RealClimate logo

Arctic Sea Ice decline in the 21st Century

Filed under: — group @ 12 January 2007 - (Français)

Guest Commentary by Cecilia Bitz, University of Washington

Last month a paper I co-authored received considerable media attention. Headlines read “Experts warn North Pole will be ‘ice free’ by 2040”, “The Big Melt: Loss of Sea Ice Snowballs“, and “Arctic Clear for Summer Sailing by 2040: Models Predict Rapid Decline of Sea Ice”. The story also reached NPR, BBC, CBC, the Discovery channel, and Fox News, among others. Dr. Marika Holland, the first author of the paper, was inundated with media attention. About a dozen journalists contacted me too. I was impressed by the questions they posed — questions that probably reflect what the public most wants to know. However, after giving lengthy interviews, I would read the resulting article and see my explanations boiled down to a few lines. In this essay, I’d like to explain the science in the paper and give my answers to the most often asked questions.

In our paper (with co-author Bruno Tremblay), we examined the September Arctic sea ice cover in the 20th and 21st centuries in climate models, and found occasional decades of very rapid retreat. The most extreme case was a decrease from 6 to 2 million square kilometers in a decade (see Fig 1). This is about 4 times faster than the decline that has been observed in the past decade.

Figure 1: (a) Northern Hemisphere sea ice extent in September from one integration of the Community Climate System Model version 3 (CCSM3) with observations from the satellite era shown in red. The light blue line is a 5-yr running mean. The three lower panels show the September ice concentration (ice floes are separated by open water) in three select decades.

It is common practice to run climate models multiple times with slight variations to the initial conditions. Because the system is chaotic, the natural variability in each run is random and uncorrelated from one run to the next. When an ensemble of runs is averaged, the natural variability is reduced in the ensemble mean, and it is easier to detect a significant trend.

An ensemble of runs offers an opportunity to evaluate rare events too, such as extreme sea ice decay. We were in search of evidence for “tipping points“, which several authors have speculated might exist in sea ice. RealClimate places sea ice in the category of systems with “known unknowns” with regard to tipping points. This means we know there are thresholds involving sea ice (e.g., it can cease to exist), but we don’t know when, or if, the climate will arrive at one.

Only one of seven ensemble members had an event as extreme as quoted above, and it resulted in near ice-free conditions for September by 2040 (see Fig 1d). (The sea ice grows back at least for some portion of winter for the duration of the 21st century.) However, every ensemble member had an event 5 years or longer at some time in the 21st century when the sea ice retreat was about 3 times faster than the observed retreat since 2001 (see Fig 2). These ensemble members took about 5–10 years longer to become nearly ice-free in September than the most extreme case.

As illustrated in Fig 1, the sea ice retreat accelerates during the 21st century as the ice decays and more sunlight is absorbed by the ocean (the positive ice-albedo feedback). Increasing ocean heat transport under the sea ice adds to the melt back. The retreat appears abrupt when natural variability in the ocean heat transport into the Arctic Ocean is anomalously high. We did not find clear evidence of a threshold, which can be difficult to identify given the variability and complexity of the climate system. Therefore we can neither verify or rule-out the existence of a tipping point. Regardless, the rapid declines seen in our runs are a serious concern.

Figure 2: Northern Hemisphere sea ice extent in September for all seven integration of the CCSM3 with observations from satellite era shown in black.

Most common questions asked by journalists

1) How does our model compare with the trend in the observed record?

The trends in the seven ensemble members for 1979-2006 span the trend in the observations: Some members retreat a little faster and some a little slower, as expected from the random natural variability in the runs (see Fig 2). The model also reproduces the mean and variance of the observations with good fidelity.

2) Other scientists are predicting an ice-free Arctic in September by the year 2060-2080, why is this model predicting it 20-40 years sooner?

First consider estimates based on extrapolation from the observational record. I’ve heard these numbers quoted in the media, but I have not seen a reference to a scientific paper that discusses the analysis in any detail. Figures 2 and 3 illustrate the danger of making an estimate of the future from the observational period. The future trend is not linear, the observational record is too short and the ice-free time is too far in the future to trust extrapolation. If one carries out such an exercise anyway, extrapolation from a linear fit to 1979–2006 gives a zero intersect (indicating the first ice-free year in the future) at about 2110 (see Fig 3). If instead one uses just the last decade, the extrapolation gives 2060. Both estimates are questionable, and so instead we turn to climate models.

Figure 3: Extrapolating into the future from the observational record.

3) Is sea ice in our model retreating faster than in other models?

Figure 4 shows September ice retreat in 16 models that were archived for the IPCC AR4. The most extreme predictions are from models that have too much or too little sea ice extent compared to observations, so it is important for a model to produce the correct sea ice coverage in the past. Some of the spread is expected from natural variability, but much depends on differing model sensitivity relating to the representation of sea ice, heat transport by the ocean, and cloud cover. It is not possible to identify the most accurate model prediction, although I think it is safe to rule out some of the outliers owing to their poor match to the observations.

About half of the models become ice-free in September during the 21st century. I included one ensemble member from our model, CCSM3, which is in the middle of the pack until about 2020. Our model run retreats faster than most after about 2020, but it isn’t radically different.

There is considerable uncertainty in future model projections, and Figs 2 and 4 illustrate why it would be better not to focus too much on the year 2040, which to our dismay was highly publicized. The more important message from models is that all but a few outliers predict enourmous sea ice retreat this century. At least a few respectable models predict a nearly ice-free Arctic by midcentury, with a retreat that may be punctuated by rapid events.

Figure 4: Northern Hemisphere sea ice extent in September from model integrations submitted to the IPCC AR4 with observations from satellite era shown in black.

4) Is it too late to save the sea ice?

The future emissions scenario discussed here is one that assumes modest increases in emissions. If humans can reduce the rise in emissions compared to this, then sea ice retreat would be slower and rapid events would be rarer, according to the IPCC AR4 models.

5) Have we crossed a tipping point?

I don’t think we have yet. If we fix the greenhouse gas and aerosol levels at year 2000 values and run the model into the 21st century, the sea ice retreats for only another decade or two and then levels off (some of the ensemble members even recover a little bit). So according to our model, the sea ice does not appear to have passed a threshold yet. We have not done an exhaustive study of any years beyond today, so unfortunately we cannot say with certainty that no tipping points exist. The bottom-line: The retreat can be surprisingly rapid even without clear evidence of a tipping point.

I thank Dr. Holland for valuable suggestions to improve this post and providing Fig 1. I thank Ian Eisenman for computing ice extent from the IPCC AR4 models shown in Fig 4. I look forward to reading your comments and questions.

127 Responses to “Arctic Sea Ice decline in the 21st Century”

  1. 101
    Jim Crabtree says:


    One gallon of gasoline weights about 6.2 lbs. It contains about 5.4 lbs of carbon (gasoline is approximately 87% carbon). This one gallon of gasoline produces 19.8 lbs of CO2. This means the average American car (assuming an average weight of 3 500 lbs and 20 mpg) produces its own weight in carbon dioxide in about 12 000 miles. If one takes the carbon cost of refining, transport of gasoline, then the 12 000 miles drops to about 7 500 miles. That is why we have a Prius for our primary vehicle. I also have a small pickup truck (better gas mileage) since I need it for carrying large stuff and it generally gets less than 3 000 miles per year on it.

    When I taught at the University of South Carolina back in the 1980’s, I used to bicycle 9 days out of 10 giving me 10 miles of exercise each day. I actually got my car insurance reduced because of this!

    Jim Crabtree

  2. 102

    Re “Perhaps not sharply focused on this post, but a question I’ve wanted to ask for some time. When ice melts, it takes up heat (80 times as much per gram as required to raise water temperature 1 degree C, if I remember right), and this must cause a cooling in ocean and/or atmosphere. Is this large enough to be a significant negative feedback on a global or regional scale?”

