An evaluation of the use of min/max temps to determine average temps, and a comparison of three observation times.
Guest essay by: Robin McMeeking
I was an Air Force weather observer in the early 1960’s and have retained an interest in weather things. Later in in my civilian career I spent many years developing software. Several months ago while reading about adjustments that had been made to the historical temperature record it piqued my interest. I was initially surprised to learn about the reliance on min/max thermometers in these early records.
But in the days before flight there wasn’t any need for regular systematic tracking of temperature. At any rate much of the historical temperature record is based on averaging min/max values. To be useful these values need to be recorded daily at reasonably close to the same time of day. This record keeping was done by volunteers reading thermometers at selected locations. The assumption is that the volunteers were fastidious in performing this task.
It’s not intuitive, but has been recognized for many years that the time of day used for the readings can exert some bias toward higher or lower average temperature determinations. It’s called Time of Observation Bias (TOB). I was not convinced by the explanations of the reasons for TOB, but I was even more dubious about the validity of relying min/max values to calculate an accurate average temp.
So, when the issues about TOB and average temps came up I was interested in performing some analysis, but didn’t have any data to work with. Some time later I discovered the network of Personal Weather Stations (PWS) on Weather Underground. Eventually I found that daily historical data from the PWSs can be downloaded (one day at a time!). Fortunately I was able to develop software to perform the downloading. And when I was almost done with my preliminary analysis I reread some of the posts on TOB and discovered the NOAA and other sources of official temperature data. Oh well!
I wasn’t trying to look at climate trends, I basically wanted to evaluate the suitability of using Min/Max temps to derive daily average temps. And I wanted to simulate a variety of observation times. I selected 16 PWS’s from around the US and downloaded all of 2015 plus 1/1/16 to allow for shifting observation times (well over 1.5 million records).
Most of these PWS’s record at 5 minute intervals, a few at 15. I calculated a mean temp from the min/max values in each 24 hour period, averaged all of the reported temps, and averaged the last reported temp in each hour to simulate what would be available from historical WBAN records (like we did in the ’60s). I evaluated TOBs of midnight, 0800 and 1600.
I selected PWS’s that are in rural locations, but I realize that I have no information on the suitability of the siting of any of these stations. That bothered me so when I finally became aware of NOAA data (after I was done with the PWS stuff) I downloaded data for one station from NOAA and performed the same analysis. Results from that data were consistent with the PWS analysis.
Below are tables of the summarized data. If you know of anyone that would like to see results from the intermediate steps (daily totals with monthly summaries) I’d be happy to share it. What I have concluded is that Min/Max method tends to yield a higher value than the more inclusive averages at all three TOBs, with 1600 being worst and 0800 being “best” of the times I tested. Averages based on hourly readings (which frequently don’t include the max &/or min) are virtually indistinguishable from averaging all readings.
The tendency of Min/Max to be high led me to wonder how much time per day is spent with temp above or below the mean. The results on that surprised me. Shown in the tables below.
Based on 12 months of data for 2015. All temps in Farenheit.
Note from Anthony: Some readers might suspect the PWS network has exposure and placement problems, like the NOAA network, and they would be right. However, since he isn’t looking at trends, just the diurnal variation of temperature and the issue of recording times, in my opinion, that issue doesn’t apply here.