records. It is particularly important to generate continuous, high temporal resolution CO2 records which 300 overlap and match the Antarctic ice-core CO2 records.