Controls on Millennial-Scale Atmospheric CO₂ Variability During the Last Glacial Period

Abstract

Changes in atmospheric CO₂ on millennial-to-centennial timescales are key components of past climate variability during the last glacial and deglacial periods (70–10 ka), yet the sources and mechanisms responsible for the CO₂ fluctuations remain largely obscure. Here we report the ¹³C/¹²C ratio of atmospheric CO₂ during a key interval of the last glacial period at submillennial resolution, with coeval histories of atmospheric CO₂, CH₄, and N₂O concentrations. The carbon isotope data suggest that the millennial-scale CO₂ variability in Marine Isotope Stage 3 is driven largely by changes in the organic carbon cycle, most likely by sequestration of respired carbon in the deep ocean. Centennial-scale CO₂ variations, distinguished by carbon isotope signatures, are associated with both abrupt hydrological change in the tropics (e.g., Heinrich events) and rapid increases in Northern Hemisphere temperature (Dansgaard-Oeschger events). These events can be linked to modes of variability during the last deglaciation, thus suggesting that drivers of millennial and centennial CO₂ variability during both periods are intimately linked to abrupt climate variability.

Plain Language Summary

Ice cores provide unique records of variations in atmospheric CO₂ prior to the instrumental era. While it is clear that changes in atmospheric CO₂ played a significant role in driving past climate change, it is unclear what in turn drove changes in atmospheric CO₂. Here we investigate enigmatic changes in atmospheric CO₂ levels during an interval of the last glacial period (~50,000 to 35,000 years ago) that are associated with abrupt changes in polar climate. To determine the sources and sinks for atmospheric CO₂, we measured the stable isotopes of carbon in CO₂ and found that the primary source of carbon to the atmosphere was an organic carbon reservoir. Most likely, this carbon was sourced from a deep ocean reservoir that waxed and waned following changes in either the productivity of the surface ocean or stratification of the deep ocean. We also found that atmospheric CO₂ can change on the centennial timescale during abrupt climate transitions in the Northern Hemisphere. This observation adds to a growing body of evidence that abrupt changes in atmospheric CO₂ are an important component of past carbon cycle variability.

1 Introduction

In order to predict the climate impacts of anthropogenic CO₂ emissions over the coming millennia (Clark et al., 2016), we must understand how the climate system and carbon cycle have interacted over these same timescales in the past. The CO₂ rise during the last deglaciation, arguably the most well-studied example of past carbon cycle variability, is likely a combination of millennial-scale climate change and glacial-interglacial shifts in temperature and ice volume, which are all amplified through a system of climate-carbon cycle feedbacks. To disentangle the millennial-scale component, we investigate carbon cycle variability of the last glacial period when climate variations were largely unaffected by changes in Northern Hemisphere insolation and ice volume that characterize glacial terminations (supporting information Figure S1).

During Marine Isotope Stage 3 (MIS3), atmospheric CO₂ varied between about 195 and 225 ppm with a roughly triangular waveform (Indermuhle et al., 2000; Figure 1e). This pattern mimics Antarctic temperature, rising during the strongest Antarctic warmings and falling during the coolings (Figure 1b). From the perspective of Greenland climate and the well-known Dansgaard-Oeschger (DO) events, CO₂ increases during cold phases (stadials; Ahn & Brook, 2008) with the most prominent increases corresponding to only those stadials associated with an enhanced flux of debris-laden ice to the North Atlantic (known as Heinrich events; these specific cold intervals are often referred to as Heinrich stadials). CO₂ decreases after a rapid switch to warmer conditions over Greenland (interstadials) and slower cooling in Antarctica (Bereiter et al., 2012). This strong relationship between Antarctic temperature and CO₂ has focused attention on the Southern Ocean (SO) as the major conduit for the transfer of carbon between the atmosphere and the ocean on millennial and glacial-interglacial timescales (Sigman et al., 2010), with possible triggers in the North Atlantic.

