Global reorganization of atmospheric circulation during Dansgaard–Oeschger cycles

Here, we focus on the amplitudes of these transitions in atmospheric circulation, investigating their global pattern and spatial distribution, as well as their impact on shifts in moisture source regions, changes in seasonality, and in the amount of precipitation, as recorded in speleothem δ¹⁸O records. We utilize δ¹⁸O records of 111 speleothems from 67 different caves distributed across all continents except Antarctica. The data were extracted from the second version of the Speleothem Isotope Synthesis and Analysis (SISAL) database (35). A detailed description of the data and choice of criteria for suitable speleothems is provided in the supplement. We analyze the time series with respect to abrupt climate warming events—i.e., detection of DO transitions from stadial to interstadial climate conditions. To detect the transitions in the different time series, we follow an established method (9), sliding two neighboring windows through the datasets of δ¹⁸O values. The largest difference of the averaged values in both windows provides an estimate for the center point of a potential abrupt transition. To be assigned as a transition, the strong change in δ¹⁸O must moreover be close to a known stadial–interstadial transition (3; for technical details, see supplement). In our study, we focus on interstadial minus stadial δ¹⁸O differences in precipitation derived from speleothem δ¹⁸O values.

In addition to speleothem δ¹⁸O time series, we analyze the global patterns of temperature and precipitation changes associated with stadial–interstadial transitions that arise naturally and unforced in a fully coupled version of the National Center for Atmospheric Research (NCAR) Community Climate System Model, version 4 (CCSM4; 36, 37). Applied boundary conditions (trace gases, orbital configuration, land-sea mask, and continental ice shields) were those of 21 ka B.P. The model was run at 1° × 1° for both the atmosphere and ocean, a resolution comparable to CMIP5 models (for details, see SI Appendix). We calculated 100-y averages of precipitation and temperature data during each stadial and interstadial state and determined the difference in order to compare them to the speleothem δ¹⁸O differences.

Speleothem δ¹⁸O is enriched compared to δ¹⁸O in cave drip water by temperature-dependent fractionation processes during carbonate deposition on speleothem surfaces. Cave air temperature is generally well represented by mean annual air temperatures. Therefore, we can account for this fractionation process by using CCSM4-derived mean annual air temperature for stadials and interstadials and applying a laboratory-based fractionation equation (38). Thus, we can derive the δ¹⁸O values for drip water during stadials and interstadials. Note that the choice of a different temperature–fractionation relationship obtained in other studies (39, 40) does not influence the results as we focus on δ¹⁸O changes instead of absolute temperatures and the temperature–fractionation slope is similar for all relevant studies. The oxygen isotope fractionation within the cave due to temperature changes is mostly small, except for Europe where temperature differences between stadials and interstadials are largest (SI Appendix, Fig. S8). Global datasets for δ¹⁸O of precipitation and cave drip water indicate that the mean drip water δ¹⁸O closely represents variations in δ¹⁸O of local precipitation (41) and especially as we investigate climate-driven changes in δ¹⁸O of two climate states, climate-independent cave site-specific offsets will cancel. Hence, drip water δ¹⁸O variations as derived by speleothem δ¹⁸O values can primarily be interpreted as a proxy for δ¹⁸O changes in precipitation, which in turn is often used as a proxy for both precipitation amounts and changes in moisture sources (19, 20), although other less important processes can impact δ¹⁸O at specific locations as well.

We find the δ¹⁸O time series to exhibit characteristic patterns of the stadial–interstadial transitions for each of the 111 speleothems. The range of the δ¹⁸O difference in precipitation for individual stadial–interstadial transitions is approximately between −3.5 and +4.9‰, where a negative sign indicates more negative δ¹⁸O values during interstadials compared to stadials, and a positive sign is related to less negative δ¹⁸O values. We take the median amplitude of these individual transitions over each cave site for a more robust data evaluation, reducing the impact of possible outliers originating from growth rate and data resolution variations. The median stadial–interstadial transition amplitudes range from −3.1 to +2.4‰ during the last glacial period. Compared to the median δ¹⁸O changes of a stadial–interstadial transition in Greenland’s NGRIP data (around 3.5‰), the absolute median values of δ¹⁸O changes during stadial–interstadial transitions are smaller for all speleothem records investigated. We observe a pronounced global pattern with spatial heterogeneity of opposite signs in speleothem δ¹⁸O changes (Fig. 2).

