GPS Water Vapour Experimental Results From Observations of the

Australian Regional GPS Network (ARGN)

Yanming FENG, Zhengdong BAI
Cooperative Research Centre for Satellite Systems
Queensland University of Technology
GPO Box 2434, Q4001, Australia
Phone 07 38641363, fax 07 38641517
y.feng@qut.edu.au, z.bai@qut.edu.au

Peng FANG
Scripps Institute of Oceanography
University of California, San Diego
San Diego, 9500 Gilman Dr., La Jolla, CA 92093
pfang@josh.ucsd.edu

Allyson WILLIAMS
Queensland Climate Centre for Applications
Queensland Department of Primary Industry
203 Tor Street, Toowoomba QLD 4350
WilliaAJ@prose.dpi.qld.gov.au

Key Words: GPS Meteorology, Integrated Precipitable Water Vapour, Radiosonde,GPS

network

Abstract

Water vapour is one of the most important components of the Earth’s Atmosphere. It is
the source of precipitation, and its latent heat is a critical ingredient in the dynamics of
most major weather events. As a greenhouse gas, water vapour also plays a critical role
in the global climate system. However, water vapour has been one of the most poorly
measured and least understood components of the Earth’s atmosphere in the past.
Recent advances in the Global Positioning System (GPS) have enabled atmospheric
parameters such as integrated water vapour (IWV) to be determined to a high degree of
accuracy on a routine basis, using continuously operating ground-based GPS receivers
and advanced GPS analysis software. Compared to the traditional water vapour
observing systems, such as radiosondes, the GPS water vapour technique is more
advantageous because of its low-cost, high-measurement accuracy, all weather
operability and long-term measurement stability.

This paper experimentally studies the GPS water vapour estimation techniques using
GPS observations from the Australian regional GPS network (ARGN), with focus on
GPS integrated precipitable vapour (IPW) estimation, radiosonde-IPW estimation (Rad-
IPW), estimation of weighted averaged mean temperature and IPW solution analysis.
Physically GPS-IPW and Rad-IPW solutions are identical. In Australia, there are 9
radiosonde sites in the Australian Upper Air Network (AUAN) located close to their

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

GPS sites in the ARGN network. Our experimental studies are based on the GPS and
radiosonde data collected from five of these sites during the period of July to September
2000 under all weather conditions. These stations are Alice Springs, Townsville,
Darwin, Hobart, and Cocos Island, and also providing half-hourly surface
meteorological data so that direct comparisons between Rad-IPW and GPS-IPW at
these sites can be conducted.

The experimental results have demonstrated that the IPW rms accuracy of 1mm has
been consistently achievable with GPS observations of the ARGN network and 1.2 mm
IPW accuracy achievable with radiosonde observations. About 500 comparisons
between GPS-IPW and Rad-IPW solutions have shown an agreement at a standard
deviation of 1.8 mm and a mean difference of -0.66 mm. Noticing that the theoretical
accuracy of each comparison is about 1.6mm, this agreement is considered very
reasonable and encouraging. Further studies may focus on the establishment of an
operational GPS-IPW estimation in existing GPS networks (governmental and private
GPS network infrastructures) in Australia and the improving the agreement or GPS-
IPW accuracy.

1. Introduction

The term ‘GPS meteorology’ is applied to the process of remote sensing of the
troposphere and stratosphere by measuring the refraction of GPS signals that propagate
through the atmosphere [1]. Traditionally, the propagation delay of GPS signals due to
the atmosphere is considered a nuisance parameter that hinders the ability to obtain
precise coordinates using the Global Positioning System. Recent development in GPS
precise positioning and orbit determination has enabled atmospheric parameters to be
determined to a high degree of accuracy on a routine basis, using continuous tracking
data from ground-based GPS receivers. The delay in the GPS signals due to the
atmospheric water vapour can be analysed and converted to Integrated precipitable
water vapour (IPW) [2, 3]. Water vapour is one of the most important constituents of
the atmosphere as it is the principal mechanism by which moisture and latent heat are
transported. As a greenhouse gas, water vapour also plays a critical role in the global
climate system: it absorbs and radiates energy from the sun, affects the formation of
clouds and aerosols, and the chemistry of the lower atmosphere. Despite its importance
in climate and weather prediction, water vapour has been one of the most poorly
measured and least understood constituents of the Earth's atmosphere. GPS can
dramatically improve this situation, because of the following distinct advantages that
the technique enjoys:
• High temporal resolution: GPS can provide IPW estimates as accurate as those
derived from the commonly used radiosonde instruments, but with a temporal
resolution of 30 seconds to 30 minutes (compared with weather balloons which are
released every 12 or 24 hours), all-weather operability and long-term measurement
stability.
• Low cost equipment/maintenance: Geodetic quality GPS receivers are
comparatively modestly priced ($15-20K at present), but operate unattended for
long periods with high reliability, and with low telemetric data transmission cost.
On the other hand, the more expensive radiosondes need regular calibration, and
there is an operational cost of $500 per release for each radiosonde station.

