SOLA, 2008, Vol. 4, 013 016, doi:10.2151/sola.

2008 004 13

Correction for Radiation Dry Bias Found in RS92 Radiosonde Data

during the MISMO Field Experiment
Kunio Yoneyama1, Mikiko Fujita1, Naoki Sato1, Masatomo Fujiwara2, Yoichi Inai2,
and Fumio Hasebe2
1
Institute of Observational Research for Global Change, JAMSTEC, Yokosuka, Japan
2
Graduate School of Environmental Science, Hokkaido University, Sapporo, Japan

Abstract data with the SW humidity data and the GPS-derived

PWV data showed a clear dry bias only in the daytime,
Atmospheric soundings using the Vaisala RS92 suggesting a solar radiation-induced dry bias.
radiosonde were intensively conducted during the field In general, the influence of solar radiation on
experiment MISMO (Mirai Indian Ocean cruise for the humidity data is larger in the upper troposphere.
Study of the MJO-convection Onset) in the central and However, the significant difference of PWVs between
eastern equatorial Indian Ocean from October to GPS and radiosonde indicates that humidity data in the
December 2006. By comparing the RS92 relative humi- middle and lower troposphere are also affected. As the
dity data with that from the Meteolabor Snow White main purpose of MISMO was to observe the atmos-
(SW) chilled-mirror dew/frost-point hygrometers pheric condition when convection in the Madden- Julian
launched on the same ship around the local noon time, Oscillation (MJO; Madden and Julian 1994) was initi-
the dry bias was found to increase significantly with ated, we cannot neglect such an error in humidity data.
height. In addition, it was also revealed that the dry bias Hence, this study evaluated and corrected the dry
had a clear diurnal variation with its maximum at local bias of RS92 humidity data taken during the MISMO
noon and near-zero at night from the comparison of experiment using data from other different observa-
precipitable water vapor (PWV) with that derived from tions. First, we estimated and reduced the daytime dry
the shipboard Global Positioning System (GPS) data. bias by utilizing the Meteolabor SW hygrometer data.
Therefore, the dry bias of the RS92 data could be attrib- We then evaluated the corrected humidity data by com-
uted to a solar radiation-induced error that was recently paring PWV with that derived from GPS data.
discussed by V
omel et al. (2007).
In this study, we developed a correction scheme for
the MISMO RS92 humidity data as a function of 2. Data
pressure and local time using SW data, and then con-
firmed its validity with GPS-derived PWV. During the MISMO field experiment, the Mirai
stayed within the range of 2°S 3°N, 79°E 82°E from
October 22 through November 25, 2006. Atmospheric
1. Introduction soundings using the Vaisala RS92 radiosonde were con-
ducted every 3 hours (277 times) during this period, and
Radiosonde observations provide basic and useful other 52 soundings were conducted along the equator in
information on the atmospheric condition. However, December. During the stationary observation period (i.e.,
many researchers found dry bias in humidity data and first period), fifteen special soundings which included
extensive efforts have been made to identify the error the Meteolabor SW hygrometers were also carried out
sources and to reduce the bias from original data (e.g., to investigate the transport and dehydration processes
Guichard et al. 2000; Wang et al. 2002; Nakamura et al. around the tropical tropopause. Since the SW hygrome-
2004; Miloshevich et al. 2006; H
aberli 2006). Recently, ter directly measures dew/frost-point based on a
V
omel et al. (2007) showed a large solar radiation- chilled-mirror principle, SW data have been used as a
induced dry bias in humidity data obtained by the reference for evaluation of the accuracy of humidity
Vaisala RS92 radiosonde which is now widely used in data taken by a capacitive thin-film sensor including
the world. This dry bias is mainly caused due to the lack the Vaisala radiosondes (e.g., Wang et al. 2003). Detailed
of radiation protective shields for RS92 model. While technical characteristics of the Meteolabor SW hygro-
manufacturers should improve the skill to produce meter and the Vaisala RS92 radiosonde were examined
reliable instruments, researchers should also be aware of in several studies (e.g., Fujiwara et al. 2003; V

