Arctic Sea Ice Loss, part 1

Although it’s useful and sometimes interesting to refute silly ideas about Arctic sea ice loss (such as the claim that it is “stabilizing” or even in “recovery”), it’s far more interesting scientifically to consider what available data actually tell us about the changes of the ice pack in the frozen north.


Three parameters are often used to characterize sea ice. Area is simply the area of ice-covered ocean. Extent is the amount of ocean which has a minimum ice coverage percent (usually 15%). It may be the case that this is easier to determine than area itself. Volume is the total volume of sea ice, which I regard as a more relevant physical parameter but has not been estimated with anywhere near the accuracy of either area or extent. This is primarily because computing volume requires estimating sea ice thickness, which as far as I know has only been directly measured intermittently.

The best volume data of which I’m aware are from PIOMAS, but that is based on computer simulations of ice thickness using weather data (calibrated with available data), and although the results have been shown to be telling the story right, they cannot be thought as accurate as either area or exent data. Future observations may improve things dramatically, but at present we can’t study ice volume with as much confidence as other measures.

We can also study some derived variables. Two in particular come to mind. First is average thickness, which we’ll define at volume divided by area. In addition, sea ice extent is larger than area because the ice pack breaks up (expecially around its edge) and spreads out over a larger area. We’ll define spread as the ratio of extent to area.

Let’s study all these quantities to estimate their changes during the satellite era. For extent, we’ll use data from NSIDC (national snow and ice data center), for area data from UIUC (University of Illinois and Urbana-Champaign, which runs the excellent website Crysophere Today), for volume data from PIOMAS. For all quantities I’ll use monthly averages, which gives plenty of time resolution and enables us to study the annual cycle itself, as well as trends for different parts of the year.

When looking for trends it helps to eliminate the annual cycle in order to isolate the longer-term changes. This is usually done by computing anomaly, the difference between the value at a given moment and the average value at the sime time of year. This is fine, although one problem does arise. If the annual cycle itself changes, then when we subtract the average annual cycle, the residual annual cycle (the difference between the present annual cycle and its long-term average) will remain in the data. This is clear in graphs of such “simple anomaly” for any of the given variables, for instance for extent:

extent_anom1

Note the large 1-year cyclic variations since about 2007. That’s the residual annual cycle, because the annual cycle since 2007 has been larger than its average throughout the time span.

We can, however, improve things by computing what I’ll call an adaptive anomaly. Instead of subtracting the average annual cycle for the entire data set, we’ll subtract the average annual cycle for years near the time in question. This is the same method used to compute anomaly for CO2 data, because the annual cycle of atmospheric CO2 has also increased over time. I’ll define adapative anomaly as the difference between a given moment’s value, and the average for the same time of year during the 5-year time span nearest the given moment of time. I’ll further define that average as the result of a 5th-order Fourier series fit. I’ll use adaptive anomaly throughout, and simply call it “anomaly.”

Using adaptive rather than simple anomaly has almost no effect at all on the trend, but it does improve the visual interpretation of anomaly graphs by isolating the trend changes from the annual-cycle changes. And, frankly, that’s what anomaly values are about.

The anomaly for extent data now looks like this:

extent_anom

Note that the recent large annual fluctuations are gone. Just out of curiosity, here’s the difference between the “simple” anomalies and the “adaptive” anomalies, which shows the changes over time in the annual cycle itself (actually the difference between the momentary and long-term average annual cycle):

anom_diff

The anomaly graph also shows (in red) a smooth which estimates the trend in the data. There are plenty of fluctuations about that trend, but there’s no statistically justifiable evidence that the trend is different than indicated by the smoothed line.

An easy way to visualize the trend is to compute annual averages:

extent_anom_1yr

I’ve smoothed these (red line) on a slightly shorter time scale than used for the monthly data, but bear in mind that the differences between the two smooths (which are small) are not statistically significant.

Note that 2007 was an unusually low year for sea ice extent, dipping well below the trend. But its dip isn’t that remarkable, it’s comparable to that in other years. The extreme low in 2007 is mostly due to the trend, but the dip well below the trend in 2007 has been attributed to “weather” — specifically, to an extreme minimum that year caused by wind patterns which tended to flush ice out of the Arctic, through the Fram Straight, into the north Atlantic where it melted away.

Note also that just as random fluctuation (weather) caused 2007 to dip below the trend, it also caused 2008 and 2009 values to rise above the trend. This was touted by fake “skeptics” as some kind of “recovery” of Arctic sea ice, some of them even called it the onset of return to prior sea ice levels. For data with this level of noise and autocorrelation, declaring trend reversals or even reductions based on such short time spans is foolish. Subsequent years have contradicted their assertions.

The record low in 2012, however, seems not to have been exaggerated by weather fluctuations (despite claims to that effect). It’s right on the trend line. That doesn’t prove that it wasn’t brought about by such fluctuations, but it does show that there’s really no evidence of that, at least not yet. According to the evidence, the trend now is as fast as it ever was.

The trend has been different at different times of year. We can use the monthly data to compute the average trend rate for each month of the year as well as its uncertainty level, which gives this:

extent_rate

All months show significant ice loss, but April extent has declined by “only” about 35,000 km^2/yr while the September average decline comes in at a whopping 87,000 km^2/yr. Furthermore, the September decline has accelerated over time (as has the annual average decline):

Ext09

The blue line shows a quadratic fit to the data, which is statistically significant and confirms acceleration of September sea ice loss. In fact by this test the months of June through December all show statistically significant acceleration of sea ice loss, none show statistically significant deceleration.

In addition to looking for trends (and their changes) in the value of sea ice extent, we can also study trends in the annual cycle itself. Windowed Fourier analysis enables us to estimate how the amplitudes and phases of the various Fourier series components change over time. Here, for instance, is the change in the amplitude of the fundamental Fourier component of the annual cycle of sea ice extent:

extent_amp

The really big change has been since 2007, with the amplitude increasing and the last couple of years showing the largest cycles yet.

The relative amplitudes are the ratios of the amplitudes of the various Fourier harmonics to the amplitude of the fundamental:

extent_relamp

Amplitude of the 2nd Fourier component is shown in black, of the 3rd in red, and the 4th in green. Changes in the relative amplitude suggest change in the shape of the annual waveform. Only the 2nd component has changed noticeably, with its relative amplitude increasing recently, so that its actual amplitude has increased even faster than that of the fundamental component. The result of this particular change is that the annual September minimum has become “pointier” recently while the annual maximum has become more “flat.”

Sea ice area shows much the same trend (and even very similar fluctuations) as sea ice extent:

area_anom

It probably shouldn’t be too much of a surprise that the changes in the rate of ice loss according to sea ice area are just about identical to those indicated by sea ice extent. Again the overall rate has accelerated and shows no evidence yet of any deceleration. Again different months exhibit different rates:

area_rate

For area data, it’s the months of July through January which show significant acceleration, none show significant deceleration.

Area data show an increase in the size of the annual cycle which is much like that in extent data. Something area data show even more strongly than extent data is change over time of the timing of the annual cycle, as indicated by the phase (i.e., time of year) of the annual maximum of the fundamental Fourier component:

area_phase

The annual maximum migrated earlier in the year by about 0.02 cycles (about a week) in the late 1980s, and has slowly returned to near its early-1980s value since.

Arctic sea ice volume shows a somewhat different pattern that that of extent and area:

volume_anom

There are still plenty of fluctuations but they’re not as large relative to the trend as with extent or area. This may be partly because the volume data involve the output of a computer model for ice thickness. There is still strong acceleration of sea ice volume loss, and no sign of deceleration.

Different months show different average rates of decline, but the seasonal differences aren’t as pronounced as with extent or area:

volume_rate

One extraordinary and dramatic thing to note is how much the September average (when it reaches its annual minimum) has declined:

Vol09

The accumulated loss is nothing short of astounding. Between 1979 and 2012, September average sea ice volume declined from 16.9 thousand km^3 to a mere 3.4 thousand km^3 — a decrease of 80%. Nothing short of astounding.