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

• High spatial resolution possible: In many countries or regions, networks of tens,
and even hundreds, of GPS receivers have been (or are being) established by
government agencies and private companies to support survey, geodetic and
navigation applications. With some additional capital investment such receiver
networks can be upgraded for water vapour measurements with unprecedented
spatial resolution.

If such a GPS-based IPW measuring system were implemented, near-term/short-range

weather forecasting and nowcasting would be possible, which may lead to reductions in
damages due to sudden severe weather conditions[4]. In addition, such data could
improve mid-term weather predictability, which is particularly useful for determining
the likelihood the onset of floods or the converse, severe drought [5]. Furthermore,
GPS-sensed tropospheric delay information can be used for image correction, for
example of satellite remote sensed SAR images[6]. Long-term climatology would also
benefit from the GPS-derived water vapour time series that would become globally
available [5].

Over the past few years, various government agencies around the world have launched
projects to establish ground-based GPS meteorological networks. For example, the
Forecasting System Laboratory (FSL), of the National Oceanic & Atmospheric
Administration, has established a GPS Integrated Precipitable Water (GPS-IPW)
Demonstration Network of 55 sites in the United States, which currently produces IPW
solutions on an hourly basis (http://www.fsl.noaa.gov). This network will be extended
to about 800 sites for operational weather forecasting purposes. In Japan,
meteorological organisations can use the GPS data from the GPS Earth Observing
Network (GEONET) of about 1000 stations to compute IPW data [7]. In Germany, a
large strategic research project known as the GPS Atmosphere Sounding Project
(GASP) commenced in 1997 to use the existing network of about 200 GPS receivers
established for surveying and navigational purposes to provide continuous water vapour
estimates for atmospheric studies. Extensive experiments have so far shown that
integrated PW can be obtained with an accuracy of better than 2mm (or at the 5% level)
from GPS observations. Hence it is now evident that the GPS-IPW technique is feasible.
The research on GPS meteorology in Australia so far has been limited to some
experimental studies, including the work on accuracy of GPS IPW solutions in
comparison to microwave water vapour radiometer (WVR) and radiosonde
measurements [8]. Recently, the Cooperative Research Centre for Satellite Systems at
QUT and Queensland Centre for Climate Applications (QCCA) have completed a joint
research project, which aims to experimentally study the GPS water vapour estimation
techniques using GPS observations from the Australian regional GPS network (ARGN).
We have developed a package of software tools and scripts for the computations
involved, such as GPS-IPW estimation, radiosonde-IPW estimation (Rad-IPW),
estimation of weighted averaged mean temperature, IPW solution analysis. Physically
GPS-IPW and Rad-IPW solutions are identical. In Australia, there are 9 collocated
radiosonde/GPS sites in Australian Upper Air Network and the ARGN network. Our
experimental studies are based on the GPS and radiosonde data collected for the period
of July to September 2000 at the five of these stations under all weather conditions.
They are Alice Springs, Townsville, Darwin, Hobart, and Cocos Island, which can also
provide half-hourly surface meteorological data so that direct comparisons between
Rad-IPW and GPS-IPW at these sites could be conducted.

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

In the following sections, we first outline the technical basis for GPS-IPW estimation
can comparison with Rad-IPW estimates, including a generalised GAMIT-based GPS
data processing scheme, GPS-IPW estimation and Rad-IPW estimation algorithms.
Secondly, we present the experimental results from these 5 co-located GPS/Rad
stations. We explore the uncertainty of GPS-IPW solutions and Rad-IPW solutions, and
we discuss the effects of the separation between GPS sites and Radiosonde sites as well
as the effects of weighted means temperature. Finally, findings of this experiment are
summarised.

2. GPS and Radiosonde Water Vapour Estimation

Determination of IPW from GPS observations

As the GPS signal travels from the satellite to the receiver, the atmosphere of the earth
causes a delay in the travel time of GPS signals. The delay caused by the charged
particles in the ionosphere can be eliminated using dual-frequency receivers. For the
signal frequency below 30GHz, propagation velocity in the neutral atmosphere depends
on the refractive index associated with temperature, pressure and water vapour. The
errors caused by slowing and bending of the GPS signals in the neutral atmosphere is
called the tropospheric signal delay. This delay reaches a value of about 2.0-2.5 metres
in the zenith direction and increases approximately with the 1/sine of the elevation angle
up to 20-28 metres at the angle of 5 degrees[9]. The delay in the slant direction can be
written as the product of a mapping function and the total delay in zenith direction. The
total zenith tropospheric delay (ZTD) can then be solved in the GPS data processing,
resulting in a time series of zenith delay values for every station [2].

