Ice Forecast Update Update

One of the blessings of NSIDC sea ice data is that they update their monthly averages promptly.

This enables me to update the forecast for the September minimum sea ice extent using the monthly averages from July. It turns out that July itself set a new record for lowest monthly average sea ice extent:

It also set a new record for lowest monthly average sea ice area:

Modeling the September minimum extent using time (including a quadratic term), July extent and area, and June area (the best model according to AIC), the new prediction is 4.22 +/- 0.62 million km^2, which would just barely break the 2007 record for all-time minimum extent:

I’ll point out that the error bars are sizeable, and this September could well fail to break the 2007 record. In fact at the moment I’d say that the odds are about 50-50. But it’s very likely that the 2011 minimum will have less extent than 2010, and will probably show at most the 2nd-lowest September average sea ice extent on record.


32 responses to “Ice Forecast Update Update

  1. What is the difference between extent and area in this context?

    [Response: Area is the total area of ocean covered by sea ice. Extent is the total area of ocean which is at least 15% covered by sea ice. Area is almost always less than extent, since extent can include regions which have a large fraction of open water.]

    • Thanks, Tamino. How, then, is extent calculated? Is the map divided into discrete squares, and squares which are 15% covered in ice considered part of the extent? Or is a different strategy used?

      [Response: Basically yes. I don’t think they’re grid *squares* however, I think they’re latitude-longitude cells, or maybe image pixels — but I’m not sure.]

  2. THanks… wait. If I stand on my head when I look at this….

  3. Last I checked, this wasn’t shaping up as a particularly warm summer. Anyone have any ideas about why this year may well break 2007? My best guess would be “thin, young ice melts fast”.

    I’m also wondering what the deniers/skeptics will argue if the 2007 record is broken. At the moment I’m betting on “no significant change in arctic sea ice extent since 2007”.

    • Andrew Dodds

      You mean, in skeptic-land, the 2007 sea ice minimum is the equivalent of the 1998 freak temperature excursion – a place to start all your graphs from.

      Of course, it took about 7 years for 1998-like temperatures to become ‘the new normal’, whereas 2007-like sea ice conditions seem to have taken about 3 years…

      • Exactly so. Some consistently refer to 2007 as a freak ‘wind-driven event,’ in order to suggest that temperature had nothing to do with it.

        Dishonest, of course, but there you are.

    • Some parts of the Arctic–particularly western Siberia–have been quite warm this summer. And big swathes experienced remarkably warm anomalies during parts of the ‘freezing season,’ surely leading to thinner ice now.

  4. The NSIDC (and I, for that matter) in computing extent do so on the grid cells. The grid cells are patches on the polar stereographic projection data grids that they and I use (my grid being slightly different). The reason for the polar stereographic projection is that the cells are of more nearly equal area than would be the case for a latitude-longitude projection.

    One of the sources of difference between NSIDC and JAXA/IARC and me is the choice of land masks. Land cells aren’t included for extent. But nature is not so kind as to put land on a grid.

  5. Tamino, have you heard about this sea ice prediciton:
    Their current prediction is 4.5 +/- 0.3 million km^2.

    How high is your r^2?
    Do you know something about their methods?

    • Their +/- 0.3 refers to +/- one standard deviation, while Tamino’s range is, I assume, for two SD.

      Another interesting link on that site is to this call for September arctic sea ice predictions:

      Predictions must be submitted by Friday Aug 5, 2011. I’m tempted to give it a shot.

      [Response: Yes my range is +/- 2 standard deviations.]

  6. We’ve had a drastic slowdown of the melt-rate over the last week or more, so it appears that the odds of seeing a lower September minimum than 2007 are dropping quickly. But Arctic weather is amazingly fickle, so that could change once again. Personally, I think that Tamino’s predictions is going to look pretty good at the end of the season.

    As to factors influencing the melt, I think that warm temperatures are indeed part of the answer. While global temps were lower in the early part of 2011, much of the Arctic stayed warm–there has been a pretty serious warm anomaly in western Siberia over most of the winter, and a good dollop of warm anomalies in the Canadian Arctic as well–albeit less persistently than the Siberian one. For instance, the Eastern Arctic stayed warm late in the fall, leading to a slow freezeup of Hudson Bay. (That led, I’m pretty sure, to some of that ‘thin ice’ Ernest K mentioned.) Currently temps in that area are once again pretty elevated.

    A good summary view of all this can be had by navigating to Neven’s blog, linked by Daniel above, and going to the “Daily Graphs” page. It has such goodies as the DMI ‘Arctic 80-degrees North’ temperature graph, which shows that the high Arctic had a pretty warm winter. (During the melt season itself there is much less variability in that area, so temps are currently pretty close to norms.) There’s also a “Regional” page, where you can see at a glance the evolution of the ice cover for each of the various seas making up the Arctic Ocean.

  7. Ernst K, at the Swedish deniers blog “The Climate Scam” they just dusted off some old newspaper stories from early last century about how warm it was in the Arctic back then. I assume this is in preparation for a new low this summer:

  8. I tried to replicate Tamino’s analysis but got slightly different figures so I may well have done something wrong.

    When we have multiple methods there seems to be two obvious things to do: Firstly compare them to decide which is the more powerful technique. Second would be to try to combine them to create a better prediction.

    This is what I got:
    Predict R^2 SE RMSE
    NSIDC avg area & ext 4.306 0.883 0.337 0.310
    CT daily area & avg Ext 4.391 0.904 0.306 0.281
    Gom & area residual 4.345 0.585 0.291 0.282
    Gom, area&Ext resids 4.263 0.596 0.292 0.278

    No surprise that all the predictions are close. My R^2 for my gompertz fit and area residual at under 0.6 looks rubbish compared to over 0.9 but this is simply because my linear regression is set a harder task of predicting residuals from fits rather than the total data.

    The lowest standard error of these is using my gompertz fit and area residual at 0.291 suggesting, I think, this is a more powerful analysis method. I may have done something wrong and a 2 standard deviation range of 0.61 might well be better? But uses more data so may not be as good?

    Is it odd that I find the extent data not to be of any use whereas with your technique, you do find the extent data to be useful?

    The daily area data is more up to date and seems more useful than monthly average. IJIS daily extent information is only a short record. Is it possible that two such extent datasets could somehow be used?

    Combining these methods to get a better prediction does not seem likely to be feasible as they just aren’t independent. Combining a model and a statistical model may seem more plausible but I fear that it would be difficult to get an understanding of the extent to which the model is tuned to give similar data to the area and so the model might not be as independant as needed?

  9. At Cryosphere today, compare the sea ice extent maps for July 30, 2011 (latest available) and July 30, 2007. Most of the remaining ice in 2007 had 80-100% concentration. This year, most of the remaining ice has 60% concentration. Does that difference imply that there could be a massive late-season decline?

    [Response: It certainly makes the ice more vulnerable to both melt and export. But keep in mind that the weather still has a potent impact. In 2007, winds and currents were “optimal” for reducing ice extent, and we probably won’t see the same conditions this year. For the moment I’ll stick with a 50-50 chance of lower extent minimum than 2007. And I’ll re-do the forecast when the August NSIDC numbers are posted.]

  10. My general impression is that weather is a so big component that almost all model will have the same power unless they are able to do some midterm weather prediction.

    • Which should be precisely the point. 40 years ago, it didn’t make a bit of difference what the weather did, because 3-4 m thick ice doesn’t respond to weather to the same degree that the newer, slimmer ice does.

      • Which is why I was so puzzled at the paper that stated that thinner ice was less susceptible to transport. That didn’t make sense to me, and still doesn’t.

        Nor does it seem particularly well-supported by this year’s IJIS-JAXA extent curve. (I know, that’s a hopelessly unscientific statement.)

