Shakun et al., 2012, Nature


The Pleistocene Ice Ages represent the largest climate changes in recent Earth history. It was their discovery in the mid-19th century, in fact, that first awoke people to the possibility of major global climate change and spurred interest in understanding how it occurs. Ironically, while this interest has since yielded a fairly coherent picture of modern climate change, the Ice Ages themselves remain to be fully explained.

Deep-sea sediment cores first provided compelling evidence in the 1970s that Ice Ages were linked to slow cyclical changes in the earth’s orbit around the sun. These orbital cycles, however, affect the hemispheres in opposite ways, warming one while cooling the other. This seemed at odds with geologic data suggesting that the whole planet goes into and out of an ice age together. A possible explanation came in the 1980s when ice cores taken from Antarctica showed that atmospheric CO2 levels also rose and fell with the Ice Ages, suggesting that CO2 may have been the ‘globalizer’ of these climate changes. Temperatures recorded in these ice cores, however, indicated that past episodes of warming began slightly before CO2 started rising – apparently relegating CO2 to being an amplifier of warming at best rather than a driver. The conventional wisdom that the hemispheres exit Ice Ages together has also been questioned, which would further downgrade CO2 on the list of Ice Age ingredients and imply it may have been more of an effect than cause of past warming.

The Antarctic ice-core record now stretches back 800,000 years and shows a remarkable correlation between temperature and CO2 over this entire time period – clearly the two are linked, but the question remains “how?” A key issue here is that while the ice cores record global atmospheric CO2 concentration (since it is well-mixed by the winds), they only reflect local temperatures in Antarctica. Just as no place today can be reliably expected to reflect the global average temperature, it is questionable if Antarctica does this over the ice ages. One basic reason for this is simple: heat moves around with winds and ocean currents. So, one place might get warmer at the expense of another place getting relatively colder, and looking at either location individually would give a skewed view of the larger temperature patterns. Therefore, the only way to reliably track changes in the total amount of heat at the surface of the planet is to average temperature data from as many locations as possible. This approach should help to cancel out these heat redistributions, removing a complicating variable, and make it easier to interpret what drove global temperature in the past.

A few simple examples illustrate the utility of averaging:
(1) If we look at daily temperatures for 80 cities around the Northern Hemisphere over the first half of 2011 (after standardizing the data to make them readily comparable by subtracting the mean and dividing by the standard deviation of each), we see a fairly messy scatter of datapoints. To be sure, a general warming trend from winter to summer is apparent, but otherwise trying to pin down the onset of warming or its pattern through time is difficult. Even more interesting, plotting the earth’s orbit around the sun (also standardized, and zeroed at the equinox) shows a general correlation with temperature, but while some temperature datapoints lag behind the orbit, others appear to lead it. At first glance, this might seem to raise some doubt if summer last year was caused by the earth going around the sun..! The answer, of course, is that reshuffling of heat around the Northern Hemisphere helped produce the large scatter in local temperatures. If we average these 80 locations together to smooth this effect out, we see a gradual shift from winter to summer as expected, and the lag of temperature behind the orbit becomes clearer.
(2) Increasing global mean temperature during the past century is evident from satellites, surface thermometers, warming oceans, rising sea level, and retreating glaciers. And yet, plotting instrumental temperatures from the 80 sites where the deglacial proxy data come from (again standardizing to mean zero, unit variance) shows a very noisy graph with a warming trend barely discernible at best. A hasty conclusion might be that global warming has not occurred, though this would miss the essence of the term global. Instead, a simple average of these 80 sites, shows a clear warming trend over the past hundred years. Additionally, this signal is nearly identical to the global mean obtained using all available data from around the world, suggesting 80 locations can provide a decently representative picture of the globe. The lesson from this plot is that heat redistributions within the climate system (associated with e.g., the Pacific Decadal Oscillation, El Nino-Southern Oscillation, etc.) can cause large temperature changes at a given site that are not necessarily representative of the planet as a whole – but averaging can reduce these regional effects and help reveal the global pattern.  

