Potential use of GPS observations for validation of numerical
models: an investigation of the diurnal humidity cycle using
radiosonde, and ECMWF Hirlam, LM models
Mariella Tomassini1 and Siebren de Haan2
1GFZ,
2KNMI,
Potsdam, Germany
De Bilt, The Netherlands
Abstract
Estimates of Integrated Water Vapour (IWV) retrieved using the Global Positioning
System (GPS), radiosondes (RS) and Numerical Weather Prediction Models (NWP)
are compared for two sites, Delft (The Netherlands) and Lindenberg (Germany). The
focus of this comparison is drawn towards the diurnal humidity cycle. Other
comparisons show root mean square difference of around 1-2 kg/m2. Hourly GPS
observations reveal a daily cycle of IWV quite different from that of radiosondes and
numerical models for the two months in summer under investigation. The major
difference occurs during daytime, with GPS indicating more water vapour than the
other systems.
Introduction
Atmospheric water vapour is highly variable in space and time depending on several
complex phenomena such as convection, precipitation, surface evapotranspiration,
turbulence etc. These phenomena are described in numerical weather prediction
(NWP) models via physical parametrisation but at the present there are very few
observations to validate the model humidity. Radiosondes are launched at few
locations and only 2 (sometimes 4) times per day, satellite observations, such as
ATOVS (Advanced TIROS Operational Vertical Sounding instruments) on board of
the polar orbiting NOAA satellites, cover the ocean well but over land regions the
sounding of humidity is difficult and the time frequency is limited. Thus, recently
available estimates of integrated water vapour (IWV) from Global Positioning System
(GPS) ground networks could provide a new precise insight into the variability of
humidity over land. GPS IWV can be derived at short time interval from a dense
network, thus with a very good spatial and temporal resolution. The precision of the
measurements is in the order of 1-2 kg/m2 shown by a large number of studies (see
e.g. Rocken, et al, 1997 and Liou et al, 2000), which is an acceptable value for
verification purposes.
In this report GPS IWV has been compared with IWV derived from radiosondes
profiles and from NWP model fields. The particular aspect, which was investigated, is
the diurnal variation of IWV. The stations Delft (DELF), The Netherlands, and
Lindenberg (LDBT), Germany, have been selected for the comparison since nearby
radiosonde launches are available four times a day. The periods for the statistic
evaluation are August 2001 (for GPS versus radiosondes) and August 2002 (GPS
versus model, and GPS versus radiosonde at Lindenberg). It must be said that one
month is too short a period to eliminate advectively induced IWV changes and to
obtain a reliable humidity diurnal cycle. However, some interesting deviations among
the different systems are found and they indicate the need of further investigations.
For example the bias found between GPS and models is not negligible and it could
cause problems when attempting to introduce the data in the model assimilation
system itself.
This report continues with a short discussion of the data sources. Next the
comparison of the diurnal humidity cycle is presented and the last part contains the
conclusions.
GPS data
The GPS data are near-real time data processed by GeoForschungsZentrum
Potsdam (GFZ), Germany, using the EPOS.P.V2 package (Gendt et al.,1995). They
are available every half-hour, although for the comparison only hourly data have
been considered. The IWV is obtained from the Zenith Total Delay (ZTD) with the aid
of the Bevis formula (Bevis et al., 1994) using the pressure and temperature
measured at the very close meteo station.
Radiosonde data
For the computation of radiosonde IWV, data of temperature and humidity at
significant levels or 10 second data have been used. For station DELF the closest
radiosonde site is De Bilt (code 06260), at 56 km distance and 27m below the level of
the GPS station. The data used in this study is 10 second data obtained with a
Vaisala RS90. Only data from August 2001 have been considered since from July
2002 the radiosonde is launched at De Bilt only at 00 UTC, and at other times at
airforce base Valkenburg, making the radiosondes measurements not homogenous.
The Lindenberg radiosonde (code 10393), Vaisala RS80, is only 2 km apart from the
GPS site and almost at the same height (only 8 m difference). For this radiosonde
only significant level information was available. The number of levels used in this
study is around 30 which introduces only small errors in IWV with respect to IWV
derived from a full level profile.
NWP Models
LM
The Lokal-Modell (LM) of the German Weather Service (DWD) is a non-hydrostatic
limited area model for central and western Europe, with a resolution of 7km on the
the horizontal and 35 levels on the vertical. The data assimilation is based on
nudging towards observations (synop, radiosonde, aircraft) and analysis fields are
available every hour (LM1AN).
ECMWF
IWV was retrieved from the ECMWF (European Centre for Medium-Range Weather
Forecasts) operational model at ~40 km resolution and with 60 levels. The
temperature and specific humidity at pressure levels are used for the computations of
IWV. These levels are from the four dimensional variational global analysis and IWV
is computed every 3 hours.