  11. Along with melt and export it means there’s a great potential for compaction if the winds are favorable. That’s one of the things that happened in 2007, along with export through the Fram straight.

    There was a mass of ice being exported through the Fram in July – extent in the Greenland Sea was increasing for a chunk of the month.

    Now that has stalled, and if you look at one of the extent reconstructions (like this one) you’ll see that there’s a lot of melting going on off eastern greenland, but no new purple fingers (on that graphic) being pushed down there from the Fram.

    A shift in wind patterns could cause extent to drop like a rock … but if conditions stay as they are, extent will continue to drop slowly, most likely.

    So we shall see …

  12. Take a good look.

    You can now circumnavigate the arctic, in a small boat, at least.

    Seems really early to me …

    • My understanding is maybe not, since there’s a lot of broken up ice (“bergy bits” etc.) that’s unimportant for counting purposes but potentially very hazardous for small boats. IIRC the official declaration of an open NWP is based on an assessment of that hazard. Bob can enlighten, hopefully.

    • At Neven’s, consensus was that the NEP was open on the earliest date so far. That’s unofficial, of course.

  13. Alas, I stay out of Arctic water, myself. Of more concern is that the satellites rely on a 15% ‘cutoff’. The ‘no ice’, as far as the satellite is concerned, is not ‘no ice’ for trying to drive your boat through it.

    On the other hand, the fact that this kind of discussion is being held speaks to a dramatic change to Arctic ice climatology.

  14. Did the people who circumnavigated in a sailboat last year wait until the NWP was officially open? I know they saw ice in places though made it through …

  15. I’ve thought the Capie-index at Nevens’ could be a predictor for cloud formation, as well as it is the measure of compactness, but I admit there’s very little if any trend in that. So the variation in Capie should be constrained to some values, changing with the progress of the melting period, f.e. the autumns are much cloudier up there (the sun lets the extra water vapor present to condense more easily,

  16. Meanwhile, the POIMAS Arctic sea ice volume anomaly is hell-bent on screaming well into ‘new record’ territory.

  17. Tamino,

    “It turns out that July itself set a new record for lowest monthly average sea ice extent”

    The drop in July is nothing short of astounding . The monthly average for July up through and including this year is 9.69 million sqkm. This year came in at 7.9 million sqkm. Almost a 20% drop. The 2008 level was 9.06 and every year since its been in a severe downward decline. From NISDC data for July it looks like the last time extent was anywhere near the current average was 2004 (9.6 million sq km).

    I see what you mean by death spiral.

    Just to stay in the prediction game for Sept extent the Kalman filter I built for Sea Ice Extent prediction using the full 32+ years data (described here: was not expecting this July measurement value predicting for July 8.95 million square km with a one sigma prediction uncertainty of .447 million sqkm. Now it is over compensating for the large July prediction residual in the Aug and Sept predictions. August is predicted at 5.44 million sqkm with 1 sigma of .5457 million sqkm September is predicted at 4.07 million sq km at 1 sigma of .8665 million sqkm.

    As an assessment of the filter performance and to see how much of an outlier this July measurement was I looked at a statistic defined by the ratio of the filter residuals to the 1 sigma predicted error over all 385 (or so) months of predictions and measurements. It has mean -.009 and 1 sigma of .86. So based on those statistics it seems to be doing OK in general. The ratio for July is over 2.

    Just to get an idea on the distribution for this statistic I ran a Lilliefors test for normality on the its distribution and it did not reject the null hypothesis at the .05 level (p value = .08). So assuming a gaussian for this distribution is probably OK (I think).

    I don’t have a convenient way to show plots of this stuff otherwise I would put some up.

    Anyway comments criticisms or suggestions are always welcome.

  18. arch stanton

    Sorry Tamino, I only wanted to link to youtube and not embed it.


    [Response: Try again!]

  19. arch stanton

    Sorry to bother you Tamino. Forget it. I give up.

    Thank you for deleting all of these.