It has been suggested many times that mean temperatures are artificial, and that I should look at trends in maximum or minimum temperatures instead. But I would say that all temperatures are artificial. None of minimum, maximum, or mean are good measures of temperature.
Describing temperature is a perennial problem. The Australian Bureau of Meteorology explains temperatures on its very informative website at http://www.bom.gov.au/climate/cdo/about/about-airtemp-data.shtml
Air temperature: The temperature of air is a measure of the average thermal energy of the molecules in the air – the higher the temperature, the higher the energy of the molecules. Historically, the most common instrument used has been a liquid-in-glass thermometer mounted in a Stevenson screen. Modern automatic weather stations generally use an electronic sensor. The daily maximum (highest) and minimum (lowest) temperature are generally recorded once each day at 9 am local time, while instantaneous measurements of air temperature may be made at various times throughout the day.
The daily (24 hour) maximum and minimum air temperatures are nominally recorded at 9 am local time. The highest temperature for the 24 hours leading up to the observation is recorded as the maximum temperature for the previous day, while the lowest temperature over the previous 24 hours is recorded as the minimum temperature for the day on which the observation was made. If, for some reason, an observation is unable to be made, the next observation is recorded as an accumulation, since there has been more than one day over which the maximum and minimum temperature may have occurred.
Temperature data prior to 1910 should be used with extreme caution as many stations, prior to that date, were exposed in non-standard shelters, some of which give readings which are several degrees warmer or cooler than those measured according to post-1910 standards.
Daily maximum temperatures usually occur in the afternoon and daily minimum temperatures overnight or near dawn. Occasionally, however, the lowest temperature in the 24 hours to prior to 9 am can occur around 9 am the previous day if the night was particularly warm.
Averaged over a month we get Monthly Mean Minimum and Maximum; averaged over the calendar year we get Annual Mean Minimum and Maximum; averaged over the entire record we get the long term Means. The average of minimum and maximum will give a daily, monthly, or annual mean.
The “average” or mean is completely artificial as it corresponds to actual temperature from twice to a few times per day. The advantages of calculating an average, however, are that
- it is a measure of central tendency;
- it is useful for revealing underlying trends;
- it removes much of the “noise” or variability.
Previously, I have focussed on the macro level of long term (century long) mean trends. In this post, however, I shall look at the noise- the micro level of a 24 hour period. I will show why the Maximum and Minimum temperatures are not very useful measures of temperature. Later I will show how this relates to the longer term measures.
In recent years many Automatic Weather Stations have been installed at BOM sites. Three-quarters of the High Quality network which is used to analyse climate trends have AWS stations. These produce continuous temperature data streams, though the interval between readings varies- some are 1 minute readings, some are much longer. Here are some examples of the data produced. I have used records from the Mackay Met Office in November 2010. This was (and continues to be) a period of changeable weather, with very unusual cloud, rain, and cooler temperatures. So far no maximum has reached the mean maximum for November of 29.2C. Mackay MO is located at Mt Bassett, a low hill about 500m from the Coral Sea in tropical North Queensland.
I should warn readers that these pages are only slightly useful. They do not give an accurate indication of temperature because the data points are at 30 minute intervals. While you may think that is quite high resolution, it’s not good enough. It nearly always completely misses daily Tmax and Tmin.
- The enormous amount of noise, even with high humidity and cloudiness. The temperature varies up and down during the day and night, as clouds come over, the sun breaks through, warm or cool air pockets pass through, rain falls, and the wind changes direction.
- Apparent Tmax and Tmin are reached only briefly. Actual Tmax and Tmin can only be for much shorter times- a few minutes at most.
I checked with the local office: the instruments are regularly checked and calibrated. If the maximum temperature on 15 November was reached between 1.00 and 1.30, then why doesn’t this show up on the graph? Where’s the spike? It’s because the graph is linked to 30 minute, not 1 minute, data.
These pages are virtually useless, so I don’t know what their purpose is, except that they do show the amount of noise at a particular location, and roughly when high and low temperatures occur. But they don’t show the full range of the temperature.
The reason I am using them: to show how briefly Tmax and (mostly) Tmin are reached. Maximum and Minimum temperatures do not give a true indication of temperatures on any day. To say that the temperature in Mackay on 15 November was 26.3C is completely misleading- the temperature reached 26.3 for a minute or so; for most of the day it was much less. Neither Tmax nor Tmin is much use, and therefore neither is the mean. Monthly mean maximum and minimum are meaningless.
I have suggested to BOM that the graphs and observations pages could be improved by including the data points for Tmax and Tmin. They didn’t want to know.
Compare Mackay’s graphs with-
Notice the low is reached during the afternoon thunderstorm- a fall of about 9 degrees in 90 minutes!
How shall we describe the temperature for each day? Temperature measurement of Tmax and Tmin at 9.00a.m. is traditional but an anachronism when we have instruments that can read the temperature continuously. It would be equally as accurate and easier for the public to understand to calculate a running 24 hour mean or median from the available AWS data. BOM will take no notice of this suggestion, however the latest weather graphs could be made a very useful and precise temperature record, if they showed 1 minute data (and therefore Tmax and Tmin would be readily visible) with the continuously calculated 24 hour running mean or median, of the previous and next 12 hours.
Gets rid of the noise, shows the central tendency, and shows the trend. What more could you want?