RealClimate logo

Clarity on Antarctic sea ice.

Filed under: — eric @ 19 December 2014

I’ve always been a skeptic when it comes to Antarctic sea ice. I’m not referring here to the tiresome (and incorrect) claim that the expansion of sea ice around Antarctica somehow cancels out the dramatic losses of sea ice in the Arctic (NB: polar bears don’t really care if there is sea ice in Antarctica or not). Rather, I’m referring to the idea that the observation of Antarctic sea ice expansion represents a major conundrum in our understanding of the climate system, something one hears even from knowledgeable commentators. In this post, I’ll try to provide some clarity on this subject, with some basic background and discussion of a couple of important recent papers.

In general, Antarctic sea ice forms near the coastline, where upwelling waters cool to the atmosphere. It melts when the winds and currents push it into areas of warmer water to the north. In the summer, it melts pretty much all the way back to the coast. An efficient way to form lots of Antarctic sea ice during the autumn growth season is to have strong winds that push the ice away from the coastline. Pushing sea ice away leaves open water that can lose heat to the atmosphere, creating more sea ice. The persistent circumpolar westerlies are critical in pushing ice toward the north, into warmer waters. (Owing to the Coriolis effect, westerly winds cause northward-flowing surface ocean currents in the Southern Hemisphere).

The importance of the winds in controlling Antarctic sea ice leads to the obvious idea that changing winds can explain the increase that has been observed over the last several decades. There has indeed been a substantial increase in the circumpolar westerlies; this is very well established from observations and is associated with the oft-discussed increase in the “Southern Annular Mode” (SAM) index2. Averaged over the year, the SAM index has increased nearly monotonically since the 1970s (e.g., Marshall et al., 2003). This has led to a fairly simple logic in explaining the recent sea ice increase: the westerly winds have increased, so sea ice has increased too. Furthermore, there is good evidence that the increasing westerlies are a response to anthropogenic climate forcing from CO2 and other greenhouse gas increases in the troposphere, along with ozone declines in the stratosphere (Thompson and Solomon, 2002; Thompson et al., 2011). This would suggest that the observed increase in Antarctic sea ice extent is anthropogenic in origin, just like the Arctic sea ice decline, but for very different reasons. In short, reduced ozone in the stratosphere, and increased CO2 in the troposphere — both climate forcings that are unequivocally anthropogenic — cause increased westerly winds, which cause Antarctic sea ice to expand.

Of course, it’s not that simple. For one thing, the average increase of Antarctic sea ice is actually a small number that is the difference of two big numbers — modest increases over a large area, mostly in the Eastern Hemisphere, and very large decreases over a smaller area in the Western Hemisphere. The map below, showing change in the length of the sea ice season over the last 30 years, illustrates this point well. In spite of the average increase, there are very rapid declines in the Bellingshausen and Amundsen Seas, comparable to sea ice declines in the Arctic. Furthermore, the only season is which there is a significant trend in the westerlies is austral summer. There is a weak positive trend in fall, but both spring and winter show no trend; the SAM trends in these seasons may even be slightly negative, depending on which data are used (Ding et al., 2012). Yet the pattern of sea ice change is quite similar in all seasons: decreasing along the Pacific coast of West Antarctica, and increasing around most of East Antarctica, and in the Ross and Weddell Seas.

Trend in the length of the sea ice season, 1979-2010. Blue and purple areas show areas where sea ice is declining, orange and red where it is increasing. Source: Maksym et al., 2012</a

Trend in the length of the sea ice season, 1979-2010. Blue and purple areas show areas where sea ice is declining, orange and red where it is increasing. Source: Maksym et al., 2012

On top of these subtleties, confusion about the role of the winds has arisen because some of the prominent modeling studies that have examined the relationship between the westerly winds and Antarctic sea ice have come up with results that appear to be in direct opposition to the observations. When fully coupled climate models are run with increased CO2 and decreased stratospheric ozone, the westerly winds increase as has been observed, but sea ice decreases around most of Antarctica. For example, Bitz and Polvani, 2012 found that the pattern of trends is the mirror image of the observations, with increases, rather than decreases in the Amundsen and Bellingshausen Seas.

Annual mean response of sea ice concentration to ozone depletion in a fully coupled climate model (CCSM3, and 1° resolution). Thick black contour shows the marks the winter edge (15% concentration); thin black lines show areas where the change is statistically significant. Note that in this figure, red means a decrease in sea ice. Source: Bitz and Polvani, 2012, Figure 1d.

