Sunday, October 20, 2013

Temperature Deviations From Normal

The last four days in Fairbanks have all been more than 2 standard deviations from the mean. This brings the annual total to 42 days that the average daily temperature has been at least 2 standard deviations (more than +2 or less than -2) from normal. The total of 42 (through October 19th) far surpasses the record for an entire calendar year – the twice set value of 37. The values from 2011 onward were calculated based on published temperature and standard deviation normals published by NCDC. The values prior to 2011 were calculated based on normal temperatures calculated using the previous NCDC third-order spline method based on 30-year monthly average temperatures. The daily standard deviation calculation for dates prior to 2011 were determined according to Arguez (2011) which were also used by NCDC to calculate 2011 and onward daily standard deviations.

If temperature data were exactly normally distributed, we would expect that 16.6 days per year would be +/- 2.0 standard deviations from the mean. In actuality, the long term average for Fairbanks (and Anchorage) is 14.6 days per year. I believe this discrepancy is due to a few large outliers pushing the standard deviation values higher than what would be expected from pure normally distributed data.

The chart below shows the daily temperature standard deviations for different decadal climate periods fr Fairbanks and Anchorage.


  1. Very interesting. Two comments: it appears the median number of +/- 2SD days was considerably higher prior to 1970 - do you have a sense for whether this is some artifact of the calculation method, or did the shape of the distribution really change?

    Second: I believe there is a minor flaw in the Arguez method of computing daily standard deviations, leading to an overestimate of the true daily variance; this would at least partly explain the deviation from expectations that you report. I am close to completing a brief paper on this subject and will forward it to you soon for your perusal.

    1. Interesting. The Arguez method is pretty unsatisfying in my opinion and not very fun to calculate. I look forward to seeing your manuscript. I added a chart of the daily standard deviation by climate period for Fairbanks. The variability is much reduced in the 1981-2010 period. This year's temperature variability is more typical of 50+ years ago.

    2. Thanks, Brian - that is rather fascinating. So the variability in the decades since 1970 has tended to be lower than in the prior periods used to calculate the reference standard deviation. There is a lot of scope for interesting research here.

    3. Yes, very interesting indeed. Certainly worth a closer examination. I updated the post to include the daily SDs for Anchorage. A similar story. Since the Arguez method was successfully replicated for the 1981-2010 time period, I feel good about the values for prior decades. With the large differences primarily occurring in Winter, we can assume that winter variation is much reduced.