    It doesn’t cause a cooling, just a change of state. Temperature will stay at 0 C until the whole ice object is melted, at which point it will start to rise again.

  3. 103
    Steve Bloom says:

    Re #92: The reference was to fixing atmospheric levels at the values they had in 2000. Obviously those levels are already history and are far lower than any realistic peak given that emissions are continuing, but the point is that the ice is not irretrievably lost if there are sharp reductions now.

    The question that is begged by all of this (but was outside the scope of the paper) is whether some sort of tipping point may be involved with a loss of ice that includes the entire summer high-insolation season (and thus allows the Arctic Ocean to start picking up some serious heat in contrast to a relatively brief ice-free period centered on the relatively low-insolation equinox) and, tipping point or not, what effect a much warmer Arctic Ocean would have on the Greenland ice sheet.

  4. 104
    Ike Solem says:


    Burn rubber, not gasoline.

  5. 105
    Cecilia Bitz says:

    Re #52 GFDL models are missing because they were not yet archived when
    the data were downloaded. They are archived now, but we haven’t gone
    back to get them. It is a lot of work to gather and process the data.

    Re#55 & #84 I said nothing about Rothrock failing to adjust for time
    of year. He’s my colleague, I know he did. I have used his data in one
    of my own papers. You are assuming that the climate is linear, and it
    is not. Dr. Muller – you might also be interested in my paper about
    ice thinning
    from a theoretical and modeling point of view.

    Re #60 & #68 I agree it is difficult to melt the very thick ridged ice
    that piles up each winter against Greenland and Ellesmere Is. This is
    why we considered the last panel in Fig 1 to be essentially ice-free,
    even though there is still some stubborn ice that remains.

    Re#64 Isn’t there enough heat beneath the sea ice to melt it? Yes
    there is enough heat in the Atlantic Layer, which is roughly 500-1000
    m deep. It is isolated from the sea ice by an intevening layer called
    the cold-halocline, which is stabilized by high river runoff and
    Pacific water inflow through the Bering Stait. A paper
    analyzed recent variations in the cold-halocline.
    These layers
    are reproduce in our model and they vary dynamically with the climate.

    Re#71 It is not reasonable to ask a model to predict THE 2006
    anomaly. Look at Fig 2 and you will see that the different ensemble
    members from the model span a large range in just one year. The model
    cannot be expected to predict the spatial pattern of the anomalies in
    individual years, which depend quite a bit on natural variability. As
    stated in the essay, our ensemble of runs agree with the recent
    Northern Hemisphere average trends from the satellite era. It is also
    worth mentioning that if we use the model to estimate natural
    variability (we have multicentury runs at “present day” forcing levels
    from our model while we only have 28 years from observations), then we
    can “detect” a significant trend in the 28 years from
    observations. However, the trend in the last decade from observations
    is not significant.

    Re #79 Robert Grumbine estimates if all the sea ice on Earth melts it
    will raise sea level by 4mm. See his report
    It is pretty trivial compared to sea leve rise expected from
    thermal expansion of land ice runoff (see response to #70).

    Re #92 The greenhouse gas _level_ is fixed, not the emissions. It is
    a senistivity experiment, not a “simulation”. You are correct that
    fixing emissions would still result in rising levels.

    Re #93 Thanks for the analog. I’ll have to try it. The chips are just
    a bonus.

    Re #95 Heat is “absorbed” by melting sea ice and warming the ocean,
    among other things. The ocean, however, has taken up about 7000 times
    more heat than what it takes to melt 40% of the Arctic sea ice,
    according to a recent study by
    Levitus and colleagues
    Ocean heat uptake below the surface mixed
    layer does slow the rate of surface warming. I wouldn’t call the heat
    used to melt sea ice a negative feedback (as noted by #102), but in
    any case, it is a pretty trivial amount of heat compared to uptake by
    the ocean.

    Re #100 Our model overestimates cloud fraction in winter and clouds in
    summer are to thick, so to speak. Every model has biases and always
    will. We are working on reducing this bias however.

    I didn’t want to rag on the media too much here. Overall I was
    impressed by my exchange with reporters. I was dismayed by a couple of
    articles that presented very little detail, especially those that
    appeared to cut-and-paste excerpts from other articles. The worst
    example of these tended only to report “scientists predict ice-free
    summer by 2040” with no explanation of the context AND two articles
    misspelled my name (this is my fate having a last name one letter
    removed from a bad beer). The more important story for me is about
    the rapid events, which ultimately lead to the 2040 prediction.

  6. 106
    Timothy says:

    This is a question for Stefan, with reference to his recent Science article. In this you state that: “there are as yet no published physically based projections of ice loss from glaciers and ice caps fringing Greenland and Antarctica.

    I was wondering whether the use of “physically based” was intended to exclude the study by Gregory and Huybrechts []. Is/was there work more simplified, relying on thermal arguments rather than the physical dynamics of the ice cap?

    I had thought that Gregory [and collaborators of] had done some work forcing a specific ice sheet model. Perhaps that isn’t published?

  7. 107
    Jim Crabtree says:

    Dr. Bitz:

    I have in front of me an article about the results of the meeting of the Canadian Arctic Shelf Exchange Study that took place last year in Winnipeg. Dr. David Barber estimates that the arctic could be ice free in about 15 years.

    I played a little with modeling back in the 1970 for a short while before I decided on a different research area. I look at models as a tool that goes through two phases. Phase 1: Does the model show that I understand the problem? Phase 2: Once the model shows I understand the problem, then I can use it for research into the unknown.

    I really question whether the current models are to the state where they can be used for prediction when there are so many “butterflies” out there and some of them are probably big “butterflies”. (For example – models that treated the Greenland ice cap as a solid mass).

    As I have taught in computer science classes (I am now retired), a model is:
    1) no better than one understands the environment they are trying to model.
    2) no better than the data they have available.
    3) no better than the mathematical relations required.
    4) no better than the code that is being written (how many bugs are there that you don’t know about – How do you know when there is a bug in a model????)
    5) no better than how well truncation or roundoff is handled (after many iterations, this could become a HUGE problem).
    and 6) all of the above combined.

    I have watched predictions of an arctic summer being ice free move from 2100 to 60 years away, to 40 years away, and from the above mentioned article to 15 years away. I personally think 15 years is much closer to being reality having watched the arctic ice sheet and looked at available data over the last 7 or so years.

    Are you familiar with Dr. David Barber’s (University of Manitoba) work? If so, what is your opinion of it?

    I don’t mean to sound too critical of your work. But being I also design and develop software, I am extremely critical of my own work and I don’t release software until I am satisfied with it. My testing is to try to find the limits of my software. I learned that from working for an engineering company that had a sign above the door of the group that did Numerical Control that read something like this: “70% will pass you in school, 99.99% will fail you here”. This was back in around 1966 and I was writing NC processors that had to work with tolerances of 0.000 1 inches. There you learn about roundoff and truncation real fast!

    Anyway, keep finding those butterflies and I hope to see more papers in the near future.

    Jim Crabtree

  8. 108
    Isaac Held says:


    Sorry to come to this thread so late, but I am curious about your abrupt decreases in ice. If you take the ensemble mean of all of your runs and then look at the departures from this mean, are these skewed in the sense that the negative excursions are larger than the positive ones? Or is this abruptness possibly just an impression from the superposition of a smooth ensemble mean decrease and symmetric noise?

  9. 109

    I’ve recently run across some weird peer-reviewed stuff from a guy named Goode who is apparently working from a NASA grant — he maintains the Earth’s bolometric Bond albedo decreased measurably from 1994 to 2002, from 0.310 to 0.295 (with error bars), which if true would create a measurable climate forcing. He’s careful to note that it would take time to work its way through the climate system. Cites (and please forgive the weird characters from my cutting and pasting):

    P. R. GOODE, E. PALL_E, V. YURCHYSHYN, J. QIU, J. HICKEY, P. MONTA~N _ES RODRIGUEZ, M.-C. CHU, E. KOLBE, C.T. BROWN AND S.E. KOONIN 2002. “Sunshine, Earthshine and Climate Change: II. Solar Origins of Variations in the Earth’s ALbedo.” J. Korean Astronomical Soc. 35:1-7.