The zenith delay can be divided in two parts, one caused by the total amount of dry gas-
hydrostatic zenith delay (HZD) and one caused by total water vapour constituent -wet
zenith delay (WZD). The hydrostatic delay can be calculated by an empirical model
using the surface pressure and temperature. There are a number of well-defined models,
such as Saastamoinen model[10], Hopfield model [11]and Black model[12]. In GPS
meteorological study, the most popular one is the Hopfield model, in which the
hydrostatic zenith delay is expressed as
P
HZD = 1.552 ⋅ (h − H )
T (1)
h ≡ 40.082 + 0.14898 ⋅ (T − 273.16)

where, T is the absolute temperature in Kevin, and h denotes the height of upper edge
troposphere in km. H is the station orthometric height in km, and P is the total air
pressure measurement in hPa at the station.

When the hydrostatic delay is subtracted from the total delay, the wet zenith delay
(WZD) is left:

WZD=ZTD-HZD (2)

WZD can be converted to Integrated Precipitable Water Vapour (IPW). Qualitatively,

the IPW can be related to the WZD by

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

 IPW = F ⋅ WZD (3)

where the mapping scale factor F is a dimensionless parameter, which is given by [3]
10 6
F= (4)
k 
ρ ⋅ Rv ⋅  3 + k 2 − w ⋅ k 1 
 Tm 
where ρ is the density of the liquid water, Rv is the specific gas constant for water

vapour, k1 , k 2 and k 3 are the atmospheric refractivity constants, and w is the mass ratio

of water vapour molecule to dry air molecule. Tm is a weighted mean temperature of

troposphere, which is defined as [1]

Tm =
∫ (e / T ) ⋅ dh (5)
∫ (e / T ) ⋅ dh
2

where e is the water vapour pressure and T is the absolute temperature in Kelvin. Water
vapor pressure e is calculated using temperature t and relative humidity RH [13]
7.5⋅t

e = RH ⋅ 6.11 ⋅ 10 t + 237.3 (6)

The vapor pressure is in hPa and t is in degree Celsius.

In practice, the upper-air sounding can provide a series of discrete temperature and
relative humidity measurements along weather balloon’s ascending path. These discrete
values profile temperature and water vapor distribution. In Equation (5), water vapor
pressure is needed. We first convert the relative humidity profile into the profile of
water vapor pressure through Equation (6). Actually, the discrete observations in a
profile, such as in temperature profile, separate the troposphere into many temperature
layers, and two sequential temperature measurements represent the temperature at the
bottom and top of each layer. If we assume temperature and water vapor pressure
variations in each layer are linear, Equation (5) can be approximately rewritten by

e
∑ T ⋅ (h i +1 − hi )
Tm = (7)
e
∑ T 2 ⋅ (hi +1 − hi )
In the above expression, the subscript i and i+1denote the bottom and the top of each
layer, h is the height above the mean sea level in meter, e and T are the average water
vapor pressure and temperature for the corresponding layer.

The core GPS data processing employes GAMIT software package[16], which was
originally designed for high precision geodetic positioning and GPS satellite orbit
determination. GAMIT uses double-differenced observables to perform weighted least

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

square adjustment on user selected unknown variables. The flowchart of our GPS-IPW
data processing using GAMIT is shown in Figure 1.

regional/local network
GPS observations
Precise orbits
Earth. Orientation
GAMIT software
Parameters (EOP)
Apriori positions Evaluation :
Radiosondes
Station surface IPW Satellite TPW
WVRs
Meteorological
NWP models
Data (T,P) NWP, climate,
geodesy,
monitoring etc )

Figure 1. GAMIT-Based GPS-IPW Data Processing Scheme

Radiosonde Integrated Precipitable Water Vapour Estimation

The Radiosonde data includes pressure, temperature and relative humidity profiles
above Radiosonde station. These data can be used to calculate the IPW. The Integrated
Water Vapor (IWV) along the path of the sounding balloon can be calculated by[1]

IWV = ∫ ρ v ⋅ dh (8)

where ρ v is the density of water vapor. Then the IWV can be mapped into IPW through
IWV
IPW = (9)
ρ
In the above expression, ρ is the density of liquid water. According to the gas state
equation, the water vapour density ρ v can be calculated by
e
ρv = W (10)
Rv ⋅ T

where Rv = 461.495 J ⋅ K −1 ⋅ kg −1 is the specific gas constant for water vapor, and eW is
the partial pressure of water vapor which can be obtained from the relative humidity
using the following formula recommended by World Meteorological Organization in its
Technical Note No. 8
eW = RH ⋅ exp(−37.2465 + 0.213166 ⋅ T − 2.56908 ⋅ 10 −4 ⋅ T 2 ) (11)
where RH is relative humidity in percentage, and T is the absolute temperature in
Kelvin.

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

The observation profiles from Radiosonde are discrete series of temperature and relative
humidity observations at difference heights. This separates the atmosphere into many
layers. Taking eW and T at each layer into (8), we can easily obtain the corresponding
water vapour density. If we assume linear water vapor density variation in each layer,
(8) can be approximated by

1
IPW =
ρ
∑ (h j +1 − h j ) ⋅ ( ρ vj +1 + ρ vj ) / 2 (12)

In the above expression, the subscript j+1 and j denote the top and bottom of each layer
for height and water vapor density.

Obviously, the more observations in the profiles of temperature and water vapour
pressure, the thinner the layer thickness is, a better approximation will be in Equation
(12) towards its theoretical definition (8).

3. GPS Water Vapour Experiments in Australian and Results

Experimental description

The purposes of the experimental study from an Australian network are threefold: first,
to establish GPS-IPW estimation capability and demonstrate feasibility of operational
GPS water vapour estimation in a Australian network; second, to test IPW estimation
algorithms from GPS zenith troposheric delay (ZTD) measurements and radiosonde
measurements, and third, to assess the performance of GPS derived IPW by comparing
GIPS-IPW with Rad-IPW solutions.

We collected GPS tracking data on a daily basis from15 ARGN stations, operated by
Australian Land Information Group (AUSLIG). Nine of these stations are located
closely to the radiosonde sites of the Australian Upper Air Network (AUAN), which
consists a total of 52 sites in Australian, including 2 sites in Macquarie and Cocos
Islands and 3 sites in Antarctica (Davis, Casey and Mawson). The GPS tracking data
were collected at 30 sec every 24 hours. These sites are operated primarily for geodetic
and geodynamic purposes, and do not provide surface meteorological data. The data for
period of 1 July to 30 September 2000 were retrieved and processed for this experiment.
The radiosonde data for the same period were provided by Australian Bureau of
Meteorology (BOM). Radiosondes are released once or twice daily at about UTC 2300
or 0030 from these 9 collocated sites. Surface meteorological data were also collected
every 30 min at Alice Springs, Cocos Island, Darwin, Hobart and Townsville. The rest 4
sites in Macquarie island and Antarctic region record the meteorological data every 3
hours. We estimate GPS zenith tropospheric delays with GAMIT software every 30 min
to match radiosonde release times and the 30 min meteorological data interval. In
principle, we can interpolate the 3-hourly meteorological observations to the radiosonde
release times (at about UTC 00300). This however may introduce additional uncertainty
in comparing both solutions. In the following discussion, the analysis focuses on the
results from the first five sites.

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

Daily GPS data processing

GPS data were processed with GAMIT version 10.03 using conventional 24 hour
session though our procedure is capable of handling variable session length starting at
arbitrary time. The zenith tropospheric delays of each site were estimated every 30
minute within a 24 hour session. In order to obtain the high quality absolute
estimates[17], four remote sites, often referred as the corner sites, were chosen at Davis,
Maquarie, Cocos and Townsville, where their positions were tightly constrained on the
apriori values obtained from SOPAC (Scripps Orbit and Permanent Array Center). The
precise GPS orbits and EOP (Earth Orientation Parameters) were obtained from IGS
(International GPS Service) and IERS (International Earth Rotation Service)
respectively. It is very important that tight enough constrains should be put on those
parameters, such orbits, EOP, and site positions, to the maximum extend of known
confidence so that zenith delays could be estimated properly, since the tropospherica
delay is strongly coupled to the site vertical component, which is in turn coupled to
orbits, while the orbits are coupled to the EOP.

Computing IPW from GPS ZTD and surface meteorological data

The GPS-IPW solutions are derived from GPS ZTD solutions and the zenith hydrostatic
delay (HZD), which is computed from the surface meteorological data as, described by
Equation (1). Ideally, the GPS antenna and surface meteorological sensor are placed in
the same location, perhaps a few metres apart, so the GPS-IPW measurement is
physically referred to the vertical column water directly above the site. Again, to
compare GPS-IPW solutions and Radiosonde solutions, the GPS sites should be
collocated to the radiosonde sites releasing the balloons vertically above. However, this
experiment studies a different case, where there are lateral and vertical separations
between GPS sites and radiosonde sites. As shown in Table 1, the GPS at Darwin site is
located up to 53 km apart from the radiosonde site.

Table 1. Separation Between ARGN GPS sites and AUAN Radiosonde Sites
Site ∆N(km) ∆E(km) ∆H(m) ∆s(km)
Alice Spring -13.7474 0.3513 -41.5060 14km
Cocos Island -0.2142 -0.0714 -1.4990 0.2km
Darwin 46.1132 -26.7622 -44.3020 53km
Hobart -9.4690 -9.0253 5.6910 13km
Townsville 2.3634 -29.9867 -22.8230 30km
Davis 0.0135 -0.0046 -10.3140 10km
Macquarie Island-0.0183 1.0454 -7.7430 8km
Casey 0.4574 0.7032 1.9080 2km
Mawson 0.3703 0.0999 -21.0550 21km

Figure 2 illustrates the separation of GPS site and the collocated radiosonde site.
Computation of GPS IPW requires the temperature and pressure measurements at the
GPS station level. However, the temperature and pressures given by BOM are referred
to the radiosonde station level and quasi-non-hydrostatic (QNH) level respectively. The

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

calculation to derive QNH pressure has been performed on the station level barometric
pressure (PQFE) value. The station height above mean sea level (Hmsl) was also
required. The following equation from Bureau of Meteorology Specification A2669
Version 4.0 Page 100 was applied.