For Asia, more negative δ¹⁸O values in precipitation during interstadials compared to stadials are observed in the speleothem records (Fig. 2). Following the general interpretation of δ¹⁸O values in paleoclimate data from the South Asian monsoon region (12, 42, 43), namely that decreases of δ¹⁸O reflect more humid trends, the amount of precipitation is higher during interstadials than stadials. This is in line with the interpretation of isotope-enabled GCM data focussing on freshwater-induced millennial-scale cooling events (44, 45) and is consistent with the CCSM4 simulations, where large parts of the southern Asian monsoon region show wetter conditions during interstadials (Fig. 2A). We note here that modeling studies focussing on the East Asian Summer Monsoon system highlight an important contribution of the precipitation history (44, 45), but also argue for variations of the precipitation source regions (46) to explain observed shifts in δ¹⁸O in precipitation.

Unfortunately, there is only little information from other speleothem precipitation-sensitive proxies available in this region. Often, the Mg/Ca ratio and Sr/Ca ratio are considered as more reliable proxies to trace changes in the amount of precipitation. To date, the only study focussing on δ¹⁸O and trace elements for millennial-scale variability during the deglacial period indeed suggests that δ¹⁸O records in the northern/eastern Asian monsoon region do not necessarily reflect changes in the amount of precipitation but possibly in the length of the annual monsoon rainfall (47).

Changes in the amount of precipitation also constitute the leading interpretation for variations in δ¹⁸O values in speleothems from the South American monsoon domain (18, 48). However, modeling studies showed that δ¹⁸O changes, especially in eastern Brazil, are also enhanced by a direct propagation of δ¹⁸O-depleted meltwater effects (49). During stadial–interstadial transitions, speleothems from South America show an increase in their stable oxygen isotopic composition, representing dryer conditions during interstadials compared to stadials. Thus, the δ¹⁸O values of the stadial–interstadial transitions in this region react in the opposite way compared to the Asian monsoon region, where the δ¹⁸O shifts have been mainly ascribed to a northward shift of the Intertropical Convergence Zone (ITCZ, (18, 48, 50). Such a shift was previously predicted by modeling studies investigating the impact of high-latitude warming and reduced sea ice cover on the position of the ITCZ (51). A northward shift in the mean position of the ITCZ goes in concert with an associated asymmetry in the Hadley circulation in the two hemispheres (18) as well as a northward displacement of the upper-level subtropical westerly jet. Hemispherically averaged, this asymmetry results in more precipitation in the northern hemisphere monsoon regions and less precipitation in the Southern Hemisphere monsoon regions during interstadial conditions (42). A similar feature is also present in the CCSM4 model data. Indeed, the northward-shifted ITCZ during interstadials is expressed by a higher (lower) rainfall amount in the northern (southern) hemispherical (sub)tropical regions (Fig. 2A).

Stable oxygen isotope and model results indicate that other regions, such as the subtropical northern hemisphere, are also affected by the shift of the ITCZ and an associated reorganization of the Hadley Circulation. For example, we see a decrease in speleothem δ¹⁸O in the Caribbean (Fig. 2A) during stadial–interstadial transitions. This has been reported to reflect an increased amount of rainfall in response to an ITCZ shift toward more northern locations during interstadials (52, 53) and is present in all speleothem records from this region. In contrast, the CCSM4 model results show a heterogeneous precipitation pattern for the Caribbean and do not indicate a clear direction and high level of agreement compared to South and South-East Asia and the northern part of South America.

For all of Europe, we observe a consistent increase in δ¹⁸O values in precipitation during stadial–interstadial transitions (Fig. 2B), which most likely has a common cause. One possible process is a change in the δ¹⁸O values and temperature of the moisture source. Stable oxygen isotope values derived from planktonic foraminifera of marine sediments in the Atlantic (Iberian Margin and off Ireland) show differences of about 1‰ with lower values during interstadials (13, 54). At the same time, they show warmer surface waters by about 10 °C. A 10 °C change of surface water (8, 55, 56) indicates that seawater δ¹⁸O values must have been about 1.2‰ higher in interstadials compared to stadials using commonly accepted oxygen isotope fractionation factors (38–40). This value is comparable to the observed averaged interstadial–stadial δ¹⁸O differences in precipitation at European cave sites (0.95‰). We conclude that δ¹⁸O changes at the moisture source seem to be the main reason to explain most of the observed δ¹⁸O differences during stadial–interstadial transitions over the European continent.

