We were able to significantly improve the frequency response by using human hair (the "Lorusso probe"!), which had a much smaller thermal mass than the cotton thread used earlier. However, maintenance of these fine-wire probes for our operation purposes still disuades us from using them routinely. We currently use the "sonic virtual temperature" from sonic anemometers, corrected for humidity fluctuations using krypton hygrometer measurements, for measurement of temperature fluctuations.
We speculate that NOAA may have used finer cotton thread or another material and thus had a smaller error than we encountered. Furthermore, the heat flux encountered at the BAO (height range 10-300m) would be dominated by lower frequencies than we expect. Thus, the BAO heat fluxes would be expected to have smaller errors than what we found.
The AIR fast temperature sensor is a PRT (platinum resistance thermomometer). The sensing element is a fine (0.5 mil?) platinum wire approximately 30 cm long wrapped around a spiral frame and has a nominal resistance of 150 ohms and can vary by +/-15 ohms depending on the individual wrapping. A (DC?) bridge circuit is used to convert the resistance (as a function of temperature) into an analog voltage in the range of +/-5 V. We are told (by the manufacturer?) that the response of this circuit is linear with temperature (since resistance is linear with temperature?). The circuit has two adjustments: a gain and an offset pot. These can be changed to set the output to the nominal value of +/-5V = +/-50C, i.e. a gain of 10 C/V and an offset of 0.
The calibration routine used with this device is called by:
c=linear(0.001525925,0.0)
which uses the linear calibration routine with a gain of
(10 C/V)(5V)/(32767 counts) = 0.0015925925
[Steve S. thinks this should be 32768 counts] and an offset of 0.0 C.
We generally followed a standard calibration procedure. During ARM91 and SJVAQS91, the foam calibration chamber was placed over each probe and the two pots adjusted to attain the nominal calibration. Ideally, the gain (which is checked by switching the bridge to introduce a change in voltages which is supposed to be equivalent to 5 C) will be read as 5.00 C and the offset (difference of 10*output_voltage from chamber temperature) will be 0.00 C. Values measured during ARM91 were:
Fast T's calibrated 2/13/91 by Kurt
ragwort:100 (4m uw) +.02 C (5.02) gain offset
+.05 temp offset
ragwort:101 (10m uw) +.02 C (5.02) gain offset
+.05 C temp offset
cosmos:100 (4m ati) -.01 C (4.99) gain offset
+.05 C temp offset
indicating that the gain can be set to 0.4% and the offset is reasonably small.
These errors were felt to be neglible so the default calibration above has been
used. Note that the FLUKE DMM with its thermocouple sensor is being used as
a reference to 0.01 C. Although the FLUKE DMM has been calibrated by Ray,
the thermocouple electronics have never been tested. Also note that during the
adjustment, it is necessary to iterate adjusting the two pots since one
adjustment affects the other. During FLAT90, these pots were never changed, but the gain calibration was checked from time to time (supposedly each time a sensor was changed - mostly due to snow kills). These values were:
NCAR NCAR NOAA NOAA NCAR uw.4m uw.7m ati.4m ati.7m ati.13m 6 Oct 4.93 4.95 4.62 4.75 4.95 17 Oct 4.86 4.58 5.00 3 Nov 4.4 4 Nov 4.66 4.86 5.03 5 Nov 5.18 5.03 8 Nov 5.44 5.03 4.56 4.88 5.00 16 Nov 5.41 5.04 4.56 4.87 5.02Note that the 8 and 16 Nov values were pre- and post-cals of the same probes, which shows that the calibration is repeatable/stable to 0.02 C. Also note that, with the exception of uw.4m, the values appear to be stable within one bridge.
For FLAT90 processing, the 6 Oct values were used for the first two operations periods and the 8 Nov values were used for the remainder of the experiment. This was done by specifying linear gain values of:
uw.4m uw.7m ati.4m ati.7m ati.13m
ops1,2 0.001504562 0.001510666 0.001409955 0.001449629 0.001510666
ops3-6 0.001580859 0.001535081 0.001422162 0.001483120 0.001535081
Offset values were never quantitatively checked during FLAT90, but were up to
10 C in some cases. This information has not (yet - will be for second pass?)
been input to the linear calibration routine. During ARM91, a comparison between the AIR instrument and the PAM psychrometers showed that the mean behavior was not necessarily linear and differences of 1 C were common. The variances of two AIRs were compared but did not show significant differences, hopefully indicating that this problem will not effect the temperature fluctuation measurements. During SJVAQS91, a loose connection in one of the bridges was fixed and improved agreement between two AIRs. After SJVAQS91, Steve Semmer has checked a bridge in the calibration lab and finds a worst-case error of 1.2 C and speculates that the DC-DC converter added to the bridge could be the culprit. Further tests as to the source of this error in the means need to be made. I don't think that the BAO people have ever counted on this instrument to measure means.
The frequency response of this sensor needs to be quantified. It has been suggested that the response is similar to a sonic anemometer because the wire is about the same length as a sonic anemometer path. Kinks in the wire as it wraps around the spiral have been found to introduce another time constant on aircraft-(Electra)mounted probes but may not be a problem here. Note that some of the high-frequency attenuation observed in FLAT90 and ARM91 probably was due to the block averaging done in the process of sampling the A/D.
Finally, it has been suggested that thermal wakes from the wire's supporting frame could interfere with the temperature measurement. We need to check with Chandran about this.