It was mentioned in recent discussion that Cohen et al. (2012) found recent winter cooling in much of the boreal (northern but not necessarily Arctic) northern hemisphere. In fact, here’s their key graph for that particular question:
Colors show the wintertime temperature trend from 1988 through 2010, with red indicating warming while blue indicates cooling, based on the HadCRUT3 gridded temperature data set. They note cooling in northern Europe, Russia, and eastern North America.
I decided to take a look at another data set for part of the area they studied, the data from NCDC (National Climate Data Center) for the 344 climate divisions covering the contiguous U.S. If we estimate the trend (by linear regression) for the winter season (Dec-Jan-Feb) only, for all U.S. climate divisions since 1988 (which is a little bit longer than the time studied in Cohen et al., since these data go through Sep.2012 while they ended with Dec.2010), it looks like this (red is warming, blue is cooling, larger dots mean faster rates):
Sure enough, the broad swath of blue in the eastern U.S. shows cooling just as reported by Cohen et al.
A natural question is whether the cooling is statistically significant. Sure, the trend estimate is negative but is that a genuine trend or could it be random fluctuation? Let’s plot the same graph again, but use open circles for the estimates that are not statistically significant and closed circles for those that are:
Clearly the vast majority of the trends are not significant, and those that are, are few enough in number that it could be the result of randomness. By the way, I did not correct for autocorrelation but it’s much smaller for regional than for global temperature, and it’s my blog so I’m allowed to get a little lazy from time to time.
That doesn’t mean that this region didn’t cool off. But it does mean that we don’t have the evidence to call it a trend. It could be — the variance of regional temperature is quite a bit larger than that for global, so it’s harder to confirm a trend significantly. But the evidence of trend isn’t really there.
In any case, the pattern for winter (DJF) differs (whether significantly or not) from that for the entire year:
The bulk of the country shows warming since 1988, with a pocket of cooling in the far Pacific northwest. If we do the “non-significant open, significant closed” version,we see that for many of the regions the trend is significant, so yes the country as a whole has warmed since 1988 with a pocket of cooling in the far Pacific northwest:
An interesting behavior emerges if we analyze each month separately (this is an animated GIF, you may have to click the graph to see the animated version):
The eastern cooling pattern doesn’t show in December, but it does during January and February. So, it’s really more a Jan-Feb cooling than the entire winter. But again, those monthly trends aren’t really significant — here are the not-significant-open, significant-closed versions for Jan and Feb:
Despite the fact that these patterns may not be genuine trends, they are at least genuine fluctuations. Cohen et al. attempt to understand them by exploring their relationship to other recent (since 1988) climate observations.
Their proposed theory is that the root cause is warmer summers and autumns. This increases Eurasian snow cover during Autumn (especially October) due to greater atmospheric water vapor content. This in turn promotes the cold phase of the Arctic Oscillation (AO) in winter, leading to colder winter temperatures over much of the northern hemisphere.
Thus, the regional NH wintertime temperature cooling is directly tied to the declining trend in the wintertime AO, which is forced partly by increases in Eurasian snow cover during the autumn, driven in response to warmer surface temperatures at high latitudes prior to and concurrent with the autumnal advance in snow cover.
They point out that a reason climate models may have predicted otherwise, is that they don’t model snow cover variability very well:
A warmer, more moisture-laden Arctic atmosphere in the autumn contributes to an increase in Eurasian snow cover during that season. This change in snow cover dynamically forces negative AO conditions the following winter. We deduce that one main reason for models failing to capture the observed wintertime cooling is probably their poor representation of snow cover variability and the associated dynamical relationships with atmospheric circulation trends (Hardiman et al 2008, Jeong et al 2011). Incorporation of the snow cover–AO relationship into seasonal forecasts is shown to greatly improve their abilities, and hence long-term climate solutions from coupled climate models may also bene?t from improved snow–AO relationships.
It’s speculative of course, as Cohen et al. themselves point out. As we’ve noted, the reported trends don’t pass statistical significance so it remains to be seen whether the observed phenomenon itself is even meaningful, let alone their purported explanation. But they do find impressive correlations between autumn snow and winter AO, and between winter AO and northern hemisphere temperatures, in just the right regions.
Those interested in more details should read their paper. As for whether or not they’re fundamentally correct … time will tell.