HIRLAM
The HIRLAM (High Resolution Limited Area Model) used in this comparison has a
resolution of 22 km and 31 levels on the vertical. The assimilation scheme used in
HIRLAM is an optimal interpolation. IWV data are retrieved every 3 hours from
(temperature and specific humidity) analysis fields.
COMPARISON OF IWV DIURNAL CYCLE
Means of IWV from GPS data, radiosondes (RS) and models at different time of the
day have been computed and compared during the month August 2001 and August
2002, and are shown in Figures 1-3:
Fig. 1 shows that around noon GPS measures higher IWV values then those from
RS, especially at station LDBT: at 12 UTC the bias GPS minus RS is 0.5 kg/m2 (Aug
01) for DELF and 1.5 kg/m2 (Aug 01) / 2 kg/m2 (Aug 02) for LDBT. Moreover at
station Lindenberg GPS and RS describe quite different daily cycles of IWV. For
example GPS IWV increases steadily from 00 UTC to 18 UTC (in August 2001 more
than 2 kg/m2), whilst the radiosonde IWV decreases from 00 UTC to 12 UTC (~1
kg/m2) and increases from 12 UTC to 18 UTC. At DELF there is a better agreement
in the diurnal variation, with both GPS and RS decreasing from 6 UTC to 12 UTC.
Also the standard deviation GPS minus RS shows variability with daytime and it is
largest at 12 UTC for Delft and at 18 UTC for Lindenberg. To be noted that the larger
standard deviation at DELF (~ 4 kg/m2 at 12 UTC) compared to LDBT (~2 kg/m2)
can be due to the larger distance between GPS and RS.
It is interesting to mention that another comparisons GPS versus radiosonde
confirms the above results. During four months in summer 2001 at site Camborne,
UK, it was found that the day time difference GPS minus RS is generally higher than
the night time difference, ~ 1 kg/m2 (John Nash, UkMet Office, personal
communication)
The comparisons of GPS against NWP models (Fig. 2 and Fig.3) show a common
aspect: all three models indicate a decrease of IWV from midnight to midday, in the
order of 1 mm, and an increase afterwards, more precisely from 12 UTC to 15 UTC
for all cases and from 15 UTC to 18 UTC for LM at Delft. The increase of IWV is in
the order of 1 kg/m2 for ECMWF and LM, whilst for HIRLAM it is almost 3 kg/m2 in
three hours. The GPS IWV has the opposite behaviour, i.e. it increases from 9 UTC
to 15 UTC, so that the GPS minus model IWV bias has a maximum around noon.
This result is very similar to that seen in the comparison GPS against radiosonde and
it is not so surprising since models do use radiosonde data.
It is interesting to note some difference in the GPS cycle at the two locations. At the
inland site LDBT a marked increase of humidity (~ 1 kg/m2) starts around 06 UTC (5
local time) and ends at 15 UTC (14 local time). The GPS cycle at DELF, which is a
site near the coast and therefore influenced by sea breeze, presents a local minimum
at 9 UTC and an increase of 0.8 kg/m2 in the following three hours.
COMMENTS
Hourly GPS observations reveal a daily cycle of IWV quite different from that of
radiosondes and numerical models for the two months in summer under
investigation. The major difference occurs during daytime, with GPS indicating more
water vapour than the other system. Radiosonde observations are known to have a
dry bias, however, in the case of Lindenberg several corrections are applied to
reduce this bias (Leiter, 1997). Besides it is not clear which would be the cause of
systematic day-night difference. Just to have an idea, a systematic error of 1 kg/m2 in
IWV corresponds approximately to a systematic error of more than 2% in relative
humidity at all levels. A known daytime dependent source of error in radiosonde
observations, the so-called sensor arm heating error, can produce lower relative
humidity in the first 200-300 m of the ascent (Wang et al., 2002). The upper limit of
this error is in the order of 5%, which would result in a very small deviation in IWV at
mid-latitude summer, in the order of 0.2 kg/m2.
Three facts can cause the distinction in difference in diurnal cycle between RS and
GPS for the two sites. Firstly, the RS IWV observed at Lindenberg is retrieved using
significant profile data and RS IWV observed at De Bilt is calculated using 10-second
data. This may introduce a bias that is expected to be more or less constant during
the day. Secondly, the RS types are different: RS Lindenberg is an RS80, while De
Bilt is RS90. The latter RS type has fewer problems due to improved humidity sensor,
which is alternately heated to reduce errors due to contamination of icing. A said
before, for RS Lindenberg several corrections are applied to reduce this bias. Thirdly,
the difference in location introduces differences in diurnal humidity: Lindenberg lies
inland and Delft lies close to the North Sea coast.
In order to use GPS data for validation of the humidity in numerical models and
to assimilate them in model analysis to improve weather forecast, it is
important to investigate those aspects of the data processing which could
cause systematic errors, such ionosphere effect, ocean tide loading,
conversion from ZTD to IWV, influence of cloud liquid water etc.
References
Bevis, M., S. Businger, S. Chiswell, T.A. Herring, R. A. Anthes, C. Rocken, and R.H.