Another cluster of speleothem records located over North America also shows an increase in δ¹⁸O values during stadial–interstadial transitions. This region is situated well within the present-day westerly wind system. The stadial–interstadial shifts in speleothem δ¹⁸O from the North American continent have originally been interpreted in terms of northward shifts in the westerly storm tracks (Fort Stanton Cave; 19) and related to changes in seasonal precipitation, with summers usually showing higher δ¹⁸O values compared to winters. The other speleothems from the North American continent show a consistent pattern compared to Fort Stanton speleothems and thus highlight the general applicability of this interpretation. The change in precipitation seasonality is also present in the model data, which show either a strong increase in summer precipitation (Midwest United States) or a reduction of winter rainfall (Western United States) for interstadial compared to stadial conditions (SI Appendix, Fig. S9). An updated explanation is provided by a recent proxy data–climate model comparison study (57). This study suggests tropical–extratropical teleconnections for observed δ¹⁸O changes as the driving mechanism for increased winter moisture supply from subtropical latitudes during stadial periods in the Western United States. This explanation again stresses the importance of precipitation seasonality.

There is relatively less data available from the Southern Hemisphere outside the tropics. Nevertheless, the negative values observed from the two cave sites in New Zealand (Fig. 2A) suggest a northward shift in the westerly wind system of the Southern Hemisphere during interstadials, which occur in concert with the derived northern hemisphere westerly and ITCZ shifts. This shift in the westerlies most likely led to a change in the source of precipitation.

Recently, there was a model setup published, which was designed to model the stable oxygen isotope composition of precipitation during the last deglaciation (isotope-enabled transient climate experiment—iTraCE, (45, 58) with different forcing (greenhouse gases, ice sheet topographies, orbital parameters, injection of freshwater). When considering the transition from the Heinrich 1 state to the Bølling/Allerød phase during the last deglaciation and using the fully forced experiment setup (greenhouse gases, orbital forcing, freshwater, and ice sheets) as an equivalent for the glacial stadial–interstadial transitions, it might be suitable to compare the speleothem data with the iTraCE results. Overall, there is excellent agreement between the speleothem-derived interstadial–stadial δ¹⁸O offsets in precipitation and the results for the stable oxygen composition of precipitation of the iTraCE model experiments (SI Appendix, Fig. S6 and section 5). This comparison, using an additional, independent model simulation, strongly corroborates our results presented in Fig. 2A.

We proceed with analyzing the cave-site-specific mean interstadial–stadial differences in δ¹⁸O with respect to latitude (Fig. 3). The northern hemisphere high to midlatitudes show strong positive values. Toward lower latitudes, the offset becomes less positive and turns into a negative shift until a minimum is reached in the latitudinal belt between 20°N and 40°N. The more southward isotope shift for the North American sites compared to the ones in Europe and Asia might likely be explained by the more southward extent of the Laurentide ice sheet as compared with the Fennoscandian one. From about 20°N toward the equator and further south (10°S to 30°S), the δ¹⁸O differences for stadial–interstadial transitions increase. In general, the absolute values of the data between 10°S to 30°S are smaller than those in the northern hemisphere mid-latitudes, albeit some large interstadial–stadial differences are observable on the South American continent. The values even further south than 30°S show less positive interstadial–stadial δ¹⁸O differences again. In the Southern Hemisphere, only a few cave sites are available and the robustness of those observations could be questioned, but the described features appear also to be valid, when accounting for all detected individual events (SI Appendix, Fig. S10).

Outside of Europe, where strong moisture source temperature and δ¹⁸O changes superimpose the precipitation δ¹⁸O changes, the δ¹⁸O pattern in precipitation is consistent with the above-discussed changes in global atmospheric circulation patterns. During a stadial–interstadial transition, the mean ITCZ position is shifted northward, resulting in decreased precipitation in South America, while at the same time the northward ITCZ shift results in lower δ¹⁸O values in the monsoon regions of the northern hemisphere — e.g., in the Caribbean, on the Indian subcontinent, and over large parts of China. At the same time, the northward ITCZ shift during the phases of strong sea ice reduction is accompanied by a northward shift of the westerlies, which contributes to the observed changes in δ¹⁸O over North America and New Zealand.

To further investigate the geographical origin of stadial–interstadial transitions, we consider the change of the absolute amplitude of interstadial–stadial δ¹⁸O differences with increasing distance from the North Atlantic region. Our analysis shows that among all records, the largest speleothem δ¹⁸O interstadial–stadial differences are from caves closest to this region (Fig. 4).