IPWrad
IPWGPS
Radiosonde
site
GPS site TRad
PQFE
PGPS TGPS

QNH level
PQNH

Figure 2. Deduction of Trad and PQNH to Trad and PGPS, which is highly sensitive to the
height difference between two sites.

PQNH = P0 {( PQFE / P0 ) Ru / g + H stn u / T0 } g / Ru (13)

where,
PQNH = QNH pressure in hPa
P0 = 1013.25 hPa
PQFE = QFE pressure in hPa
R = 287.04 m 2 / °C sec 2
u = 0.0065 m / sec 2
g 9.80655 m / sec 2
H stn = Station height in m
T0 = 288.16 ° K
We now need to deduct the temperature Trad and QNH pressure PQNH to the station level
of GPS site: TGPS and PGPS.

Although theoretically we should consider the effects of both lateral separation and
vertical difference on the temperature and pressure parameters, only the height
difference is sensitive to temperature and pressure changes. The difference between Trad
to Tgps is approximately archived by the following equation:

∆T = β • ∆H (14)
where β is the rate of change if temperature with altitude, whose measured values are on
the order of (-6±0.5 deg)/km , such as β=–6.05oK/km for tropical zone and β=–6.45 for
temperate zone and β=–6.525 for Arctic zone[14]

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

For the maximum height difference ∆H =-44.30m at Darwin station, we can calculate
∆T =β∆H=(-6.05)x(-0.0443)=0.27deg, which has little contribution to the IPW solution
errors. Therefore, the temperature changes due to the height difference were not
considered in the studies. In other words, the temperature Trad was directly used as Tgps
in the actual data analysis.

The inverse problem of the equation (12) is now derived as

PQFE = P0 {( PQNH / P0 ) Ru / g − H stn u / T0 } g / Ru (15)

To compute the PQFE for the GPS site, we simply substitute Hstn with the height above
mean sea level for the GPS site. For instance, given PQNH=1017.9 hPa, Hstn =650m at a
Radionsonde site and Hstn =606m at a GPS site, we obtain: PQFE =941.93 hPa and
PGPS=946.92 hPa respectively.

According to Equation (1), we have seen that the HZD is more sensitive to pressure
than to temperature. Deduction of PQFE to Pgps was carried out as an essential procedure
in this analysis.

Experimental Results and Discussions

We estimate the integrated water vapour (IPW) using daily GPS tracking data of 24
hours from each GPS/Radiosonde collocated site every 30 min. Figure 3 shows the
RMS accuracy of the IPW estimate over the 24 hour period. The RMS in fact reflects
the ZTD estimation accuracy by GAMIT software. We have clearly seen the session
boundary effects” of the solutions: the ZTD estimates at two boundaries have the largest
uncertainty. To demonstrate the accuracy of IPW solutions, we can compare the two
IPW solutions for the common data point at UTC 0h, which are derived from two
different GPS data sets. Figures 4a-b compares the GPS-IPW estimates at the data point
of UTC 0h (upper) and their difference (lower) for two different stations: Cocos Island
and Alice Springs, representing the worst and normal cases. We find that the standard
deviation (std) accuracy of the differences is 2.1mm and 1.4 mm for Cocos Island and
Alice Springs respectively. This implies that each estimate at UTC 0h would have the
uncertainty of 1.4mm and 1.0mm respectively. Taking the edging effect into account,
we believe that the GPS-IPW estimate at UTC 2300h would have the accuracy of 1mm
or better. This does not include the contributions of uncertainty of surface
meteorological data.

We can also estimate IPW from radiosonde data that contains pressure, temperature and
relative humidity profiles of all the layers above the site. As mentioned by other
researchers[13], the accuracy of radiosonde IPW observation is estimated at about
1.2mm. This does not account for the precision of numerical integration based on
Equation (12). Theoretically, the denser the radiosonde samples along the path and the
higher the balloon rises above the site, the more accurate the IPW approximate.
Practically, the water vapour vertical distribution varies from time to time and location
to location. Denser radiosonde sampling does not necessarily provide an accurate IPW
estimates. For information, Table 2 gives the average number of samples for each
radiosonde sites involved in our experimental studies.

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

 Table 2. Average number of Radiosonde Samples
Alice Springs Cocos Island Hobart Townsville Darwin
74.3 111.6 72.5 101.2 90.3

We determine the performance of GPS-IPW estimates by comparison with the

radiosonde IPW solutions. Figures 5a-c illustrates GPS-IPW estimates (blue solid line)
and Rad-IPW observations (red cycles) for a 3-month period at three locations: Alice
Springs, Townsville and Darwin. We have noticed that out of 498 comparisons between
O the GPS-IPW and Rad-IPW estimates for all the stations, there are 7 comparisons
whose values (Rad-IPW minuses GPS-IPW) are beyond the range of ±6mm. Otherwise
the GPS-IPW and Rad-IPW estimates agree very nicely to each other very nicely. The
half-hourly GPS IPW observations clearly resolve the many dip changes during a few to
24 hours. Figure 6 demonstrates the agreement with a scatter-plot for all the five sites
comparing GPS-IPW solutions against the radiosonde estimates with removal of these 7
outliers. The average, maximum and minimum IPW and the number of outlying
differences (greater than 6mm) for the five stations, together with statistics of the 491
comparisons (within ±6mm) are summarised in Table 3.

Table 3. Average IPW estimate and Comparison of GPS IPW estimate with radiosonde
IPW estimates during a 3 months period. A total of 491 comparisons are included in the
mean, standard deviation, root mean squares.
Average Alice Cocos Hobart Darwin Townsville
(mm) Springs Island (mm) (mm) (mm)
(mm)
Average GPS-IPW 22.27 11.55 39.66 12.65 26.99 19.19
Max (Rad-GPS) N/A 1.00 8.47 1.26 5.79 3.18
Min (Rad-GPS) N/A -2.60 -5.53 -2.97 -15.24 -10.19
# of |Rad-GPS|>6mm N/A 0 2 0 4 1
Mean (Rad-GPS) -0.66 -0.71 -0.56 -0.90 -0.11 -1.40
Stddev (Rad-GPS) 1.80 0.93 2.10 0.89 2.41 1.65
RMS (Rad-GPS) 1.92 1.17 2.17 1.27 2.41 2.16

While the stddev values in Figure 6 and Table 3 are computed relative to the mean of
the difference between GPS IPW and radiosonde IPW to show the dispersive errors of
all the sites, the RMS errors are computed relative to zero without bias removal, thus
giving a total level of the agreement. It can be seen that the GPS-IPW data agrees with
radiosonde data for all the five stations within 1.92 mm. The better agreement is seen at
Alice Springs and Hobart while the worse agreement is seen at Cocos Island, Darwin
and Townsville. In addition, there are negative biases for all the stations. The largest
bias is seen at Townsville where the radiosonde reports on average 1.40mm less than
GPS. The source of these biases is yet to be investigated although the biases are very
small.

Improving the agreement

While the GPW-IPW results based on the final IGS orbits theoretically would agree
with the radiosonde to zero biases and stddev (standard deviation)
σrad-IPW =Ra σ GPS − IPW + σ Rad − IPW = 1.0 + 1.2 = 1.56 mm
2 2 2 2

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

The actual results in this experiment agree to the mean of -0.66mm mean and the stddev
of 1.80 mm, or a RMS error of 1.92mm. This implies that the errors left in GPS-IPW
and/or Rad-IPW results contributing a bias at the order of 0.7mm and uncertainty at the
level of 0.80mm. Therefore we wish to explore the possible sources of errors in order to
improve the agreement or the accuracy of GPS-IPW solutions.

Firstly, we may question whether GPS-IPW estimates at the GPS site can represent the
GPS-IPW solutions at the radiosonde site because of the lateral separations of tens of
kilometres and height difference of tens of meters between two sites respectively. Water
vapour measured in Alice Springs may be spatially less variable than Darwin, thus
showing a better RMS agreement for the IPW estimates. In other words, there would be
a different disagreement in these stations in a different season, or different locations. An
opposite example is the Cocos Island site, where the GPS site and radiosonde site are
about 200 metres apart, but the IPW estimate agrees at 2.1mm RMS only. Therefore,
the separation is only of one of the reasons for the IPW disagreement.

Secondly, we may argue that the weight mean temperature Tm used to derive IPW from
ZTD in Equation (2). A recent water vapour study for Hong Kong region has shown
significant effect of the Tm. error on the IPW solutions. Table 4 lists the average IPW
estimate error as a function of Tm errors of 3K and 5K for the five stations. We testified
TM effect by comparing the GPS-IPW estimates with the constant Tm=295 K and Tm
time series derived from radiosonde data according to Equation (7). Figure 7 shows the
temperature observations, Tm estimate from radiosondes and against the Tm regression
model and Bevis Tm model for Alice Springs station. We have seen that radiosonde-
derived Tm is very close to Bevis Tm model, and there is an average difference of 10
degrees between the constant Tm (295K) and the estimated Tm. However the
improvement with radiosonde-Tm is not as significant as indicated by theory. The
average, maximum and minimum IPW and the number of outlying differences (greater
than 6mm) for the five stations, together with statistics of the 479 comparisons (within
±6mm) are summarised in Table 3. Figure 8 compares the GPS-IPW solutions with the
constant Tm and radiosonde Tm at Hobart station. Further efforts are yet to be made
towards effects of the weighed mean temperature.

Thirdly, we could take into account the quality and the surface meteorological data
themselves and the possible loss of accuracy when the surface temperature and QNH
pressure for the radiosonde site were transferred to these at GPS station level. The error
in pressure measurement can be related to the IPW estimate error as
δIPW ≈ 0.4 ⋅ δ Pr essure (16)

where δIPW is the error in IPW in units of mm and δPressure is the pressure error in
hPa. Thus to keep the contribution of the pressure error below 0.1 mm IPW, the total
pressure error should be better than 0.25hPa. According a research report by Bureau of
Meteorological Research Centre [15], we can conclude that the typical RMS error of an
interpolated pressure should also fall the range of less than 0.5hPa. Therefore, the total
pressure error could be below 0.2 mm in IPW.