Main results

Global temperature and CO2: The pattern of global temperature rise over the end of the last Ice Age as reconstructed from 80 proxy records around the world is strongly correlated with the increase in atmospheric CO2 concentrations recorded in ice core air bubbles. Furthermore, global warming generally appears to have lagged a few centuries behind the rise in CO2. These two points are consistent with the idea that CO2 was a major driver of global warming at the end of the ice age.
Hemispheric temperatures and AMOC: Averaging the proxy temperature records in the Northern and Southern Hemispheres separately indicates that the hemispheres warmed at different rates and times over the deglaciation. Southern Hemisphere warming appears to have led changes in CO2 - consistent with the Antarctic ice core results - while the Northern Hemisphere seems to have lagged behind. More interestingly, the differences between these two hemispheric curves are correlated with changes in the strength of the Atlantic Meridional Overturning Circulation (AMOC) recorded in marine sediments. The AMOC currently transports a large amount of heat northward across the equator, helping to explain why the Northern Hemisphere is a degree or so warmer than the Southern Hemisphere today. It follows that turning down this flow can shift the balance of heat back toward the Southern Hemisphere. Thus, oscillations in the strength of the AMOC during the deglaciation provide a likely explanation for the different patterns of warming seen in the two hemispheres. For instance, at the onset of deglaciation the AMOC appears to have collapsed, initiating warming in the south at the expense of cooling of the north. A key point here is that this can why Antarctic temperatures lead CO2, even though the global average temperature does not.
Model simulations: While the correlations between the proxy temperature curves and the CO2 and AMOC reconstructions are quite suggestive of a mechanistic link, they do not prove causation. Our paper also includes output from the first-ever transient simulation of the last deglaciation with a coupled global climate model to examine the physical basis for these correlations. The model results indicate that greenhouse gases can indeed explain the preponderance of the global warming recorded by the proxy records, while heat shifts between the hemispheres are largely attributable to variations in AMOC strength, supporting our interpretations of the data.
Triggering deglaciation: While these results suggest that collapsing ocean circulation initiated warming in Antarctica and the subsequent rise in CO2 caused global-scale warming, they do not reveal the ultimate trigger for deglaciation - what caused the AMOC to collapse in the first place? Paleo sea-level data show an initial jump in sea level around 19,000 years ago and the glacial geologic record on land indicates that this water came from melting Northern Hemisphere ice sheets (Clark et al., 2009). A probable explanation for this ice-sheet retreat is cyclical changes in the earth's orbit, which caused an increase in summer sunlight hitting the Northern Hemisphere beginning a few thousand years earlier (He et al., 2013). The resulting influx of meltwater to the North Atlantic may have placed a buoyant cap of freshwater on the critical zone where the AMOC sinks down toward the seafloor, effectively jamming the circulation...and its transport of heat.


What does this study tell us about global warming today?
In truth, not a whole lot. A large body of evidence already supports the role of CO2 in causing climate change - from theory, models, and paleoclimate. Since the ice-core record of greenhouse gases was developed, there has been little doubt that CO2 was somehow involved in the Ice Ages. This study simply suggests a leading, rather than secondary, role for greenhouse gases. In this sense though, the study provides additional evidence for the potency of CO2 as a warming agent and and gives a very tangible, real-world example of the long-term implications of rising greenhouse gas levels...e.g., the end of an ice age.

But the deglacial CO2 rise was natural?
The climate system does not care whether CO2 comes from a volcano, decomposing dirt, or a car tailpipe. Any molecule of CO2 has the same heat-trapping properties as any other molecule of CO2. Also, it is important to note that CO2 varying in the past due to natural processes does not mean that the current rise in CO2 is natural. The dominantly anthropogenic origin of rising CO2 today is clear from several lines of evidence, including its chemistry (d13C, d14C), its anticorrelation with atmospheric oxygen content, and an accounting of human fossil fuel consumption and land use changes. These past changes in CO2, however, do tell us that we need to be aware of how the carbon cycle responds to a changing climate if we are to anticipate the future trajectory of greenhouse gas levels (anthropogenic + natural).

Where did the CO2 come from?
There was probably less carbon stored on land during the Ice Age than today due to the generally drier conditions and greatly expanded ice cover then; so the land probably took up rather than emitted CO2 to the atmosphere during the deglaciation. The only other major carbon reservoir that exchanges CO2 with the atmosphere on the time scale of a deglaciation is the ocean, and so the extra CO2 must have been stored in the deep sea during the Ice Age. Current thinking points to the Southern Ocean as the key region where the CO2 was vented to the atmosphere. This, in fact, might explain why CO2 lags slightly behind Antarctic temperature - climate there has to change to open or close the Southern Ocean carbon valve.

Can 80 points represent the globe?
As the 20th century example above suggests, 80 widely distributed points may provide a decently representative picture of global mean temperature. Sampling our climate model of the last deglaciation at the 80 proxy points likewise suggests that these points capture the global mean well. The reason for this is simple - temperature changes covary over relatively long distances, such that warm years (or millennia) in Boston tend to be warm years in New York. So, a temperature record from a single point can be representative of a much larger area, and 1000 points may not actually give you much more information than 100.

