Carbon isotopes characterize rapid changes in atmospheric carbon dioxide during the last deglaciation (original) (raw)

Significance

Antarctic ice cores provide a precise, well-dated history of increasing atmospheric CO2 during the last glacial to interglacial transition. However, the mechanisms that drive the increase remain unclear. Here we reconstruct a key indicator of the sources of atmospheric CO2 by measuring the stable isotopic composition of CO2 in samples spanning the period from 22,000 to 11,000 years ago from Taylor Glacier, Antarctica. Improvements in precision and resolution allow us to fingerprint CO2 sources on the centennial scale. The data reveal two intervals of rapid CO2 rise that are plausibly driven by sources from land carbon (at 16.3 and 12.9 ka) and two others that appear fundamentally different and likely reflect a combination of sources (at 14.6 and 11.5 ka).

Keywords: ice cores, paleoclimate, carbon cycle, atmospheric CO2, last deglaciation

Abstract

An understanding of the mechanisms that control CO2 change during glacial–interglacial cycles remains elusive. Here we help to constrain changing sources with a high-precision, high-resolution deglacial record of the stable isotopic composition of carbon in CO2 (δ13C-CO2) in air extracted from ice samples from Taylor Glacier, Antarctica. During the initial rise in atmospheric CO2 from 17.6 to 15.5 ka, these data demarcate a decrease in δ13C-CO2, likely due to a weakened oceanic biological pump. From 15.5 to 11.5 ka, the continued atmospheric CO2 rise of 40 ppm is associated with small changes in δ13C-CO2, consistent with a nearly equal contribution from a further weakening of the biological pump and rising ocean temperature. These two trends, related to marine sources, are punctuated at 16.3 and 12.9 ka with abrupt, century-scale perturbations in δ13C-CO2 that suggest rapid oxidation of organic land carbon or enhanced air–sea gas exchange in the Southern Ocean. Additional century-scale increases in atmospheric CO2 coincident with increases in atmospheric CH4 and Northern Hemisphere temperature at the onset of the Bølling (14.6–14.3 ka) and Holocene (11.6–11.4 ka) intervals are associated with small changes in δ13C-CO2, suggesting a combination of sources that included rising surface ocean temperature.

Over thirty years ago ice cores provided the first clear evidence that atmospheric CO2 increased by about 75 ppm as Earth transitioned from a glacial to an interglacial state (1, 2). After decades of research, the underlying mechanisms that drive glacial–interglacial CO2 cycles are still unclear. A tentative consensus has formed that the deglaciation is characterized by a net transfer of carbon from the ocean to the atmosphere and terrestrial biosphere, through a combination of changes in ocean temperature, nutrient utilization, circulation, and alkalinity. Partitioning these changes in terms of magnitude and timing is challenging. Estimates of the glacial–interglacial carbon cycle budget are highly uncertain, ranging from 20–30 ppm for the effect of rising ocean temperature, 5–55 ppm for ocean circulation changes, and 5–30 ppm for decreasing iron fertilization (3, 4), with feedbacks from CaCO3 compensation accounting for up to 30 ppm (5, 6).

A precise history of the stable isotopic composition of atmospheric carbon dioxide (δ13C-CO2) can constrain key processes controlling atmospheric CO2 (7, 8). A low-resolution record from the Taylor Dome ice core (9) identified a decrease in δ13C-CO2 at the onset of the deglacial CO2 rise that was followed by increases in both CO2 and δ13C-CO2 (Fig. 1). A higher-resolution record from the European Project for Ice Coring in Antarctica Dome C (EDC) ice core (10) provided additional support for the rapid δ13C-CO2 decrease associated with the initial CO2 rise, and box modeling indicated that this decrease was consistent with changes in marine productivity. The record also included other rapid changes in δ13C-CO2, albeit at low precision, supporting large variations of organic carbon fluxes, notably a sharp increase in δ13C-CO2 during the Bølling–Allerød (BA) interval attributed to carbon uptake by the terrestrial biosphere. A combined record including higher-precision EDC and Talos Dome data (11) documented a δ13C-CO2 decrease beginning near 17.5 ka. This shift in δ13C-CO2 was interpreted to indicate that some process in the Southern Ocean (SO), possibly changes in upwelling, drove the initial CO2 rise. This previous work did not resolve high-frequency variability in the δ13C-CO2 records that may be essential for discerning mechanisms of change.

