Abrupt Climate Change at the End of the Last Glacial Period Inferred from Trapped Air in Polar Ice

See allHide authors and affiliations

Science  29 Oct 1999:
Vol. 286, Issue 5441, pp. 930-934
DOI: 10.1126/science.286.5441.930


The last glacial period was terminated by an abrupt warming event in the North Atlantic ∼15,000 years before the present, and warming events of similar age have been reported from low latitudes. Understanding the mechanism of this termination requires that the precise relative timing of abrupt climate warming in the tropics versus the North Atlantic be known. Nitrogen and argon isotopes in trapped air in Greenland ice show that the Greenland Summit warmed 9 ± 3°C over a period of several decades, beginning 14,672 years ago. Atmospheric methane concentrations rose abruptly over a ∼50-year period and began their increase 20 to 30 years after the onset of the abrupt Greenland warming. These data suggest that tropical climate became warmer or wetter (or both) ∼20 to 80 years after the onset of Greenland warming, supporting a North Atlantic rather than a tropical trigger for the climate event.

Evidence for extremely abrupt changes in Earth's climate has come principally from the annually layered Greenland ice cores (1–7), although the tropics collect most solar radiation and are central to Earth's heat and water vapor budgets (8). Accordingly, tropical climate records are critical for understanding abrupt climate events. Especially important are those records that reveal cause-and-effect relations by giving the precise relative timing of changes in the tropics and the high latitudes. However, high-resolution tropical climate records are few in number and for the most part equivocal on the existence and timing of abrupt climate events.

Evidence from Bolivian ice cores at 18°S shows that tropical climates warmed rapidly about 15 thousand years (ky) B.P. (before present, where present is 1950 A.D.) (9). This warming is roughly synchronous with prominent warming at 14.67 ky B.P., known as the Bølling Transition, seen in the Greenland Summit ice cores (1, 2, 5), European pollen records (10), and many North Atlantic sediment records (11, 12). Other recent studies have documented low-latitude rapid climate shifts at about this time in the Santa Barbara basin (13), Arabian Sea (14), and the Cariaco basin of Venezuela (15); however, dating uncertainties of a century or more limit the precision with which the timing of these events and their putative Arctic counterpart may be compared.

Atmospheric methane concentrations also increased abruptly at this time, as inferred from measurements of trapped air bubbles in the Greenland Summit ice cores (16, 17) and Antarctic ice cores (18, 19). This increase is thought to have been caused by an increase in wetland extent and temperature, as wetlands were the principal source of methane in the preindustrial period (17) and changes in sink strength are thought to have been small (20). Tropical wetlands have been proposed as a major contributor to the Bølling methane increase, because ice sheets covered the primary extratropical methane source areas at this time (17, 21). This hypothesis is supported by the fact that the difference between Greenland and Antarctic methane concentrations (referred to as the interpolar gradient) underwent a relatively small increase despite the ∼25% jump in concentration (21, 22). When inverted for source distribution with a three-box atmospheric transport model, methane records from Greenland and Antarctica imply a predominantly low-latitude source (23). Because the atmosphere is well mixed (mixing time of ∼1 year) relative to methane's lifetime in the atmosphere (∼10 years) (24) and wetland methane emissions are broadly correlated with temperature and precipitation, methane concentration integrates climate information over the entire tropical region. Knowledge of the precise timing of the abrupt methane increase relative to the Greenland warming should thus place constraints on the relative timing of tropical and Arctic abrupt climate change.

Two problems have prevented relating changes in temperature and methane in the ice core record. First, the age of the air trapped in bubbles is less than the age of the enclosing ice because air is occluded at some depth below the surface of the ice sheet in the bubble close-off region (25, 26). This gas age–ice age difference is not known well for past times, making uncertain the relative timing of temperature (which is obtained from the ice matrix) and atmospheric gas changes (17,26). Second, the temperature inferred from the18O/16O ratio of the ice (δ18Oice) is uncertain because factors other than local mean annual temperature may affect this ratio, such as the seasonality of precipitation at the ice core site (27) and the temperature and proximity of the water vapor source (28, 29). Borehole temperature calibrations of the δ18Oice paleothermometer have demonstrated that the modern spatial calibration underestimates the glacial cooling by a factor of two (30). These issues have raised the question of whether the abrupt increases in Greenland δ18Oice represent isotopic artifacts rather than local temperature changes as commonly inferred (29,31) and have created uncertainty about the magnitude of temperature change they may indicate.

Here we address these problems with measurements of isotopes of nitrogen and argon gas trapped in air bubbles in the GISP2 (Greenland Ice Sheet Project 2) ice core. Bubble 15N/14N and 40Ar/36Ar record a signal of rapid temperature change at the surface of the ice sheet (7). The combination of nitrogen and argon isotope measurements provides a direct estimate of the magnitude of the temperature increase, based on laboratory calibration of the isotope fractionation due to thermal diffusion. Methane and nitrogen diffuse downward through the snow layer at nearly the same speed (32) and are trapped together in the bubbles, making a precise comparison of the timing of atmospheric methane change and local temperature change possible (7).

The porous and partially consolidated layer of snow on top of polar ice sheets (known as firn) contains air that mixes by molecular diffusion with the overlying atmosphere (25). The isotopes of gases in this stagnant column of air separate as a result of temperature gradients and gravitational settling (7,33) (air in the free troposphere does not separate in this manner because of vigorous convective mixing). A gas mixture subjected to a temperature gradient will unmix slightly, with heavier gases generally moving toward colder regions (34, 35). The magnitude of the steady state effect is given by δ = ΩΔT, where δ is the fractional deviation of the isotope ratio of the cold gas parcel from the warm gas parcel, ΔT is the temperature difference between the two gas parcels, and Ω is an empirical quasi-constant we call the “thermal diffusion sensitivity” (36). We measured Ω in a known temperature gradient in the laboratory and found preliminary values of Ω = +0.0145 per mil K−1 for δ15N and +0.036 per mil K−1 for δ40Ar at a mean temperature of −43°C (estimated error of ±3%; future work may reduce this error). For example, a 10 K temperature difference will result in a +0.145 per mil enrichment in δ15N in the cold gas relative to the warm gas. For comparison, our 1σ measurement precision (defined as the reproducibility of replicate analyses of ice cut from the same depth) is ±0.007 per mil for δ15N and ±0.03 per mil for δ40Ar, corresponding to ±0.5 and ±1 K, respectively.

Gravitational settling (33) is given by δ = [e Δ mgz/RT − 1] × 103, where Δm is mass difference,g is gravitational acceleration, z is depth,R is the gas constant, and T is temperature. For example, a 92-m-deep stagnant air column at 226 K will be enriched by +0.48 per mil in δ15N relative to the top of the column (in our case, the atmosphere). Gravitational effects may be separated from thermal effects by comparing δ15N with δ40Ar/4, because the latter is only ∼60% as sensitive to thermal fractionation as δ15N. Gravitational settling scales with the mass difference (4 for40Ar/36Ar), so gravity affects δ15N and δ40Ar/4 equally (33). Measurements in modern firn air confirm the different sensitivities of nitrogen and argon isotopes to thermal diffusion (Fig. 1).

Figure 1

Depth profile of nitrogen and argon isotope ratios in air withdrawn from the snowpack at Siple Dome, Antarctica, on 14 January 1998, δ15N is defined as [(15N/14Nsample)/(15N/14Natmosphere) −1] × 103 and is expressed in units of per mil. δ40Ar is the corresponding quantity for the40Ar/36Ar ratio and is divided by 4 to facilitate comparison with δ15N. The seasonal cycle of temperature at the surface creates a strong temperature gradient in the top few meters, causing thermal fractionation. Because the profile shown was taken in summer, temperature decreases with depth. Thermal diffusion thus drives a transient “pulse” of heavy isotopes down into the colder firn, which mixes by diffusion with the underlying gas. Note the “pulse” of enriched δ15N and δ40Ar/4 centered at about 6-m depth, superimposed on the linear downward increase due to gravitational settling. The good agreement of δ15N with δ40Ar/4 at depths >20 m confirms that gravity is the only process affecting the deep firn at present at this site, whereas the weaker δ40Ar/4 response in the “pulse” is the fingerprint of thermal diffusion.

After an abrupt climate warming, a temperature gradient will persist in the firn for several hundred years (the thermal equilibration time of the firn) and will thermally fractionate the entire air column (7). Gases diffuse about 10 times as fast as heat in polar firn (37), so the isotopic signal penetrates to the bottom of the firn long before the temperature equilibrates, in about a decade (7). The bubbles thus record a signal of the climate event as an abrupt increase in δ15N, slightly smoothed by the diffusion process in the firn (38), followed by a gradual decrease in δ15N over several hundred years as the firn becomes isothermal once again.