Ware, 1994, GPS Meteorology: Mapping Zenith Wet Delays onto Precipitable Water,
Journal of applied meteorology, Vol. 33, 379-386.
Rocken, C., T. van Hove, and R. Ware, 1997, Near real-time GPS sensing of
atmospheric water vapor, Geophysical Research Letters, Vol. 24, No. 24, 3221-3224.
Liou Y.-A., C.-Y. Huang, and Y.-T. Teng, 2000, Precipitable water observed by
ground-based GPS receivers and microwave radiometry, Earth Planets Space 52,
445-450
Leiterer, U., 2002: Improvements in Radiosonde Humidity Profiles Using RS80/RS90
Radiosonde of Vaisala, Betr. Phys. Atmosph., 70, 319-336
Wang, J., H. L. Cole, D. J. Carlson, E. R. Miller, K. Beierle, A. Paukkunen, and T.K.
Laine, 2002: Corrections of Humidity Measurements Errors from the Vaisala RS80
Radiosonde – Application to Toga-Coare Data, J. Atmos. Oceanic. Tech., 19, 9811002.
Fig. 1 Monthly mean at given hours of IWV from GPS data (GPS) and IWV from
radiosonde (RS), GPS minus radiosonde (bias) and standard deviation (std) for: (a)
GPS DELF and radiosonde De Bilt, August 2001; (b) GPS LDBT and radiosonde
Lindenberg, August 2001; (c) GPS LDBT and radiosonde Lindenberg, August 2002.
GPS DELF - RS De Bilt Aug 2001
28,0
4,0
27,0
3,0
2,0
26,0
1,0
25,0
0,0
24,0
-1,0
0
6
12
18
bias & std [mm]
IWV [mm]
(a)
0
hour (UTC)
GPS
RS
bias
std
IWV [mm]
29,0
2,5
2,0
1,5
28,0
27,0
1,0
0,5
0,0
-0,5
26,0
25,0
24,0
0
6
12
18
bias & std [mm]
GPS LDBT - RS Lindenberg Aug 2001
(b)
0
hour [UTC]
GPS
(c)
RS
bias
std
31,0
2,5
30,0
2,0
29,0
1,5
28,0
1,0
27,0
0,5
26,0
0,0
0
6
12
18
0
hour [UTC]
GPS
RS
bias
std
bias & std [mm]
IWV [mm]
GPS LDBT - RS Lindenberg Aug 2002
Fig. 2 DELF, August 2002. Mean of IWV from GPS data (GPS) and IWV from models
(a) ECMWF, (b) LM1AN, (c) HL22, and mean of GPS minus model IWV (bias) and
standard deviation (std) .
IWV[mm]
30,0
4,0
29,0
3,0
28,0
2,0
27,0
1,0
26,0
0,0
25,0
bias&std[mm]
GPS DELF - ECMWF Aug 02
(a)
-1,0
0
3
6
9
12
15
18
21
hour [UTC]
GPS
ECMWF
bias
std
GPS DELF - LM1AN Aug 02
30,0
4,0
29,0
3,0
28,0
2,0
27,0
1,0
26,0
0,0
25,0
-1,0
0
3
6
9
12
15
18
bias&std[mm]
IWV[mm]
(b)
21
hour[UTC]
GPS
LM1AN
bias
std
GPS DELF - HL22 Aug 02
30,0
4,0
29,0
3,0
28,0
2,0
27,0
1,0
26,0
0,0
25,0
-1,0
0
3
6
9
12
15
18
21
hour [UTc
GPS
HL22
bias
std
bias&std[mm]
IWV[mm]
(c)
Fig. 3 LDBT, August 2002. Mean of IWV from GPS data (GPS) and IWV from models
(a) ECMWF, (b) LM1AN, (c) LM1MO,(d) Hirlam (HL22), mean of GPS minus model
IWV (bias) and standard deviation (std).
(a)
31,0
3,0
30,0
2,0
1,0
29,0
0,0
28,0
-1,0
27,0
bias&std[mm]
IWV[mm]
GPS LDBT - ECMWF Aug 02
-2,0
0
3
6
9
12
15
18
21
hour [UTC]
GPS
(b)
ECMWF
bias
std
GPS LDBT - LM1AN Aug 02
3,0
2,0
30,0
1,0
29,0
0,0
28,0
-1,0
27,0
bias&std[mm]
IWV[mm]
31,0
-2,0
0
3
6
9
12
15
18
21
hour[UTC]
GPS
LM1AN
bias
std
IWV[mm]
34,0
33,0
32,0
31,0
30,0
29,0
28,0
27,0
4,0
3,0
2,0
1,0
0,0
-1,0
-2,0
-3,0
-4,0
0
3
6
9
12
15
18
21
hour [UTc
GPS
HL22
bias
std
bias&std[mm]
GPS LDBT - HL22 Aug 02
(c)