Figure 1

Taylor Glacier CO₂, CH₄, N₂O, and δ¹³C-CO₂ (blue markers with black smoothing spline; this study). (a) NGRIP δ¹⁸O (North Greenland Ice Core Project Members, 2004). (b) WDC (WAIS Divide Ice Core) δ¹⁸O (WAIS Divide Project Members, 2015). (c) Hulu Cave δ¹⁸O (Cheng et al., 2016). (d) WDC (Rhodes et al., 2015; thin black line) and Taylor Glacier CH₄. (e) Siple Dome (Ahn & Brook, 2014; gray markers) and Taylor Glacier CO₂. (f) Taylor Glacier δ¹³C-CO₂. (g) Talos Dome (Schilt et al., 2010; gray markers) and Taylor Glacier N₂O. (h) Global Terrestrial O₂ isotopic fractionation, Δε_LAND, constrained from a combination of WDC and Siple Dome (Seltzer et al., 2017; Severinghaus et al., 2009). Dotted lines highlight the stadial to interstadial transitions, and the red bar highlights the Heinrich stadial with the associated rise in CO₂ and CH₄ and increase in Δε_LAND. The NGRIP chronology is increased by 0.63% prior to 22 ka for consistency with the WDC chronology (WAIS Divide Project Members, 2015).

The ratios of carbon isotopes differ among key carbon reservoirs, and thus, some of the sources and sinks driving past atmospheric CO₂ can be constrained using the isotopic composition of atmospheric CO₂ (Bauska et al., 2016; Eggleston et al., 2016; Köhler et al., 2006; Schmitt et al., 2012). Four major processes fractionate the carbon isotopes (ε_A-B ≅ δ¹³C_A − δ¹³C_B): photosynthesis that makes CO₂ into organic carbon on land (ε_{land-atmosphere} = approximately −18‰), photosynthesis in the surface ocean that forms particulate organic carbon from dissolved inorganic carbon (ε_POC-DIC = approximately −22‰), air-sea gas exchange (ε_{DIC-atmosphere} = approximately +8‰; decreasing by 0.1‰ for every 1 °C increase in ocean temperature), and a negligible fractionation during the formation of CaCO₃ in surface ocean (ε_DIC-CaCO3 = approximately 0‰) by measuring the time varying changes in both atmospheric CO₂ concentrations and the δ¹³C-CO₂, we can constrain the sources of the CO₂ changes. Rising atmospheric CO₂ and decreasing δ¹³C-CO₂ is consistent with organic carbon sources to the atmosphere; rising atmospheric CO₂ and increasing δ¹³C-CO₂ is consistent with a warming ocean; and rising CO₂ with no changes in δ¹³C-CO₂ could indicate a balanced contribution of rising ocean temperature and organic carbon sources or the influence of CaCO₃ or volcanic sources. Air-sea gas exchange in the high-latitude ocean may affect atmospheric δ¹³C-CO₂, with rising atmospheric CO₂ and large decreases in δ¹³C-CO₂ predicted by box model experiments that include enhanced air-sea gas exchange in the SO (Bauska et al., 2016; Köhler et al., 2006). Finally, changes in the abundance of C3 and C4 plants on land (Ehleringer et al., 1997) or ecological shifts in marine biosphere (Broecker & McGee, 2013) would change the overall photosynthetic fractionation and thus the details, but not the fundamentals, of using δ¹³C-CO₂ to fingerprint carbon sources.

Other atmospheric gases provide additional constraints on carbon cycle variability, particularly related to the response of the terrestrial biosphere to abrupt climate change. Most evidence suggests that past variations in methane are dominated by climate variability over tropical and boreal wetlands (Brook et al., 2000; Rhodes et al., 2015). N₂O is produced in oxygen-poor ocean waters and terrestrial soils. Rising N₂O most likely reflects either decreasing oxygen levels in the intermediate-depth ocean, warmer terrestrial soil temperatures, or a combination of the two (Schilt et al., 2014). The δ¹⁸O of atmospheric O₂ (δ¹⁸O-O₂) largely follows the changes in δ¹⁸O of seawater and orbitally driven changes in the Dole effect, but rapid increases in δ¹⁸O-O₂ have been argued to reflect southward shifts in low-latitude precipitation (Seltzer et al., 2017; Severinghaus et al., 2009; see supporting information for definition of Δε_LAND).