Finally, the IPW estimation errors may also contain errors contributed from Hopfield
HZD model, and effects of ground station coordinates. Further investigation into these
possible sources of errors should be conducted.

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

Figure 3. RMS error of GPS ZTD estimate for 24 hours

6 0
S t a t io n : C o c o s Is la n d
5 0

4 0

3 0

2 0
1 8 0 1 9 0 2 0 0 2 1 0 2 2 0 2 3 0 2 4 0 2 5 0 2 6 0

4 m e a n = 0 .1 m m , s td = 2 .1 m m
2

-2

-4

-6
1 8 0 1 9 0 2 0 0 2 1 0 2 2 0 2 3 0 2 4 0 2 5 0 2 6 0

G P S - IP W s o lu t io n p re c is io n (U T C 0h)

a) 25
S t a t io n : A l ic e S p r in g s
(m m )

20

15
IP W

10

0
170 180 190 200 210 220 230 240 250 260

G P S - IP W s o l u t i o n d i ffe r e n c e ( U T C 0h)
8
e rro r (m m )

4
m e a n = -0 .3 m m , s td = 1 .4 m m

2
IP W

-2

-4
170 180 190 200 210 220 230 240 250 260
D a y o f th e y e a r (2 0 0 0 )

b)
Figure 4. Comparison of the GPS-IPW estimates at the data point of UTC 0h (upper)
and their difference (lower) for two different stations: Cocos Island and Alice Springs.

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

 30

Station: Alice Springs

25

20
IPW (mm)

15

10

0
(a) 0 10 20 30 40 50 60 70 80 90 100
Time (51710+Day)

40

35 Station: Townsville
30

25
IPW(mm)

20

15

10

0
10 20 30 40 50 60 70 80 90 100
Time (51710+Day)

(b) 50

45 Station: Darwin
40

35

30
PWV (mm)

25

20
(c)
15

10

5
0 10 20 30 40 50 60 70 80 90 100
Time (51710+Day)

Figure 5. Comparison of GPS-IPW estimates (blue solid line) and Rad-IPW

observations (red cycles) for a 3-month period at the three locations: Alice Springs,
Townsville and Darwin.

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

 70

# of sites 5
60 # of comparisons 491
Mean= -0.66mm
Stddev= 1.80mm
50 RMS= 1.92mm
GPS IPW (mm)

40

30

20

10

0
0 10 20 30 40 50 60 70
Radiosonde IPW (mm)
Figure 6 A scatter-plot for all the five sites comparing GPS-IPW solutions against the
radiosonde estimates

Table 4. Average IPW estimate error as a function of Tm errors of 3K and 5K for the
five stations.
Tm error F Error Alice Hobart Cocos Townsville Darwin Average
Springs Island
3K 0.0022 0.16 0.18 0.57 0.29 0.39 0.29
5K 0.0032 0.24 0.26 0.82 0.42 0.56 0.46

Table 5. Comparison of GPS IPW estimate with radiosonde IPW estimates during a 3-
month period. A total of 479 comparisons are included A total of 491 comparisons are
included in the mean, standard deviation, root mean squares.
Average Alice Cocos Hobart Darwin Townsville
(mm) Springs Island (mm) (mm) (mm)
(mm) (mm)
Max(Rad-GPS) N/A 1.25 8.32 1.55 5.63 3.26
Min(Rad-GPS) N/A -2.45 -5.75 -2.48 -15.44 -10.42
# of |Rad-GPS|>6mm N/A 0 1 0 5 2
Mean (Rad-GPS) -0.58 -0.55 -0.66 -0.51 -0.22 -1.32
Stddev(Rad-GPS) 1.78 0.87 2.12 0.82 2.41 1.40
RMS (Rad-GPS) 1.87 1.03 2.22 0.96 2.42 1.92

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

 Tm estimation
310

305 Station Alice Springs

Ts

300 Tm (radiosonde)

295
Temperature (K)

290

285

280

275

270 Tm=158.49+0.423Ts
Tm=70+0.72Ts

265
0 100 200 300 400 500 600
Number of Measurements

Figure 7. Tm estimates from radiosonde data and regression models at Alice Springs
2
Tm estimate
1.5
Station: Hobart mean=-0.50mm
std =0.82mm
1

0.5

0
IPW error

-0.5

-1

-1.5

-2 Tm=295K
mean=-0.90mm
-2.5
std=0.90mm
-3
180 190 200 210 220 230 240 250 260 270
Day

Figure 8. Comparison of the GPS-IPW solutions with the constant Tm and radiosonde
Tm at Hobart station.

