Sea Ice Curiosity

The sea ice data available from NSIDC includes estimates of both sea ice extent and sea ice area. Extent is the area over which sea ice concentration is at least 15%, while area is … well, the area covered by ice. Necessarily, sea ice extent will be greater than sea ice area.

It’s mentioned in each data file for the northern hemisphere that there’s a “hole” in the satellite coverage right over the pole, i.e., an area which is not observed by the satellite sensor. Furthermore, this hole was larger in olden days (through June 1987) than in more modern times (since July 1987). In the before time, the hole covered 1.19 million km^2, but now it’s only 0.31 million km^2.

For estimating extent, the hole is assumed to be entirely covered with ice at a minimum of 15% concentration, i.e., the entire hole is counted as part of the ice extent. For estimating area, the hole is simply not counted, hence the NSIDC area data have a discontinuity at the June/July 1987 boundary, when the size of the hole changed. Perhaps that’s one of the reasons area estimates are not as popular as extent estimates.

In any case, one can study the area as well as extent data, and see how they compare. For that purpose, I assumed that the unobserved polar hole was entirely covered with ice at 100% concentration, i.e., that it should all be included in the area calculation. That may be unrealistic — but it does give me numbers to work with. So, for months up to and including June 1987 I added 1.19 million km^2 to the area estimates, since July 1987 I added 0.31 million km^2, to get estimates of northern hemisphere sea ice area for comparison with sea ice extent data.

The interesting comparison (or at least, one interesting comparison) is the difference between extent and area. This showed a characteristic annual pattern through 2006, with the difference being reasonably constant for much of the year at between 1.5 and 2 million km^2. However, extent minus area tended to rise in June, peak in July, decline in August, decline further to about “normal” in September, rise again in October, then decline to normal again in November. The pattern is evident here:

In 2007, the year of the great melt, an interesting thing happened. The extent minus area difference reached its peak not in July, but in June instead (2007 is plotted in blue):

Over the next 2 years the pattern returned to a more normal one. There was still a premature June increase in 2008 but not as pronounced as that of 2007, while the pattern for 2009 was easily in the “normal” range overall. Here’s the updated plot with 2008 in green and 2009 in brown:

Now we come to this year. Most of us are aware that June saw an unusually large decline in northern hemisphere sea ice extent. The interesting thing to note is that this year also shows premature increase in the extent-minus-area difference, but this time it’s less pronounced but more premature — the difference shows unusual increase in May (2010 is plotted in red):

Whether or not this is a sign, a harbinger of dramatic things to come for the 2010 summer minimum, I don’t know. Honestly: I doubt it. But it is a curiosity, and does seem to indicate that not only is the summer minimum of northern hemisphere sea ice in decline, the annual pattern is also changing. Certainly, curious.

And for those who are curious about the southern hemisphere, here’s the pattern of extent minus area throughout the year:

39 responses to “Sea Ice Curiosity

  1. The weather effects are becoming more pronounced than the climate effects during melt season? Of course climate effect is just the warming up there. Could be an indication of the instability of the polar vortex?

  2. This is a timely post! I just started playing with R yesterday. I can see why statisticians like it, even though I’m well out of my depth most of the time.

    I was just looking at ice extent from NSIDC. I in-filled missing values with a spline fit. Then I had a look at the seasonal decomposition. The trend is strongly linear, which surprised me a little. I’m not sure if the Loess smoothing was aiming for a linear trend, but that’s what it found. Completely unsurprising is the huge variation in the remainder in recent years.

    Then I found a function that Tamino hasn’t shown off much – monthplot. I used it on the raw extent, giving extent by month. No surprises there! Declining all the time, but a lot more in summer/autumn.

    I used monthplot on the remainder from the seasonal decomposition. This did surprise me a little: we hear so much about the huge winter ice extents recently (how the deniers do go on). But we see very little variation December-March. Even with the late maximum, there is nothing exceptional there at all. By April-May, we get more variation anyway. The only exceptional part is the unsurprising recent negative remainders in the summer. And that’s well documented.

