Sunday :: Mar 22, 2009

an enigma, wrapped in a riddle, shrouded in mystery


by Christina Hulbe

I exaggerate, as does the headline here. Nevertheless, an important paper published recently in the journal Nature fills in some large gaps in our knowledge of temperature trends in a remote, sparsely observed region, the Antarctic. The data are important not only as a metric by which to track the effects of global warming, but also as a tool with which to improve our understanding of climate system processes and the models we use to project future change.

The problem with evaluating temperature trends over the Antarctic continent is the data. There's not that much of it. The satellite record, which provides good spatial coverage, is short and the ground-based observational record is sparse (and often also of limited duration). Multi-annual temperature trends related to normal modes of atmospheric variability can also complicate the picture. The Antarctic Peninsula is known to have been warming as long as there have been weather stations in the region (measurements begin in 1901 at Orcadas on the South Orkney island of Laurie). Temperature changes in the interior of the continent have been less clear.

Recently, a multidisciplinary group led by a University of Washington professor, Eric Steig, made a fundamental contribution to sorting out temperature change in the Antarctic (Nature, 2009, v 457; source). Their analysis, which yields the longest and most spatially complete temperature record yet available, shows that the mean annual temperature trend from 1957 (when the data and their analysis begin) to 2006 is positive across the whole of the continent and that the warming is greater over the west Antarctic than over the east.

Why were they able to improve on past analyses? Math! More after the break...

geography basics
Antarctica can be divided into three major geographic regions. The Transantarctic Mountains divide the continent into eastern (on the Indian Ocean side) and western (on the Pacific Ocean side) regions. The East Antarctic is a high, dry, cold, vast plateau of drifting snow atop a kilometers-thick ice sheet. The ice only becomes steep near the coast.

The West Antarctic interior has a different geometry, as I have discussed here before. The West Antarctic ice sheet stores enough ice to raise sea level between 5 and 6 meters, were it to all melt and the East Antarctic Ice Sheet holds about 10 times more. (The total ice volume is larger than this but once the ice is melted, you have to fill in the hole it left behind.)

The third region, the Antarctic Peninsula (AP, map), is a narrow, mountainous chain extending north toward Drake Passage. The AP plateau is composed of a series of ice caps, which drain into mountain glaciers that eventually deliver the ice accumulating on the plateau back to the sea. About half a meter of sea level equivalent ice is stored on the AP. Temperatures on the AP are relatively mild, compared with the rest of the continent.

Here are some maps of Antarctic surface elevation and mean annual temperature.

The continent is surrounded, seasonally, by sea ice that freezes at the ocean surface. Just as in the Arctic, sea ice formation in the Antarctic is important to many parts of the Earth system, including ocean circulation and climate.

a note about variability
When looking for temperature trends, we must distinguish between changes associated with oscillations within the atmosphere (such as the tropical El Nino Southern Oscillation) and changes due to additional forcings on the system (for example, anthropogenic greenhouse gas emissions).

Outside of the tropics, the leading mode of variability in atmospheric circulation is an oscillation of atmospheric mass between mid- and high-latitudes in both the northern and southern hemispheres. Centered about the poles, these are often called annular modes. The annular modes describe variability in the atmospheric circulation that is not due to the changing seasons. They reveal themselves as distinctive multi-annual patterns in sea level pressure, temperature, wind strength, and other weather phenomena. Recent work using AWS data and climate models, summarized at this website (Monaghan and Bromwich at The Ohio State University), shows just how different Antarctic temperature trends can look over short time spans.

Ultimately, we are interested in both the long-term trend and the variability. These quantities allow us to diagnose certain aspects of global warming, put observed changes (for example, glacier retreat on the AP or changes in penguin colony sizes) into a climate system context, and test the performance of climate models.

temperature records
A quick look at this map reveals a fundamental problem in making useful assessments of Antarctic temperature change. The ground-based data are sparse. Automatic Weather Stations (AWS, picture) tend to be located near the coast and near research stations. The record is also discontinuous, with new stations added and old stations going inactive over time. Only 15 stations reach from the present back to the start of the Antarctic scientific era, the 1957/58 International Geophysical Year. The satellite record provides better spatial coverage and is in essence continuous but only begins in the late 1970's and early 1980's.

Before we get to Steig and colleagues' contribution, let's clarify the problem. The Antarctic is vast (about twice the area of the United States) with great spatial variation in its weather patterns, due to topography as well as patterns in the larger atmospheric and ocean circulation. It's the same for any continent.

Imagine that you have two temperature measurements, one on the Oregon coast and one in Chicago and you want to use them to predict the temperature in Sturgis, South Dakota. You can't simply draw a line between the two, calculate a slope for that line, and place Sturgis on the line (a procedure called linear interpolation). Even a smart line that took into account elevation differences among the three locations would get it wrong, because the regional climate patterns that affect weather in coastal Oregon and in the midwest are different. You might be better off just using Chicago as a predictor for Sturgis but you wouldn't be correct, because atmospheric processes over the north Pacific (and reflected in weather along the Oregon coast) do, in some way, affect the weather in Sturgis.

Suppose you added a few more records, perhaps in Philadelphia, Tallahasse, Galveston, and Santa Cruz, and computed an average that accounted for the varying distances between those cities and Sturgis. Your prediction for Sturgis might still be poor because you have assumed a very simple relationship between Sturgis and those cities, a relationship that doesn't take weather patterns established by atmosphere (and ocean) circulation into account. Now suppose you also want to use data archives from your handful of recording stations to make maps of past temperature change over the whole of the United States. Further suppose that individual stations in your network blink in and out over time. It's a big challenge. This is the challenge faced in the Antarctic.

The ideal data set for investigating past temperature change would be a regular grid of continuous temperature records over the whole of the the continent. Each year's grid would be a slice in time that we could compare with other years to compute both trends and variability. But we don't have that, so we must interpolate (fill in the gaps) the existing data over the whole continent and we must do so in a climatologically useful way. There are many interpolation schemes available, from simple linear schemes that assume a constant change (or slope) between neighboring data points, in our case a location and a temperature, to fitting polynomials to groups of data points, to more complicated approaches. What we need for our problem is a scheme that accounts for the regional patterns set up by topography and the larger climate system. We also need to be able to evaluate the statistical significance of any trends we calculate using our finished product.

Various approaches have been used to reconstruct complete historical temperature fields, with somewhat inconsistent results. For example, Monaghan and Bromwich, whose work I noted above, use spatial relationships that emerge in a climate model among sites where long-duration AWS records exist to interpolate between the sites (technical reference).

the new Antarctic work
Steig and colleagues use a new approach, one that does not rely on climate models. Instead, they use two different data sets, satellite observations of surface temperature (thermal infrared, THIR, "skin temperature") and all existing AWS records (not just the long duration ones) to compute a statistical measure of the spatial patterns in those data, a quantity called the spatial covariance pattern.

Covariance is a measure of the degree to which two variables change together. If one variable increases while the other does as well, there is a positive covariance between them. When the temperature in Chicago is higher than average, is it higher in Sturgis as well? When the temperature in Astoria, Oregon, is higher than average, is the temperature higher or lower than average in Sturgis?

The two different estimates of spatial covariance (from satellite and from the AWS) can then be compared. This is important because the satellite THIR and AWS do not measure exactly the same thing. It turns out that the covariance patterns agree well and the usefulness of the method developed by Steig and his colleagues is affirmed. The significance of a trend computed over a particular region using the new method is also calculated. The statistical significance tells us whether or not we are allowed to say that a computed trend is distinguishable from zero.


results of the analysis
In a nutshell, Steig and colleagues find that Antarctic surface temperature has warmed significantly since 1957, that the trend is stronger in winter and spring than other seasons, and that the trend is larger in the west Antarctic than in the east. The mean annual warming continent-wide is 0.12 +/- 0.07 degrees Celsius per decade. The west Antarctic warming trend is 0.17 +/- 0.06 degrees Celsius per decade (the Peninsula-only rate is 0.11 +/- 0.04 degrees C per decade).


context
It is important to understand this result in context. While the overall trend is warming, significant multi-annual variability exists and there is important information in that variability, for example about the role of the southern annular mode and changes in stratospheric ozone, as I discussed here a while back. It seems to need repeating that the variability is part of what defines climate.

The new result also shows that there are some things missing in the climate models, which in general do not produce the observed warming trend. Why this is the case is an important (if small) field of research. Mismatches between model and observation point the way to processes and places in need of further study. The answers must lie in the details of atmospheric circulation but also in the sea ice just offshore. As sea ice extent and season length change, so too does heat exchange between the ocean and the atmosphere (which in turn can affect temperature over the Antarctic continent). It is known that limitations in the representation of sea ice lead to underestimation of Arctic sea ice decline in climate models. It is reasonable to suspect similar problems around the Antarctic as well. Steig and colleagues conduct a nice analysis of this in their paper.

meta
If you are interested in (an interesting) "meta" discussion of this paper and its reception in the media, visit RealClimate here.

Christina Hulbe :: 6:35 PM :: Comments (20) :: Digg It!