News & Views | Published:


Glacial hiccups

Nature volume 409, pages 147148 (11 January 2001) | Download Citation


The climate instability of glacial times probably resulted from abrupt switches in ocean circulation. A computer-model simulation provides the first glimpse of the dynamics involved.

Twenty years ago, climate was thought to have remained generally stable, at least for the past few millions of years. The belief was that the ocean and the atmosphere adjusted slowly to stately variations in ice-sheet extent during glacial–interglacial cycles. But the study of palaeoclimatic records has since revealed that climate was in fact highly variable during glacial times1. It switched abruptly between cold and warm modes, with the temperature in Greenland changing by up to 10 °C in a matter of decades2. A crucial step towards understanding these glacial hiccups is presented by Ganopolski and Rahmstorf3 on page 153 of this issue.

It has long been suspected that ocean circulation in the North Atlantic is involved in the abrupt coolings and warmings during glacial periods4. The circulation depends mainly on the density of sea water, which is a function of temperature and salinity. These two properties determine the strength of the so-called thermohaline circulation, which in the North Atlantic contributes to the northwards flow of warm water on the surface, followed by heat release and sinking of the cooler water at high latitudes, with ensuing southerly flow at depth.

As early as 1961 it was proposed5 that the interplay between the effects of temperature and salinity could lead to different modes of ocean circulation. In particular, changes in the amount of fresh water in the Nordic Seas can affect deep-water formation in the North Atlantic and alter the thermohaline circulation. As shown in Fig. 1a, present-day climate corresponds to an active North Atlantic circulation. If freshwater input into the Nordic Seas rises above a threshold value (F1 in Fig. 1a), the thermohaline circulation must jump abruptly from its equilibrium (warm) branch to a different one. This new branch corresponds to a much reduced circulation, with colder temperatures at high latitude because less heat is transported there.

Figure 1: Climate (temperature) stability as a function of freshwater input at high latitudes in the North Atlantic.
Figure 1

a, Under present-day conditions, North Atlantic climate has essentially two possible equilibria. When freshwater input exceeds a threshold value F1, thermohaline circulation jumps from the upper (warm) equilibrium branch to the lower (colder) one, which corresponds to thermohaline collapse (blue line). It can return to the upper branch only if fresh water is removed (by, say, evaporation) and decreases below the threshold value F2. The hysteresis width F1F2is large. So present climate is not destabilized by weak freshwater perturbations. b, Under the conditions of the Last Glacial Maximum, the hysteresis is much narrower and so the system is much more sensitive to the input or removal of fresh water — even a slight reduction can induce abrupt warmings, and such Dansgaard–Oeschger warming events are evident in the palaeoclimate record. Large inputs of fresh water, as during Heinrich events (ice-sheet melting), will induce a relatively small cooling through thermohaline collapse. c, A guess at an intermediate situation, as pertained during isotopic stage 3, around 50,000–30,000 years ago. The warm (upper) branch is more stable than it is under LGM conditions, corresponding to the longer Dansgaard–Oeschger events that occurred at this time.

When the freshwater perturbation vanishes, the Atlantic circulation does not return to its initial behaviour, but stays inactive. Only a negative perturbation (removal of fresh water, for example by evaporation) can bring it back to normal. In other words, the return pathway is not the same as the perturbation one. The system follows a so-called hysteresis loop as shown in Fig. 1a. Model experiments6,7 have confirmed this behaviour for the Holocene — the interglacial of the past 10,000 years. But palaeoclimatic records1 show that the Holocene has been a time of essentially stable climate.

So how do we investigate the large variabilities of glacial times? A prerequisite is a reasonable computer representation of climate at the Last Glacial Maximum (LGM), which occurred around 22,000–19,000 years ago. This is a tough task for the general- circulation coupled ocean–atmosphere models of today's climate; indeed, equilibrating such models for the very different LGM climatic regime is daunting in itself. But a faster track is possible, and a few years ago Ganopolski et al.8 built an 'Earth model of intermediate complexity' that can perform integrations over thousands of years, yet can also represent the main characteristics of the ocean and atmosphere fairly well. The model's relevance was shown when its results for the LGM climate compared favourably with palaeoclimate reconstructions.

