Thermometer Biases

Discussion of Location Biases
Discussion of Intrinsic Biases
Analysis of Location Biases
Analysis of Intrinsic Biases


Those of us who put thermometers up at our homes in order to record temperatures need to worry about whether they are actually recording the true temperature. There are two possible biases: a bias due to the location of the thermometer, and the intrinsic bias of the thermometer.

The location bias means whether an accurately calibrated thermometer is recording a different temperature than the true air temperature due to the location of the thermometer. This can be caused by a large number of different effects. Examples include:

The intrinsic bias simply means whether the thermometer is properly calibrated. That is, when the thermometer reads 70°, is the temperature truly 70°, or is it actually 68° or 72° due to an error in the calibration of the thermometer?

As part of trying to understand the calibration of the official Fallbrook weather station, Gaylord Moxon and I took two portable thermometers and directly measured the location biases of several places. We also measured the intrinsic relative error of several thermometers, including the new official Fallbrook thermometer.

The next two sections will discuss the location and intrinsic biases in more detail, followed by analysis of each bias from our measurements.

Discussion of Location Biases

Nearly every location that one can think of has some possible problems. Here are some of the potential problems:

The only location left is probably an open airy area underneath shade trees whose branches are very high, so as not to corrupt the low readings at night. If you don't have such an area, you'll either have to compromise or build a weather service type enclosure and put it in the open.

Such considerations make it clear why the weather service houses its thermometers in a standard box that has excellent ventilation. The enclosed box ensures that no direct radiation falls on the thermometer, as well as eliminating any reradiation from any heated surface. The excellent ventilation, coupled with the white-painted wood material of the box, assures that the sun on the surface of the box does not cause the inner surface of the box to be above ambient temperature.

Despite these arguments, I was skeptical about how well these little boxes in the sun would work until I measured the temperature inside one to verify how well it worked. The Weather Service knows what it is talking about here!

Discussion of Intrinsic Biases

Probably most people who have bought a cheap thermometer from a hardware store have looked at all the different samples of the same model and noticed that every sample doesn't read the same temperature, despite being in essentially the same location. This is a direct observation of the relative intrinsic biases of those thermometers. That is, since it is not known what the true temperature is at the location of those thermometers, all one can observe is the error of an individual thermometer relative to the other thermometers.

Suppose there are five thermometers, which read 70, 71, 72, 73 and 74°, respectively. The average of all five thermometers is 72°. The bias of those thermometers relative to the average is -2, -1, 0, 1 and 2°, respectively. If I needed to buy one of those thermometers, I'd buy the one that read 72°, figuring that it was most likely to be properly calibrated.

The assumption behind that reasoning is that on average, the manufacturer tries to make their thermometers properly calibrated, but that manufacturing errors may produce a variation of ±2° (in this hypothetical example) for individual samples. However, without any other information, I have no way of knowing whether the manufacturer in fact has an error in their average calibration. Thus the true error of those thermometers might be -4, -3, -2, -1 and 0°, respectively, meaning that only the thermometer reading 74° is properly calibrated.

There is only one fundamental way to check the calibration of a thermometer. In the Fahrenheit scale, 32° is the freezing point of water and 212° is the boiling point of water (at a normal atmospheric pressure of 760 mm of mercury). A thermometer is usually a very linear device, meaning that it changes its reading by the same amount for every degree rise of temperature. Hence if the thermometer properly reads the freezing and boiling point of water, it is most likely correct over the entire range between the two.

Unfortunately, most weather thermometers read between -50° and 120°, since that is the expected range of temperatures encountered in nature. Thus it is possible only to measure the intrinsic thermometer calibration at 32°.

Another way to calibrate a thermometer is to measure it against a thermometer which has been properly calibrated. The National Institute of Standards and Technology (NIST) performs such measurements. One can have NIST calibrate a thermometer for $259 per point or purchase a NIST-traceable thermometer, which has been compared to a NIST-calibrated thermometer, for ~$150.

If anyone knows of a NIST-traceable thermometer in the Fallbrook area, please let me know so I can compare my thermometer to it!

Analysis of Location Biases

A Taylor 21413 analog liquid-in-glass thermometer (a cheap ~$10 "backpacking" thermometer) and an Oregon Scientific BAR-888 electronic thermometer ($130) were carried to a number of locations during a two hour period to measure temperatures there. The measurements were taken between 12:41 and 2:43 p.m. PDT on 6/13/00 in Fallbrook, California at three different locales: my house (1802 Acacia Lane, elevation 690'), Mox's house (839 Via Alegre, 840'), and the official Fallbrook weather station in downtown Fallbrook (315 E. Ivy, 699'). These sites roughly form an isosceles triangle with a base of 1.1 miles between the official site and Mox's house and sides of 3.5 miles from either location to my house.

This period contained essentially the peak temperature of the day, and hence the true temperature was nearly constant during this time (see plots below). All three locations are expected to have nearly the same temperature. Even if the maximum distance difference of 3.5 miles was along the sea-breeze direction, the expected temperature change would be only a degree or two. The elevation difference would give a maximum temperature difference of 0.8°. Further, these two effects work with opposite sign in this case, partially canceling each other out, since the highest location was also the farthest from the coast.