omel et al.
data accuracy for precise process studies as well as 2003; Miloshevich et al. 2006). Since the SW hygro-
climate studies. meters were launched separately with the RS92 radio-
In October December 2006, the field experiment sonde, the RS92 data at the nearest time were compared
MISMO (Yoneyama et al. 2006) took place in the equato- with the SW data (Table 1). For comparison, both RS92
rial Indian Ocean centered at 0°, 80.5°E using the and SW data were converted onto 5-hPa interval data
research vessel Mirai, and a total of 329 Vaisala RS92 from different sampling time data sets. Note that all the
radiosondes were launched. During the same period, RS92 radiosondes used in the MISMO were the latest
observations using the Meteolabor “Snow White (SW)” model, which can be confirmed from their serial
chilled-mirror dew/frost-point hygrometers were con- numbers listed in Table 1 (i.e., First three characters
ducted 15 times, and most of them (14 times) were (YWW) of serial number indicate the calibration year
launched near local noon. In addition, precipitable water (Y; B = 2006) and week number (WW) (e.g., Miloshevich
vapor (PWV) from the Global Positioning System (GPS) et al. 2004). So, in this case sensor calibrations at the
measurements of zenith tropospheric delay (ZTD) was manufacturer were conducted in the weeks from 34 to
also continuously observed during the entire cruise. As 38 of the year 2006, that is 1 2 months prior to the ex-
will be shown later, comparison of the RS92 humidity periment.).
In addition to the radiosonde observation, continu-
Corresponding author: Kunio Yoneyama, Institute of Observa-
ous PWV measurement was carried out using GPS
tional Research for Global Change (IORGC), Japan Agency for receiver on the ship (Rocken et al. 2005, 2006). During
Marine-Earth Science and Technology (JAMSTEC), 2-15, Na- the entire cruise, the position of GPS antenna and ZTD
tsushima-cho, Yokosuka 237-0061, Japan. E-mail: yoneyamak were measured every 5 seconds. For data processing,
@jamstec.go.jp. ©2008, the Meteorological Society of Japan. global mapping function that was recently provided by
14 Yoneyama et al., Correction for Radiation Dry Bias

Boehm et al. (2006) was used. PWV was then estimated Table 1. Launch time of RS92 and SW used for comparison.
from ZTD with surface pressure and temperature data Solar zenith angle indicates the value when RS92 radiosonde
(Bevis et al. 1992). Since the GPS-derived PWV is one of was launched.
the most reliable parameters and is often used to
evaluate the quality of radiosonde humidity data (e.g., Time (LST) Nominal RS92 Solar Zenith
No.
Nakamura et al. 2004; Van Baelen et al. 2005; Wang and Day SW RS92 Time Serial No. Angle (deg)
Zhang 2007, manuscript submitted to J. Climate), com-
parison between PWVs derived from observed/cor- 1 Oct. 29 12:02 10:47 11:00 B3540428 15.9
rected radiosonde humidity data and from GPS was 2 Oct. 30 14:01 13:25 14:00 B3551054 33.5
made to evaluate the correction. In this study, hourly- 3 Nov. 1 11:10 10:30 11:00 B3550610 19.2
mean GPS-derived PWV data were used, since it took 4 Nov. 2 11:12 10:30 11:00 B3450300 19.4
about one hour for radiosonde sounding to observe the 5 Nov. 3 11:15 10:30 11:00 B3551140 19.7
entire troposphere. 6 Nov. 4 11:17 11:02 11:00 B3551142 16.1
7 Nov. 7 13:00 13:29 14:00 B3550592 35.3
8 Nov. 8 14:00 13:30 14:00 B3550725 35.7
3. Dry bias and its correction 9 Nov. 9 14:00 13:30 14:00 B3551261 35.8
10 Nov. 11 11:12 10:30 11:00 B3550625 21.6
Figure 1 shows the relative difference of relative 11 Nov. 12 11:13 10:45 11:00 B3551073 19.9
humidity (RH) between RS92 and SW as a function of 12 Nov. 13 11:09 10:47 11:00 B3550953 19.9