Fig. 1.

Fig. 1.

Carbon isotope records during the last deglaciation. Taylor Glacier δ13C-CO2 data from this study (red). Previous work from Taylor Dome (gray open circles) (9), Grenoble EDC data (open green squares) (10), Bern EDC data (orange circles) (11, 45), sublimation measurements from EDC (blue triangles), and Talos Dome (purple squares) with an estimate of the 1-sigma uncertainty from a compilation of previous ice core δ13C-CO2 data (11).

Here we use an analytical method (12) that employs dual-inlet isotope ratio mass-spectrometry to obtain precision approaching that of modern atmospheric measurements [∼0.02‰ 1-sigma pooled SD based on replicate analysis compared with ∼0.05–0.11‰ for previous studies (9–11)]. We extracted atmospheric gases from large (400–500 g) samples taken from surface outcrops of ancient ice at Taylor Glacier, Antarctica, at an average temporal resolution of 165 y between 20 and 10 ka, and subcentury resolution during rapid change events. This resolution allows us to delineate isotopic fingerprints of rapid shifts in CO2 that were previously impossible to resolve. Our study complements recent precise observations of CO2 concentration variations during the last deglaciation, which revealed abrupt centennial-scale changes (13) (Fig. 2).

Fig. 2.

Fig. 2.

Carbon cycle changes of the last deglaciation. WAIS Divide continuous CH4 (green) (14) and discrete CO2 (blue) (13) concentration data plotted with Taylor Glacier CO2 and δ13C-CO2 data (this study) (red markers, black line is a smoothing spline), the five-point running Keeling intercept with shading indicating the R2 for each time interval. Blue bars indicate intervals of rapid CO2 rise identified in the WAIS Divide ice core (13).

During the initial 35-ppm CO2 rise from 17.6 to 15.5 ka, we find a 0.3‰ decrease in δ13C-CO2 that is interrupted by a sharp minimum coincident with rapid increases in CO2 and CH4 around 16.3 ka (13, 14) (Fig. 2). The 16.3-ka feature in the CO2 and CH4 concentration records, which corresponds to a 0.1‰ negative excursion in δ13C-CO2, has been plausibly tied to the timing of Heinrich event 1 (13, 14) and signals a mode switch in the deglacial CO2 rise. The subsequent slower rise in CO2 from 15.5 to 14.8 ka is not accompanied by large changes in δ13C-CO2. Across the Oldest Dryas to Bølling transition (14.6–14.3 ka) and coincident with a 10-ppm CO2 increase and large CH4 increase, we resolve a 0.08‰ increase in δ13C-CO2 (Fig. 2). Rapid increases in CO2 and CH4 at the Younger Dryas (YD) to Preboreal transition (11.6–11.4 ka) are associated with minor variability δ13C-CO2. On the other hand, the onset of the YD (12.8–12.5 ka) is characterized by a small rise in CO2 associated with a 0.15‰ decrease in δ13C-CO2 that appears tightly coupled to the timing of the large CH4 decrease. The recovery from this excursion is characterized by increasing CO2 and δ13C-CO2. Broadly, our data confirm the results of Schmitt et al. (11) (Fig. 1). However, some of the large swings in δ13C-CO2 indicated by the earlier EDC record (10), may be inaccurate and require reexamination.