  3. Fascinating. So in June 2007 the ice was more spread out than any previous year … but by Sept. 2007 it was more compact than any previous year.

    This year’s progression is also turning out to be remarkable.

  4. Sorry, meant more pronounced than before.

  5. “Necessarily, sea ice extent will be greater than sea ice area.”

    Technically… not true. I could imagine a giant ocean filled with 10% concentration of ice, that will have positive sea ice area but zero sea ice extent. (Assuming I’m getting the definition of area right). In practice, of course, sure. Angels on the head of a pin and all that.

    [Response: Quite right. My mistake.]

    Extent minus area is an interesting metric… I feel like there should be something there that could lead to improved predictions of melt seasons. We know that extent in May is not well correlated with extent in September (I presume that excludes the long term trend): but what if we use both extent _and_ area for data, giving us an idea about how spread out the ice is?

  6. “Whether or not this is a sign, a harbinger of dramatic things to come for the 2010 summer minimum, I don’t know. Honestly: I doubt it. ”

    Especially given that last data point: current extent is larger than 2007’s extent at this date, and if the extent minus area is smaller, than the total area must be _much_ larger than 2007’s area. Based on this graph showing 2010 a bit behind 2008, and the knowledge that current extent is a little less than 2008, I’d make 2008 extent as a best guess for final melt season, with a large error bar because of unpredictable weather patterns.

    [Response: Although current extent is larger than on this data in 2007, I don’t know about area or extent minus area. The data in this post are monthly averages from NSIDC, which only go up to June (for which 2010 extent was the lowest on record).]

  7. M: Tamino was right about area always being less than extent – for the NSIDC dataset anyway! I imagine he remembered reading in the documentation “The total area is always less than total extent.” This is because “The values for ice area are obtained by summing the concentration of ice within each pixel over the entire ice extent.” This excludes areas of <15% concentration from the area as well as the extent.

  8. J: I was looking at the Cryosphere Today animation. Over the last 2-3 days, the ice has begun to compact. I don’t know if this will last, but it does agree with the NSIDC prediction that the dipole anomaly is reforming.

    Anyway, I’m getting off-point. As the ice is compacting, it has the opposite effect for all the broken-up peripheral ice. It gets spread over a huge area as the core ice pack moves together. That’s what happened in June 2007, enabling all the spread out ice to melt rapidly by September.

    It is too late for the same thing to happen this year, but I expect ice area and extent to start a late decline similar to 2008. Assuming the weather doesn’t change yet again, of course.

  9. Two more data points:

    JAXA area

    JAXA extent

    I don’t know if they do the same 15% cutoff for area as well as extent as is done by NSIDC, but I do know the area *graphic* is always smaller than the nice big extent graph!

    As you can see, JAXA has area plummeting while extent is drifting downwards less impressively. This is consistent with the stalled Beaufort Gyre tending to spread ice out rather than compact it. But conditions appear to be moving back towards favoring compaction, which would lead to a more rapid decrease in extent.

    We’ll see. As Dirk says over at RC, the thin ice cover of the past few years makes it much more difficult to predict what extent will be, as it’s so dependent on how this thinner, weaker ice is moved around.

    • dhogaza, thx for uncovering that JAXA Area sheet. Slam dunk on those who hang on to JAXA Extent near real time now slightly above 2009… meantime Area clearly dipping away from 2009… The Ice is Rotten as the Amundsen Captain wrote recently while traversing the Arctic this year, hitting a 14 meter thick, yes fourteen meter thicks shelf and it disintegrating.

  10. Didactylos: Ah, yes, that would be a key detail in the definition of “area”. Under that definition, Tamino is correct.

    Tamino: Ah, if those graphs are for the June average, then yes, a prediction based on June extent and those graphs would be that we would be approaching something like 2007 (with large error bars) due to a) smaller extent, but b) less extent/area discrepancy.