pressure height for 14 cases (No. 1 14 in Table 1). The 13 Nov. 14 11:13 10:31 11:00 B3610667 22.1
observations were conducted near local noon when the 14 Nov. 15 11:10 10:30 11:00 B3611007 22.5
influence of solar radiation was large. It is obvious that 15 Nov. 20 17:02 16:41 17:00 B3830442 80.0
RS92 data have a dry bias compared to SW data and
their difference increases with height. While the differ-
ence is almost zero at the surface, it reaches over 40% in
the upper troposphere above 150-hPa level. Figure 2 il-
lustrates the RH profiles of SW and RS92 for the case of
local evening launch (No. 15 in Table 1). It shows no sig-
nificant dry bias (mean difference calculated over the
troposphere is 0.6% RH). These results strongly suggest
that the dry bias found in Fig. 1 is attributed to the in-
fluence of solar radiation as discussed by V
omel et al.
(2007). It is worth noting that the mean bias at the
surface is almost zero in this study, while relative differ-
ence of about 9% was found in the study of V
omel et al.
(2007, see Figs. 6 and 8 of their paper). At present, we
cannot provide a definitive explanation on this differ-
ence. One possible factor, however, may relate to the dif-
ference of observational location; one is on land, while
the other is at sea, because of their different albedos.
Further studies may be needed. Note that we launched
radiosondes from the launcher system situated in the “
sea-container” on-board the Mirai. Thus, there should be
no influence of the so-called sensor arm heating error Fig. 1. Relative difference between RS92 RH and SW RH (= 100
(e.g., Wang et al. 2002) which is attributed to the radia- × (RHRS92 RHSW)/RHSW, black dot). Results are shown for 14 sets
of near local noon observation. The red line indicates the mean
tion heating of the senor arm prior to the launch and in- profile, and blue horizontal lines indicate the standard devia-
fluences the data only taken during the first minute tions. The green curve indicates the profile obtained by a poly-
from the launch. nomial fitting to the mean profile. See text for details.
As mentioned, Figure 1 is obtained from observa-
tions conducted near local noon, and we do not have
enough data at other local times for SW data. Thus, we
compare the PWVs derived from RS92 and from GPS to
examine the diurnal variation of dry bias. Figure 3a
shows the difference of PWV between RS92 and GPS as
a function of local standard time (LST; UTC+5). For
comparison, all the RS92 and GPS data from October 22
through November 25 were used. One should note that
similar results were obtained in both cases, when we
limited the analysis period to the one between October
29 and November 15 when SWs were launched, and
when we extended the period to December 10 including
entire 329 radiosonde data (not shown). In Fig. 3a, the
difference is significant only in the daytime, suggesting
again that the dry bias is primarily caused by solar ra-
diation. While the daily-mean difference of PWV is 1.6
mm with +/ 2.6 mm standard deviation, their mean dif-
ference at 1100 LST and 1400 LST are 3.7 mm and 4.1
mm, respectively. Note that we also calculated the PWV
from 14 SW near local noon data to confirm the rela-
tionship between two different data sets. In result, we
confirmed that they showed a good agreement having
their mean difference (SW-derived PWV minus GPS-
derived PWV) of 0.5 mm with +/ 1.5 mm standard de-
viation.
Based on these results as well as the work of V
omel Fig. 2. RH profiles observed by SW (red) and RS92 (blue) which
et al. (2007), we will then determine the correction factor were launched in the local evening (No. 15 in Table 1).
 SOLA, 2008, Vol. 4, 013 016, doi:10.2151/sola.2008 004 15