    E. Palle, P. Monta~nes Rodriguez, P.R. Goode, J. Qiu, V. Yurchyshyn, J. Hickey, M.-C. Chu, E. Kolbe, C.T. Brown, S.E. Koonin 2004. “The Earthshine Project: update on photometric and spectroscopic measurements”
    Advances in Space Research, preprint.

    Any comments?

  10. 110
    Eric Swanson says:

    Re: #109

    There’s been considerable discussion in the literature about “earthshine”. A google search returned many hits, including a link to your second reference at:

    Other papers appeared in the JGR, at these links:

    There was also some discussion also in SCIENCE. Another piece in the puzzle, I’m sorry to say.

  11. 111
    Sashka says:

    Looking at Fig. 2, it seems to me that (apart from one year in mid-90s) the red line representing observations sticks to the low side of the envelope generated by the ensemble of model runs. Does it not suggest a bias in the model?

  12. 112
    Cecilia Bitz says:

    Re #108 Hi Isaac! No significant skewness in the A1B scenario. Another reason to dismiss the tipping point speculation. Thanks for the suggestion.

  13. 113
    Joseph O'Sullivan says:

    Thanks to Dr Bitz for participating. I saw the news stories and did not have the chance to track down the original paper. It was easier just to read RC to get the full story!

    Oh, one more thing, “having a last name one letter removed from a bad beer” do you mean schlitz? I remember TV commercials for schlitz that had a taste test with bigger name beers. Schlitz lost but not by much, and they ended the commercial by saying hey we’re not that bad ;)

  14. 114
    Mark Ritzenhein says:

    Re 107

    Jim Crabtree, Thank you for your short list of errors. I am glad to know that you hold yourself to a tight distrust of the natural inaccuracy of models. I, too, believe that the Arctice ice sheet will melt suddenly and rapidly.

    A general question: did climate models predict the sudden collaps of the Larsen ice shelf in the Antarctic? If not, why not? Second, does it really matter if the polar ice sheet melts in seventeen or one hundred years? This is relatively unimportant to the fact of its loss.

  15. 115
    Hank Roberts says:

    For those still wondering:

  16. 116
    Nigel Williams says:

    So, in relation to tipping points of knowledge, (ie we know enough to know it will happen.. etc) if we see 1 metre sea rise by 2100, that implies we see at least 12mm rise per year from now to then. So if we see 12mm per year, is that a fair enough portent that we will see 1m by 2100, and is it then fair to say that if we get 1m rise by some future date, then we will get the rest as well? It seems unlikely that sea level rise will stop until it is done, doesnt it?

    And from that, if one was working on a plan for a new (or relocated) town, how far above present day sea level would it be prudent to build that town, if it was to have a life of 500 years or so? 10m? 20m? ??

  17. 117
    llewelly says:

    Nigel, I suspect the majority of buildings being built today are unlikely to remain livable for 500 years.

  18. 118
    David B. Benson says:

    Nigel — Work out the sea stand rise if all of Greenland and all of West Antarctica should melt. Establish your new town some meters above that.

    While it is true that few buildings last 500 years, some do and I have certainly seen older ones than that in Europe. But the town itself has a value greater than just the structures and infrastructure in it. By all means plan ahead for being safely elevated.

  19. 119
    S. Molnar says:

    Re #117: Yes, but the life of a town can reasonably be expected to exceed the life of individual structures in the town.