Processes Controlling Atmospheric δ13C-CO2

To visualize the constraints provided by δ13C-CO2 on the processes controlling CO2 we use a cross-plot of CO2 and δ13C-CO2 (referred to as a Keeling plot when the x axis is equal to 1/CO2). When the classic Keeling plot is applied to a two-component system, and assuming conservation of mass, the y axis intercept of a linear regression to the data (_y_0) approximates the δ13C signature of a secondary external source mixing with a primary source (15). In the more complex mixing between the atmosphere, ocean, and terrestrial biosphere, _y_0 is still indicative of the source reservoir’s δ13C signature but interpretation requires a model of the carbon cycle to account for processes such as air–sea gas exchange, ocean mixing, and ocean–sediment interactions, which buffer the atmospheric δ13C-CO2 signature over long timescales and reduce the slope of δ13C:1/[CO2] (7). We use a simple box model of the carbon cycle, previously published box model experiments (7) and intermediate complexity models to account for these effects and deconvolve the processes responsible for the deglacial rise in CO2. We divide the processes into the following categories: ocean productivity and circulation, land carbon storage (particularly rapid changes), ocean temperature, the CaCO3 cycle, and air–sea gas exchange. Individual model Keeling plot intercepts are listed in SI Appendix, Table S1; the ranges within each category are represented graphically in Fig. 3_A_.

Fig. 3.

Fig. 3.

Cross-plot of data constraints and model experiments. (A) Shaded lines show the range of model-based constraints on various carbon cycle processes as listed in SI Appendix, Table S1 (changes in ocean biological pump/circulation, yellow; deglacial increase in SST, blue; rapid release of land carbon, green; rapid change in Southern Ocean gas exchange, purple; CaCO3 cycle, gray). (B) All Taylor Glacier data with arrows as guides to the approximate time path. (C) The data divided into the early HS1 (yellow) and later deglaciation (blue) modes of variability. Colored markers divide the data by time period and the shaded vectors indicate the linear regressions of the data with the 1-sigma uncertainty. (D) Further division of the data into the abrupt changes at The 16.3-ka event and onset of the YD (red). See SI Appendix, Table S2 for statistics.

An oceanic δ13C-DIC depth gradient is controlled by a combination of ocean mixing and export from the surface of isotopically light organic carbon. Decreased carbon export or increased ocean ventilation during the last deglaciation would decrease the δ13C-DIC gradient, leading to a rise in atmospheric CO2 and decrease in atmospheric δ13C-CO2. First, we simulate a plausible signature of a glacial–interglacial weakening of the marine biological pump by forcing a decrease in the strength of the Subantarctic biological pump from near full efficiency (PO4 = 0.2 mmol m−3) to preindustrial level (PO4 = 1.4 mmol m−3). The timing of this change is scaled to the decrease in dust delivery to Antarctica (16). This leads to an increase in CO2 and decrease of δ13C-CO2; the relationship is characterized in our model with _y_0 equal to −8.6‰. For comparison, factorial experiments with the Bern 3D model of the last glacial–interglacial cycle isolate the effect of iron fertilization in the model (17) with a resultant Keeling intercept of −9.6‰. Second, we vary the rate of SO upwelling in our box model. Greater ocean ventilation raises CO2 and lowers δ13C-CO2 resulting in a Keeling plot intercept of −8.4‰. Experiments with the Bern 3D model where wind stress over the SO was varied (20–180%) show that atmospheric CO2 positively correlates with the rate of ocean overturning (18). The simulated CO2 and δ13C-CO2 produce a Keeling plot intercept between _−_7.6‰ [when the Atlantic Meridional Overturning Circulation (AMOC) is in an “on” state] and −8.6‰ (when AMOC is in an “off” state). Finally, decreased AMOC has been hypothesized to slow the delivery of low preformed nutrient water to the deep ocean and consequently drive a weakening of the biological pump. Experiments with the Model of Ocean Biogeochemistry and Isotopes/University of Victoria climate model of intermediate complexity that simulated an interval of collapsed AMOC show atmospheric CO2 rising and δ13C-CO2 decreasing (19) (_y_0 = −8.5‰). Combining all these experiments provides plausible constraints on oceanic sources to the atmosphere (_y_0 ∼= −7.4 to −9.6‰), which include both productivity driven changes (e.g., iron fertilization) and deep-ocean ventilation change (e.g., enhanced turnover of deep water masses) (Fig. 3_A_).