    My prediction not based on those graphs, and taking into account July so far, is that we probably won’t drop below 2007… but I wouldn’t rule it out. It seems highly unlikely that the 2007 record will last more than 2 or 3 more years, though.

  11. Wouldn’t it make more sense to look at (area / extent)? That should be a reasonable approximation of the average ice concentration across the pack: i.e. how healthy the ice is.

  12. Michael Hauber

    From extensive looking at maps of concentration etc I note that in 2007 there was a quite massive area of reduced concentration in the Beaufort sea, that only lasted a few days. This was followed by a fast retreat of the sea ice boundary through the Bering Strait. At first this retreat did not result in massive extent or area reductions as the retreating ice edge was contrained within the narrow Bering Strait. But this was followed by a much earlier retreat of the ice edge along a very long front north of Russia, which no other year has been able to come close to.

    Towards the end of June 2010 a perfect melt pattern set up with a large high in the Beaufort Sea and the postive dipole. This resulted in massive areas of reduced ice concentration, and a drop in Crysosphere Today area anomaly of close to -2 million sq km, which is equal to the anomaly during 2007 minimum (anomaly reached -3 milllion sq km after the minimum).

    From MODIS visual imagery I’m quite sure the large areas of reduced concentration in late June were due to surface melt, and that the ice was mostly in fact fairly solid. The appearance of the ice was quite bluish (indicating melt) with the dark areas of ocean water restricted to near the edge of the ice sheet.

    However recently there have been some fairly significant areas of reduced ice concentration, and this does appear to correspond to areas of ocean appearing within the ice sheet, although visual views in MODIS are becoming much rarer due to the low pressure dominated conditions.

    Another thing I’ve noticed is that in 2006 and 2009 the extent reduction rates appear a little as if a switch was thrown on a partciular date from fast reduction to slow reduction. A similar switch appears to have been thrown in late June this year. In 2009 and this year the switch seems to correspond to a change to cloudier lower pressure conditions, and perhaps this is a semi-normal Arctic phenomena where summer can often start out dryish and fine, until the development of the ‘wet season’.

    PIOMAS also shows very low volume anomalies. With concentrations and PIOMAS modelling both hinting at a thinner than ever before ice sheet, it will be interesting to see how the rest of this season turns out….

  13. One thing I’ve been curious about is the long time-series of sea ice area anomaly shown at Cryosphere Today (

    Is there any significance to the annual cycle in anomalies that seems to have developed after the minimum sea ice area observed in 2007? Any ideas or references to this?

    [Response: Anomaly is the difference between observation and the *average* annual cycle. The annual cycle is quite different since the great melt of 2007, and that net change in the annual cycle persists in the anomaly signal.]

  14. The next week will be interesting. It was this week in 2008 when the late melt happened. Area anomaly is already low for this time of year. Not as low as a month ago, but still -1.4

    If it gets to -1.8 then it will be comparable to 2007 and 2008.

  15. Michael Hauber

    The annual cycle in anomalies reflects the fact that summer minimum has been reducing faster than winter maximum. One contributer to this is that at summer minimum there is a wide area of ice edge within the Arctic basin that can move towards the pole if a little extra heat is applied. In winter the ice edge is much more contrained by land, or the boundaries with much warmer waters in the North Pacific and North Atlantic.

    Most of the long term reduction in winter extent seems to be on the Atlantic side, and summer extent more on the Pacific side.

  16. I have made some animations of different sea ice concentration maps in the last 2 weeks or so. Looks quite interesting, but nothing conclusive can be drawn from it as of yet.

    BTW, Tamino, I have linked to both your Arctic sea ice posts in my latest Sea Ice Extent update.

  17. Tamino, or other statistics experts:

    Is there a different minimum period to discern Arctic sea ice trend than for temperature? I read on here aways back that the absolute minimum length for global temperature is 14/15 years*, elsewhere that 20 – 30 years is a good standard and I know that the classic (WMO) period is 30 years.