As some caves and speleothems might not detect the full δ¹⁸O amplitude of stadial–interstadial transitions (SI Appendix, section 6), we consider only the amplitude for those cave sites that represent the largest 33% of all values per 2,500 km distance bin in the following. The mean δ¹⁸O difference of the speleothems within the group of the maximum amplitude sites for a given distance decreases with increasing distance from the high northern latitudes (Fig. 4). This is a robust feature, independent of the applied cut-off value (SI Appendix, supplemental text and Fig. S7) and with a correlation coefficient between the considered δ¹⁸O differences and distance from Greenland of −0.62 (p < 0.01). Modeled relative absolute precipitation changes between interstadials and stadials at the cave sites show a similar behavior. This proxy data- and model-based observation provides additional information on the geographical origin of the DO events. From a potential source region where the transitions are triggered, any signal is expected to attenuate with increasing distance from that origin. We hypothesize that this is also a valid feature in the complex and highly nonlinear climate system, because for longer distances, the influence of other processes modifying the signal will increase. We hence present empirical evidence that the origin of DO events can be traced back to the northern high latitudes.

A similar feature is also present for the Heinrich 1—Bølling/Allerød transition in the iTraCE experiment. The δ¹⁸O differences in precipitation between the warm Bølling/Allerød and the preceding cold Heinrich 1 event show a similar decreasing δ¹⁸O difference with increasing distance from the North Atlantic core region (SI Appendix, Fig. S7) providing support for our speleothem-derived observation. It is expected that the direct response in the mid-latitudes of the northern hemisphere is larger than the response in the Southern Hemisphere because of atmospheric wave propagation. This could explain why there exist some regions closer to the North Atlantic core region in both simulation products (e.g., the Mediterranean sites), which show a smaller response than some locations farther away (e.g., southern American sites). Nevertheless, the statistically highly significant relationship between signal and distance from the North Atlantic core region suggests the existence of a general physical mechanism responsible for the signal attenuation despite deviations from the overall trend due to local characteristics. We propose that the signal attenuation is driven by temperature and temperature gradient changes during stadial–interstadial transitions being largest close to the North Atlantic. As relevant variables for precipitation and its isotopic signature such as sea level pressure, evaporation, and winds are strongly influenced by the temperature and temperature gradient changes (59, 60), the signal attenuation could be imprinted onto precipitation and its oxygen isotope signature. This would lead to these two variables showing a similar signal attenuation structure as temperatures. However, the exact processes have yet to be investigated.

We have thus provided a consistent global analysis of stadial–interstadial transitions with a focus on associated changes in atmospheric circulation and precipitation. The maximum amplitudes support the hypothesis that the North Atlantic region is the trigger of the stadial–interstadial transitions. We were able to obtain a global picture of changes in δ¹⁸O of precipitation related to stadial–interstadial transitions, which helps to define the geographical constraints on the maximum shifts of the mean position of the ITCZ and an associated relocation in the mid-latitude westerlies. Our proxy-based findings are in remarkable agreement with simulations of unforced DO-type oscillations with a state-of-the-art, high-resolution climate model and simulated oxygen isotope changes in a fully forced transition between Heinrich stadial 1 and the Bølling/Allerød phase. Our results will be used as benchmarks for other isotope-enabled general circulation models to investigate the stadial–interstadial transitions in future research, aiming for a more complete picture of abrupt climate changes in comprehensive climate models.

This is TiPES (Tipping Points in the Earth System) contribution #103; the TiPES project has received funding from the European Union’s Horizon 2020 research and innovation program under grant agreement No. 820970. N.B. acknowledges further funding by the Volkswagen Foundation. M.B.-Y. and N.B. acknowledge additional funding by the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska–Curie grant agreement No. 956170. C.V.-P. acknowledges the financial support of the Portuguese Foundation of Science and Technology to Centre for Marine and Environmental Research through UID/00350/2020 CIMA. N.W. and K.R. acknowledge funding by the Deutsche Forschungsgemeinschaft project no. 395588486 and the German federal ministry of Education and Research through the Palmod project (code 01LP1926C). This study includes data compiled by SISAL, a working group of the Past Global Changes (PAGES) project and is in part inspired by discussions at a SISAL workshop in Xi'an (China) in 2019. PAGES received support from the Swiss Academy of Sciences and the Chinese Academy of Sciences.