During the deglaciation, the release of oceanic carbon to the atmosphere is likely partially offset by the gradual accumulation of carbon on land, primarily during the later part of the deglaciation when the major ice sheets are small and dwindling (14–10 ka) (20). This land carbon uptake would lower atmospheric CO2 and increase δ13C-CO2. Because the magnitude of carbon isotopic fractionation by photosynthesis is very similar in the marine and terrestrial regimes, changes in organic carbon cycling between the land, atmosphere, and ocean that are slower than the timescale of ocean mixing are broadly indistinguishable from the atmospheric data alone. However, for changes in organic land carbon storage that are rapid relative to the mixing time of carbon in the ocean–atmosphere system, the atmospheric signal will more closely reflect the isotopic signature of organic carbon and then diminish as it is buffered by exchange with the deep ocean. We drive changes in atmospheric CO2 of about 10 ppm by varying the land-to-atmosphere carbon flux in our model at periodicities of 50, 500, and 5,000 y. The change in δ13C-CO2 decreases with increasing periodicity resulting in _y_0 of −13.4‰, −10.9‰, and −9.8‰ at the 50, 500, and 5,000 y periodicities, respectively. Fast land carbon fluxes to the atmosphere can therefore be distinguished from changes in the ocean biological pump with high-resolution atmospheric data (_y_0 = ∼<−10.9‰; Fig. 3_A_).

Increasing ocean temperature decreases both the solubility of CO2 in seawater and the magnitude of isotopic fractionation during air–sea gas exchange. Rising atmospheric CO2 and increasing δ13C-CO2 are therefore consistent with ocean warming. Forcing our model with latitudinal temperature stacks (21) for the deglaciation results in a 35-ppm rise in atmospheric CO2 with an increase of about 0.3‰ in δ13C-CO2 with an apparent _y_0 from this effect of −4.5‰ (Fig. 3_A_).

Carbon isotopic fractionation during CaCO3 formation from seawater is very small compared with that of photosynthesis, and the δ13C of CO2 from volcanic emissions, though poorly constrained, is very similar to atmospheric values. Processes like CaCO3 compensation, reef building, and volcanic emissions are thus consistent with rising CO2 and little to no change in δ13C-CO2 (_y_0 = initial atmospheric δ13C-CO2; Fig. 3_A_). Moreover, a decrease in the amount of respired carbon in the deep ocean, driven by either oceanic changes or land carbon regrowth, will trigger increases in CaCO3 preservation (and corresponding production of CO2) that act to restore [CO32−] over multimillennial timescales (22, 23). The indirect effect of a weakened biological pump or land carbon regrowth would thus be an increase in CO2 with little change in δ13C-CO2 over thousands to tens of thousands of years. In Keeling plot space the inferred intercept of a weakened biological pump would slowly asymptote to the CaCO3 intercept. On even longer timescales the δ13C-CO2 signature of all processes is dampened toward a steady state determined by the input of volcanic and weathering fluxes of carbon to the atmosphere/ocean (105 y) (24). The box model experiments presented here largely exclude CaCO3 feedbacks (SI Appendix) and thus represent only the direct effect of various carbon cycle processes that are important in constraining the isotope signature on the centennial-to-millennial timescale, at the expense of underestimating the long-term feedbacks that are significant on glacial–interglacial timescales.

Changes in air–sea gas exchange, via changes in sea-ice extent or wind speed in the SO, are hypothesized to have a significant impact on CO2 and δ13C-CO2, generally with increased air–sea gas exchange leading to increases in atmospheric CO2 and large decreases in δ13C-CO2 (7, 25). In our model, varying the air–sea gas exchange coefficient over the SO by ±50% over periods of 50, 500, and 5,000 y (but keeping ocean mixing constant) produces small rises in CO2 and a sharp decrease in δ13C-CO2 when air–sea gas exchange is enhanced. Keeling plot intercepts vary greatly with the periodicity of forcing (−37‰ to −18‰) but are consistent with results from the Box model of the Isotopic Carbon cYCLE (BICYLE) (SI Appendix, Table S1). Changes in air–sea gas exchange could therefore contribute to rapid δ13C-CO2 variability.

Identifying and Diagnosing Deglacial δ13C-CO2 Variability

All of the above processes can work in combination, leading to a system that is fundamentally underconstrained by the CO2-to-δ13C-CO2 relationship. Nonetheless, the changing relationship between CO2 and δ13C-CO2 with time can be combined with the model constraints to divide the data into time intervals based on the dominant processes (Fig. 3_B_). We identify four major patterns of variability of the carbon cycle spanning the length of our record: (i) a millennial-scale increase in CO2 and decrease in δ13C-CO2 during the early part of Heinrich Stadial 1 (17.6–15.5 ka); (ii) rising CO2 with generally increasing δ13C-CO2 during the later portion of Heinrich Stadial 1 (15.5–14.6 ka) and later portion of the YD (12.8–11.5 ka); (iii) rising CO2 with centennial-scale negative isotopic excursions at 16.3 and 12.8 ka; and (iv) centennial-scale CO2 rises with minor changes in δ13C-CO2 at 14.6 and 11.5 ka. SI Appendix, Table S2 provides the Keeling plot intercepts for some of these intervals and Fig. 3 C and D show the divisions graphically.

From 17.6 to 15.5 ka δ13C-CO2 decreases by about 0.3‰ and CO2 increases by about 35 ppm. This distinct phase of the deglacial CO2 rise was previously identified with less precise data and attributed to an increase in SO upwelling (11). Excluding the excursion around 16.3 ka (see below) the strong relationship between CO2 and δ13C-CO2 across this interval (_R_2 = 0.97, _y_0 = −8.6‰; Fig. 3_C_) is consistent with the bulk of the CO2 increase being driven by a weakening of the efficiency of the biological pump. The decrease in atmospheric δ13C-CO2 is thus consistent with either increased ocean ventilation (11, 26, 27), an increase in the ocean preformed nutrient content driven by a decrease of North Atlantic Deep Water (NADW) formation (28), or a decrease in the export of organic carbon to the deep ocean (29).

The timing of the δ13C-CO2 decrease coincides with the deglacial decrease in the dust flux over Antarctica and may lead the inferred maximum in SO upwelling. By 16.0 ka, the non-sea-salt calcium flux at the Talos Dome ice core site (30) had decreased to near interglacial levels, whereas the SO opal flux recorded at 53.2°S, 5.1°E (31) was still increasing to values that peaked between 15.5 and 14.5 ka (Fig. 4_A_). Our data thus support the hypothesis that a decrease in iron fertilization was important during the earliest stages of the last deglaciation (32). The magnitude of the direct effect of iron fertilization is partially constrained by the data, suggesting an upper limit of 35 ppm CO2 change from this mechanism. This is consistent with empirically derived estimates from the relationship between atmospheric CO2 and Subantarctic productivity over multiple glacial–interglacial cycles (∼40 ppm) (33) and the coupling of CO2 and ice core proxies for dust delivery during the last glacial–interglacial cycle (≤ 40 ppm) (30). However, state-of-the-art biogeochemistry models simulate smaller glacial–interglacial changes due to iron fertilization between 8 and 15 ppm (34–36). Model and empirical estimates could be reconciled if other mechanisms for lowering the efficiency of the biological pump (e.g., ocean ventilation) are working in concert with iron fertilization during this interval and account for part of the 35 ppm increase in atmospheric CO2. Possibly, a fast response of the carbon cycle to iron fertilization is superimposed on a slower change driven by upwelling, resulting in the two distinct rates of CO2 rise during HS1.

Fig. 4.

Fig. 4.

Climate and carbon cycle changes during the last deglaciation. (A) Proxies for Greenland temperature (46) (purple), West Antarctic temperature (37, 47) (blue), East Asian precipitation (48) (green), dust delivery to Antarctica (30) (yellow), Southern Ocean upwelling (31) (blue markers), global temperature relative to the early Holocene (blue banding) (21), and the Taylor Glacier CO2 and δ13C-CO2 data (red). The red bars indicate periods of rapid δ13C-CO2 decreases; blue bars indicate rapid CO2 increases with slight increases or little change in δ13C-CO2. B and C highlight the changes in temperature and atmospheric CO2 at the centennial scale.

From 15.5 to 11 ka, atmospheric CO2 increases by 40 ppm and δ13C-CO2 gradually increases (_y_0 = −6.2‰; Fig. 3_C_) with a plateau during the BA (∼14.6–13.0 ka) at values of 244 ± 2 ppm and −6.63 ± 0.04‰ for CO2 and δ13C-CO2, respectively. Broadly, the large increase in CO2 and small overall change in δ13C-CO2 is consistent with the atmospheric CO2 increase being driven by changes in the CaCO3 cycle, volcanic emissions, or concurrent changes in organic carbon cycle and ocean temperature. Most studies conclude that CaCO3 feedbacks account for up to 30 ppm of glacial–interglacial CO2 change (with an e-folding timescale of ∼5,000 y; refs. 5, 6). Though certainly significant in controlling the glacial–interglacial CO2 and δ13C-CO2 differences, these effects are too slow to explain the rapid increases in atmospheric CO2 around 14.75 ka (∼10 ppm over 200 y) and 12.9–11.5 ka (∼30 ppm over 1,500 y). Moreover, a CO2 increase driven by only the CaCO3 cycle or volcanic emissions would produce no variability in δ13C-CO2. Instead, changes in δ13C-CO2 of about 0.1‰ on the centennial timescale suggest that these changes are in part driven by a combination of rising SST and 13C-depleted sources (i.e., ocean ventilation, land carbon). Allowing for relatively small contributions from CaCO3 cycling and volcanic emissions, the δ13C-CO2 data from 15.5 to 11.0 ka are consistent with a roughly equal mix of sources from rising ocean temperature and a weakened biological pump. The trend to more positive δ13C-CO2 suggests that the temperature effect was slightly greater (60 ± 10%, assuming two end-member mixing). Note that any sources of atmospheric CO2 from the CaCO3 cycle or volcanic emissions would decrease the inferred absolute changes of these two processes but have little effect on their relative contribution. The greater importance for temperature-driven changes in the later compared with the earlier part of the CO2 rise is consistent with the increase in global surface temperature lagging the increase in atmospheric CO2 (21).

A recent high-resolution record from the West Antarctic Ice Sheet (WAIS) Divide ice core demonstrated that rapid increases in CO2 of about 12 ppm at both the onset of the BA (14.6 ka) and end of the YD (11.5 ka) occurred exactly coincident with abrupt increases in CH4 and Northern Hemisphere (NH) temperature (13). Our data suggest that the effect of rising sea surface temperature (SST) on atmospheric CO2 may be most pronounced during these two distinct intervals. Moreover, the WAIS Divide ice core revealed that Antarctic temperature remained stable or even continued to warm until ∼200 y after the onset of NH warming (37) (Fig. 4 B and C). A global temperature reconstruction, though uncertain on centennial timescales, records temperature increases of ∼1 and ∼0.5 °C at the onset of the BA and end of the YD, respectively (21). At the onset of the BA, our records show a 12-ppm increase in CO2 and a 0.1‰ increase in δ13C-CO2, consistent with SST dominating the atmospheric CO2 budget. At the end of the YD, we observe very little change in δ13C-CO2 during a 10-ppm rise in atmospheric CO2, suggesting a balanced contribution from 13C-depleted carbon sources and rising SST. This relationship between ocean warming and rising CO2 suggests an important positive climate–carbon feedback that may be operating on the centennial timescale. These observations also constrain hypotheses that organic carbon sources explain the atmospheric CO2 increases associated with NH temperature rise. Thawing NH permafrost at the onset of the BA (38) or ocean “flushing” events tied to the resumption of AMOC (39) would need to be compensated by carbon sinks that are more depleted in 13C (i.e., a steeper vector in the Keeling plot that leads to a net increase in CO2 and slight increase in δ13C-CO2) or accompanied by sources that enrich the atmosphere in 13C.

