Summer Ice

RealClimate has an interesting post on arctic sea ice from Dirk Notz at the Max Planck Institute, which discusses (among other things) the obsession with the summer minimum extent. He adroitly points out several interesting points. One is that weather conditions strongly affect the summer minimum, so it’s really too early to know whether this summer’s minimum will be a record-breaker or just ho-hum. Another is that the decline in arctic sea ice thickness is where the real story lies, but for the moment thickness is not nearly so well-observed as extent so it’s extent that captures the most attention.

In spite of the uncertainties (or perhaps because of them), it’s still fun to predict what the future will bring. Everybody seems to want to get a piece of the action; the June outlook from SEARCH offers predictions from a number of researchers:

We can’t let professional arctic researchers have all the fun! After all, Shemp has his forecast for the summer minimum in arctic sea ice extent. Here’s mine.

NSIDC has posted their monthly figures through June, and of course we have access to daily data (for the northern hemisphere at least) from JAXA, although those data don’t begin until 2002. In my opinion, it’s important to incorporate the long-term trend, which just isn’t as well defined by the briefer JAXA data. I also note that the September average from NSIDC is a pretty good match to the recent summer minima from JAXA:

There are notable differences, but those are mainly in winter. I think (but I’m not sure, perhaps some knowledgeable reader can enlighten) that the JAXA group doesn’t count certain southern areas that only freeze over in winter, so there’s a notable difference in winter maxima but little in summer minima. In any case, the NSIDC September average is a pretty good proxy for the JAXA summer minimum — and it goes back over 30 years, not just 8.

What has the September average done for the last three decades? This:

The decline is obvious. What may not be obvious (without the regression line) but is still clear, is that the decline has accelerated. That’s why I’ve fit a quadratic trend line. So here’s a simple projection: I’ve extrapolated the trend line to this year. Which gives a predicted September extent of 4.78 million km^2 (plotted as the red dot on the graph). I’ll call that my forecast for the summer minimum.

We can also use the scatter in the residuals from the trend-line fit to estimate the uncertainty in that prediction. Make that 4.78 +/- 0.95 million km^2 (plotted as the “error bar” on the red dot in the graph).

That puts my forecast for the summer minimum extent at somewhere between 3.83 and 5.73 million km^2. My low end is lower than any of the pros prognosticated (except for Wilson, that Maverick!), but my high end is about equal to the most conservative professional opinions.

If the range I’ve outlined seems kinda uncertain, well, that’s the way forecasts are. As Dirk Notz emphasizes in his RC article, this summer’s weather will have a lot to do with the minimum, and that’s something we really can’t predict. The uncertainty is in the system.

Nonetheless, I expect this summer’s extent to be less than last year’s. I’ll give that an 89% chance of happening. As for who’s prediction will be closest to reality — time will tell.

38 responses to “Summer Ice

  1. If you think there is an 89% chance of lower than last year then you want InTrade.

  2. FYI, there are two non-Pro predictions in that chart. Wilson is obviously one of them. The submission process was open to anyone. I’m rather surprised there weren’t more non-Pros.

  3. Rattus Norvegicus

    I would have like to have seen Goddard’s!

  4. I used to watch the Navy’s page but it hasn’t been updated since 2003. The multitude of factors going into modeling were sketched out there.

  5. Goddard first said “more than 2006” (nearly 6 million km^2), then said “5.5” some time later. When people tried to hold him to his original “more than 2006” figure, he said “5.5 was my first *numerical* estimate”.

    Whatta slimeball …

    “As Dirk Notz emphasizes in his RC article, this summer’s weather will have a lot to do with the minimum, and that’s something we really can’t predict. The uncertainty is in the system.”

    And, as he points out, has increased in recent years due to the ice being, on average, so much easier, therefore much more prone to being compacted or spread out by wind conditions.

    We’ve seen that over the last couple of weeks, with the Beaufort Gyre stalling, then reversing (and ice extent decreasing at a lower rate than seen in the JAXA data in any previous year as a result).

