One of the interesting things about being a scientist is seeing how unexpected observations can galvanize the community into looking at a problem in a different way than before. A good example of this is the unexpectedly low Arctic sea ice minimum in 2007 and the near-repeat in 2008. What was unexpected was not the long term decline of summer ice (this has long been a robust prediction), but the size of 2007 and 2008 decreases which were much larger than any model had hinted at. This model-data mismatch raises a number of obvious questions – were the data reliable? are the models missing some key physics? is the comparison being done appropriately? – and some less obvious ones – to what extent is the summer sea ice minimum even predictable? what is the role of pre-conditioning from the previous year vs. the stochastic nature of the weather patterns in any particular summer?
The concentration of polar expertise on the last couple of questions has increased enormously in the last couple of years, and the summer minimum of 2009 will be a good test of some of the ideas that are being discussed. The point is that whether 2009 is or is not a record-setting or near-record setting minimum, the science behind what happens is going to be a lot more interesting than the September headline.
In the wake of the 2007 minimum, a lot of energy went in to discussing what this meant for 2008. Had the Arctic moved into a different regime where such minima would become normal or was this an outlier caused by exceptional weather patterns? Actually this is a bit of false dichotomy since they aren’t exclusive. Exceptional patterns of winds are always going to be the proximate cause of any extreme ice extent, but the regime provides a background upon which those patterns act. For instance, in the paper by Nghiem et al, they showed the influence of wind patterns in moving a lot of thick ice out of the Arctic in early 2007, but also showed that similar patterns had not had the same impact in other years with higher background amounts of ice.
This ‘background’ influence implies that there might indeed be the possibility of forecasting the sea ice minimum a few months ahead of time. And anytime there is the potential to make and test predictions in seasonal forecasting, scientists usually jump at the chance. So it proved for 2008.
Some forecasting efforts were organised through the SEARCH group of polar researchers, and I am aware of at least two informal betting pools that were set up. Another group of forecasts can be found from the Arctic ice forecasting center at the University of Colorado. I personally don’t think that the intrinsic worth of a successful prediction of overall sea ice extent or area is that societally relevant – interest in open shipping lanes that might be commercially important need much more fine-grained information for instance – but I think the predictions are interesting for improving understanding of Arctic processes themselves (and hopefully that improved understanding will eventually feed into the models and provide better tests and targets for their simulations).
What was particularly interesting about last years forecasts was the vast range of forecasting strategies. Some were just expert guestimates, some people used linear regression on past data, some were simply based on persistence, or persistence of the trend. In more mature forecasting endeavours, the methods tend to be more clustered around one or two proven strategies, but in this case the background work is still underway.
Estimates made in June 2008 for the September minimum extent showed a wide range – from around 2.9 to 5.6 M km2. One of the lowest estimates assumed that the key criteria was the survivability of first year ice. If one took that to be a fixed percentage based on past behaviour, then because there was so much first year ice around in early 2008, the minimum would be very low (see also Drobot et al, 2008). This turned out not to be a great approach – much more first year ice survived than was predicted by this method. The key difference was the much greater amount of first year ice there was near the pole. Some of the higher values assumed a simple reversion to trend (i.e. extrapolation forward from the long-term trend to 2008).
Only a couple of the forecasts used physics-based models to make the prediction (for instance, Zhang et al, 2008). This is somewhat surprising until one realises how much work is needed to do this properly. You need real time data to initialise the models, you need to do multiple realisations to average over any sensitivity to the weather, and even then you might not get a range of values that was tight enough to provide useful information.
So how did people do? The actual 2008 September minimum was 4.7 M km2, which was close to the median of the June forecasts (4.4 M km2) – and remember that the 2007 minimum was 4.3 M km2. However, the spread was quite wide. The best estimates used both numerical models and statistical predictors (for instance the amount of ice thicker than 1m). But have these approaches matured this time around?
In this year’s June outlook, there is significantly more clustering around the median, and a smaller spread (3.2 to 5.0 M km2) than last year. As with last year, the lowest forecast is based on a low survivability criteria for first year ice and I expect that this (as with last year) will not pan out – things have changed too much for previous decades’ statistical fits on this metric to be applicable. However, the group with the low forecast have put in a ‘less aggressive’ forecast (4.7 M km2) which is right at the median. That would be equal to last year’s minimum, but not a new record. It would still be well below the sea ice trend expected by the IPCC AR4 models (Stroeve et al, 2008).
There is an obvious excitement related to how this will pan out, but it’s important that the thrill of getting a prediction right doesn’t translate into actually wanting the situation to get worse. Arctic ice cover is not just a number, but rather a metric of a profound and disruptive change in an important ecosystem and element of the climate. While it doesn’t look at all likely, the best outcome would be for all the estimates to be too low.