In their new paper3, Ganopolski and Rahmstorf use an improved version of this model to describe the sensitivity of glacial climate to small changes in the amount of fresh water in the Nordic seas. The most notable result is the very different shape of the hysteresis curve under LGM conditions (Fig. 1b). The hysteresis is wide for present-day climate, and can account for the stability of the Holocene because only large freshwater perturbations can destabilize the system. But under LGM conditions, this hysteresis is much narrower, bringing the thresholds for abrupt change closer to the 'unperturbed state'. This explains why glacial climate was so unstable. The LGM equilibrium is located to the right of this hysteresis loop: so even a small loss of fresh water (through increased evaporation, decreased precipitation or run-off from land) induces an abrupt warming, as manifest in the events, known as Dansgaard–Oeschger warmings, recorded in ice cores1.

The LGM equilibrium is stable if only small amounts of fresh water are added. But, in contrast to the situation today, the lower branch of the LGM hysteresis is not flat. It does not initially correspond to a collapse of the thermohaline circulation, but only to a colder climatic regime where deep-water formation takes place south of Iceland instead of in the Nordic seas. Consequently, there is room for additional cooling through complete thermohaline collapse if there is a substantial addition of fresh water. This is what happened during so-called Heinrich events in the North Atlantic9 — times, as reflected in sediment cores, when there was large-scale release, and subsequent melting, of icebergs from the polar ice sheets.

But we also need to simulate these rapid warming and cooling events under different external 'forcings'. To start with, we can make a guess for conditions intermediate between a full glacial and a full interglacial (that is, between Fig. 1a and Fig. 1b). Figure 1c shows my own guess for such an intermediate period, 'isotopic stage 3' (50,000–30,000 years ago). Here the hysteresis is wider than in LGM conditions and, like today, the unperturbed state lies inside the hysteresis loop. Warm modes would last longer, as is evident in certain Dansgaard–Oeschger events during this time. So the dynamical picture provided by Ganopolski and Rahmstorf3 nicely accounts for the phenomenon of glacial variability. The temporal and geographical patterns of events that emerge from the model, in particular the phase relationship between warmings in Greenland and Antarctica, also compare rather well with the palaeoclimatic data10.

Plenty of questions remain, of course, most notably that concerning the initial causes of the instabilities. Ganopolski and Rahmstorf carefully avoid the problem by applying a weak, but unexplained, periodic forcing to generate Dansgaard–Oeschger oscillations in the model. They mention solar variability as a possible cause. But little is known of such variability, and invoking it is more pulling out a wild card than providing a solid explanation. The thermal response of the climate model (about 7 °C warming over Greenland) is also weaker than ice-core records indicate (about 10 °C). So other feedbacks probably need to be taken into account — changes in the concentration of atmospheric methane10, for instance. More broadly, the next step in this line of research will require the coupling of climate models with an ice-sheet model that can simulate the storage and release of large amounts of fresh water over centuries or millennia.

The topic of climate stability is high on both scientific and political agendas, and looks set to stay there. A faithful representation of the Earth system in computer-model form is needed to clarify events of the past, better to predict the future. But the route to that end still lies mostly ahead of us.


  1. 1.

    et al. Nature 364, 218–220 (1993).

  2. 2.

    et al. Nature 391, 141–146 (1998).

  3. 3.

    & Nature 409, 153–158 (2001).

  4. 4.

    , & Nature 315, 21– 26 (1985).

  5. 5.

    Tellus 13, 224–230 ( 1961).

  6. 6.

    & Nature 351, 729–732 (1991).

  7. 7.

    & Nature 378, 165–167 (1995).

  8. 8.

    et al. Nature 391, 351–356 (1998).

  9. 9.

    & Paleoceanography 14, 716–724 (1999).

  10. 10.

    et al. Nature 394, 739–743 (1998).

Download references

Author information

Corresponding author

Correspondence to Didier Paillard.

About this article

Publication history




  1. Search for Didier Paillard in:

Further reading


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Newsletter Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing