Arctic sea ice has become one of our bellwethers, in part because for the past 30 years we have been able to watch it expanding and retreating nearly in real time, but also because it definitely obeys the laws of physics.
Recently the extent of Arctic sea ice has decreased fairly steadily. It is also notable that 2007 was a year of record minimum extent (in September), prompting conjecture about the Arctic becoming ice-free soon, with an ice-free September perhaps as early as 2013. However 2008 did not quite match the 2007 record. This year, and the next couple of years, will show us whether 2007 was a blip or the start of something even more serious than the 30-year trend.
I still remember our geography teacher at school starting to teach us climatology by saying “The climate is the average weather”. He went on to tell us about the convention of presenting the climate as “normals”, or averages, over 30 years. One bad summer simply doesn’t add up to a climatic change. But it seems that it is human nature both to get rattled when a record is broken and to forget about the problem when the record-breaking behaviour is not repeated.
Equally, and unfortunately, it is not in human nature to worry much about systems like the climate that evolve over 30-year time scales. It is too easy to make the mistake of thinking that bad things aren’t going to happen for 30 years.
Knowing that sea ice obeys the laws of physics, we should expect replacing bright sea ice with dark open water to illustrate the idea of feedback. The less reflective the surface, the more the radiative heating, and the more the loss by melting, and the less reflective the surface … . In other words, the future of sea ice could start to look much, much worse quite suddenly. However, knowing too that we don’t know everything, we should be concerned about the inability of climate models to agree on the future of sea ice.
Sea-ice predictions come from a couple of dozen large-scale climate models. They diverge wildly over the next century. Many of them don’t even reproduce the recent evolution of sea ice, a clear indication that the modellers have development work to do. Two recent analyses show just how unsure we are about what is coming, but both also offer interesting ideas about how to cope with uncertainty as manifested in poor model performance.
Boé and colleagues point out that the models that do the best job of simulating the past 30 years are also the models that predict the earliest disappearance of sea ice. Using the A1B scenario for greenhouse-gas emissions, considered middle-of-the-road, this happens in about 2060-80 if we agree that disappearance means dropping below 10% of the 1979-2007 average in September. Boé and colleagues explain this observation in terms of the models’ accuracy in describing the proportion of the ice that was thin to begin with, and therefore more at risk of disappearing altogether.
Wang and Overland selected the best-performing models by requiring them not only to come within 20% of the September observations for 1980 to 1999, but to beat the same target for the range of extents observed during the whole year. The thinking is that a model is likely to be more trustworthy if it matches more of the firm evidence. They find, with the six models that qualify, that September sea-ice extent is most likely to drop below about 10% of the recent average in 2030-2050.
Comparing the two studies, Mat Collins prefers to put his money on the later Boé estimate than on Wang and Overland. His reasons are cogent, but I think that the crucial point is to exclude the models that obviously get the recent history of sea ice wrong, something Boé and colleagues do not do.
Our understanding of the laws of physics already gives us the message “Sooner or later”, but focusing on the models that are not obviously wrong – not the same thing as obviously right, of course – the message becomes “Sooner rather than later”.