2 Methods and Materials

We present new atmospheric histories of CO₂, N₂O, CH₄, and δ¹³C-CO₂ spanning 47–35 ka from the Taylor Glacier, Antarctica, blue ice area (Figures 1d–1g). The chronology (Baggenstos et al., 2017) is constructed by synchronizing the Taylor Glacier CH₄ and δ¹⁸O-O₂ records to the WAIS (West Antarctic Ice Sheet) Divide Ice Core (WDC) timescale (Buizert et al., 2015) and thus linking to the radiometrically dated Hulu Cave record (Cheng et al., 2016; Figure 1c). The concentrations of CO₂ and N₂O and the δ¹³C-CO₂ were measured on the same sample using the Oregon State University ice grater system (Bauska et al., 2014). The long-term reproducibly for CO₂, N₂O, and δ¹³C-CO₂ are ±1 ppm, ±5 ppb, and +0.02‰, respectively (Bauska et al., 2015, 2016). The combination of increased resolution and precision equates to a significant improvement over previous reconstructions which focused on longer-term changes (Eggleston et al., 2016; Figure S1). Additionally, previous reconstructions in this time interval were limited by offsets between cores (see Eggleston et al., 2016 for detailed discussion and Figure S2). Using the interval from 47 to 43 ka as a baseline for comparison, the Talos Dome δ¹³C-CO₂ record has significantly lower and more variable values (mean ± s.d. = −6.72 ± 0.23‰; n = 12) than data from the EPICA Dronning Maude Land (−6.50 ± 0.12‰; n = 8) and EPICA Dome C (−6.59 ± 0.16‰; n = 5) ice cores. In the same interval, the Taylor Glacier record averages −6.55 ± 0.07‰ (n = 14) in broad agreement with EPICA Dronning Maude Land and EPICA Dome C. Although we can confidently interpret relative changes in our new record because of the improvement in precision and resolution, addressing the absolute accuracy of the δ¹³C-CO₂ values requires additional interlaboratory and intercore comparison.

3 Results

3.1 Ice Core Constraints on Greenhouse Gas Variability

The most salient mode of variability in atmospheric CO₂, the millennial-scale rising and falling with Antarctic temperature, is accompanied by an inversely correlated change in δ¹³C-CO₂ (Figures 1e and 1f). Atmospheric CO₂ ranges from about 195 to 215 ppm with the corresponding variability in δ¹³C-CO₂ spanning −6.45 to −6.65‰ (a −0.1‰ change for every +10 ppm). At finer scales, we observe several other modes of variability. The CO₂ rise during Heinrich stadial 4 (40.2–38.4 ka; HS4) starts off slowly, rising 3–4 ppm, while δ¹³C-CO₂ decreases by ~0.05‰. As noted previously in the Siple Dome ice core (Ahn & Brook, 2014), the rise then accelerates in a sharp jump of about 8 ppm. The rapid phase of the rise is coincident with the midstadial rise in CH₄ noted in the WAIS Divide Core (Rhodes et al., 2015; Figure 1d, red vertical line) and the increase in Δε_LAND first observed in the Siple Dome Core (Severinghaus et al., 2009) and confirmed in the WDC record (Seltzer et al., 2017; Figure 1h). No change in N₂O is resolved in the Taylor Glacier data set, consistent with the Talos Dome record (Schilt et al., 2010; Figure 1g). The resolution of the carbon isotope measurements prevents a clear fingerprinting of the source but a replicated sample clearly falls to more negative values off the more gradual trend by about 0.1‰ (Figure 1f). In the later part of the HS4, CO₂ continues to rise slowly by 3 ppm and δ¹³C-CO₂ decreases by another 0.1‰.