    Statistically speaking, is it about the same for Arctic sea ice?

    (* Although, the hoo-hah about Jones’ ‘no statistically significant warming since 1995’ makes me wonder if 14/15 years is a little short)

  18. The weather effects are becoming more pronounced than the climate effects during melt season? Of course climate effect is just the warming up there. Could be an indication of the instability of the polar vortex?

    First post gets it right, IMO. Much arctic sea ice is thinner, less buttressed by thicker ice, etc, structurally weaker. Pushed around by wind much more than older ice.

    Of course, I’m making the same argument that was misinterpreted post-2007. Weakened ice combined with exceptional weather led to the huge 2007 drop.

    The denialsphere mistakenly claims that 2007 was due only to weather, discounting the fact that ice was far weaker and thinner due to increased annual melt.

  19. Wave action is going to have a greater affect as the ice becomes weaker.
    As more ocean surface is exposed, wave action will increase and the current weaker/thinner ice will be less resistant to breakup than the old tougher/thick ice.
    Of course broken up ice presents more surface area and can melt even faster.

  20. Thanks, Hank, but that’s about temperature trends (and I read it before).

    I’m wondering if there is a different minimum period for statistical significance for other climate phenomena – in this case, Arctic sea ice. I know this can be assessed using observed data, but I don’t have the skill to do it.

    [Response: Yes. IIRC it takes less time for arctic sea ice because the signal-to-noise ratio is higher.]

  21. Thanks, Tamino, that was my hunch. It would be good to have a resource for what the minimum period is when engaging skeptics on sea ice trends. I’ve not been able to find any on the web (resources, that is :-).

  22. I have a badly written blog post up that relates to what is being discussed in this post by Tamino.

    Amongst other things in the blog post I agree with what Michael Hauber wrote here with regards to melt ponds probably causing the early peaks of 2007 and 2010:

    When air temperatures get high enough and the sun reaches its maximum elevation around summer solstice melt ponds start to form on the ice, fooling satellite sensors into believing this is open water. For extent numbers this isn’t so much a problem because a melt pond would have to get really big to cover a grid cell of 625 square km in such a way (more than 85%) for it to not be counted. With area however every percent of the surface that doesn’t look like ice gets added to the total, making sea ice area numbers go much lower than extent numbers. But at this point in the melting season melt ponds aren’t that big any longer, having been drained or starting to freeze over again, so we are mostly looking at the spreading of ice when fiddling around with area and extent numbers.

  23. Also on Neven’s blog, I have a couple of related graphs comparing area with extent:
    (1) a cycle plot of annual trends in area and extent, by month; and
    (2) a time plot of monthly ratios of area to extent since 1978.

  24. Barry, Grumbine is teaching how to take a data set — any data set — and look at how much variation is in it, and from that, how to decide how much more data to collect to have a good chance to detect a trend if there is one. It’s one of the very basic lessons of statistics. He illustrates it with temperature data, but you can follow his examples using other data sets.

    Did you work the example he gives? If so you should have an idea how this is done.

  25. DeWitt Payne

    The Cryosphere Today area time series for Arctic, Antarctic and Global are available on line. Apparently, they correct for the Arctic hole as there is no discontinuity in 1987. The anomaly column for these series is based on the 1979-2008 average. The 1979-2000 anomaly and average data is also still available by deleting the ‘.1979-2008’ from the end of the URL.

    As far as area calculation, area is also not counted for pixels with concentration below the site’s cutoff, normally 15% but as high as 30% for Arctic ROOS.

  26. Hank, I’m no good at maths. I don’t know what program to open a ‘tar file in. I can get anomalies downloaded to Excel, but I just don’t have the skill to follow Grumbine’s method. I could maybe work it all out in a week, full time. Maybe.

    But thanks again. It’d be nice to be able to do it, but I’m a complete novice when it comes to statistics. Tamino’s site is good because I can grasp the processes conceptually, if not actually replicate them. Other of Grumbine’s posts on deriving signal also made sense, conceptually.

