Do We Really Know What The Temperature Is?
December 6, 2010Posted by on
How many times have you seen temperature graphs similar to this:
In figure 1 it is a graph of the average yearly anomaly’s for the USHCN Raw dataset for the station located at State College Pa, but it could just as well be a station record from GISS or from GHCN or from any place around the world and the basic properties of the graph would be the same. You would have start and end dates, data points, a possible trend line and so forth. It all looks fine except for one little problem that I have yet to seen addressed in Climate Science: Instrument Error.
Every device that man has built to measure something has a built in error range and it is something that can only be accounted and know if you compare your device against one of known value or against something of a known physical property. For Thermometers they typically compare them to the known freezing and boiling points of water. This is called calibration and even after that there is still a error band, which is the accuracy of the thermometer. With today’s technology we can whittle down that error band sometimes as low as +/- .001 but for something like that it takes a lot more money so typically you won’t get near that level. However what about in the past? How accurate were those thermometers back in the late 1800’s and early 1900’s? What was the accuracy of ones even in the 40’s and 50’s?
If you look around the internet you find that Liquid in Glass (LIG) are still for sale and they have error ranges as much as .5°, others at .3° and you can buy Platinum Resistance Thermistors (PRT) that have accuracy’s of .1°, which brings us back to what was it like back in the olden days before the people taking these readings were worried about raises in temperature of a fraction of a degree. So would it be reasonable to assume that early Max/Min thermometers had a +/- accuracy of at least .5°? I would say yes.
Now you may ask what does that got to do with anything? The answer is a lot. You see that +/- rating of .5° means that when you see a daily max/min temperatures in those old records it isn’t exactly true. Example if it is written that the Max was 50°F and the Min was 39°F that means that the Max could actually be up to 50.5° and as low as 49.5°, with the Min being actually between 39.5° and 38.5°. This in turn affects the mean for day and consequently the average mean for the month by the same amount. Here is an example: Lets say that the Thermometer in State College Pa during the year of 1939 had an accuracy of +/- .5° F . Below you will see how much that can affect the means.
|Oct||Max||Min||Mean||Max +.5°||Min +.5°||Alt Mean #1||Max -.5°||Min -.5°||Alt Mean #2|
This turn gives 3 different monthly means: 52.3°F, 52.8°F and 51.8°F.
Now lets assume that Thermometer had been there for the years 1930-39 and that the thermometers before and after it are perfect, ie an error of zero.
As you can see in Figure 2 there is a change in the trend lines just by factoring in a 10 year instrument error. If the Thermometer actually was reading high by .5° the trend is 1.0° F if it was -.5° the trend increases to slightly over 1.1°F. However we know that the thermometers before and after those years have their own built-in errors. We also know from the metadata from stations that thermometers go bad, get broken and sometimes are just changed for more modern and more accurate equipment. So lets simulate what would happen if the record started in 1895 with a Max/Min thermometer with an accuracy of +/- .5°F, in 1940 it was replaced with one that had and accuracy of .3°F and in 1990 was replaced with an automated system that uses a PRT with an accuracy of .1°F
As you can see in Figure 3 the trend lines have a significant change. For an actual reading of all three thermometers of max negative error the trend is 1.7° F and with all three thermometers reading max positive error the trend is .6°F. That is over 1°F in difference and that error range has not been accounted for nor shown in the temperature graphs we see. Of course the actual errors could cancel out or they could actually be worse than demonstrated since I didn’t introduce a new thermometer with a different accuracy in the baseline period. Also not taken into account is wear on the thermometers over time, especially the older ones. As the Thermometers get older they get less accurate and with quality standards in the 1800’s not what they are today we have no idea what the actual temperature was back then and we only have a SWAG of what it actually is today. To demonstrate this read up on the early history of one of the stations in the NCDC database:
HISTORY OF WEATHER OBSERVATIONS Fort Bayard, New Mexico 1867 – 1893
The January 1886 observation form contained a note stating the office did not have a serviceable maximum thermometer. It stated a replacement from the Army Surgeon General’s Office had arrived broken and that the office had requested a replacement, but it had not arrived. The note also stated the minimum thermometer was not in good working condition prior to 8 January 1886. However, minimum temperatures continued to be recorded. A replacement maximum thermometer was received and readings were commenced 19 February 1886.
Beginning January 1888 neither maximum nor minimum temperatures were recorded. A note on the January 1888 observation form stated, “The Maximum and Minimum thermometers are so unreliable from age and service that no observations with them have been made during the month.”
As shown in that paper you got multiple changes in thermometer and we don’t know the error range on any of them are. Now extrapolate from there and with what is shown in the graphs above and realize that EVERY Station and thermometer have this exact same problem and supposedly we KNOW that temperatures have risen roughly .8°C over the last century or so. Doesn’t seem likely now does it with that much built-in error to the instruments that can not be accounted for nor removed from the record.
Data taken from the USHCN v2 Raw dataset found here for monthly data: http://cdiac.ornl.gov/ftp/ushcn_v2_monthly/
The data for the table can be found at this address: http://www7.ncdc.noaa.gov/IPS/coop/coop.html
Here is a local copy of the PDF page used: