Atmospheric 14C/12C changes during the last glacial period from Hulu Cave

See allHide authors and affiliations

Science  14 Dec 2018:
Vol. 362, Issue 6420, pp. 1293-1297
DOI: 10.1126/science.aau0747

The whole story

An accurate, precise record of the carbon-14 (14C) content of the atmosphere is important for developing chronologies in climate change, archaeology, and many other disciplines. Cheng et al. provide a record that covers the full range of the 14C dating method (∼54,000 years), using paired measurements of 14C/12C and thorium-230 (230Th) ages from two stalagmites from Hulu Cave, China. The advantage of matching absolute 230Th ages and 14C/12C allowed the authors to fashion a seamless record from a single source with low uncertainties, particularly in the older sections.

Science, this issue p. 1293


Paired measurements of 14C/12C and 230Th ages from two Hulu Cave stalagmites complete a precise record of atmospheric 14C covering the full range of the 14C dating method (~54,000 years). Over the last glacial period, atmospheric 14C/12C ranges from values similar to modern values to values 1.70 times higher (42,000 to 39,000 years ago). The latter correspond to 14C ages 5200 years less than calibrated ages and correlate with the Laschamp geomagnetic excursion followed by Heinrich Stadial 4. Millennial-scale variations are largely attributable to Earth’s magnetic field changes and in part to climate-related changes in the oceanic carbon cycle. A progressive shift to lower 14C/12C values between 25,000 and 11,000 years ago is likely related, in part, to progressively increasing ocean ventilation rates.

Libby pioneered the 14C dating method (1), which revolutionized a number of scientific disciplines, most notably archeology and climatology. However, variations in atmospheric 14C, likely caused by changes in the shielding of cosmic rays induced by the Earth’s and Sun’s magnetic fields and/or the redistribution of 14C among different carbon reservoirs, were soon recognized (2). These changes necessitate the calibration of 14C ages against a calendar time scale. A precise and accurate 14C calibration is considered the Holy Grail of radiocarbon dating.

Our ability to calibrate the 14C time scale has been limited by our ability to establish the absolute age of a material that contains information about atmospheric 14C/12C. By the late 1980s, the most recent portion of the 14C time scale [last ~10 thousand years (ka)] was calibrated extremely precisely using dendrochronology. The development of mass spectrometric 230Th dating methods (3) and their continued refinement (4) opened up the possibility of extending the calibration much deeper in time, led to the first large extension of the calibration well back into the Pleistocene (5), and ultimately has led to the current contribution. However, the 230Th dating approach has its own constraints. Corals, which are good materials for 230Th dating, do not accumulate continuously over thousands of years and are difficult to collect since those in the time range of interest are now largely submerged. Stalagmites, which can be excellent choices for 230Th dating, typically contain a significant fraction of carbon ultimately derived from limestone bedrock, which is essentially 14C-free. Stalagmite-based calibrations must therefore correct for a dead carbon fraction (DCF), which can be large and variable and is typically the main hurdle in such efforts (6, 7).

Southon et al. (8) demonstrated that the DCF in one Hulu Cave (32°30ʹN, 119°10ʹE) stalagmite, H82, was unusually small and stable, allowing a precise and accurate 14C calibration in the 26.8 to 10.6 ka B.P. (before the present; “present” is 1950 CE) interval (fig. S1). Here, we show that older Hulu Cave stalagmites, MSD and MSL, have similarly low and stable DCFs (Figs. 1 and 2), which allow for precise and accurate 14C calibration for the remainder of the 14C time scale back to ~54 ka B.P.

Fig. 1 Hulu speleothem 14C versus 230Th ages and comparison between Hulu and IntCal13 14C ages.

(A) Hulu [olive-brown, H82 (8); blue, MSD, and green, MSL (this study), and IntCal13 14C (17)] vs. 230Th ages. 14C error bars are 1σ. For clarity, uncertainties in IntCal13 are not shown. The floating tree ring Δ14C datasets (purple) (14, 15) are tuned to the Hulu 14C record (11). The red square (1σ) is the independent data point based on 14C measurements on wood associated with the Ar-Ar dated Campanian Ignimbrite (13). (B) 14C age difference (black) between Hulu dataset and IntCal13 (17). The gray envelope shows the uncertainty (1σ). Hulu 14C ages are corrected for the DCF (450 ± 70 years) (8). (C) Calendar age minus IntCal13 (red)/Hulu (blue) 14C age. The light blue envelope shows the uncertainty (1σ). The three Hulu sample datasets replicate over contemporary growth periods. Hulu Cave 14C data are consistent with IntCal13 between ~10.6 and 33.3 ka B.P. but lower in 14C ages between ~33.3 and 42 ka B.P. and higher between 42 and 53 ka B.P.

Fig. 2 Comparison of Hulu Δ14C data with IntCal13.

Hulu Δ14C data are shown with error bars with the same color codes as in Fig. 1. IntCal13 and its dataset (17) are shown in the gray envelope and gray bars. 14C error bars are 1σ. Hulu data overlap with IntCal13 between ~10.6 and 33.3 ka B.P.; however, there are substantial offsets, particularly before 30 ka B.P., and the Hulu record exhibits substantial previously unknown millennial-scale structure. The purple error bars and red square are the floating tree ring series and Campanian Ignimbrite data, as in Fig. 1.

All three Hulu stalagmites record climatic conditions in their oxygen isotopic compositions (9, 10), including Asian monsoon equivalents of the stadial and interstadial events recorded in Greenland and the Heinrich Stadials recorded in North Atlantic sediments. Thus, we are able to compare our final 14C/12C record to the major climate events of the last glacial period, with negligible stratigraphic uncertainty.

Here, we present ~300 pairs of 14C and 230Th dates from MSD (51 to 18.5 ka B.P.) and MSL (analyzed between 54 and 36 ka B.P.), extending the 14C record back to 54 ka B.P. (Fig. 1, figs. S2 to S5, and tables S1 and S2). Temporal resolution per pair is ~170 years. We drilled sequential powders for 230Th dating, leaving a ridge of solid calcite behind for 14C dating (figs. S2 and S3). This procedure avoids use of a powdered sample for 14C analysis, which can lead to 14C contamination (8). Methods are described in the supplementary materials (11). The large overlaps in ages between MSD and MSL (15 ka) and between H82 (8) and MSD (8 ka) (Figs. 1 and 2 and figs. S4 and S5) allow us to test for precision, accuracy, differential contamination/diagenesis, and differential changes in the DCF.

Through comparison to the dendrochronology record, DCF in H82 is low and constant within tight bounds, even across major climate boundaries, equivalent to a 14C age offset of merely 450 ± 70 years (8) (fig. S1). With the same DCF correction for MSD and MSL, we observe strong agreement between the overlapping portions of Δ14C records from MSD and H82 as well as for MSD and MSL (Figs. 1 and 2 and fig. S6). Although we cannot rule out scenarios where, for example, the DFC shifts similarly in pairs of stalagmites, the replication among stalagmites is consistent with small DCF for all three speleothems and DCF stability within tight bounds over the period of our extended record (fig. S6). We therefore adopt the H82 DCF correction of 450 ±70 years for the entire record. 14C data from modern drip-waters (figs. S7 to S9) suggest that the soil above portions of the cave is characterized by open system conditions, which together with an unusual sandstone ceiling above the three samples provide a possible explanation for the low DCF that we infer for the three stalagmites (11, 12).

A number of arguments support the accuracy of the record. The younger portion of the H82 record agrees with the dendrochronology record (8). The overlapping portions of the three stalagmite records are internally consistent. There is agreement between one of the highest values in our record (Δ14C = 700‰, at ~39.85 ka B.P.) with a precisely and carefully determined independent data point based upon wood associated with the Campanian Ignimbrite and precise Ar-Ar dating (13). Finally, two floating dendrochronology sections (14, 15) can be placed on the Hulu calibration in such a way that overall trends and finer-scale features match the Hulu curve (11) (Figs. 1 and 2 and fig. S10). We should point out, however, that others have previously proposed a placement later by ~1 ka for one of these floating chronologies (16); see the supplementary materials for a discussion of this issue.

Considering the full record, there is a general correspondence with the latest IntCal compilation (17) (Figs. 1 and 2) within fairly large uncertainties, confirming the general validity of the compilation. However, for the portion older than 30 ka B.P., clear differences emerge. The Hulu record has less uncertainty and resolves previously unknown fine-scale structure. Between 33.5 and 42.5 ka B.P., the Hulu record indicates larger offsets between 230Th ages and 14C ages than IntCal13, with offsets between the records as high as 1 ka, corresponding to a higher Δ14C by as much as 170‰ as recorded at Hulu. Conversely, from 42.5 ka B.P. to the end of the IntCal curve at 50 ka B.P., the Hulu record indicates smaller offsets between 230Th and 14C ages, by ~1 ka, which corresponds to ~140‰ lower Δ14C. From 50 to 54 ka B.P., the Hulu curve indicates similar though nominally higher Δ14C than during the subsequent few millennia. Another notable difference is the sharper and higher amplitude increase in Δ14C around 42.5 ka B.P. A notable similarity is the lack of a prominent low Δ14C excursion around 31 ka B.P. This low, present in Cariaco sediment and Bahamas speleothem datasets (7, 18), was omitted from IntCal13 because of its absence from the Lake Suigetsu record (19). The Hulu data support this omission.

14C ages are generally less than calendar ages throughout the full record, reaching a maximum offset of ~5200 years between ~39.3 and ~40.8 ka B.P. (Fig. 1). The offset is largely due to higher atmospheric Δ14C, although there is also a progressive offset of 2.83% of the age due to the use of the Libby half-life in calculating the 14C age. Between 54 and 43 ka B.P., Δ14C values range between 0 and 300‰, then increase sharply to values exceeding 600‰ by 42 ka B.P. (Fig. 2). High values continue until 38.8 ka B.P., reaching the highest values in the full record of 700‰ at 40.8 and 39.3 ka B.P. Between 38.8 and 38.0 ka B.P., Δ14C decreases sharply to values around 500%. Between 38.0 and 25.0 ka B.P., Δ14C values exhibit millennial-scale variability with highs around 600% and lows around 400%. Notable is a relative high of about 600% at 33.8 ka B.P. From 25.0 ka B.P. to the mid-19th century (as previously known), Δ14C values gradually diminish from around 500% to 0, with significant changes in slope between 16 and 11 ka B.P.

The new data provide critical constraints on the causes of changes in Δ14C during the last 54 ka. The millennial-scale pattern of Δ14C variations (Fig. 3) has similarities to the geomagnetic record (Virtual Axial Dipole Moment data) (20), suggesting that changes in shielding of cosmic rays by the geomagnetic field are responsible for much of the millennial-scale variation in Δ14C. Of note is the coincidence within tight age uncertainties between the abrupt increase in Hulu Δ14C and the onset of the Laschamp magnetic excursion at ~42.3 ka B.P. (21), as well as between the period of weakest geomagnetic field during the Laschamp (~41.1 ka B.P.) (21), which correlates with the highest Δ14C values over the past 54 ka. This suggests that the Laschamp is responsible for both of these features. Additionally, a second prominent peak in the Hulu record at ~34 ka B.P. is consistent with the timing of the Mono Lake excursion (22), suggesting that this excursion is responsible for the Δ14C peak (Fig. 3).

Fig. 3 Comparison of 10Be flux, geomagnetic field, model and Hulu 14C data.

(A) Greenland 10Be flux (36). (B) Stacked geomagnetic field (gray, 1σ envelope) (20). (C) The model Δ14C record (11) (gray, 1σ envelope) based on 14C production inferred from the geomagnetic field (20). (D) Blue and red envelopes (1σ) are composite Hulu (10.6 to 54.0 ka B.P.) and IntCal13 (0 to 10.6 ka B.P.) Δ14C data, respectively. (E) The ΔΔ14C is the residual obtained by subtracting the model Δ14C result from the Hulu/Intcal13 Δ14C data. The gray envelope shows the uncertainty from Hulu data and model uncertainties (1σ). Two vertical bars show the Laschamp and Mono Lake excursions. The arrow indicates the large decline in Δ14C from ~25 to 11 ka B.P. See also fig. S10.

We estimated the component of Δ14C variability caused by geomagnetic field changes by using a magnetic record (20), a cosmogenic production model (23), and the MESMO-2 Earth system model (24). The output simulates that component of atmospheric Δ14C variability caused by geomagnetic field changes alone (11) (Fig. 3C). We subtracted this model curve from the observed Hulu Δ14C record to obtain a model–observation residual curve (ΔΔ14C), which shows the component of the observed variability not captured by our model, likely due to some combination of uncertainties in the input magnetic field data, inaccuracies in the model itself, solar modulation of production, and changes in the carbon cycle (Fig. 3E). We cannot use this residual as a quantitative target curve for, say, a model with a changing carbon cycle, as there are nonlinearities in the overall problem (25). Nevertheless, we consider the residual curve useful for the remaining discussion, because it guides us to the magnitude and direction of observation–model differences.

The residual is characterized by a series of millennial-scale events during the last glacial period (Fig. 4B). Given uncertainties, we have not attempted to assign a one-to-one correspondence between climate events and features in the residual trace. However, we highlight two cases where temporal constraints are robust and where the trace shows a prominent feature, the Younger Dryas (YD) and Heinrich Stadial 4 (HS 4). In both cases, residual highs correlate with cold anomalies in the North Atlantic region. For the YD, this observation confirms earlier work (2629). These studies all explained the relatively high Δ14C by invoking carbon cycle changes associated with climate change with, in one case (29), an additional contribution from solar modulation during the early YD. For HS 4, temporal constraints place the end of the Laschamp (16, 21) ~1 ka well before the prominent residual peak that correlates with HS 4. Even that long after the end of the Laschamp, one would expect high atmospheric 14C, because the e-folding time for reaching isotopic steady state after a production change is on the order of thousands of years, the time scale of deep ocean ventilation. However, the time scale for the initial significant diminution of atmospheric 14C following a production drop is a few hundred years, a time scale tied to reaching isotopic steady state with the upper portion of the ocean. Our model captures this, as evidenced by the few-hundred-year difference between production shift (Fig. 3B) and Δ14C response (Fig. 3C) for numerous production changes. Since Δ14C does not fall in the centuries after the Laschamp but instead rises slightly to a high value that correlates with HS 4, we conclude that another factor besides magnetic field change has contributed to these high values, likely carbon cycle changes associated with climate change.

Fig. 4 Comparison of the ΔΔ 14C record with other climate proxy records.

(A) Antarctic ice core dust flux record (EDC) (37). (B) The ΔΔ14C record (the residual as determined for Fig. 3E) and composite atmospheric CO2 record (yellow) (38). (C) Detrended ΔΔ14C record (11). (D) The Hulu δ18O record (10). (E) Greenland ice core δ18O record (NGRIP) (39). Vertical light yellow bars indicate HS 4 and YD. The arrow shows the Δ14C trend, as in Fig. 3.

Given the general character of the millennial-scale variability in the residual trace, it is plausible that the relationships that we observe for the YD and HS 4 are more general features of the last glacial period climate and carbon cycle. The YD, HSs, and Greenland stadials (GSs) correspond to weak modes in the Atlantic Meridional Overturning Circulation (AMOC), as inferred from the 231Pa/230Th record (30). A weak mode may increase atmospheric 14C due to diminished flux of 14C to the intermediate and/or deep ocean, as supported by observed increases in radiocarbon-based ventilation ages during HS 1 and the YD in the western equatorial Atlantic (31). Regardless of the specific mechanisms, there is clear evidence at the millennial scale for elevated Δ14C at specific cold times in the North Atlantic, perhaps associated with AMOC slowdown.

We now consider the long-term gradual lowering of Δ14C, from ~500‰ 25 ka B.P. to ~150‰ 11 ka B.P. Bard et al. (5) attributed much of the decline to steady increase in Earth’s magnetic field, with some (100 to 150‰) plausibly caused by carbon cycle changes. Köhler et al. (25) reached similar conclusions. Notable was their use of the ice core 10Be record to predict production-related changes in Δ14C. This strategy takes into account production changes caused both by the terrestrial magnetic field and by solar modulation. They reached a similar conclusion as Bard et al. (5), i.e., that production changes could not explain the full Δ14C shift over this interval and that carbon cycle changes could account for up to 100‰ of the shift. Our work confirms some of these conclusions, as our residual trace shows a significant decline after accounting for magnetic field-related production changes.

The broad lowering of Δ14C throughout this interval could plausibly result from progressively increasing ocean ventilation. All other factors being equal, the shorter the mixing time, the less time for 14C to decay, the more 14C in deep waters and, by mass balance, the lower the Δ14C of the atmosphere. Presuming an average deep water age of 1000 years at 11 ka B.P. and a 60:1 ratio of deep water to atmospheric carbon, the lowering of atmospheric Δ14C over this time period can be explained by a progressive shift in deep water age from about 3000 years at 25 ka B.P. to the assumed 1000-year value at 11 ka B.P.

There is some support for the inference of increasing ventilation with time, as observations indicate that the deep Southern Ocean and South Pacific were poorly ventilated at the last glacial maximum (3234). Deep ocean Δ14C data for times since the last glacial maximum (35) do not clearly resolve pre-Holocene from Holocene ventilation ages, but they also do not preclude large pre-Holocene ventilation ages. Thus, while it is likely that deep ocean ventilation change accounts for a portion of the residual 25 to 11 ka B.P. Δ14C drop, it is still not clear whether it can account for the full drop. Further work is needed to close the loop on this critical issue.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S10

Tables S1 and S2

References (4055)

References and Notes

  1. Materials and methods are available as supplementary materials.
Acknowledgments: We thank W. Broecker for inspiring and supporting this work over decades. K. Tokos ran numerical model experiments using the facilities of the UM Minnesota Supercomputing Institute following the experiment design by K.M. Funding: This work was supported by China NSFC grants 41888101, 41731174, 4157020432, and 41561144003; U.S. NSF grants 0502535, 1103404, 0823554, 1003690, 1137693, 1337693, and 1702816; and the Comer Science and Education Foundation. Author contributions: H.C., R.L.E., and J.S. proposed and directed the research. R.L.E. led the writing and revision of the manuscript. H.C. and J.S. performed 230Th dating and 14C analysis of the speleothem samples, respectively. K.M. designed and performed the model simulation. W.J.Z. performed the 14C analysis of drip-water. H.Y.L., X.L.L., Y.X., Y.F.N., A.S., J.M.F., and H.C. performed subsampling work and data analyses. S.T.C., M.T., Q.W., Y.J.W., and H.C. collected samples and performed cave monitoring and fieldwork. All authors discussed the results and commented on the manuscript. Competing interests: The authors declare no competing financial interests. Data and materials availability: All data are available in the manuscript or the supplementary materials.
View Abstract

Stay Connected to Science

Navigate This Article