Two prominent abrupt increases in GISP2 δ18Oice occurred during the last deglaciation, at 14.67 and 11.64 ky B.P. on the layer-counted time scale, and were accompanied by abrupt methane increases and δ15N excursions (Fig. 2). δ15N may be thought of as a temperature-change indicator that is only sensitive to rapid temperature change (7). The δ15N signal thus confirms that the δ18Oice increases were indeed abrupt warming events and not isotopic artifacts. The event at 14.67 ky B.P. separates the last glacial period from the Bølling warm interval and is known as the Bølling Transition (5) (in the ice core record, this interval is also referred to as Interstadial 1). The cold period immediately preceding the transition is referred to as the Oldest Dryas (5). The gas time scale presented in Fig. 2 is approximate and is based on model estimates of the gas age–ice age difference that may be in error by several hundred years (39), so the slight apparent lead of δ15N over δ18Oice is not significant. A detailed expansion of the record from 14.5 to 14.9 ky B.P., covering the Bølling Transition, is shown in Fig. 3. In contrast to Fig. 2, the gas ages shown are deduced on the basis of our inference of a thermal diffusion signature in the gas isotopes (40). Individual replicate analyses of δ15N and δ40Ar/4 in pieces of ice cut from the same depth in the ice core are shown. Means of replicate methane analyses are given. Analytical techniques for gas isotopes and methane are described elsewhere (41).

Figure 2

GISP2 accumulation (51), oxygen isotopes (5), methane, and nitrogen isotopes of air in bubbles during the last deglaciation, with Taylor Dome (Antarctica) methane shown for comparison (21). Low-resolution nitrogen data are from (54).

Figure 3

The 400-year window encompassing the Bølling Transition, showing high-resolution measurements of oxygen isotopes of ice (5) and nitrogen, argon, and methane measurements made on trapped air bubbles. The gas age is deduced by assuming that the abrupt change in nitrogen isotopes marked by the dashed vertical line corresponds to the shift in oxygen isotopes at 14.65 ky B.P. Nitrogen isotope measurements made at URI and the Scripps Institution of Oceanography (SIO) appear to have a ∼0.02 per mil offset in the age range 14.59 to 14.64 ky B.P., which we do not understand. The mismatch between model and observed gas isotopes in the range 14.55 to 14.62 ky B.P. is likely due to gravitational enrichment from transient firn thickening brought about by the increase in snow accumulation, which was not included in the model.

δ15N rises across the transition from baseline Oldest Dryas values of +0.48 per mil to peak values of +0.63 per mil (Fig. 3). Methane appears to rise over a ∼50-year period from 14.65 to 14.60 ky B.P. δ15N first rises clearly beyond the envelope of Oldest Dryas variability at 14.672 ky B.P. (1821.16-m depth). We infer that this point marks the onset of the warming and may correspond to the 1.7 per mil steplike increase in δ18Oice (Fig. 3) and the abrupt doubling of snow accumulation (5) at this time. However, methane concentration remains within its range of Oldest Dryas variability [470 to 510 parts per billion by volume (ppbv)] in this ice sample and in the subsequent sample at 14.655 ky B.P. (Table 1). Methane first exceeds this range at 14.645 ky B.P. (1820.17 m), ∼27 years after the corresponding change in δ15N. We infer that methane began rising at some time during the 10-year interval between the latter two samples and thus conclude that atmospheric methane concentration started rising ∼20 to 30 years after the onset of the Greenland warming. Annual layers are discernable in this interval (3), making these estimates robust. Absolute time scale uncertainty (∼2%) (42) does not affect these conclusions, because they are based on comparisons of gas isotopes and methane from the same depths.

Table 1

Depths and deduced gas ages of inferred Bølling Transition.

View this table:

The separation between δ15N and δ40Ar/4 values in the Bølling interval (Fig. 3) confirms that thermal diffusion rather than gravity is responsible for the δ15N anomaly and implies that air temperatures warmed by 9 ± 3°C by 14.60 ky B.P. (Fig. 4). δ18Oiceincreased by 3.4 per mil across the transition (difference between the means of the 60-year period after 14.65 ky B.P. and the 100-year period before 14.67 ky B.P.) (5), implying an oxygen isotope temperature sensitivity (α) across the transition of 0.38 per mil K−1 (+0.19 or −0.09 per mil K−1). This is close to the borehole temperature-calibrated value of 0.33 per mil K−1 (30), providing independent verification of the borehole calibration.

Figure 4

Plot of measured δ15Nexcess = δ15N − δ40Ar/4, which is not affected by gravitational settling, compared with model results forced by 8°, 10°, and 12°C “step function” increases in surface temperature. Error bars show quadrature sum of pooled RMS deviations of nitrogen and argon replicate measurements. Measurements made at URI and SIO appear to have a ∼0.01 per mil systematic offset, equivalent to 2°C, possibly due to a calibration error. We adopt an estimate of 9 ± 3°C for the magnitude of the abrupt warming.

Climate-driven changes in methane emissions could have preceded the observed atmospheric methane concentration change by no more than a decade, the approximate lifetime of methane in the atmosphere (24). Interpretation of the methane rise is complicated by evidence for two methane source regions during the transition. Part of the increase in methane concentration from the Last Glacial Maximum to the Bølling has been attributed to boreal sources (21–23). During the ∼50-year interval in which methane concentrations were rising, we cannot know the interpolar gradient because of time scale uncertainties (arising from the fact that the rapid transition in methane concentration is used to synchronize Antarctica and Greenland ice core time scales). Thus, no source attributions based on the interpolar gradient are possible within this period. For example, the abrupt increase in methane concentration between 14.655 and 14.645 ky B.P. (Fig. 3) cannot be firmly attributed to a tropical climate event, because it is possible that this was the result of the boreal portion of the source increase. However, some part of the entire methane concentration increase from 14.655 to 14.595 ky B.P. must have been tropical in origin; otherwise, the interpolar gradient would have been steeper than observed just after 14.595 ky B.P. (Fig. 2) (21–23, 43). Thus, we infer that there must have been a change in tropical temperatures or precipitation (or both) at some time within the interval 14.655 to 14.595 ky B.P. This brackets the possible time of the tropical climate event to within ∼20 to 80 years after the onset of the Greenland warming.

Methane production might not respond instantaneously to environmental change, however. For example, it might take several decades for water tables to rise and fill previously drained wetlands, or sustained organic matter production might be required to create the anoxia necessary for methane production. At least some immediate response of methane production to increased tropical temperature or precipitation is expected and probably inescapable, though. Soils become waterlogged and plant growth responds to precipitation on subannual time scales, and methane emissions from modern wetlands respond to changes in climate conditions on similar time scales (44,21). Thus, we take the absence of any substantial methane increase before 14.65 ky B.P. as an indication that the tropical climate change probably lagged Greenland warming by at least ∼20 years.

The implication that high-latitude North Atlantic warming led tropical climate by several decades may provide a test for proposed mechanisms of abrupt climate events. Two alternative mechanisms may be imagined. In one scenario (the “North Atlantic trigger”), North Atlantic thermohaline overturning abruptly deepened and intensified, drawing heat poleward and warming Greenland (45), which warmed the tropics by some as yet poorly understood mechanism. Increased poleward heat transport might have reduced the Northern Hemisphere equator-to-pole temperature gradient, decreasing the driving force for the winds (46). Slackening of the winds would have reduced oceanic upwelling, wind mixing, evaporative cooling, and the gyre circulation, leading to an abrupt warming of tropical and subtropical sea surface temperature (SST). Much evidence points to a sudden drop in wind speed at the time of the Bølling Transition, including records from laminated sediments in the Cariaco basin of Venezuela (15) and dust particle size and concentration in GISP2 ice (47). Also, warmer North Atlantic SST has been shown in atmospheric model studies to produce strengthening of the Asian monsoon (27, 48), which may have warmed the tropics. In a second, speculative scenario (the “tropical trigger”), the tropics warmed first, perhaps because the temperature of upwelling water in the tropics crossed a threshold that caused the tropical ocean-atmosphere system to abruptly reorganize into a warmer state. In this scenario, the deepening of North Atlantic thermohaline overturning and warming of Greenland would be responses to tropical warming rather than primary causes. Our phasing data would appear to favor a North Atlantic trigger.


View Abstract

Stay Connected to Science

Navigate This Article