Fig. 5. Frequency distributions of RH binned by 5% obtained

from SW (blue), observed RS92 (black), and corrected RS92
(red) for 14 cases (No. 1 14 in Table 1).

RHdif = 5.61(ln P)2 + 88.17(ln P) 341.51. (1)

Note that in this study the time-lag correction
(Miloshevich et al. 2004) was not applied, since we did
not have data on sensor time-constant. However, as
V
omel et al. (2007) mentioned, the time-lag error became
noticeable at the uppermost troposphere. Therefore, we
could obtain Eq. (1) using only the data below the 150-
hPa level. Since the solar radiation is primarily a
function of solar zenith angle, we assume that the cor-
rection factor is proportional to the solar zenith angle

as,
Fig. 3. PWV difference (gray triangle) between RS92 and GPS c = cos
/cos
m, (2)
data (RS92 minus GPS) as a function of LST for RS92 data (a)
before, and (b) after the correction. Blue diamond mark indi- where c is a multiple coefficient for correction profile
cates the mean value and light blue bar indicates standard de- RHdif and
m is the mean solar zenith angle of 14 data
viation. As for the correction, see the text. (= 24.1°, Table 1). c is set to zero at nighttime data. Solar
zenith angle
is obtained from the local time and
position for each radiosonde observation. Using these
correction factors, the corrected RH, RHcor , can be ob-
tained from,
RHcor = (100/(100 + c × RHdif)) × RHobs, (3)
where RHobs is the observed RH.
As shown in Fig. 4, where the mean difference
between the corrected RS92 data and SW data varies
around 0% through the entire troposphere and it is
much smaller than the designed accuracy of radiosonde
data, the corrections were properly conducted. The
result of correction for all sounding data can be evalu-
ated by comparing the frequency distributions of RH
binned by 5% between observed/corrected RS92 data
and SW data (Fig. 5). The frequency distribution of
RS92 RH data better agrees with that of SW RH after
the correction. In the higher RH range (mainly corre-
sponding to the lower troposphere), the peak of fre-
quency is located at 70 85% in the observed RS92 data.
However, it shifts to 80 90% in the corrected one. In the
lower RH range (corresponding to the higher tropo-
Fig. 4. Difference of RH between corrected RS92 and SW data sphere), on the other hand, a peak at 10 20% in the
(black dots). The red line indicates the mean difference and observed data almost disappeared in the corrected one.
blue lines indicate the standard deviations. In order to evaluate the correction using different
data source, PWV was recalculated using corrected
RS92 humidity data and compared with GPS-derived
PWV (Fig. 3b). Although there still remain small dry
for the MISMO RS92 humidity data using SW data. biases in daytime data (1100 and 1400 LST), their mean
First, as shown in the vertical profile of relative dif- differences ( 1.3 mm at 1100 LST and 1.8 mm at 1400
ference of RH in Fig. 1, we obtained the correction LST) are comparable to that of nighttime data. As for
profile, RHdif , by applying a polynomial fitting (green the morning (0800 LST) and evening (1700 LST) cases,
solid line) to the mean profile (red solid line) as a their differences become close to zero. Note that we con-
function of pressure P in hPa as follows. firmed the correction without regard to the solar zenith
16 Yoneyama et al., Correction for Radiation Dry Bias

angle dependence expressed as Eq. (2). While we could Boehm, J., A. Niell, P. Tregoning, and H. Schuh, 2006: Global
also obtain similar results for near local noon data, Mapping Function (GMF): A new empirical mapping
morning and evening data show rather wet bias (not function based on numerical weather model data.
shown). Therefore, taking Eq. (2) into the correction Geophys. Res. Lett., 33, L07304, doi:10.1029/2005GL
scheme is effective mainly for low solar zenith angle 025546.
data. As for the remaining differences, it can be attrib- Fujiwara, M., M. Shiotani, F. Hasebe, H. V
omel, S. J. Oltmans, P.
uted to various possible factors such as spatial and W. Ruppert, T. Horinouchi, and T. Tsuda, 2003: Perform-
temporal sampling error for the two different measure- ance of the Meteolabor “Snow White” chilled-mirror hy-
grometer in the tropical troposphere: Comparisons with
ments, uncertainties in GPS-derived PWV, insufficient the Vaisala RS80 A/H-Humicap sensors. J. Atmos. Oceanic
correction scheme without calibration and time-lag cor- Technol., 20, 1534 1542.
rections (Miloshevich et al. 2006), and so on. However, in Guichard, F., D. B. Parsons, and E. R. Miller, 2000: Thermody-
general, the present results show the good correspon- namic and radiative impact of the correction of sounding
dence of PWVs derived from different data sources. humidity bias in the Tropics. J. Climate, 13, 3611 3624.
H