The measurements with the Taylor thermometer were made by holding the insulated end of the thermometer. The measurements made with the Oregon Scientific thermometer were made by holding the thermometer, which may have slightly biased the readings since there was no equivalent "insulated" end to hold. In particular, the field of view of Mox's thermometer contained a larger solid angle from Mox's hand, due to the method in which it was held, which may have slightly biased the readings.

The intrinsic bias of these thermometers is irrelevant to the analysis here, since only the relative readings from these locations matter. However, our measurements may still be influenced by the fact that we were holding the thermometers, slightly altering the location seen by our thermometers and the thermometer to which we were comparing. Furthermore, since the housings of the sensors for the thermometers are different, each thermometer may sample the environment somewhat differently if there are temperature gradients across the field of view from a given location. There is evidence of this presented below.

The measurements are consistent with all three locales having the same true temperature, and with the true temperature changing by at most 1.5° during this period. As will be seen, however, the measured temperature shows significant biases due to the precise thermometer location at each locale.

The measurement locations at my house were:

The measurement locations at Gaylord Moxon's house were:

The measurement locations at the downtown Fallbrook fire station were:

The bias due to thermometer location is derived separately from my Taylor thermometer and from Mox's Oregon Scientific thermometer. My thermometer has a resolution of 0.5° at best, so differences of 1° are not significant.

I've classified the above locations into the following categories:

Note the consistent offsets derived from both thermometers in the plots of the bias from both thermometers (remember that the absolute readings don't matter here, only the offset from the coolest readings):

The measurements showed clearly that only deep shade or the official weather station box produced an unbiased temperature reading. Even the ~20' diameter circular shade provided by a small tree still produced a bias of 1-2° due to the heated surfaces outside the shaded area.

There is a clearly increased bias at the locations of the thermometers that Mox and I use to measure our Fallbrook high temperatures, due to the influence of heated concrete for my measurements and the influence of the arbor roof on Mox's measurements. This bias is consistent with what was found previously in a comparison of our temperatures to the official Fallbrook high temperatures for a period of several years. Hence this implies that our bias is mostly due to the locations of our thermometers, and not to biases in our thermometers themselves.

Analysis of Intrinsic Biases

Unfortunately, the data do not unambiguously allow the deduction of the intrinsic biases of the thermometers. I suspect that the reason is due to the location biases, which are different due to the observer being present in the field of view of the thermometers and due to the thermometers effectively seeing different environments due to different shielding about the sensors.

The raw data, giving the difference in the readings of one thermometer relative to another at the same location, show the problem:

ThermometerBias Relative to
TaylorOregon Scientific
tjc+2, +2-4.4
mox+2.9, +4.40.5, -0.8
Taylor--2.4, -2.8, -2.0, -5.4, -4.6, -5.2
Oregon Scientific+2.4, +2.8, +2.0, +5.4, +4.6, +5.2-

The differences between the Taylor and Oregon Scientific thermometers clearly vary with time. There is a ~1.5 hour gap between the differences of {-2.4, -2.8, -2.0} and {-5.4, -4.6, -5.2}. The location bias definitely changed during that interval. Hence it is likely that this change shows that the thermometers do in fact effectively see different environments, despite being physically at the same location.

If I arbitrarily assume that the larger differences are caused by location, and thus ignore them here, I arrive at the following relative calibrations:

ThermometerBias Relative to
Taylor1996 Official
Oregon Scientific+2.5+2
Current Official-2.5-3
1996 Official+0.5-

The 1996 Official bias comes from my previous analysis of the calibration of the official Fallbrook weather station.

However, I have little confidence in the validity of this relative calibration for the tjc and mox sensors due to the probable remaining location bias. In fact, previous comparisons combined with this analysis make it likely that there is very little intrinsic bias of either sensor relative to the 1996 Official Fallbrook temperatures. That is, virtually the entire source of the previously-deduced bias is due to the location bias.

The comparison of the remaining thermometers is probably valid, since it derives from the relative bias at the location of the official Fallbrook temperature, using observations inside the weather box. These comparisons make better sense than assuming that the current official temperature is unbiased, which would make the Taylor thermometer biased high by 2.5° and the Oregon Scientific thermometer biased high by 5°, values that are probably too high to be expected from miscalibration.

I checked the calibration of my Taylor thermometer in an ice-water bath, and it correctly read 32°. However, this does not prove that the thermometer is accurate at 80° without having another calibrated data point.

Overall, I consider it likely, although not definitively proven, that the current official Fallbrook high temperatures are ~3° too low. What is known is that the current official Fallbrook high temperatures are ~3° too low as compared to the 1997 and 1998 official Fallbrook high temperatures.


Go to Fallbrook Weather

Copyright © 2000 by Tom Chester.
Permission is freely granted to reproduce any or all of this page as long as credit is given to me at this source:
Comments and feedback: Tom Chester
Last update: 24 June 2000.