Are there measurable differences between air temperature at sites with different microclimates?
| One of the basic questions for climate science is whether surface air temperatures are increasing, either globally or regionally.
Although measuring air temperature seems like a perfectly straightforward undertaking, it is actually a difficult measurement to do
correctly if one is looking for small differences over long periods of time. One important consideration is the siting of
temperature monitoring stations. Air temperatures are supposed to
be measured about 1.5 m above the ground, over a natural surface that is representative of the surrounding area and that is far away from trees, buildings, and other artificial
sources of heating or cooling. It has been documented convincingly that many "official" weather stations do not meet these requirements. The image at the right shows a station in Forest
Grove, Oregon, that is part of NOAA's U.S. Historical Climate Network (USHCN), one source of data about air temperatures in the U.S. that is used
by climate modelers.
But, this site is on what looks like a gravel or asphalt pad with a
window air conditioner on the side of a building only 10' away,
venting toward the temperature sensor (mounted perhaps a little over 2 m above the ground in the "stacked pie plate" radiation shield).
(See this site for this and other examples of poorly sited weather stations.) By any standard,
this is a terrible site for monitoring climate-related long-term
temperature changes!
Data interpretation problems can exist even with well-sited instruments. There are always questions about whether even a "good" site is truly representative of an entire area. These questions are more likely to be important on hilly terrain rather than flat terrain, for example. Therefore, a reasonable research project is to examine the effects of microclimates and temperature sensor sitings on measurements of air temperature. |  |
|
Is it possible to use relatively inexpensive thermistor-based temperature sensors for such a project?
The TMC6-HD sensor from Onset Computer Corporation costs $35 (as of early 2012). Its characteristics are shown in the image at the right.
The accuracy of this sensor with a U12 data logger (the logger CSRES uses for recording all its measurements) is given as ±0.21°C for cold temperatures and
±0.38°C for warmer temperatures; that is, the accuracy of the measurement is a function
of the ambient air temperature itself. Even though the accuracy is specified as "±", the error is much more likely to be systematic in one direction or
the other than random.
Temperature differences due to microclimate conditions or sensor siting may be in the same range as (or even less than) the specified accuracy of these temperature sensors. So, any reliable experiment using TMC6-HD sensors needs to characterize carefully the relative performance of "identical" devices to make sure that instrument differences can be separated from real temperature differences. Experiments using these sensors can produce reliable results only if their characteristics are understood, and if it can be shown that the differences among sensors are systematic and do not change randomly with time. | 
|
Considering the relatively low absolute accuracy of the sensors described above compared to the temperature differences that may exist,
it should be clear that it is by no means sufficient simply to mount "identical" sensors in different places, record
data, and look at the differences; this simple plan will inseparably combine instrument and site differences. This may result in concluding that
temperature differences between sites exist when they do not, or in masking real temperature differences. To avoid these problems,
it is necessary to design an experiment that will provide ongoing intercomparisons among sensors.
Suppose there are two sites, Site1 and Site2. Comparing temperatures at these sites requires three separate sensors (1, 2, and 3); make sure the sensors
are labeled so that there will no doubt which one is which!
An experiment procedure could look like this:
 |  |
The two graphs show data collected at 5-minute intervals with three temperature sensors for 5 days,
starting on January 9, 2012. The
sensors were fastened together with a rubber band and placed inside an unvented wooden enclosure where data loggers for other
experiments are kept. This is not an appropriate place for measuring air temperature, but it is a reasonable place to collect data for
comparing these three sensors. The differences among the sensors are very small — possibly smaller
than might always be the case! But, in any event, Excel's trendline-generating function can be used to determine what is assumed to be
the linear relationship between sensors 1 and 2 and sensor 3. It is OK if the differences among sensors are larger than seen here, as long
as those differences are stable over time.
Using a third sensor moved back and forth between Site1 and Site2 will provide
an ongoing check on whether the
relative performance of the Site1 and Site2 sensors is changing over time. Hopefully, the relative calibrations do not change during the experiment!
If they do, it probably means that these temperature
sensors are not reliable enough for this kind of experiment. If the relative calibrations do not change significantly, choose one
sensor as the "reference" — we have chosen sensor 3 as the reference. Then all the data collected can be related to this reference
and Site2 temperatures can be compared to
Site1 temperatures. If the differences between two sites are small, statistical tests may need to be applied to
determine quantitatively whether those differences are significant.
Two sites have been used for a test of how this kind of experiment might work. Site1 is on a wooden arbor at the border between a grassy area and an area planted with flowers. Site2 is on the support column for a deck at the back of a house. The radiation shields are pointing approximately south in both locations. Site2 is about 5 m lower in elevation than Site1 and it is more protected from winds. Behind the Site2 installation is a poured concrete wall for an unheated basement.
 |  |
| Site 1 (The third temperature sensor at the right, mouned below the two on the left, is not part of this study. The reference sensor that moves back and forth between Sites 1 and 2 is on the left.) | Site 2 |
| The graph shows some temperatures sampled at 5-minute intervals at Sites 1 and 2 during January 2012. There are some differences of a few degrees between these sites, especially during cold nights. Perhaps this is due to the fact that Site 2 is much closer to a building and the site is more protected from winds. But, determining how these differences are related to site, seasonal, and weather conditions is what makes this an interesting research investigation! |  |