Two significant features in our record are the sharp century-scale minima in δ13C-CO2 centered at 16.3 and 12.8 ka. These two events are associated with significant increases in atmospheric CO2 of about 7 ppm and very rapid decreases in δ13C-CO2 of nearly 0.2‰ (Fig. 3_D_). The higher resolution WAIS Divide record (13) indicates that the atmospheric CO2 increase during the 16.3-ka event is greater in magnitude (∼12 ppm) and more rapid than our Taylor Glacier data resolves. The abrupt CO2 increase at 16.3 ka could plausibly be interpreted as an event superimposed on a 3-kyr-long trend of rising atmospheric CO2 from about 17.6–14.75 ka, whereas the 12.9-ka excursion appears to occur near the beginning of a relatively rapid atmospheric CO2 increase from 12.9 to 11.5 ka. The 16.3-ka event is also associated with a small CH4 increase that has been attributed to a rapid increase in Southern Hemisphere (SH) methane sources, perhaps associated with a southward excursion of the intertropical convergence zone (ITCZ) associated with Heinrich event 1 (14).

Our modeling experiments show that these features are consistent with a rapid release of terrestrial carbon to the atmosphere over a period of a few hundred years or less (accounting for the smoothing effect of gas trapping in the firn; SI Appendix, Fig. S10). These two events occurred during intervals of very weak monsoon strength in the northern tropics and some of the coldest conditions in the high-latitude NH (Fig. 4). Model experiments suggest that colder and drier conditions following collapse of the AMOC can drive decreases in land carbon stocks in the high-latitude NH (40). Although the global net balance in the model depends on the background climate and vegetation, net increases in atmospheric CO2 occur under glacial conditions. Our data thus suggest a possible link between tropical CH4 production and high-latitude terrestrial carbon pools driven by centennial-scale cold periods and/or drought during the deglaciation.

Alternatively, the minima in δ13C-CO2 are consistent with rapid CO2 increases driven in part by periods of enhanced air–sea gas exchange. As shown earlier, δ13C-CO2 can be highly sensitive to changes in air–sea gas exchange. The precipitous drops in δ13C-CO2 may reflect intervals of enhanced air–sea gas exchange that, in combination with other 13C-depleted sources, drive increases in atmospheric CO2.

Another possible scenario that can produce a Keeling plot intercept of less than the typical oceanic end member involves a combination of a weakening biological pump source that is moderated by a smaller CO2 sink from decreasing SST. The combination of the two vectors could produce an intercept that is more negative (<−8.6‰) than the biological pump signature alone. Although this scenario is unlikely for the 16.3-ka event where we observe a significant and rapid increase in atmospheric CO2 with no large SST decreases, the δ13C-CO2 decrease at the onset of the YD is associated with a strong winter cooling in the NH, particularly in the North Atlantic. The subsequent recovery of δ13C-CO2 following the minimum would likely require a source of atmospheric CO2 from increasing SST, possibly from a delayed warming in the SH.

