1
Accuracy Assessment and Correction of Vaisala RS92 Radiosonde
Water Vapor Measurements
by
Larry M. Miloshevich
National Center for Atmospheric Research
*
Boulder, Colorado
Holger Vömel
University of Colorado (CIRES)
Boulder, Colorado
David N. Whiteman
NASA/GSFC
Greenbelt, Maryland
Thierry Leblanc
NASA/JPL
Wrightwood, California
*
The National Center for Atmospheric Research (NCAR) is sponsored by the National Science
Foundation.
2
Abstract
Relative humidity (RH) measurements from Vaisala RS92 radiosondes are widely used in
both research and operational applications, although the measurement accuracy is not well
characterized as a function of its known dependences on height, RH, and time of day (or solar
altitude angle). This study characterizes RS92 mean bias error as a function of its dependences
by comparing simultaneous measurements from RS92 radiosondes and from three reference
instruments of known accuracy. The cryogenic frostpoint hygrometer (CFH) gives the RS92
accuracy above the 700 mb level; the ARM microwave radiometer gives the RS92 accuracy in
the lower troposphere; and the ARM SurTHref system gives the RS92 accuracy at the surface
using 6 RH probes with NIST-traceable calibrations. These RS92 assessments are combined
using the principle of Consensus Referencing to yield a detailed estimate of RS92 accuracy from
the surface to the lowermost stratosphere. An empirical bias correction is derived to remove the
mean bias error, yielding corrected RS92 measurements whose mean accuracy is estimated to be
±3% of the measured RH value for nighttime soundings and ±4% for daytime soundings, plus an
RH offset uncertainty of ±0.5%RH that is significant for dry conditions. The accuracy of
individual RS92 soundings is further characterized by the 1-σ “production variability,” estimated
to be ±1.5% of the measured RH value. The daytime bias correction should not be applied to
cloudy daytime soundings, because clouds affect the solar radiation error in a complicated and
uncharacterized way.
3
1. Introduction
Atmospheric water vapor measurements are used in a wide variety of both operational
and research applications, including as input to forecast models and radiative transfer
calculations, for validation of ground-based and satellite remote sensor retrievals, and for
development of water vapor and cloud parameterizations, among other applications. The high
vertical resolution of radiosonde measurements is well suited to these measurement needs, if not
for their substantial inaccuracy under certain atmospheric conditions, especially in the upper
troposphere (UT) and lower stratosphere (LS). As with all measurements, their scientific value
is tied to estimates of uncertainty in the measurements, which must be known if uncertainty in
subsequent results is to be estimated. Unfortunately, the accuracy of radiosonde relative humidity
(RH) measurements differs between measurement technologies, between radiosonde
manufacturers and models, and even with time for a given model due to hardware,
manufacturing, or calibration changes. Furthermore, the accuracy of radiosonde RH
measurements has been shown to vary with height, RH, and time of day (or solar altitude angle).
The aim of this paper is to characterize the accuracy of Vaisala RS92 radiosonde water vapor
measurements as a function of its dependences, and then develop and evaluate an empirical
correction that removes the mean bias error.
Several methods of characterizing and improving the accuracy of RS92 RH
measurements have been developed, although these methods generally address only a subset of
the main sources of measurement error. These sources include calibration error that reflects the
accuracy of the Vaisala calibration model and calibration references, solar radiation error (SRE)
caused by solar heating of the RH sensor, and time-lag error caused by slow sensor response at
low temperatures. Vömel et al. [2007] used dual soundings of RS92 and the cryogenic frostpoint
hygrometer (CFH) to show that daytime RS92 measurements suffer from SRE that increases
with height from 9% near the surface to 50% at the tropical tropopause for high solar altitude
angles (α>60°), and they derived a height (pressure) dependent correction that removes this
mean bias. Miloshevich et al. [2006] (hereafter “M06”) compared ARM microwave radiometer
(MWR) and RS90 radiosonde measurements and found that the SRE is 6-8% in terms of
precipitable water vapor (PW), where this bias represents an average over all solar altitude
4
angles and RH conditions in the lower troposphere (LT). M06 also used dual RS92/CFH
soundings to characterize the mean bias error for nighttime RS92 soundings (i.e., zero SRE), and
derived a correction that is dependent on height (temperature) and on RH. Cady-Pereira et al.
