A simple temperature reading - not so simple...
Posted July 11, 2012
This past Sunday, we had a situation arise wherein the all-time record high temperature for Raleigh, 105 F, was reached for the third time in a few weeks, and the sixth time ever. That in itself was of interest, but if you were checking our Facebook page or our website, you may have seen an update declaring that we had broken that all-time record by reaching 106 degrees at RDU, a declaration that was later corrected as the official high for the day was held at 105 degrees.
It might seem that simply recording the highest temperature of the calendar day wouldn't be subject to much misinterpretation, but as with many things, there is much more going on than it may appear on the surface, and there are more complexities to deal with than one might readily assume. First, it's worth noting that while we think of temperature changing rather smoothly and being fairly homogeneous over a short distance, and in our typical experience and the way our bodies respond to it, that is a reasonable way to look at it. However, if you use highly sensitive, short response-time measuring equipment, like fine-wire resistance thermometers, for example, to measure how temperature actually fluctuates, it can be quite variable over very short times and short distances due to turbulent irregularities. An example is shown in the image above from an old journal article from the 1950s (Gerhardt, et al, "Fluctuations of Atmospheric Temperature as a Measure of the Scale and Intensity of Turbulence Near the Earth's Surface," Journal of Meteorology, Oct 1952). It shows the temperature trace from three different thermometers, all located at the same height around 4 feet off the ground at midday, with the sensors separated by a few inches to a couple of feet. The entire time of the graph is a 5-minute span, and each of the heavy horizontal lines marks .6 degrees C. You can see that the temperature remains about the same on average for the 5 minute period, but on a scale of a few seconds is varying greatly, almost as much as a couple degrees either way from the midpoint, and a detailed look shows that the variations can be notably different from one nearby sensor to another.
The reason I mention all that is to note that it becomes important to consider how often temperature is sampled and what conventions are chosen in reporting it and maintaining climate records. For doing turbulence studies, making very high resolution measurements is critical, but for supporting aviation operations or general weather forecasting, for example, these high frequency fluctuations are not necessarily as appropriate as a more smoothly varying value. These kinds of questions had to be considered by the designers and operators of the temperature sensing equipment used at airports around the country, and in the case of RDU, serving as the "station of record" for the Raleigh area.
At RDU, the weather station is of a type known as "ASOS," for Automated Surface Observing System. The system has to be able to operate continuously with relatively little maintenance, and to provide reliable and useful data any time of day. In the case of its temperature sensor, the designers chose to use the following protocol: an instantaneous temperature sample is read from the resistance thermometer once every 10 seconds, then at the end of one minute the six resulting values are combined into a one-minute average reading. At that same time, an average of the most recent 5 one minute readings is calculated to produce a 5-minute average value, which is rounded to the nearest degree F and also converted to the nearest tenth of a degree C. That 5-minute, reported each minute, becomes the basis of all further processing and reporting by the system. Near the top of each hour, the 5-minute average that exists at that time is reported as the "hourly" value. In addition, the system checks each 5-minute average to keep track of the highest one each hour, and eventually to report (just before midnight) the highest one that occurs all day, and similarly track the lowest temperatures. If you think about the variable trace in the image above, a 5-minute average would have a lower value than one of the peaks of those fluctuations, but would also be more representative of what most of us would experience over the course of that 5-minute period. On Sunday, it is quite possible that some of the 10-second readings, and even a few of the one-minute averages, may have reached 106 or even a bit higher, but since those were fluctuations that did not last very long, the 5-minute average never rose higher than 105, which eventually was reported as the high for the day.
In addition to the process above, one other issue that crept into the temperature reporting on Sunday was how information is encoded for public dissemination. The ASOS system sends out weather observation in a code called METAR, and there are two places in the observation that show the current temperature, both in degrees C. The first rounds off the hourly value to the nearest full degree, but there is a second group toward the end of the observation that carries a third significant digit for more precision. Here is the observation from 3 PM Sunday - METAR KRDU 081951Z 22009KT 10SM SCT085 BKN250 41/17 A2987 RMK AO2 SLP107 T04060172 - the "41/17" about halfway through indicates the rounded temperature is 41 C and dew point of 17 C, while the "T04060172" at the end provides more detail about temperature and dew point, in this case showing that the temperature was actually 40.6 C. If you convert those Celsius values to Fahrenheit, they work out to be 105.8 and 105.08, respectively, and thereby round off to 106 and 105. Some temperature reports that used the rounded off value thus showed up as 106 on some phone applications, and led to a mistaken report of the new record being set on our web site and Facebook page. These were corrected a short time later, but it points up the importance of checking that second temperature listing in a METAR report, or being sure that any application converting those reports to Fahrenheit readings is using that second "T" group in the observation.
Happily, the rest of this week should be free of any triple-digit heat around here, but we will be warming up by the weekend, and could easily be in the mid 90s (compared to a normal around 90) again by early next week...