The onsets of DO interstadials are accompanied by small rises in CO₂ (Figure 1e, gray dashed lines). This variability has been noted in other cores and described as either a lagged response to Antarctic temperature (Bereiter et al., 2012) or, in the case of the weaker DO events, a lagged response to Greenland stadial conditions that are too short to impart a significant change in CO₂ (Ahn & Brook, 2014). In our record, we note that atmospheric CO₂ appears to increase along with CH₄ and N₂O with no discernable lead or lag. This is most prominent at DO8 when CH₄ rises by about 120 ppb, N₂O rises by 35 ppb, and CO₂ rises by 6 ppm (see also Figures 4n–4p). The other DO events (7, 9, and 10) are near the detection limit of our record and difficult to quantity. Across these events, δ¹³C-CO₂ either increases slightly (~0.08‰ at DO8) or shows no substantial change.

4 Discussion

4.1 Millennial-Scale Carbon Cycle Variability

The overall negative correlation between CO₂ and δ¹³C-CO₂ rules out changes in ocean temperature, the CaCO₃ cycle, or volcanic input having a dominant role in driving millennial-scale CO₂ (Figure 2). Instead, the data are consistent with changes in terrestrial carbon storage or the strength of the ocean's biological pump. If terrestrial sources were dominant, whole ocean δ¹³C would follow atmospheric δ¹³C-CO₂. If oceanic sources from changes in the strength of the biological pump or shifts in ocean ventilation were dominant, the surface-to-deep gradient in δ¹³C in inorganic carbon in the ocean would decrease along with atmospheric δ¹³C-CO₂.

Figure 2

Cross-plot of atmospheric CO₂ and δ¹³C-CO₂ spanning the MIS3 section (blue, this study) and the last deglaciation, which is subdivided into the LGM to HS1 data (red) and late deglaciation (gray). Solid lines indicate linear regressions through the MIS3 (blue) and HS1 (red) data. MIS3 = Marine Isotope Stage 3; LGM = Last Glacial Maximum; HS = Heinrich stadial.

Millennial-scale decreases in the vertical gradient of δ¹³C in the SO have been tentatively correlated to maxima in atmospheric CO₂ (Charles et al., 2010; Ziegler et al., 2013) and, therefore, in light of our new data, minima in δ¹³C-CO₂ (Figures 3a and 3b). If this coupling of the oceanic gradient of δ¹³C and atmospheric δ¹³C-CO₂ could be demonstrated to be precisely in-phase, it would provide clear evidence of an oceanic source controlling atmospheric CO₂. Taking the current data and chronology at face value, the decreases in the oceanic δ¹³C gradient are broadly associated with minima in atmospheric δ¹³C-CO₂, yet the coupling is clearly not one-to-one, possibly due to chronological errors in the marine sediment record. To explore this hypothesis further, we combined existing benthic carbon isotope records from deep SO (~42°S, 10°E, 4,600 m; Charles et al., 1996; Hodell et al., 2001, 2003; Ninnemann et al., 2004). These records use previously established age models that are tied to the Greenland ice core records using carbonate preservation and confirmed by ¹⁴C data during the deglaciation and the Laschamp paleomagnetic event during MIS3 (~42 ka; Barker & Diz, 2014). We include a minor increase in the absolute age of 0.63% to account for the possible undercounting of annual layers in the Greenland Ice Core chronologies relative to the WDC timescale (Buizert et al., 2015).

Figure 3

Comparison of atmospheric record of δ¹³C-CO₂ with oceanic records of δ¹³C and nutrient utilization. (a) Atmospheric δ¹³C-CO₂ (this study) on an inverted y axis. (b) The difference in δ¹³C between subantarctic mode water (SAMW) and circumpolar deepwater (CDW; Ziegler et al., 2013). A stack of benthic δ¹³C (black line) from a combination of record in the deep Southern Ocean (SO; gray circles; Charles et al., 1996; Hodell et al., 2001, 2003; Ninnemann et al., 2004). Also indicated is the uncertainty on these chronologies (Barker & Diz, 2014) based on synchronization between carbonate preservation and Greenland temperature (black boxes) and independent age control based on the Laschamp event (blue boxes). (d) Foraminfera-bound δ¹⁵N indicating nitrate consumption in sub-Antarctic (Martínez-García et al., 2014). Red bars highlight the broad millennia-scale correlations.