How do we know temperatures 20,000 years ago?
Many things in the environment respond strongly to temperature, and if they are preserved, dateable, and can be quantitatively linked to temperature, they can serve as a paleothermometer. For example, the magnesium content of plankton seashells is a strong function of the temperature of the water they live in, and these shells sink down into a layer of seafloor mud when the plankton die. Likewise, the isotopic composition of precipitation around the world today shows a strong correlation with the temperature of the location where it falls, and in the case of an ice sheet, gets recorded in each year's layer of snow. Also, the type of vegetation in a region is obviously strongly influenced by temperature, and one can reconstruct past vegetation communities from pollen assemblages recorded in the sediments of a nearby lake. Several other techniques rely on similar logic. While each proxy is also influenced by other non-temperature effects, these confounding factors are hopefully different for each proxy. Therefore, when lots of proxy records are amalgamated together into a global reconstruction, the common pattern that emerges is presumably dominated by temperature. In other words, the signal-to-noise ratio increases, which provides a more meaningful temperature metric than could be obtained from any single record.

How sure can we be of the timing?
The major issue for dating the proxy records is that most are from the ocean. Since old waters from the deep ocean mix to some extent with surface waters, a typical sample of seawater from the ocean's surface has a radiocarbon - or so-called reservoir - age of a few centuries. This reservoir age needs to corrected for when developing a proxy record, and unfortunately, reservoir ages do not have to be constant through time. Nonetheless, one observation provides confidence that we have used reasonable reservoir corrections. Climate signals from land-based records, where no reservoir correction is needed, are very similar to those from the ocean records and thus the timing of the ocean signals is probably close to correct. The trick with dating the CO2 record from Antarctic ice cores is that, while ice-core layers can often be counted, the gas bubbles are younger than the surrounding ice because it takes time for the pores between snowflakes to completely close off and become isolated from the atmosphere. This is a particularly big problem in parts of Antarctica where snow accumulates so slowly that this process can take millennia. Fortunately, Greenland is snowier and the problem is therefore smaller there. Abrupt global jumps in methane allow air bubbles in Antarctic ice to be linked into the much more precise Greenland chronology, and this timescale can then be used for the Antarctic CO2 record. The resulting error bars are reasonably small, though not insignificant (a couple centuries). Fortunately, a new ice core has recently been drilled in Antarctica at WAIS Divide where snow accumulates very rapidly to develop a much better-dated CO2 record.

What does the global temperature lag behind CO2 mean?
The climate does not equilibrate instantly with rising CO2, because there are a couple big sources of thermal inertia - the ice sheets and ocean, which take a long time to melt away and warm up. Therefore, it would make sense that global temperature lags behind CO2 during the last deglaciation. This is the same reason that even if CO2 concentrations were held constant today, the earth would still continue warming - the so-called "warming in the pipeline".

Did CO2 drive previous deglaciations?
The last deglaciation is the end of only the most recent Ice Age; dozens more Ice Ages preceded it. It is unclear if CO2 played a similar role during earlier deglaciation, and this is a more difficult question to answer because radiocarbon dating cannot be used for earlier deglaciations.

Does CO2 always drive climate?
No. The climate is a complex system with myriad knobs and levers that control it. This study though, in addition to many others, suggests that CO2 is perhaps the biggest control knob. As an analogy, the economy is also a complex system subject to a multitude of factors, but clearly some (e.g., technological innovation) have a larger overall effect than others (e.g., Massachusetts sales tax).

Was the climate model tuned to match the proxy data?
Mostly not. The climate model was driven by four factors: (1) variations in the earth's orbit around the sun (known precisely), (2) the retreat of ice sheets (known reasonably well from geologic records), (3) rising greenhouse gases (known from ice-core records), and (4) freshwater poured into the ocean associated with the melting of the ice sheets. We know approximately how much freshwater went into the ocean during the deglaciation from sea-level records, but it is less certain where exactly the freshwater went in through time (e.g., Antarctica vs. North America; Hudson R. vs. Mississippi R.). The location is important because it controls if/how the AMOC responds to the freshwater addition. Several reasonable freshwater scenarios were tried, and the one that yielded North Atlantic climate oscillations in best agreement with the Greenland ice core records was used. It is worth reiterating that while freshwater-AMOC changes affect how heat is redistributed around the planet, they have relatively little impact on the global mean temperature.