Science Briefs

Modeling an Abrupt Climate Change

An important reason to understand past climate changes is to help improve climate predictions for the future. This can be a difficult task due to the disparities in time scales (millions of years for past climate compared to decades for future climate), difficulties in the interpretation of past climate records, and the often very different focuses of the paleoclimate and modeling communities. Occasionally, though, a particular past period or event presents itself as an almost ideal case study for model-data comparisons.

Map showing North Atlantic circulation patterns Figure 1: Circulation patterns in the North Atlantic Ocean. Cold, dense water is shown in blue, flowing south from upper latitudes, while warm, less dense water flows north. Click for large GIF. (Source: Jack Cook for Ocean and Climate Change Institute, Woods Hole Oceanographic Institution.)

One such case was an abrupt cooling event across the Northern Hemisphere which occurred about 8200 years (8.2 kyr) ago and which is documented by multiple types of paleoclimate records as lasting several decades to a few centuries. Separate geologic lines of evidence document the catastrophic drainage of the prehistoric glacial Lakes Agassiz and Ojibway in Canada into the Hudson Bay at approximately the same time. This fresh water pulse may have been the catalyst for a decrease of meridional overturning circulation (MOC) in the North Atlantic (see Figure 1). (MOC is the flow of dense and bottom water away from high-latitude sources and the compensating return flow of less dense upper-ocean water. It is sometimes called the "conveyor belt".) This decrease may have lead to subsequent cooling around the Northern Hemisphere. However, this idea remains to be tested quantitatively.

Turning to the future, the complex coupled ocean-atmosphere models being analyzed for the Intergovernmental Panel on Climate Change (IPCC) 4th Assessment Report need to be validated against as much data as possible. In particular, modellers need to be able to evaluate their models' ocean responses to climate change (see Figure 2). The spread in model projections for the North Atlantic MOC as a function of increasing greenhouse gases is extremely large, ranging from an almost 50% decrease to a small increase by 2100. In part, this uncertainty stems from modellers tuning for the existence of a stable North Atlantic circulation, but not being able to tune for its sensitivity for lack of appropriate data.

Figure showing change of water transport by Atlantic conveyer belt as simulated by several climate models. Figure 2: Simulated water-volume transport change of the Atlantic "conveyor belt" in a range of global warming scenarios computed by different climate research centers. Shown is the annual mean relative to the mean of the years 1961 to 1990 (units SV = 106 m3/s). The past forcings are only due to greenhouse gases and aerosols. Click for large GIF or PDF. (Source: Chapter 9 of IPCC report Climate Change 2001: The Scientific Basis)

Useful model-data comparisons have a number of pre-requisites. Wide-spread and clear data on the event is crucial, but even more important is the existence of a plausible and interesting candidate for the cause of the climate change. Changes in MOC have been deduced for many paleoclimatic periods, but a further set of constraints prevents modellers being able to take full advantage of that data. In short, for fully coupled climate models the initial conditions need to be close to those of the present day, and the duration of the event needs to be short enough to be tractable with current computer resources (i.e., decades to centuries rather than millennia). This mitigates against the use of climate variability such as Dansgaard-Oeschger and Heinrich events during the last glacial period and possibly the Younger Dryas cooling (a good example of an effect with an as-yet-unquantified cause) about 10,000 years ago.

The 8.2 kyr abrupt climate change event would therefore appear to have it all: well-dated and widespread data, a relatively short duration, base climate close to modern (only remnant ice sheets, minor differences in greenhouse gases compared to the pre-industrial, and relatively small insolation differences), a potentially important ocean response, and crucially, a quantifiable hypothesis for a cause — the catastrophic draining of Lakes Agassiz and Ojibway. Our simulations of this event using GISS ModelE therefore take the hypothesized cause and try to reproduce the response. (This model is exactly the same as that used for IPCC simulations of future climate.) The cause is straightforwardly done; we add appropriate volumes of water to the Hudson Bay in very short bursts (six months to a year).

Figure showing climate model response to simulated 8.2kyr freshwater forcing event. Figure 3: GISS ModelE simulated 8.2 kyr climate response to freshwater forcing in the Hudson Bay (2.5 to 5 Sv years, or 25-50 cm of equivalent sea level). Panels (a) and (b) show the model's simulated changes in surface air temperature and precipitation, respectively. Panels (c) and (d) show changes in the simulated tracers directly compared to oxygen isotopes (or temperature and precipitation proxies) in the paleoclimate record. Click for large GIF or PDF. (Source: NASA GISS)

Judging the response is harder. To have a more consistent match to the data, we used multiple "tracers" in the model (water isotopes, methane, dust, and other aerosols) to "forward model" the same past climate records that we see in past climate proxies such as ice cores, cave records, and ocean and lake sediments. What do we see? It turns out that the amount of water in the lake can cause sufficient changes in the density of the North Atlantic to cause a slowdown in the MOC of between 30 and 60%, large enough to cause significant cooling (up to 2-3°C, or 3.5 to 5.5°F, in areas of the North Atlantic) and shifts in rainfall bands to the south in both the North Atlantic and the tropics.

The climate changes affect the tracers in predictable ways. Water isotopes become more depleted in line with the cooling (as observed), and because of the reduced rainfall in the region, dust and aerosol concentrations increase in Greenland snow (as observed). Methane emissions respond to the drying — reducing wetland extent, and cooling — reducing the basic anaerobic respiration by bacteria (again, as observed). Put together, the match to the observed data is very good over a whole range of proxies with very different biogeochemical behavior.

What does this imply? Given the very different physics of each of the proxies, it seems unlikely that the model would give good results for all of them if its simulated climate response was completely off. Therefore, the results demonstrate that a multi-decadal period of reduced overturning in the ocean is very consistent with the whole pattern of the observed event. By scaling the model's response a little, we estimate that a reduction of about 50% in the MOC is the most consistent with the data.

Thus, we have now been able to test the ocean's response to freshwater forcing. While the 8.2 kyr event is not an analog for what may happen in the future, a slowdown in the MOC is predicted by our model (and others) for a future world, partly as a function of ocean warming and partly as a function of increased freshening from ice melt and increased rainfall. As more models perform these kinds of experiments, it may be possible to narrow the uncertainties in the future projections based on how well they simulate the 8.2 kyr event.

Also See

News Release: Scientists Confirm Historic Massive Flood in Climate Change


LeGrande, A.N., G.A. Schmidt, D.T. Shindell, C.V. Field, R.L. Miller, D.M. Koch, G. Faluvegi, and G. Hoffmann 2006. Consistent simulations of multiple proxy responses to an abrupt climate change event. Proc. Natl. Acad. Sci. 103, 837-842, doi:10.1073pnas.0510095103.

Schmidt, G.A., and A.N. LeGrande 2005. The Goldilocks abrupt climate change event. Quaternary Sci. Rev. 24, 1109-1110, doi:10.1016/j.quascirev.2005.01.015.


Please address all inquiries about this research to Allegra LeGrande.