aberli, C., 2006: Assessment, correction and impact of the dry
bias in radiosonde humidity data during the MAP SOP.
4. Summary and concluding remarks Quart. J. Roy. Meteor. Soc., 132, 2827 2852.
Madden, R. A., and P. P. Julian, 1994: Observations of the 40
This paper described the dry bias found in Vaisala 50-day tropical oscillation A review. Mon. Wea. Rev.,
RS92 radiosonde data obtained during the MISMO field 122, 814 837.
experiment in October December 2006. First, a correc- Miloshevich, L. M., A. Paukkunen, H. V
omel, and S. J. Oltmans,
tion scheme was developed using Meteolabor SW 2004: Development and validation of a time-lag correction
chilled-mirror hygrometer data, and then the results of for Vaisala radiosonde humidity measurements. J. Atmos.
correction were evaluated using GPS-derived PWV. In Oceanic Technol., 21, 1305 1327.
this study, although we had only near local noon SW Miloshevich, L. M., H. V
omel, D. N. Whiteman, B. M. Lesht, F. J.
data, correction was applied to all daytime humidity Schmidlin, and F. Russo, 2006: Absolute accuracy of
water vapor measurements from six operational radio-
data by assuming that solar radiation error was propor- sonde types launched during AWEX-G and implications
tional to solar zenith angle. Since independent measure- for AIRS validation. J. Geophys. Res., 111, D09S10, doi:
ments (SW and GPS) showed a consistent result, it could 10.1029/2005JD006083.
be regarded that the correction scheme developed in Nakamura, H., H. Seko, Y. Shoji, Aerological Observatory, and
this study including the assumption of Eq. (2) was Meteorological Instruments Center, 2004: Dry biases of
properly performed. In other words, if GPS-derived humidity measurements from the Vaisala RS80-A and
PWV data which are much easier to obtain than simul- Meisei RS2-91 radiosondes and from ground-based GPS.
taneous chilled-mirror hygrometer data are available, it J. Meteor. Soc. Japan, 82, 277 299.
is possible to confirm the existence of dry bias in Rocken, C., J. Johnson, T. Van Hove, and T. Iwabuchi, 2005: At-
Vaisala RS92 data, and it might be possible to reduce mospheric water vapor and geoid measurements in the
them using the relationship obtained in this study from open ocean with GPS. Geophys. Res. Lett., 32, L12813,
SW data. Similar results (bias at the surface is almost doi:10.1029/2005GL022573.
zero, and about 30% at 150-hPa level) were obtained in Rocken, C., Z. Lukes, L. Mervart, J. Johnson, T. Iwabuchi, and M.
the comparison between RS92 and SW data of 11 quasi- Kanzaki, 2006: Real-time ionospheric and atmospheric
simultaneous observations during another Mirai MR04- corrections for wide area single frequency carrier phase
08 cruise that took place off Palau Islands in the tropical ambiguity resolution. Proc. ION-GNSS 19th Intl. Technical
western Pacific Ocean from December 2004 to January Meeting of the Satellite Division, Inst. of Navigation, Fort
Worth, Texas. 1208 1218.
2005 (not shown here, the cruise report of MR04-08 is Van Baelen, J., J.-P. Aubagnac, and A. Dabas, 2005: Comparison
available at http://www.jamstec.go.jp/mirai/). of near-real time estimates of integrated water vapor
As noted in Yoneyama et al. (2008, manuscript sub- derived with GPS, radiosondes, and microwave radiome-
mitted to Bull. Amer. Meteor. Soc.), MISMO data have ter. J. Atmos. Oceanic Technol., 22, 201 210.
been opened to public through the MISMO web site at V

omel, H., M. Fujiwara, M. Shiotani, F. Hasebe, S. J. Oltmans,
http://www.jamstec.go.jp/iorgc/mismo/ since January and J. E. Barnes, 2003: The behavior of the Snow White
2008, so that any researchers can use these data for me- chilled-mirror hygrometer in extremely dry conditions. J.
teorological and climatological studies over the Indian Atmos. Oceanic Technol., 20, 1560 1567.
Ocean. As for the radiosonde data, corrected data shown V

omel, H., H. Selkirk, L. Miloshevich, J. Valverde-Canossa, J.
in this study are available for use. Vald es, E. Kyr
o, R. Kivi, W. Stolz, G. Peng, and J. A. Diaz,
2007: Radiation dry bias of the Vaisala RS92 humidity
sensor. J. Atmos. Oceanic Technol., 24, 953 963.
Acknowledgments Wang, J., H. L. Cole, D. J. Carlson, E. R. Miller, K. Beierle, A.
Paukkunen, and T. K. Laine, 2002: Corrections of
The authors would like to express their sincere humidity measurement errors from the Vaisala RS80
thanks to technical staff of Global Ocean Development Radiosonde-application to TOGA COARE data. J. Atmos.
Inc. for their observation support on-board the Mirai. Oceanic Technol., 19, 981 1002.
Wang, J., D. J. Carlson, D. B. Parsons, T. F. Hock, D. Lauritsen, H.
They are also grateful to Dr. Junhong Wang of the L. Cole, K. Beierle, and E. Chamberlain, 2003: Performance
National Center for Atmospheric Research for providing of operational radiosonde humidity sensors in direct com-
useful comments on dry bias. Thanks are extended to parison with a chilled dew-point hygrometer and its
Ms. N. Miyamoto for her assistance with the preparation climate implication. Geophys. Res. Lett., 30, 1860, doi:
of this manuscript. Comments from two anonymous re- 10.1029/2003GL016985.
viewers helped to improve the manuscript. Yoneyama, K., Y. Masumoto, Y. Kuroda, M. Katsumata, and K.
Mizuno, 2006: MISMO: Mirai Indian Ocean cruise for the
Study of the MJO-convection Onset. CLIVAR Exchanges,
References 39, 8 10, 16. (Available from http://www.clivar.org/)

Bevis, M., S. Businger, T. A. Herring, C. Rocken, R. A. Anthes, Manuscript received 31 October 2007, accepted 6 February 2008
and R. H. Ware, 1992: GPS meteorology: Remote sensing SOLA: http://www.jstage.jst.go.jp/browse/sola/
of atmospheric water vapor using the Global Positioning
System. J. Geophys. Res., 97, 15787 15801.