[2008] determined the dependence of SRE on solar altitude angle by comparing ARM MWR PW
measurements to column-integrated RS90 and RS92 PW measurements, and derived a correction
that removes the mean bias in the LT as a function of the solar altitude angle. Miloshevich et al.
[2004] (hereafter “M04”) used laboratory measurements of the sensor time-constant as a function
of temperature to derive a correction for sensor time-lag error caused by slow sensor response at
low temperatures, and Vömel et al. [2007] showed that addressing time-lag error markedly
improves the agreement between RS92 and CFH in the UT.
This study will evaluate the accuracy of RS92 water vapor measurements using the
principle of Consensus Referencing [Fitzgibbon, 2008; Facundo and Fitzgibbon, 2007], whereby
RS92 measurements are compared to measurements from different reference-quality instruments
for the conditions under which each instrument performs well and its accuracy is known.
Simultaneous measurements from the RS92 and 3 reference instruments --- CFH, MWR, and
calibrated RH probes --- are compared in the next section. The comparisons are synthesized in
section 3 to produce an estimate of the RS92 accuracy as a function of height, RH, and solar
altitude angle. An empirical correction that removes the RS92 mean bias relative to the
reference instruments is derived in section 4, and the accuracy of the corrected data are
evaluated.
2. Instrumentation and Data
a. RS92
Vaisala radiosondes use thin-film capacitance RH sensors, where a hydrophilic polymer
layer on a glass substrate acts as the dielectric of a capacitor. The capacitance measured by the
radiosonde is proportional to the number of water molecules captured at binding sites in the
polymer structure, which in turn is proportional to the ambient water vapor concentration. The
sensor calibration relates the measured capacitance to the RH with respect to liquid water at
+25°C, and then compensates for temperature using a sensor temperature-dependence model.
5
The calibration is optionally adjusted during the ground check (GC) procedure prior to launch,
where the RH sensor is sealed in a container of desiccant assumed to be at 0.0% RH, and the
radiosonde measurement under these conditions is used as an RH offset correction in the
calibration. Unlike earlier RS80 radiosondes, RS90 and RS92 radiosondes use dual RH sensors
that are alternately heated while the other sensor measures the ambient RH, which eliminates the
problem of sensor icing in liquid water clouds or ice-supersaturated conditions. The operational
principles of Vaisala radiosonde RH sensors are described further by M06 and references therein.
The following sources of RS92 measurement error are considered in this study:
• Mean calibration bias reflects the absolute accuracy of the Vaisala calibration references and
their variation with time, which Paukkunen et al. [2001] estimates as 0.6-2% RH over the range
0-90% RH. Calibration bias also arises from inaccuracy in the Vaisala calibration model,
including curve-fit error that is inherently a function of RH and T.
• Production variability is the random sensor-to-sensor variability relative to the mean
calibration accuracy that reflects such things as manufacturing variability and inhomogeneous
conditions within the calibration chamber. A complete accuracy specification requires not only
an estimate of the mean (bias) uncertainty but also a measure of the variability, such as the
standard deviation of differences from the mean for a batch of sensors.
• Time-lag error arises from slow sensor response to changing RH conditions at low
temperatures, which has the effect of “smoothing” the RH profile in the UT and LS. This study
uses the time-lag correction described by M04, which is a numerical inversion algorithm that
recovers the “true” shape of the RH profile from the measured RH and T profiles based on
measurements of the sensor time-constant and its dependence on temperature.
• Solar radiation error is a dry bias in daytime measurements caused by solar heating of the RH
sensor [Vömel et al., 2007]. The magnitude of the SRE depends on the incident solar flux and is
therefore a function of numerous factors including the solar altitude angle (α), height (or
pressure), the angle between the sun and the sensor normal, the cloud optical depth along a sun-
sensor line, and the transmissivity of the airmass. The net heating of the RH sensor is also
![Figure 17. RS92 percentage difference from CFH for the WAVES/MOHAVE dataset after applying an earlier RS92 correction derived from TCSP dual soundings [Vömel et al., 2007]. (top) Daytime comparison, and (bottom) Nighttime comparison. (a and d) Left panels show the altitude dependence of the percentage difference for the original RS92 data. (b and e) Center panels show the altitude dependence of the difference after applying the time-lag and TCSP](/figures/figure-17-rs92-percentage-difference-from-cfh-for-the-waves-2gna6spy.webp)
