| Description: | Ongoing investigation of differences between preciptiable water vapor
measured by the microwave radiometer and precipitable water vapor estimated
by integrating the radiosonde-measured profile of water vapor density, along
with isolated periods of anomalous behavior in the AERI/LBLRTM QME results
led to the suggestion that the calibration of some radiosondes may be suspect.
One group of radiosondes, calibrated in November 1994, seemed particularly
irregular. I met with the manufacturer on February 7 to show them my
results and discuss this issue with them. The manufacturer agreed that
the data were suggestive of a calibration problem and offered to review
their calibration records for the period in question.
The initial results of the manufacture's investigation were sent to me
on February 27. The manufacturer found that radiosondes assembled
between 10/26/94 and 11/22/94 were incorrectly calibrated. The incorrect
calibration affected only the relative humidity measurement; the nature
of the calibration error is non-linear, with the radiosondes meeting
the specified +/- 2% RH accuracy at both 20%RH (+45 degC) and 0%RH (room
temperature). The error in calibration occurs at high RH and low temperature
(approximate saturation at -30 degC) and can be as large as 10%RH with
the radiosonde reading high. The direction of this error agrees with
the observation of negative integrated residuals in the AERI ch 1 region
made by Tony Clough and the tendency toward greater than 'normal' differences
in PWV when compared to the MWR when these sondes were used.
The cause of the calibration error was the addition of new, high RH,
quality control check (-30 degC) into the manufacturing stream that
required a change in the calibration software. The new software mistakenly
loaded an incorrect parameter file that caused the calibration to be
off at high RH - this was not noticed at the time because the radiosondes
passed the regular production QC checks and the failure to agree at the
new QC value was thought to result from a problem with the new reference
standard. Troubleshooting the entire calibration system in an effort
to incorporate the new QC check was conducted along with normal production,
and took about a month to complete. The troubleshooting including adjustment
of the calibration software, at which time the correct parameter file
happened to be reloaded, and after which the radiosondes met the desired
specification at the high RH value.
According to the manufacturer, approximately 1900 radiosondes were calibrated
during the period in question. Sondes from this period were used at all
CART facilities. At least 1178 soundings involving these sondes were
included in the QME results I looked at (some QME results were discarded
because of problems with either the MWR data or the sounding data). The
distribution of these 1178 by facility is:
Facility Number of Sondes First Used Last Used
(based on QME)
CF 316 1/20/95 (11:30) 9/22/95 (11:30)
B1 143 2/21/95 (17:30) 9/29/95 ( 8:30)
B4 313 4/ 6/95 (17:30) 11/ 7/95 (17:28)
B5 153 4/14/95 (17:30) 12/14/95 (17:30)
B6 253 3/22/95 (17:30) 10/19/95 (17:30)
The sondes can be recognized by decoding the serial_number metadata word
in the sounding netCDF file. The serial number of the sondes in question
is a number (8 or 9 digit) of the form DDMMYTTPP where DD is the day of
the month (leading zero missing in the netCDF data), MM is the month +80
(e.g. 91 for November), Y is the last digit of the year (4 for 1994),
TT is the calibration tray ID, and PP is the position of the sonde in
the calibration tray. The sondes in question will therefore have serial
numbers coded between 26904TTPP and 22914TTPP. |