Forecasts, predictions, projections, there are some nuanced distinctions between all these terms. But 'forecasts' alliterates best with 'first', and I'm a sucker for catchy, alliterative titles. Besides, the forecasts for this year's Arctic sea ice extent minimum that I'm going to write about, are meant as stimulating statistical exercises. No more, no less. Let that be a careful caveat.
Arctic Sea Ice Extent Forecast for September 2011
In this post, I use a quadratic regression model to forecast the September, 2011 Arctic Sea Ice Extent. The model was developed with 1980 – 2010 data. Links to the R script, source data and how-to article on polynomial regression are provided.
Kelly's forecast - based on a quadratic regression model of September sea ice extent for the 1980 – 2010 period - is that September 2011 SIE will decline 0.36 below 2010 levels, to 4.54 million km^2, with a confidence band of +- 0.59. The forecast was inspired by Tamino‘s post on Arctic Sea Ice decline called 'Death Spiral', that some of us might recall. Tamino's prediction for the 2010 minimum SIE made in July came in very close to the actual number (edit: Tamino made a prediction for the 2011 SIE minimum back in October 2010, see update at the end of this blog post):
We have a couple of 'amateur' statisticians over on the Arctic Sea Ice blog as well, who often enliven comment sections with their statistical speculations. One of them is Larry Hamilton (not an amateur, but a real statistician) who contributed some inspiring graphs to last year's Area vs Extent blog post (and number 2), one of this blog's community effort highlights so far in my humble opinion.
Well, Larry has been at it again in the last SIE update blog post and posted a link to another, slightly different graph for the 2011 September average:
I'll reproduce Larry's accompanying explanation:
It's just curve-fitting, but with a slight difference. Others tend to use linear, quadratic, exponential or logistic curves. Quadratics actually rise in the early years, then later crash below zero. Logistics have the more plausible property of approaching zero at a slowing rate, but the deceleration and acceleration phases must have the same rate.
What this graph shows is a Gompertz curve, still relatively simple but yielding an assymetrical S. It looks similar to a quadratic but the differences are improvements: no rise in the early years, no crash below zero.
The Gompertz predicts a slightly lower value than the quadratic for 2011: 4.4 instead of 4.6. It crosses the 1.0 line a few years sooner, but after that approaches 0 asymptotically.
NOT that such curve fitting is more than a what-if exercise!
So there you have it, the first forecasts. I'm sure others will come soon, as we eagerly await the first Sea Ice Outlook. Just as for the SIO, Kelly's and Larry's graph are based on NSIDC monthly extent numbers for September.
Note also that you can vote on what you think will be this year's IJIS daily SIE minimum on the poll in the right side bar (and here). More on that in a few weeks time.
UPDATE: Somehow I missed a previous blog post by Tamino, or forgot about it, but he was the first to forecast the 2011 September SIE minimum back in October 2010 by fitting a quadratic trend to the September average data from NSIDC. He used "that trend to predict that next year’s September average from NSIDC, and next year’s JAXA minimum, will be 4.63 +/- 0.9 million km^2."
The blog post is called Go Ice Go! … Going … Going … Gone!!! and here's Tamino's graph: