2012 Arctic sea ice minimum, part 3: Arctic sea ice death spiral continues

The eleventh domino has  fallen.

The extraordinary 2012 Arctic sea ice melt has resulted in a September average sea ice extent of 3.61 million sq km, according to the latest monthly data from the National Snow and Ice Data Center (NSIDC), smashing the previous record of 4.30 million sq km set in 2007.

Today, I’ll quickly review the last month’s progression. I’ll then examine the plausible future course of the Arctic sea ice “death spiral” that is likely to see the Arctic virtually free of sea ice by the 2030s if not sooner, culminating with a new graphic representation of the Arctic sea ice death spiral.

At the end of August, I projected September sea ice extent of 3.56 (+/- 0.13) million sq km, based on a simple fit of  the most recent available 5-day value to the  eventual September extent, a model that had given good results over the last decade. This year was no exception, although an early flattening and moderate refreezing over the last half of September (see below) pulled the extent up slightly, albeit well within the uncertainty range of the simple model.

As the month progressed, the short-term projection (based on a fit to month-to-date described and justified as preferable in a comment here), climbed to just above 3.6 million sq km and stayed there.

So what now? When will the Arctic be virtually free of sea ice (defined as a September sea ice extent below 1 million sq km)?

First of all, I doubt that 2012 represents as clear a “regime” change as 2007 did, although it seemed to catch experts somewhat by surprise.  The 2007 September extent smashed the previous 2005 record by 23%, a scant two years later. This year’s record eclipsed 2007 by a lesser percentage (16%), and came five years on. Moreover, a quadratic fit of observations through 2012 (orange dashed curve in the chart below) is indistinguishable from that up to 2008 (green curve).

One surprise (for me anyway) was how well the 5-year average matches the quadratic fit over the last 20 years. It’s also interesting to note that 2012 is well below the quadratic curve, but not nearly as much as 2007 (or at least two years in the 1990s for that matter).

And it was already clear before 2012 that sea ice loss was accelerating much faster than the projections of the CMIP3 set of sea ice models used in IPCC AR4, (or even the CMIP5 AR5 models for that matter). 2012 does confirm that the downward trend is still accelerating, however.

The NSIDC team recently published an overview of the CMIP3, CMIP5 and observational record that makes the salient key points (Trends in Arctic sea ice extent from CMIP5, CMIP3 and observations, Stroeve et al, GRL, Aug 25, 2012).

• CMIP5 models continue to underestimate rate of sea ice loss
• CMIP5 models are more consistent with observations than CMIP3
• CMIP5 suggests 60% of 1979-2011 rate of decline is externally forced

A recent presentation by Julienne Stroeve and Andrew Barrett, Assessment of Arctic Sea Ice in the CMIP5 Climate Models, contains this key graphic comparing model simulations and observations of September sea ice ice extent, to which I’ve added the September 2012 value.

Although CMIP5 (red line) does match quite a bit better than CMIP3 (blue), both 2007 and 2012 lie outside the uncertainty range of the CMIP5 model ensemble.

It’s also worth noting that the model ensembles show continued acceleration beyond the 4 million mark, so it is reasonable to postulate some continued acceleration of sea ice loss. But the models also show marked deceleration as ice free conditions are approached.

With these observations in mind, I propose a simple projection based on extending the current observational record that could be useful for discussion, if nothing else. My starting point is a quadratic fit to observations that Tamino proposed at his Open Mind blog as a way to project September sea ice extent a year in advance. However Tamino has also been quite explicit that such a model is not appropriate to project long term trends. Indeed, simple extrapolation of that trend would result in zero sea ice extent just after 2025, a decidedly unphysical result.

Clearly any reasonable projection or extrapolation will need to incorporate a deceleration of sea ice loss similar to that seen in the models.

To explore this idea further, I now turn to a just-published analysis by Muyin Wang and James Overland (A sea ice free summer Arctic within 30 years: An update from CMIP5 models, GRL Sept 25, 2012).

Using a seven-model subset subset of the CMIP5 that matched observations well from 1980 to 2005, they concluded that “a nearly sea ice free summer Arctic is projected by the 2030s”. Wang and Overland found a median value for breach of the  1 million threshold in 2035 (i.e. 28 years from the baseline of 2007, with a range across selected models from 2021 to 2041).

In my emulation of Wang and Overland’s median projection, I’ve assumed that the quadratic fit for 1980-2005 provides a reasonable extrapolation until about 2.5 million sq km. From that point on, the projection slows considerably, but it breaches the 1 million mark in 2035, just like Wang and Overland. This approach does seem to match the acceleration/deceleration profile of the CMIP5 subset fairly well, while retaining a close fit with the observations to 2005. The result can be seen in the following chart.

However, even though Overland and Wang use only the best matching CMIP5 models, there is still a marked divergence after 2005. The current fit to observations stands 700,000 sq km lower  than the emulated median projection, with 2012 still another 700,000 sq km lower. Mind you, the spread of the Overland and Wang subset is considerable, and they do not rule out nearly ice free conditions in the 2020s.

Now I’ll use the same approach as above, but this time I’ll use the quadratic fit to 2012, instead of 2005, for the first part of the projection. (As previously noted, this fit has quite stable since 2007). In this scenario, deceleration of ice loss starts in 2020, which is when the projection reaches about 2.5 million sq km.

This suggests that nearly ice free conditions are plausible, even likely, before 2030. If we postulate a similar uncertainty spread as in Wang and Overland, nearly ice free conditions could even come by 2020 (but could also be delayed into the 2030s).

The term “death spiral” was first invoked by NSIDC director Mark Serreze to describe the rapid and shocking sea ice loss in the north. Recently, Andy Revkin argued that a descent to nearly ice free Arctic seas by 2030 or later, should not really be characterized as a “death spiral”.  But when one considers the ongoing rapid loss, which has reduced sea ice to levels not seen in hundreds of years or even millenia, it seems absurd to argue about whether such a term should be reserved for the case of a 40-year period to breach the 1 million mark (as Revkin appears to imply), rather than 50 or 60 years.

For me, portraying the above observed fit and projection scenario as an actual spiral over 50 years has driven home this point. In the following chart, each revolution of the spiral represents one decade.

Would an extra spiral turn to get to the “nearly ice free” centre change anything? I don’t think so.

The serious implications of rapid Arctic warming seem clear enough to most scientists,  but have yet to penetrate public consciousness. That is surely due in no small part to concerted past and recent efforts to downplay the significance of rapid Arctic sea ice disappearance. In future posts, I’ll cover some of those efforts by the Fraser Institute, the Frontier Centre for Public Policy, as well as Lawrence Solomon and, yes, Richard Muller.

They and many other naysayers refuse to face two basic facts: Arctic sea ice has fallen to levels unprecedented in millenia [Kinnard et al, 2011, Nature] and that loss is mainly due to anthropogenic global warming   [ J J Day et al, 2012 Environ. Res. Lett.].

And, oh yes, Arctic sea ice is indeed in a death spiral.