  27. I don’t think you can use Grumbine’s method on satellite sea ice data, since he discards the first and last 31 years so that he can compute forwards, backwards and centred trends. Ice only has 31 years in the whole dataset, so you have an immediate problem.

    I honestly don’t believe that the ice dataset on its own permits determining what is a “significant climate trend”. If the extent data is extended back to the 1950s with observational data, then the significance becomes much clearer – and ice extent is fairly reliable that far back, since it only relies on the edges. I don’t trust ice area to the same degree, because the interior was never* directly observed and the concentration was inferred from a variety of sources.

    However, I need to run the numbers and see if I’m right…..

    *never, except for when it was, of course.

  28. I have posted a follow-up to the first blog post on Area vs Extent. It has a link to a spreadsheet I made with all the CT area and IJIS extent data from the period 2006-2010.

  29. Based on this Arctic Sea Ice dataset there is a new concentration low for July of 65.5%. The previous low for month 7 was … 2007 with 65.68%.

  30. I am interested in a tangential point. To what extent has planetary albedo been effected by the recent changes in sea ice area. The proposition has been made to me that increases in antarctic sea ice balance out the changes in total albedo that artic decreases have made.

    This is not an idea that I find credible.

    Indeed even if it were true (that the increases in antarctic sea ice are a substantial negative feedback) decreases in ice mass for the antarctic ice sheets and glaciers argue that at most sea ice increases are only marginally decreasing the changes afoot in the antarctic.

    The distinction between the trends for sea ice versus glaciers in the antarctic seems to cause a great deal of confusion in some circles.

    In the NSIDC frequently asked questions the following point is made:

    Unlike Arctic sea ice, Antarctic sea ice disappears almost completely during the summer, and has since scientists have studied it. Earth’s climate system over thousands of years has been “in tune” with this annual summertime disappearance of Antarctic sea ice. However, satellite records and pre-satellite records indicate that the Arctic has not been free of summertime sea ice for at least 5,500 years and possibly for 125,000 years.

    All of this seems very clear, but being a curious guy I was hoping someone might be willing to point out what has been published in the peer reviewed literature which would be of interest.

    • I imagine that the increase in Antarctic sea ice does slightly compensate for albedo changes in the Arctic.

      The key word is “slightly”, and for two reasons: 1) albedo is less of an issue during polar night, so changes during the summer have more impact than during the winter. 2) Albedo changes have a very local impact. It is exactly the area of water formerly covered by ice that receives more solar radiation and warms more. The Southern Ocean is huge. The Arctic Basin is a semi-closed system and relatively small.

  31. I added the July NSIDC numbers to update my graph showing extent/area ratio 1978-present, over on Neven’s Sea Ice blog.

    August or July have historically been the low points for concentration defined as area/extent ratio, and these low points (unlike the high points) have been dropping. The July 2010 ratio most resembles that of 2007.

  32. I have updated my web site on GCM reliability. I have added references and full citations for each prediction and each confirmation (17 pairs in all).

  33. there is a new concentration low

    Yes, the stalled Beaufort Gyre and wind patterns have spread out the ice somewhat, there’s been some discussion over at Neven’s excellent arctic ice blog.

    It’s susceptible to compaction and therefore a fairly steep drop in extent, depending on wind conditions. The lower ice to sea water ratio should mean it’s more susceptible to melting, too, all other factors being equal.

  34. This is my first time on the blog, so please bear with me.

    I’m now writing an introductory survey on sea level rise and, being a mere generalist, I don’t want to stray too far off the scientific reservation.

    Toward that end, although I can’t pay anything, I’m aways looking for readers who are willing to give me their brutal comments on what I draft. At present, three sections of this book are available in clean drafts. They include my thoughts on (1) the IPCC, (2) a primer on global warming, and (3) why sea level rise is important to us.

    Should you wish to read any or all of them, please contact me off list at