Conclusion

Many possible scenarios can explain the evolution of atmospheric CO2 and δ13C-CO2 during the deglaciation. Narrowing the range of scenarios is possible if the observed changes in carbon cycle can be consistently coupled to the climate history. We conclude by outlining one possible scenario that links our observations of centennial-scale features and the broader millennial-scale changes within the context of the deglacial climate transition (Fig. 4_A_). During the early part of HS1, the collapse of the AMOC (41) decreased heat transport to the North Atlantic. In response, large areas of the NH cooled and the SH warmed; possibly the ITCZ shifted southward and SH westerlies shifted southward or strengthened (42–44). We hypothesize that a shift of the westerlies off the SH continents and/or increased SH precipitation led to a precipitous decline in dust delivery over the Subantarctic ocean, driving up to 35 ppm of the CO2 rise from about 17.6–15.5 ka. The southward migration of the ITCZ also led to drying in parts of the NH, possibly causing a reduction in organic land carbon, most notably around 16.3 ka. Alternatively, or additionally, the changing SH westerlies reached a threshold around 16.3 ka, in which wind speed over the SO increased, leading to enhanced air–sea gas exchange and possibly greater upwelling.

During the later half of HS1 (15.5–14.6 ka), dust deposition in the SO had fallen to interglacial levels and further CO2 rise was driven mostly by warming ocean temperatures and an additional weakening of the biological pump, with a peak in SO upwelling or an extended interval of AMOC collapse as two possible mechanisms. During the YD (12.9–11.5 ka) most of the CO2 increase was driven by similar processes to the later half of HS1. However, the initial rise in CO2 during the YD could have been driven by either a second loss of land carbon or renewed enhancement of SH westerlies. At the onset of the BA (14.6 ka) and end of the YD (11.5 ka) significant warmings in the NH and continued warming around Antarctica likely contributed to the centennial-scale increases in atmospheric CO2.

The δ13C-CO2 record shows that the deglacial increase in atmospheric CO2 occurred in a series of steps, each with a δ13C fingerprint that suggests that different mechanisms may have been triggered at various times during the deglacial transition. Early in the transition, the decrease in δ13C-CO2 is consistent with (albeit not uniquely) a combination of atmospheric CO2 sources from respired organic carbon that exceeded sources from rising ocean temperature. This suggests that the initial trigger for the deglacial CO2 rise involved either an ocean circulation or ocean biological process. Later in the transition, the relatively stable δ13C-CO2, punctuated by centennial-scale changes, suggests a combination of sources that could include changes in the CaCO3 cycle or volcanic emissions, but most likely reflects a balanced contribution of respired organic carbon and rising ocean temperature that strengthens and weakens over time. At least twice during the deglaciation a rapid release of 13C-depleted carbon to the atmosphere may have occurred over a few centuries, suggesting that abrupt and significant releases of CO2 to the atmosphere may be common nonlinear features of Earth’s carbon cycle. Further work on defining the isotopic signature of glacial–interglacial CO2 mechanisms across a suite of carbon cycle models could yield a more precise understanding of CO2 sources during the deglaciation.

Supplementary Material

Supplementary File

Acknowledgments

We thank Tanner Kuhl, Robb Kulin, and Paul Rose for assistance in the field and Fortunat Joos, Laurie Menviel, and Andreas Schmittner for sharing model output. We thank Mathis Hain and an anonymous reviewer for comments that improved the paper. We are grateful for technical support from NSF-funded Ice Drilling Design and Operations (University of Wisconsin) and the OSU/CEOAS Stable Isotope Laboratory, in particular Andy Ross. This work was funded by NSF Grants ANT 0838936 (Oregon State University) and ANT 0839031 (Scripps Institution of Oceanography). Further support came from the Marsden Fund Council from New Zealand Government funding, administered by the Royal Society of New Zealand and NIWA under Climate and Atmosphere Research Programme CAAC1504 (to H.S.). T.K.B. was partially supported by the Comer Science and Education Foundation.

Footnotes

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

References

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary File