Introduction

Dansgaard–Oeschger events are still a much-discussed topic within the palaeoclimate research community. More palaeo-data are gradually building up a picture of the large, very rapid climate changes that took place some 24–30 times during the last glacial. Ocean sediment cores have helped to improve our understanding of the role of the oceans in these different climate changes. However, the dating is still inadequate for determining in what order the changes occurred.

We present here isotopic data from the two Greenland Summit ice cores, Greenland Icecore Project (GRIP; Reference DansgaardDansgaard and others, 1993) and Greenland Ice Sheet Project II (GISP2; Reference Grootes, Stuiver, White, Johnsen and JouzelGrootes and others, 1993).

We present a continuous (GRIP) δ¹⁸O record with a resolution of 5 years going back to 50 kyr BP and of 20 years until 100 kyr BP. the GRIP δ¹⁸O data are compared with other high-resolution records, i.e. the GISP2 chemistry record (25 years until 50 kyr BP) and the GRIP Ca²⁺ record (3 years until 100 kyr BP). for purposes of comparison, all records are converted to the GISP2 Meese/Sowers time-scale.

The high-resolution records are separated into interstadials and stadials, defined by the GRIP and GISP2 δ¹⁸O records. the signal analysis performed will be compared to previous studies of the high-resolution Ca record from GRIP and the GISP2 ion records and the new high-resolution GISP2 δ¹⁸O record. the present study is generally based on much higher-resolution data than were used in previous studies dating back to 50 kyr BP (Reference MayewskiMayewski and others, 1994; Reference Fuhrer, Wolff and Johnsen.Fuhrer and others, 1999; Reference JohnsenJohnsen and others, 2001). Finally the data will be compared to model simulations of Dansgaard–Oeschger events performed by the CLIMBER-2 model (Reference PetoukhovPetoukhov and others, 1998).

Ice-Core Data

The data used in this study originate from the two deep ice cores GRIP (Reference FuhrerFuhrer,1991; GRIP Project Members,1993) and GISP2 (Reference Grootes, Stuiver, White, Johnsen and JouzelGrootes and others, 1993; Reference MayewskiMayewski and others, 1997; GISP2–GRIP compendium volume from the Wolfboro Workshop,1995 (Journal of Geophysical Research 102(C12),1997)).The data have been sampled at very high resolution, making possible a detailed study of the transitions from stadials to interstadials and vice versa. Here we will focus mainly on the stadial-to-interstadial transitions.

Figure 1 shows the resolution of the raw data discussed in this paper. There are a few holes in the records which we have bridged by interpolation (see Reference MogensenMogensen, 2001). the raw data series were resampled at the resolution shown in Table 1 by using occasional minor over-sampling (Reference MogensenMogensen, 2001).

Fig. 1 The resolution of the raw data (sample size) used in this paper. (a) Sample size for the GRIP δ¹⁸O record (Reference JohnsenJohnsen and others, 1997); (b) sample size for the GRIP calcium record (Reference FuhrerFuhrer, 1995); (c) GISP2 sample size (Reference Stuiver and Grootes.Stuiver and Grootes, 2000); and (d) sample size of the GISP2 ion records (Reference MayewskiMayewski and others, 1997).

Table 1. Resolution of the data series discussed in this paper

In order to be able to work with data from the two ice cores, the GRIP data have been converted to the GISP2 time-scale (also named GISP2 Meese/Sowers time-scale; Reference BenderBender and others, 1994); for details on how this conversion was done see Reference MogensenMogensen (2001). This approach allows us to work with a greater number of continuous high-resolution datasets. the errors in the relative timing are dependent on the resolution and the noise of the data used to match the transitions in the two cores. the δ¹⁸O records used for matching the two records had a resolution of 20 years until 50 kyr BP and 50 years until 110 kyr BP for GISP2, and 20 years for the GRIP δ¹⁸O record. Hence we expect to ``hit’’ the same event in the two records within 2–3 data points, i.e. ~50 years back to 50 kyr BP and ~150 years for the rest of the record. Using higher resolution than 20 year averages for tying the two time-scales together results in too much noise for the method used, and would therefore not increase the ``hit-rate’’ substantially.

We will first describe how the trend of the GRIP δ¹⁸O record and the trend of the GRIP Ca record correlate with the duration of the stadials and interstadials, respectively. We will then present the transitions at the onset of Dansgaard–Oeschger events. These very sharp transitions are easier to distinguish than the less sharp transitions from interstadials to stadials.

Methods

The δ¹⁸O record in ice cores is the primary palaeo-data proxy for the Dansgaard–Oeschger events, and therefore has been chosen to define the onset and termination of these events. All the Dansgaard–Oeschgerevents (Reference DansgaardDansgaard and others,1993) were then located using the 20 year sampled δ¹⁸O datasets. the transition marks were placed at the steepest gradient for the onset and termination of each event. As tie points for transferring the GISP2 chronology to the GRIP records we used a total of 68 rapid transitions including all of the 48 Dansgaard– Oeschger transitions.

To classify the Dansgaard–Oeschger events we determined the linear trends or slopes of the GRIP δ¹⁸O record and GRIP Ca record for each stadial and interstadial. Figure 2 shows the GRIP δ¹⁸O record and the GRIP Ca record and the linear segments found. We see that for most interstadials the trend for the δ¹⁸O record decreases with time (negative trends). for each sub-period of the entire Wisconsin glacial, the linear trend was plotted against the duration of the event (see Fig. 3). Figure 3a shows the relationship for the interstadials, and Figure 3b for the stadials. We find that the shorter an interstadial is, the steeper is its trend in both δ¹⁸O and Ca. the longest interstadials are found at the beginning of the last glacial, i.e. marine isotope stages 4, 5a and 5b. the shorter events are found in marine isotope stage 3, which is dominated by Bond cycles (Reference BroeckerBroecker, 1994) consisting of one long (41000 years) and 3–4 shorter Dansgaard–Oeschger events (few hundred years). Lastly, marine isotope stage 2 is dominated by the very long Last Glacial Maximum, perturbed by only two events, one of which is classified as a Dansgaard– Oeschger event (GI-2, which lasts only about 300 years).

Fig. 2 The GRIP δ¹⁸O record and calcium record for the last glacial. Warm periods are shaded and the linear segments for each interstadial and stadial are plotted in light grey.

Fig. 3 The trends/slope of the linear segments shown in Figure 2, plotted against the duration of the corresponding event. (a) δ¹⁸O and calcium trends for the interstadials, and (b) δ¹⁸O and calcium trends for the stadials. It is clear that the strong anticorrelation between Ca and δ¹⁸O is very pronounced during the interstadials.

In Figure 3a the trends for the calcium record almost mirror the δ¹⁸O trends. This agrees well with previous results, which detect a high anticorrelation between δ¹⁸O and Ca (GRIP Project Members, 1993; Reference MayewskiMayewski and others, 1994; Reference Fuhrer, Wolff and Johnsen.Fuhrer and others, 1999). for stadials the picture is somewhat more blurred, but the averaged trend for both δ¹⁸O and calcium is more or less zero.

Results

We present here a close-up of the 13 warm transitions from 10 to 50 kyr BP (Fig. 4), and discuss the averaged transition. This is compared to CLIMBER-2 model results of changes in Greenland temperature, deep Antarctic water in the Atlantic, North Atlantic Deep Water (NADW) formation, and the Nordic Seas salinity. the CLIMBER-2 model produces annual means for a variety of variables, and we have chosen those that are of most interest with respect to the ice-core data.

Fig. 4 (a) Five-year smoothed Greenland temperature and in grey the Antarctic water flow into the Atlantic. (b) Five-year smoothed North Atlantic Deep Water formation in Sv, and in grey the Nordic Seas salinity. (c–f) Shown in light grey are the 13 transitions to Dansgaard–Oeschger events from 50 kyr BP for GRIP δ¹⁸O (c), GRIP Ca²⁺(d), GISP2 Na⁺(e) and GISP2 Mg²⁺. (f) the averages of these transitions are shown in black.

Figure 4c–f show the 13 warm transitions that occurred from 10 kyr BP to 50 kyr BP, leaving out transition 1 (Wisconsin to Holocene). They show that the switch from stadial to interstadial takes about 50–70years from the last cold value to the maximum. One large step seems to be taken during each transition, with a single jump in the δ¹⁸O values of 2.3‰. Since the most rapid shifts in δ¹⁸O are used to define the transitions, we can expect to find this single jump in the averaged record as well. If one examines the δ¹⁸O values separately (not shown here) during a transition, there is a pronounced jump in the δ¹⁸O value in most transitions. Sometimes this jump is succeeded by a temporary drop in the δ¹⁸O value superimposed on the general transition climb.

