You are on page 1of 12

Regional Tropospheric Tomography Based on

Real-Time Double Difference Observables


N. A. Nicholson, S. Skone, M.E. Cannon, G. Lachapelle and N. Luo
Department of Geomatics Engineering
University of Calgary
Canada

BIOGRAPHIES
Natalya Nicholson is a Ph.D. candidate in the Department
of Geomatics Engineering, University of Calgary. She is
working under the joint supervision of Drs. Susan Skone
and Elizabeth Cannon. She received a B.Sc. in Physics
and Geophysics from McGill University (1999) and an
M.Sc. in Space Physics from the University of Calgary
(2001). Her research focuses on regional atmospheric
modeling to mitigate errors for precise GPS positioning
and GPS meteorology.
Susan Skone, Ph.D., is an Associate Professor in
Geomatics Engineering at the University of Calgary. She
has a background in space physics and has conducted
extensive research in modeling ionospheric and
tropospheric effects on GPS. She has developed licensed
software for atmospheric research, and is co-chair of an
international study group focused on remote sensing of
the atmosphere using GPS. She is currently chair of the
Canadian Navigation Society.
Dr. Elizabeth Cannon is Professor and Head of the
Department of Geomatics Engineering at the University
of Calgary. She has been involved with GPS research
since 1984 and has published numerous papers on static
and kinematic positioning. She is a Past President of the
ION and the 2001 Kepler Award winner.
Dr. Grard Lachapelle is a Professor of Geomatics
Engineering at the University of Calgary where he is
responsible for teaching and research related to location,
positioning, and navigation. He has been involved with
GPS developments and applications since 1980. He has
held a Canada Research Chair/iCORE Chair in wireless
location since 2001.
Dr. Ning Luo has a Ph.D. in Geomatics Engineering from
the University of Calgary and a Ph.D. in Electrical
Engineering from the University of Electronic Science
and Technology of China. He is currently a senior
research associate in the Department of Geomatics
Engineering at the University of Calgary. His major areas

of interest include GNSS high-accuracy positioning,


GNSS integrated systems and wireless location.
ABSTRACT
Tomography techniques have been developed for
recovering 3-D wet refractivity distributions over regional
GPS networks. Most techniques employ single path slant
wet delay (SWD) observations as model input. In order to
recover SWD observables with sufficient accuracy, a
number of error sources must be removed including
orbital uncertainty, satellite and receiver clock biases, and
ionospheric propagation delays. Double difference,
ionosphere-free GPS observations with fixed L1 and L2
ambiguities are necessary as well as precise orbit products
and station coordinates. After error sources apart from the
neutral atmospheric range delay are mitigated, single path
delays must be recovered from the double-differenced
observations. One technique to recover single-path SWD
information uses a zero-mean assumption while another
employs a batch estimate of the average total atmospheric
delay (and average gradients) overlying a reference
station. In both cases, limiting assumptions are made
about spatial variations of properties in the neutral
atmosphere. This paper investigates the feasibility of an
alternate approach - using double difference residuals as
tomography model input directly without first recovering
single-path SWD information.
A double difference tomography model has been
developed at the University of Calgary. Simulations
indicate that integrating through estimated wet refractivity
fields to estimate zenith wet delay agree with truth to
better than 1 mm when network station heights vary by
400 m. Regular and inversion atmospheric profiles are
recovered using the model. Simulations using flat network
geometry recover smoothed vertical profiles but estimated
zenith wet delay values still agree with truth to within 3.0
mm.
GPS data from the Southern Alberta network is processed
with MultiRefTM Network RTK Software to form double
difference SWD observables. Tests using real data infer
that the magnitude of positioning misclosures can be

Presented at the ION GNSS-05 Conference, Long Beach, CA, September 13 16, 2005.