Volume data also show a larger change of phase than extent or area, which began in earnest in 2009:

volume_phase

The timing of the fundamental Fourier component of the annual cycle is coming earlier by about 10 days. The shape of the annual cycle is also changing. This is revealed by a change in both the relative amplitudes:

volume_relamp

with the harmonics becoming less pronounced relative to the fundamental, and change of the relative phases, which are the phases of the Fourier harmonics when the fundamental is at phase zero:

volume_relfaz

All in all, the three direct measures of Arctic sea ice show that the ice pack is in a “death spiral,” disappearing right before our eyes. The decline has accelerated since satellites began detailed observations. There is no evidence of any recent deceleration, despite the faulty opinions of a few. And there are some interesting changes in the timing and shape of the annual cycle. I don’t pretend to understand the physical reasons for these changes.

In another post soon, we’ll take a closer look at the changes to the annual cycle of sea ice volume, and we’ll also look at the trend and cyclic changes of sea ice “spread” and thickness.

About these ads

26 responses to “Arctic Sea Ice Loss, part 1

  1. I would argue the amount of fluctuation of 2D data is high because of the Arctic geography and wind, while volume is mostly driven by the energy budget and thus fluctuates less. Have a look at the ice mass balance buoys. Ice growth as measured by 2013A, 2012G, 2012H give a straight line, despite different position, initial thickness and weather. All three agree on 40cm in the last 2 month.

  2. Eli has always thought that the proper way to look at this is that there are large areas where the ice ALWAYS melts, e.g. south of the Bering Strait and South of Svalbard and Nova Zemelya, more or less the Arctic Sea. A sensitive analysis of melting would look at the variation over the years in the anomaly in that area and has the advantage that at the peak of winter the anomaly would always be zero because it is full with ice every year.

  3. Fram Strait (no gh) to be pedantic.
    Thank you for another wonderful post teaching me more statistics.

  4. Decline of 80% in the September average volume in 33 years is truly terrifying. If the rate of loss were linear that would mean eight years until the September average volume is zero. But, you’ve shown that the decline is accelerating. Truly, truly terrifying.

    • Andrew Dodds

      Yes.. at least according to PIOMAS, we will shortly be reaching the point where the remaining September volume is in the same region as the bigger year-on-year volume drops – i.e. the whole lot could go.

      Of course, it also has to be observed that even a near-zero PIOMAS volume measurement could still see a significant extent of fragmented ice floes.

  5. >”There is no evidence of any recent deceleration, despite the faulty opinions of a few.”

    Only a few? I thought there were a lot of modelers who believe their models’ conclusions of a deceleration despite the data. When there is such a discrepancy between models and data, what should you do? Throw away the models and just use trend analysis? Attempt to correct model biases? Rely only on data assimilating models like PIOMAS? or something else?

    Trend extrapolation is often dismissed as not physical. I tend to think volumes melted and max volume extrapolations get a bit nearer to being physical based on heat available to melt ice and on the equilibrium thermal balance thickness considerations than an extrapolation of area or extent.

    So how should you view such dismissal of trend analysis as unphysical when the models appear to be performing poorly?

  6. Regarding UIUC No. Hem. (Arctic ice area) Rate of Change, you wrote “… the months of July through January … show significant acceleration ….” Did you mean July through October?

    [Response: No.]

  7. Question: When you are doing your quadratic fit to the PIOMAS NH data, you are getting a curve which rises in the early years then falls. Wouldn’t it be better to fit only the monotonically decreasing side of the quadratic? (i.e., fit only half the parabola from zero only to further negative values?)

    [Response: Depends on your purpose. If you're trying to mimic the detailed changes, perhaps -- but a smoothing method (like Lowess) is better. But if you're testing whether or not there's a change in slope, perhaps not.]

