Arctic Sea Ice

Most discussion of the amount of sea ice in the Arctic focuses on the annual minimum (there’s a very strong seasonal cycle). It certainly has declined dramatically, in a way unprecedented for at least more than a thousand years:

But the summer/fall reduction isn’t the whole story; wintertime sea ice has declined too, although not as much as the summer/fall extent. This year has brought winter sea ice extent to new lows.

The recent failure of one of the polar-observing satellites has made it harder to track Arctic sea ice according to NSIDC (National Snow and Ice Data Center). Fortunately, the data from JAXA (Japan Aerospace eXploration Agency) doesn’t depend on that particular satellite, so they’ve continued to report on a daily basis. Unfortunately, the JAXA data don’t go back as far in time as the NSIDC data.

I decided to use JAXA data to fill in the missing recent stuff from NSIDC. This can’t be done by simply aligning the records, because their difference isn’t a simple constant — as is clearly seen by a direct comparison:


NSIDC data is in black, JAXA in red. We can see their difference more clearly by expanding the time axis to highlight their period of overlap:


Clearly, the difference is larger when sea ice is at its maximum than when it’s at minimum. If we compare the two directly, we see that they track each other quite well:


By comparing the JAXA extent to the difference, we see that it depends on how much ice there is:


The red line is a lowess smooth, and I used it to estimate a conversion from JAXA extent to NSIDC extent, then completed the NSIDC data record based on JAXA data during the missing recent period. That gave this:


The short red segment at the end is the extended section.

With the extended NSIDC data covering the time span from 1979 to the present, we can plot Arctic sea ice extent year-on-year:


The most recent data is shown by the thick red line, and reveals that this year (2016) we’ve seen less Arctic sea ice during winter than in previous years. We can transform sea ice extent to extent anomaly, removing the annual cycle (I used the entire time span as a baseline period), and see this even more clearly:


This also shows the greater degree of variation during summer/fall than winter/spring. Arctic sea ice extent, at a million km^2 below “average” right now, is lowest on record for this date, but doesn’t match the autumn record low, nearly three million km^2 below “average.”

Whether or not the record-setting low winter extent is a harbinger of a new extreme when we reach this year’s minimum, remains to be seen.

Generally, one computes anomaly by estimating what the average seasonal cycle looks like, then subtracting that from the data to remove the seasonality. When we do so, we see that the recent anomalies show quite a bit of fluctuation:


That’s because the annual cycle itself has changed. When we subtract the average annual cycle to generate anomalies, the difference between today’s annual cycle and its average still remains — there’s a “residual” annual cycle to the anomalies themselves recently.

We can remove that effect by computing an “adaptive” seasonal cycle, one which changes over time to mimic the changes in the annual cycle itself. It looks like this:


When we remove that from the data, we have a less variable anomaly estimate:


This blog is made possible by readers like you; join others by donating at Peaseblossom’s Closet.

6 responses to “Arctic Sea Ice

  1. It’s also worth noting that your first graph, from SkS, is already out of date, since late-summer minimums are about 5 million km² these days, a LOT lower than even that reverse-hockey-stick shows.

    [Response: The opening graph is from a paper by Kinnard et al. If I recall correctly, it’s of “summer mean” rather than late-summer minimum, but I could be mistaken.]

  2. Isn’t it almost time for the how low can it go prediction contest for this year?

  3. Can you give more details on how you calculate an “adaptive” seasonal cycle?

    [Response: There are many ways. For example: the folks who compute anomalies for CO2 data from Mauna Loa base the annual cycle on the nearest 6 years’ data for each value, hence the estimated seasonal cycle changes over time. What I do is estimate a time-variable cycle by fitting a multi-frequency (rather than single-frequency) Fourier series with the multiple frequencies closely spaced; that leads to amplitude modulation of the estimate. One can also directly estimate a smoothed version of the annual cycle’s amplitude and shape (say, with wavelet or windowed Fourier analysis), then directly reconstruct a time-variable estimate of the seasonal pattern.]

  4. You really are very good this R stuff :-)

    Clear and informative, as ever.

  5. NSIDC’s ‘charctic’ interactive sea ice chart is running again with provisional data for the last month and a half. It’s a simple interactive that plots daily sea ice extent for each year since since 1979.