  20. 120
    Nigel Williams says:

    Ah Llewelly, it all depends whether or not you are perched on a crag and your house is made of slate, no? :)

    Its not the buildings, its the infrastructure and land tenure that is the resource thats hard to relocate. For sure, permits for building require certainty of security of land and life of structures of 50 years. But if one is doing some long term strategic planning for an urban area, to plan for a high tide 50 years hence seems rather short sighted.

  21. 121
    David B. Benson says:

    Nigel — An elevation of 30 m ought to suffice…

  22. 122
    Nigel Williams says:

    Thank you gentlemen! Ah, the simple pleasures to be found in the appreciation and acceptance of a simple fact. 30m.

    The 30m contour puts my entire 150-year old city and its surrounding small towns below high tide – about 400,000 people. It also swamps most of the high quality horticultural soils, and finds us stranded on a relatively infertile plain of galcial outwash gravels, and separated from our only possible port for 150km either way along the coast too. Oh darn!

    We are in the process of preparing a new Urban Development Strategy for the city and environs to lead us sensibly through the next few hundred years. Looks like we shouldnt cling too fondly to the home we know, eh! I wonder – will our politicians heed this call? Mmmm..
    Thanks again for the info.

  23. 123
    Julian Flood says:

    Re 109:

    It’s a shame we don’t have albedo measurements going back 150 years, particularly ones which distinguish between high and low level cloud. The climate effects of low level oceanic stratocumulus are huge — simple wind-powered trimarans evaporating seawater to generate hygroscopic nuclei have been proposed by Salter, Latham et al in order to buy time to correct our GW errors. A few hundred million dollars per year could offset all the calculated AGW for that year and, unlike the aerosol suggestions I’ve seen this week (was it NASA?), the trimarans can be switched off overnight if we find we have made a miscalculation.

    I’d like to see albedo measurements from the waters just off China and India. I predict a fall over the last thirty or forty years with the last ten or twenty showing a marked drop as surfactant and oil spill pollution transported down the rivers and entrained in the local surface currents suppresses the production of LL strato-cu.

    Do the models incorporate changes in the ocean surface due to pollution?


  24. 124
    David B. Benson says:

    Nigel — To be precise, I have assumed 7m sea stand rise due to the melting of the entire Greenland ice sheet and 20m sea stand rise due to the melting of the entire West Antarctic ice sheet. The first number is quite solid and the second is conservatively high. If correct, this suggests the sea stand might rise as much as 27m, assuming nothing is done to amerliorate AGW.

  25. 125
    simon willace says:

    Everyone seems to agree that the ice is melting at an accelerated rate. Globally all frozen water is melting which must dilute the salts in the surface body of the sea efectivly weakening surface currents.

    In the arctic the present ice cover marks the exact position where the Gulf Stream descends into the deep before it makes its way south. As this ice sheet may soon be gone, will the slower Gulf Stream just carry on over the top dissipate then disappear?

  26. 126

    If you want ice for the arctic, please take it from Colorado. We have had record cold and snow here.

    “Last time city was white for so long? 1914
    Long, cold spell dredges up images of long gone storms

    If last month felt like the coldest in more than 16 years in Fort Collins, that’s because it was.

    January 2007 was the sixth-coldest in the past 50 years. It averaged 23.7 degrees last month, more than 14 degrees colder than 12 months earlier and 5 degrees colder than usual.

    And that 11-inch blanket of snow covering much of the city is probably unmatched since the winter of 1913-14, said State Climatologist Nolan Doesken. “

  27. 127
    Steven Soleri says:


    Cecilia, you do not have to go to ice field data collection to determine there have been recent and rapid deterioration of ice fields and glaciers in the artic. Read the letters of gold rush miners thankful that the temperatures have warmed over what the native peoples recalled just 20 year prior. Another warming period is well reported by locals in the 30’s. And you want to run an alarm on your data. A researcher that overlooks the obvious–there are first hand accounts now available of changing artic temperatures. Cecilia reports that we can turn back the CO2 clock and save the earth??? If CO2 is the problem then the Chinese and India have our perverbial behind in a vice. And I can guarantee you they have no interest in a CO2 or feedback hypothesis.