  8. A quibble: “The record low in 2012, however, seems not to have been exaggerated by weather fluctuations (despite claims to that effect). It’s right on the trend line.” This statement is referring to the graph of average values, but the 2012 record low usually refers to the September minimum extent/September monthly average, which, in all the other graphs you show, is still below the trend line…

  9. What scares me is — I can imagine that, a few decades from now, the shipping industry and governments will react with shock and horror to proposals from a few ice-huggers when they propose implementing some newly developed approach that’s actually promising to begin removing CO2 from the oceans and atmosphere.

    Imagine those bleeding-hearts wanting to protect the baby krill, and restore the Polar Sea Bears, at the huge and unacceptable cost of _freezing_over_ our now enormously valuable Arctic open water shipping lanes. Think of what it would add to the cost of everything that can be valued in dollars, eh?

    Besides, think of the danger all that ice would cause. No, I’m sure they’ll be thinking that profit and warmth are allied in this best of all possible worlds.

    You don’t know what you’ve got til it’s gone.
    Your grandchildren won’t know what you had nor why you cared.

  10. Pete Dunkelberg

    Your grandchildren won’t know what you had nor why you cared.

    Riight. Who needs New York City?

  11. Arctic summer sea ice loss won’t greatly affect New York City.

    Sea ice loss has begun to change ecologies, and we don’t know how that will work out. http://onlinelibrary.wiley.com/doi/10.1029/2008GL035028/abstract suggests the change will increase seasonal plankton blooms, at least for a while. Then what? Most of the big fish and whales (the top predators) are gone, since our grandparents’ time. Do we miss them?
    Predator diversity dampens trophic cascades
    Without them, changes will probably be exaggerated over the baseline http://scholar.google.com/scholar?q=%22primary+productivity%22+%22top+predator%22
    But we don’t know enough, as we never had a good baseline study.

    Time will tell.

    • Sceptical Wombat

      “Arctic Summer Sea Ice won’t greatly affect New York City”
      It won’t flood Manhattan but it may well result in major changes to its weather. Judith Curry and Co for instance have claimed to show a strong relationship between open water in the Arctic in Autumn (Fall if you like) and snowfall in North America and Eurasia. There is at least a theory that the warming of the air over the open water is resulting in a slow down of the Arctic jet stream and a consequent increase in its meandering and a reduction in the speed in which these meanders move around the globe. This in turn can result in prolonged hot and cold spells depending on which side of the jet stream you find yourself on.

  12. Pete Dunkelberg

    When the sea ice goes, it is much easier for the land ice to slide into the sea, and then sea level rises.

  13. Put the rate of change into millennium-scale perspective…

    (even not yet updated for 2012)

  14. “Arctic summer sea ice loss won’t greatly affect New York City.”

    I know the idea’s a bit speculative, and even if true a ‘one-off’ (probably!)–but if the atmospheric circulation late last October was indeed ‘set’ to some degree by the low ice extent, then Arctic summer sea ice loss already did greatly affect New York City…

    • Andrew Dodds

      Here in the UK, it’s almost as if the local climate went through a ‘step change’ – we’ve suddenly had a series of cold winters after a longish period of mild ones.

      The UK is an interesting bellwether for this, because we are a surprisingly long way north (As is Hudson bay-ish); so in winter the weather is dominated by heat transport rather than isolation. So it looks like we are already seeing effects from Arctic sea ice loss..

  15. Why limit ice survey to the pole? All ice in the northern hemisphere would be a more meaningful measure.

    [Response: Despite the title of the post, these data actually do include all sea ice in the northern hemisphere. They do not include landfast ice.]

  16. Shifting baselines: The proper baseline for ASI is circa 1953, which is when it started to decline, as you can see here: http://tinyurl.com/bhvhahx

  17. Sorry for two posts on the same topic, but notice extent minimum had already dropped 2m km2 by 1979.

    We have the same problem with measuring biota populations post-industrial – and even long before – too. We should measure the baseline no more than before 1850 or so for biota.

  18. Peter Wright

    Speaking as a biologist, watching the catastrophic collapse of the arctic ecosystem and seeing what rises out of the ashes will be fascinating. Intellectual curiosity is such a good way of keeping the fury and the heartbreak within limits, don’t you find?