    So the best hope for a low minimum this year would be for spread out ice continuing to be melted, some advection through the Fram and Nares straights, and then, at the very end of the melt season …

    Favorable winds that compact the ice so there’s lots of 80%, 90% etc concentration and very little 20%, 25% etc concentration …

  6. carrot eater

    Have you a link for Goddard’s original prediction? Archive it as well.

  7. Have you a link for Goddard’s original prediction? Archive it as well.

    Ugh, I’d have to go trolling through the WUWT archives. I wandered over there a couple of times during his PIP pixel counting and similarly braindead posts due to links here and elsewhere saying stuff like “ya gotta see this to believe it!!!”

    I don’t think it’s particularly important, the prediction game is, after all, just a game. As Dirk says in his RC post, the thin ice is so susceptible to being pushed around that predicting the minimum is a real crap-shoot.

  8. Goddard wants to be able to shout “Recovery!” so that is why he was predicting a return to 2006 extent.

    At the moment he is preparing to shout if it is anything above last year.

  9. I think I’ve got his original prediction.

    Steve Goddard, 2010-06-02, on this year’s Arctic sea ice minimum:

    Conclusion : Should we expect a nice recovery this summer due to the thicker ice? You bet ya. Even if all the ice less than 2.5 metres thick melted this summer, we would still see a record high minimum in the DMI charts.

    • Does DMI produce pan-arctic data? I can’t find it. Is Goddard being tricky again? If the Nares ice dam isn’t there, then we can expect more ice busy exiting the Arctic, but temporarily showing up on DMI’s Greenland charts.

      You know this is the kind of slimy bait and switch that Goddard lives for.

  10. Who is Charles Wilson? His justification for his guess* is a bit patchy, and seems to completely ignore the importance of the dipole anomaly. Plus, his discussion READS like the ravings of a CRAZY person. =DOES NOT inspire Confidence.

    I don’t think wild guesses are particularly productive, and they run the risk of being seized on by deniers, saying “but you said it would all melt this year!” (but they do that anyway, bless their confuzzled little heads).

    * It’s not quantitative in its derivation, so it’s a guess. Calling it “statistical and heuristic” is just being nice. His “statistical” part seems to be based on faulty mathematics. To be honest, I have great difficulty following it. How can you use ice volume change to predict extent?

    • carrot eater

      Wilson is a curiosity. He posts comments at WUWT, so if you want to piece together more, you can look for him there.

  11. I think I’ve got his original prediction.

    Good find – DMI only goes back to 2006, so his “DMI record” means a higher minimum extent than in 2006.

    • Hmm, that sure puts a different face on his casual toss-off that “2006 had the highest minimum (and smallest maximum) in the DMI record”, doesn’t it?

      If I wanted to play games, I’d bug him for the code used to generate the histograms in his Fig. 1. Due diligence, after all.

    • Hmm, that sure puts a different face on his casual toss-off that “2006 had the highest minimum (and smallest maximum) in the DMI record”, doesn’t it?

      Well, yes. Imagine the hours spent by SG figuring out which is the “most accurate” dataset, before settling on one that starts in 2006 :)

  12. Tony O'Brien

    Eyeballing the satellite pictures of the ice cap, so much looks chopped up. But it would appear that is now normal for this time of year.

    I have no idea what the minimum will look like, but Goddard’s 5.5 to 6 million sq km’s is way too high.

    I suggest that summer sea ice is now very vulnerable to weather events.

  13. Tamino

    You show the arctic volume sea-ice anomaly.
    It’s not thickness anomaly.
    You can’t compute easily thickness anomaly from area and volume anomalies.

    • How to reply? I assume you are talking about Tamino’s link to PIOMAS. Actually, you can compute thickness anomaly rather easily from area and volume. This tells you nothing about the distribution – no clue whether the thinning is uniform. But the average? That’s easy.

  14. didactylos

    YES I’m talking about the anomaly volume of arctic ice by PIOMAS.
    The volume anomaly is v’-v = t’*a’-t*a
    where t’ and t = thickness and a’ and a = areas
    The thickness anomaly is ( t’-t) and the area anomaly is (a’-a).
    You know (v’-v) and (a’-a), but how do you compute (t’-t), the thickness anomaly, with these relations?
    To do this you must have some absolute value.
    Undoubtfully , these values exist and I’m not doubting that the thickness is decreasing , but the best should be to show the anomaly thickness and not the anomaly volume.

    • We know more than that. We start with absolute values for v, v’, a and a’. That’s the data used to calculate the anomalies. We also know v = a*t and v’ = a’*t’

      v’ and a’ are the average over the baseline period used to calculate the anomaly. These figures are usually published in the same place as the anomaly data – such as the PIOMAS baseline

      I shouldn’t get sucked into these debates. No doubt all you meant was “Wouldn’t it be nice if Tamino had linked to submarine thickness data rather than PIOMAS”.

      • For the last time, it’s an evidence that, with the absolute values of thickness, we know the trend of thickness.
        But not, only, with the volume anomaly curve.
        It isn’t too difficult to understand it.
        So, if Tamino wants to be credible, he must be rigourous.

  15. Horatio Algeranon

    Given that most of the summer melt has already occurred this year, isn’t there actually more information available for an estimate of this year’s minimum than is contained in the trend line shown above?

    Assuming that what has happened in the “July 26 through minimum” time period of the last 8 years (a mean loss of about 2.2 million km^2 with a standard deviation of about 0.35 million km^2) is an indication of what is likely to happen over the remaining melt period this year*, one gets a minimum of about 5.0 million km^2 (obtained by simply subtracting the mean loss for the remaining summer melt period from this year’s extent on July 26.)

    The 4.78 “prediction” above would be very close to 5.0 million km^2 (within one std deviation), but the “low end” above (3.83 million km^2) would be more than 3 std deviations below that 5.0 value (using the 0.35 std deviation)

    Finding this year’s minimum by simply subtracting the mean change for the Aug/Sept melt period from the extent in late July is admittedly crude and 0ne could fit a polynomial to this year’s curve to estimate the minimum, but it’s not clear that the result would be a real improvement, particularly since it would not account for “weather related behavior” in Aug/Sept.

    The JAXA graphs for some of the previous years (2002-2009) seem to show “unexpected” behavior (eg, discontinuities in the slope) in the Aug through minimum period, undoubtedly related to weather.

  16. Horatio: It was to answer that exact question that I downloaded R yesterday, and started playing around. However, I have little clue what I am doing.

    The bad news for our hypothesis is that the summer months show the most irregular variation in extent (as you suspected). I’m sure the forecast can be tightened a little, but I don’t know how. Hopefully the pros can show us how it is done.

  17. Tamino,

    Could you kindly extend and post your quadratic fit graph to the point where ice extent is zero (and labeling that year on the X axis)? If you have the time, add a red dot for the projected ice extent for each year from now until then.

    I’m just curious, as to how it will look, when this particular pattern implies an end to summer arctic ice, and as something to reference in future years.

    My own personal expectation matches that of at least one arctic scientist (I can’t find the quote, dang it) that the actual end will come before your projected quadratic fit/x-axis intersection, and that the final “holy cow, it’s gone already” melt will come rapidly and unexpectedly.

    [Response: I wouldn’t trust such an extrapolation. Such a purely statistical model can only be valuable during the time span that we expect “statistical persistence” to be stronger than the physical processes which will alter the future trend. Statistics can only take you so far — to see further into the future requires physics.]

  18. sphaerica: about 2030. Comparable to other extreme projections, but equally (un)reliable.

  19. A few years ago I registered to challenge sceptics with real cash bets. Arctic ice is a good race to bet on. I had hoped to find a bookmaker to take the bets but I haven’t managed that – yet. Help appreciated.

    More seriously, how accurate is the headline, “Controlling Soot Might Quickly Reverse a Century of Global Warming” on

    Can it really slow the Arctic melt?

  20. To which ice extent product does Goddard’s 5.5 prediction apply?

    I mention this because DMI uses a different definition of extent (30% concentration) rather than the more common 15% used by NSIDC. On their scale, we already passed 5.5 million sq km.

    At the time of year when he made his “prediction”, concentration wasn’t low enough for DMI to show the same picture as other products, but in August, both 2008 and 2009 saw a massive drop in the DMI data – thin ice has changed the game, and I expect the same to happen this year.

    DMI served his purpose then, but come September, and Goddard will suddenly decide that JAXA is his new bestest friend.

    Watching ice melt: one of the world’s slowest spectator sports….

    • I suspect your predictions will come true. There are vast swathes of ca. 50% concentration (per CT) now; I think the chances of much of this area dropping through the 30% are strong.

      I’m wondering, too, how much of this may even hit the 15% “knee”–there could (I think) be a big late-season drop if it does. Whether that happens will, of course, be largely up to what weather gets dealt.

  21. To which ice extent product does Goddard’s 5.5 prediction apply?


  22. Watching ice melt: one of the world’s slowest spectator sports….

    I have a slower one …

    … watching to see if Lake Mead hits a new record low.

  23. Well, Goddard’s prediction is predictably wrong, no matter which of his forecasts you pick. I have no intention of going near his home den of iniquity, but I’m a little curious: what hogwash is he selling now?

  24. Dunno about Goddard–I’m pretty allergic to WUWT myself–but if his prediction was 5.5 million km2, it’s wrongness is now more predicted than predictable, since we’re currently at 5.346. A lower minimum than last year seems highly probable, since 2009 saw 5.250 on September 13. It’s conceivable that we’ll see less than 96 K of melt in two weeks, I suppose–but if anyone wants to make that bet please contact me offline. . .

    On the other hand, 2008 hit 4.719 million km2 on September 17. The exact difference (all of this is IJIS data, BTW) is 627,344 km2. To hit that number on September 17 of this year would require a mean daily melt–really meaning, of course, ice extent loss due to any cause–of just over 33 K. That’s a lot this late in the season, but perhaps not out of the question.

    Of course, it’s not strictly necessary that the minimum be reached by that date; the latest minimum (no surprise here!) was that of 2007, which occurred on September 24. The number for that date (rounded as I’ve been mostly doing throughout this comment) was 4.255 million km2.

  25. Tamino, so far your forecast for min. Arctic ice extent looks to verifying– on 6 September ice extent from NSIDC is 4.89 million km^2 (which is lower than the 2009 minimum) and is still slowly decreasing. Sea ice area seems to be almost the same as in 2008.

    Goddard snuck in a late entry to SEARCH of wait for it, 5.1 million km^2.

    You can read more at Joe Romm’s place.

  26. Goddard snuck in a late entry to SEARCH of wait for it, 5.1 million km^2.

    That’s for JAXA, which currently is at 5.027 million km^2. Let’s not forget that Goddard originally predicted “recovery to 2006 levels”, i.e. 5.9 or so, then dropped that to 5.5, then 5.1.

  27. Horatio Algeranon

    Unfortunately for Goddard, “recovering” sea ice is more than a little like a recovering alcoholic.

    Odds are good that the drink will eventually do both of them in.

  28. Although this an old topic… I try to compute this myself with R – but I’m just stuck, my basic maths are probably sooooo far away…

    Isn’t a linear regression model supposed to describe the relationship between two variables, x and y ? Obviously Y variable is the Sea Ice Extent, but what is X ? Years ? I’m sorry, but I can’t see the relationship between the number “2010” and an extent value.

    So, there must be something unsaid and obvious – I’d really appreciate if you can give me a hint. Thanks!

    [Response: In a linear regression model of sea ice extent, the dependent variable is of course extent, and the independent variable is time. This can be expressed in years, months, seconds, or any unit, and the zero point (starting point defining time=0) can be any value as well. In this case, years is a convenient unit.]