Publications and Research
|Related Links & Resources
Turn an engineer loose with data....
The NOAA Website http://www.cdc.noaa.gov/Boulder/data.daily.html has files with complete Boulder temperature data from 1898, and with both temperature and precipitation since Aug 1, 1948. Recent data are updated monthly. There are a few scattered months over this period with no data in the file, but the data set is good enough to allow me to investigate a couple of statistical properties I wondered about.
To smooth out the day-to-day variability, I decided to lump data into half-month blocks. So, for example, all the daily lows for the first through fifteenth of a given month would be averaged into one number, and the lows for the rest of the month would be averaged into another value. I decided to include temperature data from Jan 1, 1931 to use some of the data from before the time (1948-1957) when the measurements were made on the roof of the fire station. Because the measurements were influenced by solar heating of the roof, those years are clearly anomalous. (See the hot summers graph.)
My first question was, "What is the average spread between high and low for the day, and how does it vary over the year?" The graph below shows that the average spread varies from about 24 deg during the short days of winter to nearly 30 deg during the heat of summer and into early fall..
Because of the changes of the location of the official temperature measurements, Boulder is an exceptionally poor place to look for real climate trends. It's interesting to make the calculations, nonetheless. One simply shouldn't take the results too seriously.
For each of the half-month data sets, I've calculated the best fit to a straight line over the period to ascertain possible trends. I've scaled the results to show the indicated change in deg Fahrenheit per decade.
Global-warming models and measurements suggest that there is less cooling during the nights, so the minimum termperatures should show a rising trend, especially in winter. The low-temperature trends in late February and all of March agree with that general observation (but may still be meaningless). The more pronounced trend is the consistent increase of average highs from late February through early June. This may be a local indication of the widespread observation that spring has been arriving earlier than in the past. Also interesting is the apparent slight decrease of summertime lows over the 76-year period. A dip in the average highs in November could be due to increasing average cloudiness, for example.
Interested in more details? Look here for a scatter plot of the data for the first half of March to see the points contributing to the trend calculation.
Finally, we can also derive a version of "normal" highs and lows from this data set. Published normal values don't always use the same base data. For example, the years 1970-2000 are sometimes used. The plot below shows the values in half-month steps derived from the present data set.