Annual mean response of sea ice concentration to ozone depletion in a fully coupled climate model (CCSM3, and 1° resolution). Thick black contour shows the marks the winter edge (15% concentration); thin black lines show areas where the change is statistically significant. Note that in this figure, red means a decrease in sea ice. Source: Bitz and Polvani, 2012, Figure 1d.

So what’s really going on? One idea is that changes in ocean stratification might be important. There has been a huge increase in the amount of fresh water getting into the Southern Ocean from melting glaciers, especially in the Amundsen Sea (see, e.g., the latest data from Sutterly et al., 2014). Fresh water forms a sort of buoyant lid on the ocean, limiting the ability of heat from the warmer water below to get to the sea ice and melt it. A study by Bintanja et al. (2013) showed that it was a least plausible that this explains the Antarctic sea ice change. A basic problem, though, is that the greatest discharge of meltwater is occurring in the Amundsen Sea, exactly where sea ice is declining, so while this probably is part of the story, I doubt it’s very dominant.

As it turns out, comparing observations with the results of model experiments like those of Bitz and Polvani (2012) is misleading. Most such experiments are equilibrium experiments: What’s done is to run a model under “preindustrial” conditions, and then to run it again with reduced ozone and increased CO2, and to look at the difference. This provide a measure of what will eventually happen (at least in the model) after many decades or centuries. But when you look at the transient response to changes in the circumpolar winds, as Marshall et al (2014) have done, it turns out that two important things happen. The winds tend to push the sea ice boundary northward, as we would have expected. But the winds push the surface ocean northward too, and cause a slow rise in the isopycnal surfaces (surfaces of constant density). This brings relatively warm deep water closer to the surface, eventually melting sea ice after a period of a few decades, countering the initial increase in sea ice. These results explain why equilibrium model calculations find sea ice decreasing in response to ozone forced changes in the circumpolar winds, and also why observations show the opposite. Not enough time has passed for the equilibrium response to be manifested. These results suggest that some time in the next few decades, there will reverse, and average sea ice will begin to decline.

Furthermore, there’s a whole lot more going on with the winds than just “increased westerlies”. In the areas where the big sea ice losses have occurred, the concept of “circumpolar westerlies” isn’t very relevant. A far more important measure of wind variability in the Amundsen and Bellingshausen Seas is the Amundsen Sea Low (ASL).5 The ASL describes the average location of storms systems the bring heat and moisture into West Antarctica. Changes in the ASL may occur for myriad reasons, but one big hammer that can make it ring is the propagation of atmospheric planetary wave arising out of the tropics, more-or-or less associated with ENSO (El Niño-Southern Oscillation) variability. It’s been clear for many years that ENSO variability play a significant role in sea ice variability in those regions, and recent work shows that this can explain the trends pretty well too (e.g. Yuan and Li, 2008; Stammerjohn et al., 2008). Not incidentally, the adjacent land areas of the Antarctic Peninsula and the West Antarctic Ice Sheet have warmed significantly over the last few decades (Steig et al, 2009; Orsi et al., 2013Bromwich et al, 2013), and those changes can also be attributed largely to tropical climate variability (Schneider and Steig, 2008; Ding et al., 2011; Schneider et al., 2012; Steig et al., 2013). The cause of temperature and sea ice change is the same: more warm air is being steered into West Antarctica, and the atmospheric flow tends to push sea ice against the continent, keeping it from expanding.

So, do we get the right answer if we take into account all of the wind changes that have occurred over the last few decades? The answer is yes. This is nicely illustrated in a study by Holland and Kwok (2012), who showed that wind, ice motion, and ice concentration changes match each other remarkably well. Where the wind has been increasingly northward, concentrations are increasing; where wind and ice motion changes are toward the continent, ice concentrations are decreasing. And this year, Holland et al. (2014), showed that when they drive an ocean and sea ice model with observed winds — not just increased westerlies, but the full range of wind changes, as calculated by the ECMWF (European Center for Medium Range Weather Forecasting) –- they correctly simulate the overall expansion of sea ice, and they also get the pattern of changes pretty much spot-on. To be sure, the authors note that not all the details are explained, and they highlight the possibly greater importance of thermodynamic consideration (i.e. ocean temperature/stratification) in some areas than in others.  Also, the period they study (1992-2010 only) is pretty short.  The results are nevertheless pretty compelling. Just like the observations, the calculations show large decreases in the Amundsen and Bellinghausen seas, but increases nearly everywhere else.7

Modeled (top) vs. observed (bottom) sea ice concentration changes.  Source: Holland et al., 2014.

Modeled (top) vs. observed (bottom) sea ice changes. Source: Holland et al. (2014).

Taken as a whole, these results show that there is no significant contradiction between our understanding of Antarctic sea ice and the observation that it is, in average, expanding. We can explain sea ice trends in the Antarctic rather well if we take into account the full range of changes in winds that have occurred. The average expansion of Antarctic sea ice was not anticipated, but it hardly represents any sort of existential threat to our fundamental understanding of the climate system as a whole. It’s merely an interesting scientific challenge.

Not incidentally, changing winds also have a lot to do with what’s been happening to the Antarctic ice sheet (meaning the land-based glaciers, distinct from the sea ice). I’ll have another post on that later this month, or in the New Year.

For a more in-depth version of this post, see Climate Change National Forum / Making Sense of Antarctic Sea Ice Changes.

108 Responses to “Clarity on Antarctic sea ice.”

  1. 101
    Hank Roberts says:

    > CO2 and global temperature are cointegrated.
    > There is no stronger evidence of a connection …

    Do you mean you aren’t aware of any other strong or stronger evidence?
    Or do you mean you think it’s not possible there could be any stronger evidence?

  2. 102
    MARodger says:

    t marvell @100.
    This is getting rather painful. If “Certainly you would never ask for a visual representation of a climate science computer model.” what then are these representations here? We may not be able to present a graphical representation of the full complexity of a modelled system, but all models have system inputs, system outputs & internal system variables, many of which will be of great interest and easily graphed.
    Consider your Granger analysis. From #94, you appear to be using a value y(t) = log(SIE[t])-log(SIE[t-1]) to create your monthly dependent variable time series. (I will ignore your mention of percentages @61. But do correct me if I have misunderstood.) This is a heavily seasonal series but the Granger formalism overcomes this by negating autoregression. This I assume is your “monthly dummies” for which you say you use seventeen values, presumably a(0) to a(17), leaving the residual[t] which you say contains the signal from Arctic SIE.
    The input variable, the Arctic SIE signal (I assume x(t) = log(SIE[t])-log(SIE[t-1])) you tell us is spread over nine months, each Arctic month weighted by the factors b[4] to b[12], with a final residual[T] to account for model ‘noise’.
    Now, if you cannot illustrate your finding graphically, provide a(0) to a(17) and b(4) to b(12) and let somebody else graph them for you!!!

  3. 103
    t marvell says:

    Re: 102 Rodger.
    You asked for a graph. I don’t think in terms of graphs, but I’ve tried. The best I can do is to graph the outcome of the Granger test. I prepared one in STATA, but could not figure out how to upload it. So I give you this crude one:

    —-Percent change in Antarctic ice extent ——
    lag .15 .10 .05 .00 -.05 -.10 -.15 -.20 -.25
    01 …………..
    02 ………………**
    03 ……..
    04 ………………*******************
    05 ………………******************
    06 ………………****************************
    07 ………………*********************
    08 ………………********************
    09 ………………***************
    10 ………………**************
    11 ………………**************
    12 ………………*********
    13 ………………******************
    14 ………………************
    15 ………………*************
    16 ……………..
    17 ………………************
    18 ………

    The data are percent change in antarctic ice following a 1% increase in arctic ice, from 1 to 18 months later. (Asterisks show lags with negative changes.) So six months after a 1% increase in arctic ice, seasonally adjusted, there is a .26% decrease in antarctic ice, seasonally adjusted (SE = .09), and visa versa. The total effect of a one percent increase in arctic ice (in any one month) is the sum of the above, or roughly a little more than a one percent decline over 18 months in antarctic. That is, antarctic ice goes up at about the same rate that arctic ice goes down.

    The more important statistical result is the cointegration of northern and southern ice. In my experience, time series are rarely cointegrated, so this is something special. I don’t know of any way to graph cointegration. For those familiar with the procedure, the T for the lagged error term is -9.2 in the ADF with monthly data, and -4.4 with year data (1979-2014).

    The simple correlation between the two is about -.5 (p=.0001). Again, the correlation is meaningful, in spite of the fact that the variables are I(1), because they are cointegrated.

