Southern Discomfort

In AR4 (the 4th assessment report of the Intergovernmental Panel on Climate Change) the trend in Antarctic (southern hemisphere) sea ice was reported as small (5.6 +/- 9.2 thousand km^2/yr) and not statistically significant, but in AR5 (the 5th assessment report) it is reported as both statistically significant and much larger (16.5 +/- 3.5 thousand km^2/yr). Even at this rate the Arctic is still losing sea ice 3 times as fast as the Antarctic is gaining it, but the larger trend is still surprising; such a large rate of increase is, more and more, turning out to be incompatible with computer model simulations.

A new paper submitted to the Cryosphere Discussion (Eisenman, I., Meier, W. N., and Norris, J. R.: A spurious jump in the satellite record: is Antarctic sea ice really expanding?, The Cryosphere Discuss., 8, 273-288, doi:10.5194/tcd-8-273-2014, 2014) suggests that much, if not most, of the upward trend in southern hemisphere sea ice may be due to a spurious jump caused by an undocumented change to how the data are processed. It also explains the dramatic difference in the state of affairs between what was reported in AR4 and what was in AR5 just a few years later.

It turns out that the difference between AR4 and AR5 results was not just because of the addition of a few extra years’ data, it was also (in fact primarily) because the data set itself was revised. The latest is called “version 2” and was published in 2008 by Comiso and Nishio (Comiso, J. C. and Nishio, F., 2008. Trends in the sea ice cover using enhanced and compatible AMSR-E, SSM/I, and SMMR data, J. Geophys. Res., 113, C02S07, doi:10.1029/2007JC004257). They reported at the time that it made little difference to the trend in Antarctic sea ice. But comparison with the trend reported in AR4, and deduced from a “version 1” data set, show that this is not the case. The conclusion is that there was another revision, to version 1 only, which came after the research on which AR4 based its results but before Comiso and Nishio implemented the revision which brought us to version 2.

Eisenman et al. directly compared the available data for version 1 (the pre-revision data) and version 2, plotting the difference (v2 minus v1) in their figure 2:

The “step change” nature of the difference is visually evident — quite strikingly so — and easily confirmed statistically. Clearly there is a difference between data before and after December 1991 which is due to the change in the way the raw data were processed. Perhaps most suggestive is that the step occurs at precisely one of the moments when there was a transition between different sensors (indicated by the vertical dotted lines). This is too implausible to be considered a coincidence. They point out that

… subtracting 0.15×10^6 km2 from Version 2 in all months after December 1991 causes the trend to be nearly equivalent to Version 1 for the range of endpoints plotted in Fig. 1b (not shown). Hence the issue appears to be associated with erroneous calibration across the December 1991 sensor change in one of the two Bootstrap versions.

The big question is, as they say, “whether the undocumented change from Version 1 to Version 2 introduced an error or removed one,” and they try to determine which is the case but none of their methods resolves the issue unequivocally. They do speculate, pointing out that if v1 is correct then there is less growth of Antarctic sea ice and therefore greater agreement with climate models, which favors v1 being more correct.

I don’t find that argument particularly persuasive. I’m not one to bash climate models, but as far as I know, the change in sea ice is one of their weaknesses, not their strenghts.

I have, however, found what may be evidence to support their contention. I took the southern hemisphere sea ice extent from NSIDC (Nation Snow and Ice Data Center) which uses the version 2 bootstrap algorithm they studied, computed anomaly (to remove the seasonal cycle), and smoothed it on a short time scale (about 5 years). I got this:

Note that there appears to be a step change at about the time of the step change in difference between v1 and v2, and this is derived from studying the v2 data all by itself.

The step change in the smooth looks to be closer to 1994 than 1992, but if there were an actual dip around 1993, that’s exactly what we would see. I even took the v2 data and subtracted 0.2 million km^2 from all data after December 1991 (even more than the 0.15 million km^2 subtracted by Eisenman et al.) to see whether this “adjusted” data (with even larger adjustment than they suggest) would throw things out of whack. It did the opposite, making things more consistent and constant:

I consider the time evolution shown in the adjusted data to be more plausible than the unadjusted data. This suggests that it’s more likely the error was introduced in the undocumented revision just prior to v2, so there is still an erroneous transition between sensors around December 1991.

On a different topic entirely: if you’re reading this but not at its blog of origin (tamino.wordpress.com), and at the top is a line saying “Written by Amego Sando,” then you should be aware that that is a lie. This post was NOT written by Amego Sando, he’s just a thief who steals other people’s blog posts and re-posts them while claiming credit for authorship. Oh — and if you actually happen to be Amego Sando and you’re reading this: va te faire mettre.

I do take exception with one statement in the Eisenman et al. paper. As early as the abstract they state:

The results of this analysis raise the possibility that this expansion may be a spurious artifact of an error in the satellite observations, and that the actual Antarctic sea ice cover may not be expanding at all.

Note: they don’t claim that there’s no increase in Antarctic sea ice, they only raise the possibility. But I don’t consider that realistic at all. If I apply the 0.2 million km^2 adjustment, an even bigger “correction” (if it is one) than they suggest, it reduces the trend but doesn’t eliminate it. Furthermore, the increase is still statistically significant. Using the v2 data without adjustment I get a linear trend rate of 17.6 +/- 8.0 thousand km^2/yr, when the even-bigger adjustment is applied the rate decreases to 9.7 +/- 4.7 thousand km^2/yr — far less growth, but still statistically significant.

Bottom line: it seems to me that the evidence shows convincingly there is a discontinuity in at least one of the data sets at the moment of one of the sensor transitions. There is the distinct possibility, maybe even a “more likely than not,” that the data currently favored (v2) has a spurious increase which exaggerates the apparent trend. But even if that is the case, Antarctic sea ice still shows statistically significant increase since satellite observations began. And it’s well not to forget that however much sea ice the Antarctic may have increased, the Arctic has lost a whole lot more sea ice than the Antarctic has gained.