Accessing Deep Cold in Fairbanks

Reader Gary asked about quantifying deep cold in Fairbanks, i.e. when it's cold enough to require extra effort by hardly Fairbanks folk to stay warm and keep things running. Monthly mean temperatures only hint at the question, since means, by definition, smooth out the day to day variation. So here's an effort at addressing this question.

I start by calculating a "Bitter Degree Day" (BDD) statistic for each day. I've chosen a daily mean of -30F as the threshold, based on my own experience: daily means colder than -30f require some extra work. Recall that the daily mean is simply the average of the high and low, so a day with a high of- 30F and a low of -40F would have a mean of -35F. Like most degree day calculations, values outside of the threshold are set to zero: there are no negative degree days. So for each day in the winter, if the mean temperature is ≥ -30F, the BDD is zero. If the mean is less than -30F then I subtract -30 from that value and the remainder is the BDD.
Doing this uncovers some interesting nuggets. For example, this past December, there were 73 BDD, a smidgen more than the 70 BDD in December 1980, even though the monthly mean last month was nearly 7 degrees warmer than December 1980. This reflects the fact that the coldest days this past month were colder than in December 1980 AND the cold snap broke after Solstice last month, but not til the 30th in 1980. In fact, the January correlation between monthly mean temperature and BDD is a less than might be expected 0.69: good, but not great (the correlation of, say, mean temperature and conventional Heating Degrees Days would be 1.00, since the daily mean is always less than 65F in January). 
 
Winter totals have ranged from none eight times to 570 in 1933-34. In the past 30 years the highest total is 172 in 188-89, all of which were in that memorable January. Here is a plot of the winter (Oct-Mar) totals for each winter 1929-30 through this winter (thus far):   

This plot reveals that almost 20% of winters have very few (or no) BDDs, while very few (and none since 1971) have had more than 200. Except for the outlier high values, there is not much of trend, though the positive PDO phase (1977-2007), with many very BDD totals, is evident.

Since 1929-30, the monthly BDD median and maximum values are:

Nov:  0, 18 in 1994
Dec: 6, 237 in 1961
Jan: 24, 369 in 1934
Feb: 0, 84 in 1950
Mar: 0, 8 in 1956 (only March with >0 BDD)


Now the daily BDD is a measure of the mean temperature on the coldest days, and there is no doubt that the growth of the Fairbanks urban area and modern transportation have impacted very cold temperatures (mostly by providing water vapor for ice fog development). So instead of total BDD, how about number of days with a daily mean of <-30f: p="p">

The correlation between number of days and the seasonal total is quite good: 0.91 for the whole period, and still 0.88 since 1977, which can arguably be considerable the modern era of Fairbanks urban influences on temperatures.

So, this little exercise demonstrates that a focused statistic (like BDD) can get context specific information: climatology has a lot more to offer the informed user than just averages and normals.