If we compare the CLIMBER-2 Dansgaard–Oeschger transitions and the ice-core data, as done in Reference Ganopolski and RahmstorfGanopolski and Rahmstorf (2001) (Fig. 4b), we see some striking similarities. the model variables are 5 year averages, and Figure 4a and b show a Dansgaard–Oeschger transition spanning 200 years, as for the GRIP data. the variables shown were chosen because they all show clear changes through a transition. the model was forced with a sinusodial fresh-water flux into the North Atlantic region, as was done in the Reference Ganopolski and RahmstorfGanopolski and Rahmstorf (2001) study. Our forcing did not, however, include the strong fresh-water pulse introduced by Ganopolski and Rahmstorf to simulate a Heinrich event. Figure 4a shows the Greenland temperature shift and the amount of Antarctic deep water that enters the Atlantic. the CLIMBER-2 Greenland temperature bears a fair resemblance to the Greenland δ¹⁸O profile (Reference Ganopolski and RahmstorfGanopolski and Rahmstorf, 2001). the Antarctic deep water entering the Atlantic is also shown, since this variable precedes the temperature shift in Greenland. It is not possible to trace directly the Antarctic deep water inflow in the Greenland ice-core data. We do see that the initial drop in calcium precedes both the jump in δ¹⁸O and the onset of the influx of Antarctic water. This is most likely due to an artefact in the averaged Ca data as discussed below.

The early influx of Antarctic deep water in the CLIMBER-2 model is in agreement with data from ocean sediment cores off the coast of Portugal, which show that Antarctic bottom water flows into the Atlantic before the Dansgaard– Oeschger events (Reference ShackletonShackleton, 2001). Figure 4b shows the modelled NADW flux and the Nordic Seas salinity. on entering the interstadial the NADW formation is restored, and the salinity of the Nordic Seas increases, as the occurrence of convection shifts from south of Iceland to north of Iceland (Reference Ganopolski and RahmstorfGanopolski and Rahmstorf, 2001). In the CLIMBER-2 output the transition to an interstadial is quicker than shown in the δ¹⁸O data. This is probably due to the atmospheric model’s very rapid reaction to changes in ocean conditions.

The onset of the thermohaline circulation decreases the meridional temperature and pressure gradient, which again decreases the zonal winds in CLIMBER-2 (not shown). This should lead to a rapid fall in the calcium concentrations, which is not mirrored in the more slowly changing calcium data.

The Ca record (Fig. 4d) does not show the same large jumps as the δ¹⁸O, even though it is often suggested that dust/Ca²⁺ reacts faster to shifts in the climate system (e.g. transition 1, where the dust reaches Holocene values in only 20 years, whereas it takes the δ¹⁸O record about 50 years (Reference Dansgaard, White and Johnsen.Dansgaard and others, 1989)).

The Ca record could be influenced by larger noise, which smooths any sharp transition when making an averaged record. Indeed, in the individual transitions (shown in grey) we do see in some cases a pronounced step right around or before the δ¹⁸O step (this can be seen for individual transitions, i.e. Reference Fuhrer, Wolff and Johnsen.Fuhrer and others (1999) and Figure 4b and c (Reference Dansgaard, Hansen and TakahashiDansgaard and others, 1984)). Also the sodium and magnesium transitions are smoother than the δ¹⁸O record. Sodium is usually considered a proxy for the sea salts in the ice, and magnesium and calcium are considered tracers for land dust blown onto the ice.

The smooth transitions in the sodium and magnesium records may originate from imprecise tie points between the two ice-core records. Similar tie-point errors may also help explain why the averaged GRIP calcium record is even smoother in the transition even though some transitions in the calcium record are quite sharp (Reference Fuhrer, Wolff and Johnsen.Fuhrer and others, 1999).

Conclusion

We have here used high-resolution ice-core data and the CLIMBER-2 model to investigate in detail what happens during the rapid warming transitions to Dansgaard– Oeschger events.

In Figure 4, GRIP calcium, GISP2 sodium and magnesium all show a decrease prior to the large δ¹⁸O jump. We consider this early decrease to be an artefact of the averaged ionic data. We conclude that weakening of the atmospheric circulation (zonal wind speed) (Reference MayewskiMayewski and others, 1997) leading to the drop in ionic concentrations most likely occurs in phase with the isotopic shifts in Greenland. This conclusion is supported by the CLIMBER-2 model runs presented here. the CLIMBER-2 model runs show even faster shifts into a Dansgaard–Oeschger event than are evidenced by the δ¹⁸O data.