We note that the atmosphere and deep ocean δ¹³C are anticorrelated during MIS3 (Figure 3c). This supports the hypothesis that the waxing and waning of respired organic carbon source in the deep SO controlled atmospheric CO₂ and significantly limits the possibility of contributions from terrestrial sources. However, this relationship alone cannot delineate oceanic sources between changes in export productivity and changes in stratification. Evidence to support changes in export productivity comes from the correlation of the ice core data and foraminifera-bound δ¹⁵N which indicates lower nitrate utilization during periods of high atmospheric CO₂ (Martínez-García et al., 2014; Figure 3d). Iron deposition rates and export production in the sub-Antarctic are also closely coupled, suggesting that the extent of iron limitation may have played a role in this enhanced nutrient utilization (Jaccard et al., 2016). Evidence in support of ventilation changes stems from radiocarbon constraints in the South Atlantic that are closely coupled with deep ocean oxygen levels, suggesting that both export productivity and ocean circulation were working in concert over this period (Gottschalk et al., 2016). A quantitative description of this coupling requires study with isotope-enabled Earth system models; however, additional insight can be gained by comparing to periods when these processes may have become uncoupled.

Our new high-resolution MIS3 data now provide a similar sequence of abrupt climate change events to contrast with the last deglaciation. First, we compare the MIS3 data to the last deglaciation utilizing a cross plot of CO₂ and δ¹³C-CO₂ (Figure 2). We see that the MIS3 data fall along the same trend observed during HS1 of the last deglaciation (~18–15.0 ka) yet only span about 50% of range (note that the low δ¹³C-CO₂ values that fall off this trend are due to centennial-scale variability at ~16.3 ka). This suggests that millennial-scale CO₂ variations in MIS3 can be linked mechanistically to the more pronounced variability during the deglaciation. As discussed in previous work, this covariability of CO₂ and δ¹³C-CO₂ is consistent with changes in SO ventilation (Köhler et al., 2006; Menviel, Mouchet, et al., 2015; Schmitt et al., 2012; Tschumi et al., 2011) changes in export production (particularly the SO; Bauska et al., 2016; Menviel et al., 2012) or a weakening of the ocean's biological pump by a reduction in North Atlantic Deep-water (NADW) formation (Schmittner & Lund, 2015).

Second, we compare the two intervals in time with the greenhouse gas records plotted over 8-ka intervals (Figures 4a–4c). Similar patterns of millennial-scale variability in CH₄ and N₂O are observed, but the changes in CO₂ appear fundamentally different. First, the CO₂ rise during HS1 is about 20 ppm greater than the rise in HS4 (or ~2×) and, second, CO₂ remains elevated after the switch to interstadial conditions during the Bølling-Allerød (BA) rather than decreasing as observed after the onset of DO8 (Figure 3c). The δ¹³C-CO₂ follows a similar pattern. In HS1, δ¹³C-CO₂ decreases by 0.2‰ more than HS4 and stabilizes during the BA as opposed to abrupt increases observed during DO8 (Figure 4d). What was different about the deglaciation that allowed more respired carbon into the atmosphere during HS1 than HS4 and prevented, or compensated for, a potential uptake of carbon during the BA?