  4. 104
    MARodger says:

    t marvell @103.
    We creep forwards (mainly) in getting to grips with your claim of a Granger-dependency between monthly Arctic & Antarctic SIE.
    Let me explain what is needed to illustrate this dependency. If data can be derived from Arctic SIE and used to generate the wobbles as seen in the data representing (or derived from) Antarctic SIE then we will be on our way. You present a graph of factors involved in that ‘generation’ which is the step in the right direction. Sadly this use of such factors is not what I interpret your words above as saying, but hey-ho.
    You add @103 that the impact of a 1% rise in Arctic SIE would result in “roughly a little more than a one percent decline over 18 months in antarctic. “ I certainly would not describe the 1.67% decline that your factors yield in such a way. I am hoping it is your description and not my scaling of your factors that is in error.
    Also there is a difference between the description “percentage change” you use @103 and the description “I log and difference the variables” that you use @94. (You must see they are different – a ‘difference’ subtracts while a ‘percentage’ divides.) This needs to be cleared up.
    Further, you appear to be adjusting for seasonal variation but not using the Granger method (who uses autoregression for such purposes). I am assuming this to be the case because you say “seasonally adjusted (SE = .09)” which suggests a rather simple process of adjustment. (I should say that a factor of 0.09 attaching to, I assume, June is not one I can spot in any of the various factors I have derived to address seasonal cycles.)

    All this allows me to say that using a variety of seasonal adjustments for wobble and/or spread on percentage changes and the factors you graph @103 does not yield any sign whatever of any dependency. Seasonal adjustment, if performed well, should greatly reduce the impact of any autoregressions that might still be lunking in your employment of Granger although your seasonal adjustment may still be incomplete. I post here temporarily a graph of the sort of stuff I’m finding. I think you will agree that there is not a lot of dependency evident in this graph.

  5. 105
    Hank Roberts says:

    Nature Climate Change | Letter

    Evolution of the Southern Annular Mode during the past millennium

    Nerilie J. Abram, Robert Mulvaney, Françoise Vimeux, Steven J. Phipps, John Turner & Matthew H. England

    Nature Climate Change 4, 564–569(2014)

    Published online 11 May 2014

  6. 106
    Hank Roberts says:

    I’m reminded that some people are young enough to expect to see the real troubles arriving in their lifetimes.

    Sorry, kids. We had a good time, and left you to pay.

    Terrifying Paper #1, by Rignot et al. A decent chunk of West Antarctica, called the Amundsen Sea Sector, is melting particularly quickly….
    Terrifying Paper #2, by Mengel & Levermann, examines the Wilkes Basin Sector of East Antarctica.

    Links to the papers at the original post.

  7. 107
    t marvell says:

    #104 Rodger. Thanks for your interest.
    1) the exact “elasticity” is -1.54; it is very inexact.
    2) percent changes are very similar to differences of logs (times 100). “Percent change” is more intuitive, but the actual regression is run with differences of logs.
    3) seasonal correction is mainly by entering dummies for each month. I also include lagged DVs, which are necessary with the Granger test; they do some seasonal correction, but not enough. I also did the Granger with 12-month differencing, and got similar results, but with the impact sooner.
    4) I gather that your plot uses current year percent change in the north and current year percent change in the south. There seems to be no connection. I also get no connection for the current year (The impact could hardly be that quick).

    It would be hard to plot the Granger results because of the seasonal effects and because the apparent impact is spread out over about 10 years. You might see some connection with the Antarctic ice lagged 6 mo, the strongest connection.

    The granger test with differenced data shows a sizeable negative relationship between north (64-90 degrees) and south temperatures one and two years later (year anomaly data, 1904-2013). The relationship in levels, of course, is positive, which means a long-term relationship in the opposite direction from the short-term relationship. Something odd is going on. The north temperatures rose about twice as much as the south temperatures.

    In all, these statistical associations don’t mean anything without a credible mechanism that might drive them. The best I can think of is changes in the THC deep currents in the Atlantic. That is suggested by the fact that Antarctic waters are cooling down where the THC comes in (at zero degrees), but increasing elsewhere. The THC theory depends on the deep currents either cooling or speeding up. As far as I can tell, very little is known about this, although buoys are being placed there now.

  8. 108
    MARodger says:

    t marvell @107.
    I may if I get the urge revisit this to examine changes in Ant&Arc SIE year-to-year, what I interpret your “12-month differencing” to mean. But otherwise I call a halt to this.
    Mathematical analysis requires things being ‘exact’. “Very similiar” is not good enough.
    I should also point out that the graph I linked @104 gave a plot of seasonally adjusted percentage change of monthly SIE, Antarctic SIE against a factorised Arctic SIE, this latter calculated using the lags from your graph @103. And by ‘percentage change’ I mean (X(i)-X(i-1)]/X(i-1). This is different from your description of it @107 4).