Roger Pielke’s post which we criticized now has seven updates. Seven! He has protested, I would even say whined, that I and my readers have treated him unfairly. He has accused me of a lack of “professional courtesy” for such horrible deeds as blogging under a pseudonym. He even went to the trouble to dig up my real name and post my hometown location on his blog. How professionally courteous of you, Roger. That certainly advances our understanding of sea ice trends.
What he still hasn’t done is: the math.
Let’s be absolutely clear what Pielke’s post, and my criticism, are about. He’s trying to test the predictions of trends in northern hemisphere annual average sea ice extent from Vinnikov et al. (1999, Global Warming and Northern Hemisphere Sea Ice Extent, Science, 286, 1934-1937, doi:10.1126/science.286.5446.1934). Pielke claims that since 2006, the trend has “stopped and even reversed.”
Pielke didn’t even look at data for sea ice extent like Vinnikov et al., he only showed data for sea ice area, and he never looked at annual averages like Vinnikov et al., he only looked at daily data, and he never looked at raw area data he only looked at anomalies. None of that bothers me because Vinnikov et al. are interested in the trends, and the trends in anomaly are pretty much the same as those in annual averages, the trends in area are pretty much the same as those in extent. What does bother me is that Pielke saw fit to criticize a reader’s comment because the annual averages of the data Pielke used didn’t match the observations, or values for the GFDL model, plotted by Vinnikov et al. (who explicitly state that the estimates from the Hadley model are too low). Apparently he wasn’t even aware that he was averaging area and comparing that to extent, and for their absolute annual averages they tend to differ by about 2 million km^2. He’s aware of it now — but that seems to be the only mistake he’s willing to own up to.
Now Pielke suggests that I should analyze area data as well as extent data. I already did. Re-read the post and look at the 2nd and 3rd graphs. The 3rd is based on extent data from NSIDC, but the 2nd is based on area data from Cryosphere Today — the same data Pielke used.
Pielke has also protested that I’m ignoring Antarctic sea ice, and insists that I should analyze “insolation-weighted sea ice.” Why are you trying so hard to change the subject? Reminder: it’s about trends estimated by Vinnikov et al. 1999 of northern hemisphere annual average sea ice extent. Not Antarctica, not area, not insolation-weighted area, and not short-term trends anyway. Another reminder: you are the one who chose the subject. Considering that your claims have been shown false, even foolish, I guess we already know why you’re trying so hard to change the subject.
It’s very revealing that Pielke chooses to focus on the data since 2006. That’s a total of just a smidgen over 6 years of data. Yet he seems to want to compare recent short-term trends to those quoted in Vinnikov et al., for which the shortest time span having a reported trend was 16.8 years, and the computer model trends (which Pielke himself quotes for comparison purposes) are based on at least 21 years data. It’s the same old fake-skeptic trick we warned of, be especially wary of time spans that are too brief and areas that are too small.
The hilarious part is that even if you do allow this time span which is too brief, the trend is still squarely in the range given for recent trends by Vinnikov et al. I know because I already did the math. ROTFLMAO!
But let’s give Pielke credit for one thing. He did identify an interesting point in time related to Arctic sea ice extent. He utterly failed to realize that it’s not a “breakpoint” of the trend and certainly not a “stopped and even reversed” point of the trend, because he never bothered to do the math and find out what the trends are (in fact when we did it for him he refused to believe it).
It’s time to ignore Pielke altogether, because we actually are interested in what has been happening with Arctic sea ice. First let’s do as a reader suggested and compare observations to the model output reported in Vinnikov et al. 1999. Here’s Vinnikov’s graph, on which I’ve superimposed annual average extent from NSIDC data (in red):
I also put in a dashed line at 1999. Clearly, since then Arctic sea ice has declined more than expected according to Vinnikov et al. Quite a bit more. And although it shows up-and-down fluctuations on short time scales — just like the model output does! — there’s no sign of a stop, or reversal, of the trend. If you don’t believe me, do the math.
Nonetheless, something interesting happened around 2006. This is evident from the graph of anomalies:
The pattern of changes is different since that time. But is that a sign of a fundamental change in the trend? Not according to the math.
We can discover what it really means with a well-chosen wavelet analysis of the extent data (not anomalies but actual extent). Here’s the mean value according to that analysis:
No surprise there. We see short-term fluctuations — just like in the model output — but the trend continues downward. Now look at the plot of the amplitude (actually semi-amplitude) of the annual cycle (actually of the best-fit sinusoid to the annual cycle):
Bingo! That’s what really changed most recently (actually from about 2006 to 2008) — the amplitude of the annual cycle has increased by about 1 million km^2 (semi-amplitude increased by about 0.5 million km^2).
We can note the same thing by de-trending the data, then plotting each year on top of all the others, to compare the annual cycles apart from the trend. We’ll plot the cycles since 2006 in red:
Again we see that it’s the size of the annual cycle which changed.
This change is visible in the plot of anomalies because of the way anomalies are calculated. The usual way is to take each value and subtract the average value for the same time of year. For monthly data, for instance (and I’m using monthly extent data from NSIDC), we take each month’s value and subtract the average for that same month. That gives the anomaly plot already shown.
This has the effect of removing the average annual cycle from the data. But if the annual cycle changes, we won’t actually have subtracted the current annual cycle, we’ll have removed the average — and the difference between the current annual cycle and the long-term average annual cycle will remain. That’s why, after about 2006, the anomaly graph itself appears to show an annual cycle. We’re seeing the “leftover” annual cycle after removing its average.
There are ways to estimate a time-varying annual cycle, and remove that from the data, in order to compute anomaly values which take this effect into account. Here’s one example of anomalies computed this way:
We can plainly see what has happened to Arctic sea ice extent. There is a long-term decline — the trend — which continues, and which is faster now than it was just a few decades ago. There was an extreme dip in 2007, when the summer minimum reached record-low values. There was a “hiccup” after 2007, very much like the short-term fluctuations seen throughout the computer model results. And: there was a substantial increase in the size of the annual cycle.
Finally, what about testing the predictions of Vinnikov et al. 1999? Mostly the paper is about comparing computer model simulations to observed trends (using data available as of 1999). There’s actually very little discussion of prediction for future changes in Arctic sea ice extent. In fact, just about the only thing Vinnikov et al. say about that is this:
“Both models predict continued substantial sea ice extent and thickness decreases in the next century.”
That’s one computer model prediction which, so far at least, has been completely correct.