Figure 4

(left) Comparison of MIS3 data (this study) and last deglaciation (Bauska et al., 2016; Marcott et al., 2014; Rhodes et al., 2015) greenhouse gas variability. (a) WDC CH₄ (Rhodes et al., 2015). (b) Taylor Glacier N₂O. (c) Taylor glacier CO₂. (d) Taylor Glacier δ¹³C-CO₂. (e) Pa/Th, a proxy for the strength of NADW formation (Henry et al., 2016; McManus et al., 2004; orange), (f) the opal flux in the SO (Anderson et al., 2009), a proxy for upwelling (green; note the two different cores and axes are used in the comparison), (g) the benthic to atmosphere ¹⁴C age reconstructions for the subantarctic zone (Gottschalk et al., 2016; Skinner et al., 2010), (h) nssCa flux from the EDML ice core (Fischer et al., 2007), a proxy for dust delivery to Antarctica (brown), (i) opal flux in the subantarctic zone (Gottschalk et al., 2016), a proxy for productivity (from the same core as the ventilation reconstruction). Yellow and gray bars highlight the Heinrich Stadial jumps, black bars indicted the initial CO₂ rises, blue and purple bars highlight the onset of interstadials. (right) Centennial-scale variability during the last glacial period from Taylor glacier and WDC. Anomalies in the concentration and isotopic composition of the greenhouse gases are plotted with the relative timing determined by the midpoint in the CH₄ rise. Marker color corresponds to dashed lines in left panel. MIS3 = Marine Isotope Stage 3; NADW = North Atlantic Deep-water; SO = Southern Ocean; HS = Heinrich stadial; DO = Dansgaard-Oeschger.

During the Heinrich stadials, NADW weakens (Henry et al., 2016), and subantarctic productivity decreases (Anderson et al., 2014; Gottschalk et al., 2016) tracking dust delivery to Antarctica (Fischer et al., 2007; Figures 4e, 4h, and 4i). Ventilation of the SO (as inferred from radiocarbon) improves (Gottschalk et al., 2016; Skinner et al., 2010), and SO upwelling (as inferred from Antarctic productivity) increases (Anderson et al., 2009), although these changes could be more pronounced in the later part of the Heinrich stadials (Figures 4f and 4g). During MIS3, all of these changes are largely symmetric around, and consistent with, the minimum in δ¹³C-CO₂ and thus plausible drivers the change in CO₂.

During the deglaciation, some variables trend back toward Last Glacial Maximum values after HS1 (NADW and SO upwelling), while others show near permanent shifts to interglacial levels during HS1 (subantarctic productivity and SO ventilation). This decoupling allows us to partially disentangle which processes control atmospheric CO₂. Based on relationship between the proxies and atmospheric data in MIS3, we would expect that if changes in NADW and/or SO upwelling were to control atmospheric CO₂, we would observe a large CO₂ decrease and δ¹³C-CO₂ increase in the BA. This is clearly not the case and requires either a muted response to these forcings or another source of carbon in the BA that compensated for the apparent carbon sink. Conversely, if changes in subantarctic productivity and SO ventilation dominated the atmospheric CO₂ budget, we could expect CO₂ to remain elevated during the BA and δ¹³C-CO₂ to plateau at lower values, a scenario that is in much better agreement with the data. Thus, subantarctic productivity and SO ventilation appear to have a more consistent link with atmospheric CO₂ in both MIS3 and the Last Deglaciation and are strong candidates for contributing significantly to glacial-interglacial CO₂ change.

4.2 Centennial-Scale Carbon Cycle Variability

On the centennial scale, our new observations in MIS3 can be combined with recently identified variability in Last Deglaciation to suggest a ubiquitous and consistent coupling of the carbon cycle with abrupt climate change events. In Figures 4j–4q, we plot variability in greenhouse gases for two categories of centennial-scale events: the onset of interstadials (D08, the BA, and the preboreal) and mid-Heinrich stadials events (H4 and H1). The changes are plotted as anomalies, and the timing is set relative to the midpoint of the rise in CH₄. Note that the Taylor Glacier chronology is synchronized with the WAIS Divide record via CH₄, and thus, the phasing of the two ice cores cannot be interpreted.