4. Summary

Our research efforts on GPS water vapor estimation in the ARGN network have led to
establishment of the technical capability of GPS-IPW estimation on a daily basis. GPS
data from the ARGN network are collected and processed daily and GPS-IPW solutions
are produced daily if surface meteorological data are also available. We have developed

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

a package of software tools and scripts for the computations of GPS-IPW estimation,
Rad-IPW estimation, weighted mean temperature estimation and IPW solution analysis
etc.
We have used the data from the ARGN GPS sites and 5 co-located radiosonde sites
during the period of 3 months to carry out the experimental studies. The extensive
results have shown that
• GPS-IPW estimation accuracy of 1mm was achieved
• GPS-IPW estimates agree with Rad-IPW solutions at an average mean
difference of 0.66 mm and RMS of 1.92mm by about 500 comparisons.
The agreement perhaps is slightly worse than these theoretical 1.56mm rms. However,
because the quoted results were obtained under the circumstance where the GPS and
Radiosonde/surface meteorological data sites are collocated tens of kilometers apart,
this agreement level is considered very reasonable. The experimental study also shows a
possible way to develop GPS meteorology and applications with the existing
meteorological data network, instead of new meteorological sensors. This could save
significant costs in installation of GPS-Met sensors.

The future research will focus on hourly GPS-IPW operational estimation in the GPS
networks (Australia and overseas), improvement of the performance GPS IPW
estimates and investigating of GPS-IPW applications.

Acknowledgment

This work was carried out in the Cooperative Research Centre for Satellite Systems
with financial partial support from the Commonwealth of Australia and the Queensland
State Government through the Cooperative Research Centre program, and funding from
QUT industry collaborative research grant with the Queensland Department of Primary
Industry (DPI).

The authors acknowledge the Australian Land Information Group (AUSLIG) and
Bureau of Meteorology (BOM) for providing GPS tracking data and
meteorological/radiosonde data in this work.

References

1. Bevis, M., et al., GPS meteorology: remote sensing of atmospheric water vapour
using the global positioning system. Journal of Geophysical Research, 1992.
97(D14): p. 15,787-15,801.
2. Rocken, C. and et al., Sensing atmospheric water vapor with the Global
Positioning System. Geophysical Research Letters, 1993. 20(23): p. 2631-2634.
3. Bevis, M. and et al., GPS meteorology: mapping zenith wet delays onto
precipitable water. Journal of Applied Meteorology, 1994. 33(3): p. 379-386.
4. Naito, I., et al., Global Positioning System project to improve Japanese weather,
earthquake predictions. Eos, 1998. 79(26): p. 301+308-311.
5. Yuan, L.L. and et al., Sensing climate change using the global positioning
system. Journal of Geophysical Research, 1993. 98(D8): p. 14,925-14,937.
6. Businger, S. and et al., The promise of GPS in atmospheric monitoring. Bulletin
- American Meteorological Society, 1996. 77(1): p. 5-18.

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress

7. Iwabuchi, T., I. Naito, and N. Mannoji, A comparison of Global Positioning
System retrieved precipitable water vapor with the numerical weather prediction
data over the Japanese Islands. Journal of Geophysical Research D:
Atmospheres, 2000. 105(4): p. 4573-4585.
8. Tregoning, P., et al., Accuracy of absolute precipitable water vapor estimates
from GPS observations. Journal of Geophysical Research D: Atmospheres,
1999. 103(D22): p. 28,701-28,710.
9. Brunner, F.K. and W.M. Welsch, Effect of the troposphere on GPS
measurements. GPS World, 1993. 4(1): p. 42,44,46,48,50-51.
10. Saastamoinen, J., Contributions to the theory of atmospheric refraction Part II,
Refraction corrections in satellite geodesy. Bulletin Geodesique, 1973(107): p.
13-14.
11. Hopfield, H.S., Tropospheric effect on electromagnetically measured range:
Prediction from surface weather data. Radio Science, 1971. 6(3): p. 357-367.
12. Black, H.D. and A. Eisner, Correcting satellite Doppler data for tropospheric
effects. Journal of Geophysical Research, 1984. 89(D2): p. 2616-2626.
13. Liu, Y., Y. Chen, and H. Baki Iz, Precision of precipitable water vapor from
radiosonde data for GPS solutions. Geomatica, 2000. 54(2): p. 171-175.
14 Spiler Jr, J. J., Tropospheric effects on GPS, Global Positioning System: Theory
and Applications(Ed. Parkinson, Spilker Jr), Published by AIAA, Vol 1, p517-
545, 1996.
15 Glowacki, T. J, P.E Powers and T D. Keenan, The Mesoscale Surface Analysis
System (MSAS) Performance of The Analysis Method, Bureau of Meteorology
Research Centre Research Report No. 64 (Ed .Meiggen P.J., and J.D. Jasper),
September 1997.
16 King, R. W. and Y. Bock, Documentation of the GAMIT GPS analysis software
version 10.03, Mass. INst. of Technol., Cambridge, 2001
17 Duan, J. P., et al., GPS meteorology Direct estimation of the absolute value of
precipitable water, J. Appl. Meteorol., 35, 830-938, 1996.

2001 - A Spatial Odyssey : 42nd Australian Surveyors Congress