It was pointed out in comments recently that Wang et al. (2012) found a cooling trend during the winter season in the Arctic (defined as the area from latitude 60N to the pole) from 1982 through 2004, using estimates of surface skin temperature from AVHRR (Advanced Very-High Resolution Radiometer) instruments aboard satellites.
In the last post we saw that none of three other data sets exhibit winter Arctic cooling. However, two of those data sets (satellite data from RSS and UAH) were measures of lower-troposphere rather than surface temperature, and for all three the time span studied started earlier than 1982 and ended later than 2004. Furthermore, the region studied wasn’t exactly the same, since I used NCAR/NCEP data from north of the arctic circle, and the satellite data sets from RSS and UAH miss the very-near polar region (although that’s only a small fraction of the Arctic). It may be fruitful to examine data for surface temperature only, for the same time span studied in Wang et al., for the same region.
To that end, we’ll examine the third data set used previously, the NCAR/NCEP reanalysis data, but this time we’ll use the data from latitude 60N to the pole. We’ll also look at the Arctic temperature estimates of zonal temperature from NASA. This covers the region from latitude 64N to the pole, and is only annual averages, but it won’t be the main source for comparison, I’ll just use it for a “reality check.”
Some people mistakenly regard “satellite” as a magic word, indicating that its data are unimpeachable, or at least clearly superior to other sources. One need only compare the data from RSS and UAH to dispel that notion. Despite the immense effort invested to correct for such issues as instrument calibration, sensor drift, and orbit variation, the two records — derived from exactly the same satellites — are in notable disagreement. It’s also well to bear in mind that AVHRR is not a satellite, it’s a type of instrument, one which has changed over the years, and that the AVHRR record is not from a single satellite but from instruments flown aboard 15 different satellites. The complications inherent in piecing together data from so many different satellites are enormous (as the RSS and UAH people can testify). And AVHRR doesn’t measure temperature, it measures reflectance in a variety of wide spectral bands; transforming that to surface temperature is a nontrivial effort. Its temperature estimates cannot be regarded as gospel, but only as estimates.
The NCAR/NCEP reanalysis data are the output of a computer model of the atmosphere, but one which is driven (and essentially controlled) by direct observations from surface stations, weather balloons, and other sources. Because it is driven by direct observations of relevant variables (like temperature measurements from thermometers), I regard it as at least as reliable as AVHRR data.
Here’s the data used in Wang et al. (temperature data in blue), for each season and annual averages, from 1982 through 2004 inclusive (winter is Dec-Jan-Feb, spring Mar-Apr-May, summer Jun-Jul-Aug, autumn Sep-Oct-Nov):
I digitized the graphs in order to obtain the numerical values used for temperature. I also computed seasonal and annual averages for the NCAR/NCEP data. First, here’s the comparison of all four seasons for the time span studied (NCAR in red, AVHRR in blue):
There’s quite a seasonal difference. The AVHRR data are considerably warmer in summer and colder in winter, hence show a much larger seasonal cycle. Clearly there’s notable disagreement between the two. In fact, here’s the difference between them:
Note the strong annual cycle in the difference, which has gotten larger over the years.
If we compare winter (DJF) season averages from the two we see that NCAR/NCEP shows a warming trend while AVHRR shows cooling:
In fact the difference has grown rather steadily:
In spring the two records are much closer to each other, not only their absolute temperature estimates but their trends as well:
In summer we find AVHRR hotter than NCAR/NCEP:
By autumn the two again show very different trends, with NCAR/NCEP indicating faster warming that AVHRR:
An interesting comparison is their estimated annual averages:
The AVHRR data show less, and less steady, warming. It also shows greater year-to-year variability. We can compare these annual averages to those from NASA GISS, which are based on actual temperature measurements but must interpolate over much of the Arctic (GISS data shown in black, offset to have the same mean value as the combined NCAR/AVHRR data):
The GISS data are in much better agreement with NCAR/NCEP than with AVHRR, especially the most recent years. I regard this as reason to put more confidence in the NCAR/NCEP data than the AVHRR data.
We can even extend that comparison to a longer time span, although I don’t have the AVHRR data outside the 1982-2004 interval:
The agreement between GISS and NCAR is outstanding, while that with AVHRR is poor. Again, this gives me much greater confidence in the NCAR/NCEP data than in the AVHRR data. It also indicates continued warming of the Arctic from 2004 to the present.
Considering this, and the results from both the RSS and UAH satellite data sets for lower-troposphere temperature, I regard the claim of winter cooling in the Arctic (60N to the pole) to be very implausible. I don’t regard it as impossible. There’s a chance — small I think, but not zero — that the limited amount of direct observations in the very northernmost part of the Arctic makes the interpolation/model simulation overestimate wintertime temperature dramatically. It’s also possible that “skin temperature” estimated by AVHRR is sufficiently different from surface air temperature (reported by NCAR/NCEP and NASA GISS) to account for the differences. But on the whole, when it comes to winter cooling in the Arctic I’d say I just don’t believe it.
Mention was also made of a paper by Cohen et al. titled Arctic warming, increasing snow cover and widespread boreal winter cooling, in the context of possible Arctic winter cooling. But Cohen et al. doesn’t really claim Arctic cooling in winter, they claim boreal (i.e., northern) winter cooling over land areas, mainly eastern North America and northern Eurasia. That’s a different subject, one which I may look into soon.