The rapid 8-ppm CO₂ rise during HS4 at 39.5 ka likely shares a common origin with a similar event during the deglaciation with HS1 at 16.3 ka (Marcott et al., 2014; Figure 4l). Both events are associated with carbon isotope minima (Figure 4m). It has been suggested that these midstadial events can be tied to the timing of the Heinrich events and may represent a rapid release of terrestrial carbon to the atmosphere driven by a cooling and drying of the Northern Hemisphere (Bauska et al., 2016). The event in HS4 is consistent with a terrestrial origin as it coincides with an increase in CH₄ (Rhodes et al., 2015; Figure 4j), which may indicate enhanced precipitation of the Southern Hemisphere tropics, and an increase in Δε_LAND, which suggests decreased precipitation in the Northern Hemisphere (Seltzer et al., 2017; Severinghaus et al., 2009; Figure 1h). Constant N₂O indicates either that changes in terrestrial soil temperatures may have been small on the global scale, thus suggesting that precipitation was the dominant driver of the terrestrial carbon loss, or that oxygen-minimum zones in intermediate ocean were relatively stable, possibly indicating the absence of a change in ocean circulation across this event (Figure 4k).

The 6-ppm CO₂ rise at the onset of DO8 shares common features with the onset of the BA and preboreal (~11.5 ka) during the deglaciation (Figure 4p). All three events exhibit simultaneous increases in CO₂, CH₄, and N₂O that coincide with abrupt Northern Hemisphere warmings, continued warming, or at least stable temperatures in Antarctica (WAIS Divide Project Members, 2015) and greater NADW formation (Henry et al., 2016; McManus et al., 2004). The δ¹³C-CO₂ across all events is variable but shows no secular trend (preboreal) or slight increase of ~0.08‰ (BA and DO8; Figure 4q). At face value, this pattern of increasing CO₂ and increasing δ¹³C-CO₂ indicates that rising ocean temperature contributed to the CO₂ rise with additional (but limited) changes in the net flux of organic carbon. This simplest of scenarios is somewhat surprising given that changes in the ocean's biological pump may accompany the large and abrupt reorganization of ocean circulation, and changes in terrestrial carbon reservoirs are clearly indicted by the large increases in CH₄ and N₂O. Recently, the LOVECLIM model, which predicts a small positive relationship between CO₂ and δ¹³C-CO₂ during reduced NADW (Menviel, Spence, et al., 2015), also predicts increases in CO₂ of 10 to 15 ppm upon the resumption of NADW (with the effect of solubility contributing about 50% to CO₂ variability; Menviel, Mouchet, et al., 2015). Moreover, a precisely dated coral record shows that these events during the last deglaciation are associated with brief intervals of enhanced overturning in the Atlantic (Chen et al., 2015). This integrated response to the onset of an interstadial is consistent with the CO₂ and δ¹³C-CO₂ data and may be a pervasive feature of last glacial period CO₂ variability but requires ground-truthing with additional, high-resolution MIS3 marine records.

5 Conclusions

Carbon isotope data from the last deglaciation and last glacial period clearly show that CO₂ variability is the sum of multiple mechanisms, many of which are triggered by abrupt climate change. During both periods, millennial-scale variability is present and likely associated with the release of respired organic carbon from the deep ocean. Superimposed on these oscillations are two types of centennial-scale changes, (i) CO₂ increases and δ¹³C-CO₂ decreases in the middle of Heinrich stadials and (ii) CO₂ increases and small changes in δ¹³C-CO₂ that are in-phase with rapid increases in NH temperature. During the deglaciation, the millennial-scale component is enhanced, and an additional carbon source is required to sustain the CO₂ rise through the entire deglaciation. This suggests that although abrupt climate variability is not the sole driver of the deglacial CO₂ rise, it may be a prerequisite. These potential links can now be tested with model experiments that use the MIS3 data to constrain the sensitivity to centennial and millennial-scale components and the deglacial data to evaluate how these mechanisms interact with changes in insolation, ice volume, and global temperatures.

Acknowledgments

This work was funded by NSF grants ANT 0838936 (Oregon State University) and ANT 0839031 (Scripps Institution of Oceanography). Data will be made available online at National Climate Data Center (https://www.ncdc.noaa.gov/paleo/study/24170).