Following up to yesterday's quick lunchtime ice volume post. This morning I processed the data a bit differently by computing the average for each month in each year and then plotting the monthly series separately. The results are above. For the months July-December I fit a quadratic - which gives a decent fit in all cases.
The extrapolations show the Arctic ice free for six months out of the year by 2025.
A word about extrapolations, which are always dangerous. They get more dangerous when:
- There is no particular theoretical basis for the curve used as a model.
- The model has a lot of parameters (so that it has the opportunity to fit tightly in the sample period but go haywire outside of it).
- The extrapolation goes for a long way relative to the interval over which there is data.
- There is reason to suspect the underlying dynamics of the system might change during the extrapolated period.
- The underlying data are suspect, particularly if there are systematic errors that the fit is sensitive too.
In this case, we have no theoretical model - this is purely an empirical fit. (The theoretical models of the Arctic are climate models, which uniformly failed to predict the rapid and early collapse in sea-ice volume that has actually been observed - so they are presumably missing something important about the dynamics). On the plus side, we only have three parameters - a quadratic is the simplest possible model of an accelerating process, and this process is clearly accelerating. Furthermore, the quadratic gives a decent fit for all months, it's not just a fluke that it happened to fit in September. The resulting extrapolations tell a fairly consistent plausible story: ice will disappear first in September, with October and August very close behind, and then November and July, followed by December.
For the extrapolation interval, it appears to be quite short for August - October - around 15% of the data interval. We don't need the system to keep behaving the same way for very long at all in order for the extrapolated curve to go to zero. For the succeeding months the extrapolations obviously get more and more speculative.
We don't particularly have any reason to expect the dynamics to change soon. The Arctic has long been expected to warm dramatically under climate change, and all indications are that it has indeed been warming and will continue to warm. There is no historical evidence of climate fluctuations this large in Arctic ice. However, it's possible that part of the recent move is a fluctuation that could revert and delay the inevitable. In particular, look at the recovery in ice volume in the early 1980s - a similar recovery in the next few years could push things out by a decade.
Another way to assess the extrapolation is to do a stability analysis by fitting only a subset of the data. If we use all September data (ie to 2011 as of today) then we extrapolate to 2017, but if we stop the fit in 2010, or 2005, what do we get? That looks like this:
Back in the late 1990s, the prediction was very unstable year to year. However, by the 2000s it had settled into the early 2020s, and as more data has come in, the prediction has actually been coming down slightly to its present 2017 (with the caveat that we have 2012 still to come in September). When the blue curve hits the black curve, then it's over. You can see that commenter Stephen B's instinct that this could even be before 2015 is not crazy - though it's hardly a near-certainty.
Finally, the data quality is probably worth a post of its own at some point. In the meantime you can see Piomas's own discussion of their uncertainties here (also see this RealClimate post). There certainly is some uncertainty, but they believe their trend is likely conservative and in any case the uncertainty is not large enough to change the qualitative picture of a fully watery north pole in late summer pretty soon.