reduced by 18 20 % using model refractivity fields to


form estimates of DD SWD. The development and
validation of incorporating double differenced
observables directly into tomography models expands the
real-time capabilities of existing network processing
software at the University of Calgary, by allowing an
option to estimate 3-D wet refractivity distributions of the
overlying region.
INTRODUCTION
The Global Positioning System (GPS) is a ranging system
based on the accurate estimation of signal travel times
between GPS satellites and receivers. Electromagnetic
signals are assumed to propagate through space at the
speed of light and by definition in a medium with an
index of refraction, n, equal to 1. Any deviations from this
value are attributed to neutral atmospheric and
ionospheric effects that must be mitigated. In addition to
the neutral atmosphere and ionosphere delays, GPS
ranging measurements are affected by orbital errors,
satellite and receiver clock errors, receiver noise,
multipath (tracking indirect or reflected GPS signals) and
unresolved phase ambiguities. Satellite and receiver clock
errors and biases can be removed by double differencing
(DD) observations across two satellites and two receivers.
Effects of multipath can be minimized with careful site
selection or use of multipath maps and receiver noise is
minimized if carrier phase measurements are used. Orbital
error magnitudes are also reduced by double differencing
and further mitigated by using orbit products from the
International GNSS Service (IGS).
Ionospheric range delay is related to total electron content
(TEC). This delay can be estimated and 99 % of the effect
can be removed by taking advantage of the dispersive
properties of the ionosphere at GPS frequencies [Brunner
and Gu, 1991]. GPS currently transmits on two
frequencies, 1575.42 MHz (L1) and 1227.60 MHz (L2),
and a linear combination of the ranges measured on L1
and L2, called the Ionosphere Free (IF) combination, can
be formed to remove the majority of the ionospheric
effect.
Atmospheric delay is due to the presence of dry gases and
water vapor in the neutral atmosphere. The neutral
atmosphere does not have the same dispersive properties
as the ionosphere; therefore users cannot take advantage
of the two GPS frequencies to mitigate this range error.
The neutral atmospheric propagation delay along a GPS
ray path is related to neutral refractivity, N, where:
N = 106(n1)

The total atmospheric path delay can be expressed in


terms of a hydrostatic component called the slant
hydrostatic delay (SHD) and a non-hydrostatic component

commonly referred to as the slant wet delay (SWD):

SHD + SWD = 10 6 Nh ( P ) ds
+ 10 6

2
w (e, T )ds

where P is the pressure, T is the temperature, e is the


partial pressure of water vapor, Nh is the hydrostatic
refractivity and Nw is the non-hydrostatic or wet
refractivity [cf. Ware et al., 1997].
Existing atmospheric models can estimate the SHD term
with an accuracy of a few millimeters if precise surface
pressure information is available and all other error
sources, apart from the non-hydrostatic range error, have
been mitigated [Bevis et al., 1992]. Atmospheric models
estimating SWD are not as accurate since the partial
pressure of water vapor at the surface is not a reliable
indicator of the partial pressure of water vapor at altitude.
Therefore, unless in-situ measurements of either water
vapor (i.e. measured via atmospheric soundings), or
remotely-sensed SWD observations (i.e. measured via
profiling microwave radiometers) are available at all
receiver locations, alternative means are required to
effectively determine this quantity. In GPS software
packages wet delay is commonly treated as a nuisance
parameter in positioning algorithms. For GPS
meteorological applications, however, it is the remaining
SWD quantity that is of interest due to its relationship to
atmospheric water vapor.
Water vapor is one of the dominant greenhouse gases
with a high latent heat [Tsonis, 2002] and plays a key role
in the development of atmospheric dynamics. It is a
highly variable component that cannot be accurately
determined from surface data like atmospheric
temperature and pressure. At higher altitudes, information
can be obtained from atmospheric soundings (i.e.
radiosondes) and microwave radiometers. The operational
costs of a radiosonde network generally limits routine
launches to sparse networks and short-term scientific
campaigns. Similarly the current cost of a microwave
radiometer may limit its use to research institutions and
large forecasting centers. SWD observations derived for
each satellite within the field of view of a GPS receiver
provide information about the overlying atmosphere.
Over the past decade GPS has become a cost-effective
tool to continuously measure water vapor with accuracies
comparable to that of a microwave radiometer [Rocken et
al., 1997; 2000; Ware et al., 1997; 2000; Jerrett and Nash,
2001].
To utilize the spatial information of the SWD
observations, tomography techniques commonly used in
seismic research and medical imaging have been
developed for GPS applications to estimate 3-D wet

Presented at the ION GNSS-05 Conference, Long Beach, CA, September 13 16, 2005.

refractivity fields above reference networks [Flores, et al.


2000a, b, 2001; Troller et al., 2002]. The estimated fields
can be used as an additional source of data input to
numerical weather prediction models [Kuo et al., 1993;
Kuo et al., 1996]. Although the effect of the wet delay
component is an order of magnitude smaller than the
hydrostatic component, its variability can create problems
for high-precision GPS applications. With many GPS
receivers providing wet delay information, modeling of
wet refractivity can be performed over a regional
network. The model distributions have the potential to be
incorporated into real-time kinematic (RTK) positioning
software packages to improve error mitigation for users
within a reference network.
Input observables for tomography are generally formed
from GPS data by post-processing with GPS software
packages (i.e. Bernese). Total atmospheric delay is
normally provided as a batch-estimated zenith value after
precise station coordinates are determined. The software
package
may
incorporate
orbit
information,
meteorological data, or other auxiliary measurements to
improve the GPS ambiguity resolution and positioning
solutions. In order to reduce range errors such as clock
errors and biases, double differencing across satellites and
receivers is performed. When the ambiguities to all
satellites have been resolved and station positions have
been estimated, an estimate of the zenith total
atmospheric delay (and gradients) above reference
stations is performed. Tomography input is derived from
these estimates. Most approaches derive single ray path
SWD information for tomography input using a zerodifference retrieval technique [Alber et al., 2000]). Troller
et al. [2002] successfully employed DD SWD estimates as
tomography input.
Simple tomography models employ a functional approach
to estimate wet refractivity values in discrete vertical
layers. Coefficients of a low-order expansion in latitude
and longitude describe the horizontal variation within
each layer [Skone and Shrestha, 2003]. Hoyle et al.
[2004] incorporated radiosonde measurements as an
additional observable to strengthen the solution with
higher altitude observations. This is particularly helpful
when the topography of the region spanned by the GPS
network is relatively flat.
A second approach estimates wet refractivity as discrete
values in voxels (volume boxes/pixels). This technique
offers more flexibility as voxel dimensions can be tailored
for specific networks. The addition of constraint equations
to the design matrix and the method used of populating
the variance-covariance matrix can help strengthen the
least-squares estimation solution [Hirahara, 2000; Flores
et al., 2000a, 2000b, 2001; Gradinarsky and Jarlemark,
2002; Liou et al., 2003].

These tomography approaches employ zero-difference


SWD observables; however this is not conducive to realtime implementation because of the batch-estimation by
which the total zenith atmospheric delay is obtained.
Troller et al. [2002] circumvented the formation of zerodifference residuals and used the DD residuals directly.
GPS data from a Hawaiian network was processed with
Bernese Processing Software to produce DD SWD. Data
from one hour of observations was incorporated into the
batch estimation of the wet refractivity field. The
Hawaiian network provided a unique test location for a
tomographic inversion method because of the exceptional
distribution of station heights which ranged from sea level
to over 4000 m. Comparision with soundings launched
within the network indicate statistical wet refractivity
agreement of 5 20 ppm; however due to the smoothing
effect of constraints and hour-long batch estimation, rapid
temporal variations were not detected accurately.
A precise positioning software package developed at the
University of Calgary, MultiRef, has the capability of
providing real-time DD SWD observables. The objective
of this paper is to provide 3D wet refractivity fields above
a regional GPS network for precise positioning and
meteorological applications using MultiRef-derived
DD SWD input. The results of employing a recently
developed DD voxel tomography approach with the
MultiRef estimates is assessed in two ways. First, the
feasibility of using DD observables as input is
investigated through simulations. Simulated observables
are used to derive Zenith Wet Delay (ZWD) model
predictions and these are compared to truth values over a
users location within the network. Second, the
tomography model is validated using real data from a
regional GPS network. MultiRef-derived DD
observables are used as input. Wet delay corrections are
derived from the output 3D wet refractivity field. Validity
of these corrections is assessed by comparing DD SWD
recreated from the estimated wet refractivity fields to the
input observables.
The advantage of this tomography method is that it is
based on existing real-time positioning software and wet
refractivity estimation can be added as a component that
directly incorporates DD SWD into the solution. Output of
the atmosphere estimation module can in turn be used as a
means of providing network corrections to users. In
addition, this method makes no assumptions about the
azimuthal symmetry of the refractivity field around a
reference station in the derivation of the input observables
and does not require the estimation of gradients.
MULTIREF DD SWD & DD TOMOGRAPHY
APPROACH
GPS network observations are collected for validating
tomography models implemented at the University of

Presented at the ION GNSS-05 Conference, Long Beach, CA, September 13 16, 2005.

Calgary. GPS data is post-processed in simulated real


time using the MultiRef Network RTK Software. This
software package is able to produce corrections for
regional network and virtual reference stations
[University Technologies, 2004]. MultiRef employs
ambiguity resolution techniques to solve for L1 and L2
ambiguities and least squares collocation techniques to
provide corrections for virtual reference stations.
To simulate a real-time environment, GPS data is
processed using broadcast orbits and an elevation angle
cut-off of 10 degrees is applied. Pre-determined precise
reference station coordinates and pre-selected baselines
are required as input. The double-differencing approach
removes satellite and receiver clock error and biases and
minimizes un-modeled orbital errors. MultiRef
provides DD L1, L2, and IF ambiguities for all station
baselines and the base and receiver satellites in view.
Only DD observations with both L1 and L2 fixed
ambiguities are used for tomography input.
An IF combination is used to mitigate ionospheric error
from the GPS measurements before DD IF range
residuals, or misclosures, are output. These are the basis
of the DD SWD observables. The level of ionospheric
activity and DD IF residuals are assessed to ensure that
residual ionospheric error does not contaminate the
observables.
The design matrix of the tomography inversion problem is
expressed as the right hand side of equation 3. The wet
refractivity values, Nw, are the unknown parameters to be
estimated in a least-squares adjustment from the input
observables,

SWDabmn .

6
SWDmn
N w dsan
ab = 10
i =1 i i

N wk dsamk

k =1

N w j dsbnj

j =1

N wl dsbml

l =1

3.

are the refractivity values of the voxels traversed along


the line of sight between receiver-satellite pairs: a-n, b-n,
a-m, b-m, with respective distances through each voxel
by

dsani , dsbni , dsami , dsbmi

The

All observations must remain within the defined model


space and exit through the top of the highest layer. As
there is potential for some voxels to be under-observed,
nearest-neighbor averaging is performed in horizontal
layers. Voxels in the top layer are constrained to have wet
refractivity values of 0 mm/km to strengthen the solution.
Voxels are vertically related via cross-covariance values
in the variance-covariance matrix of the least-squares
adjustment as follows:
a h
Cxi, j = a1e 2 i , j
4.
where hi,j is the distance between the centroid of voxels i
and j, and a1 and a2 are user-defined constants.
Temporal variation between epochs is modeled using a
Kalman filter approach and a 1st order Gauss-Markov
relationship with a correlation time of 30 minutes. Voxel
values are initialized with wet refractivity values of 0
mm/km. Estimation of wet refractivity parameters is
performed in a batch adjustment using five minutes of
observations in each batch. For simulations, a sample
interval of 30 s was used while for real network data the
sample interval was reduced to 10 s.
SIMULATIONS

where a is the reference receiver, b is the rover receiver,


m is the base satellite, and n is the non-base satellite
depicted in Figure 1b. Parameters N wi , N w j , N wk , N wl

given

Figure 1: (a) Schematic diagram of SWD


approximated by Equation 3; (b) Schematic of range
measurements that form DD observables.

observable

SWDabmn is the double difference SWD observable


between the receivers a, b and satellites m, n. Figure 1a
depicts an example of a ray path passing through voxels
1, 3, and 4 which corresponds to one of the summation
terms on the right hand side of Equation 3.

DD SWD input is simulated and used to test the


performance of the DD tomography model. Observations
for 16 stations (4 x 4), spaced 0.7 apart in latitude and
longitude, are simulated for regular and inversion
atmospheric wet refractivity profiles. Station heights are
simulated as a flat network as well as a network with a
height variation of ~400 m. This is comparable to the
height variation of the Southern Alberta Network used for
tests with DD SWD observables formed from GPS data
processed with MultiRef software.
Figure 2 is a plan view of the 2 x 2 x 11 voxel model
space used in the testing where each voxel has dimensions
of 1.5 x 1.5 x 1000 m. The simulated station plotted in
green is the primary reference station and baseline pairs

Presented at the ION GNSS-05 Conference, Long Beach, CA, September 13 16, 2005.

are formed with the remaining stations, plotted in red. The


satellite with the highest elevation angle is selected as the
base satellite. DD SWD observables are formed for the
satellites in view and all baseline pairs with an elevation
mask of 10 applied.

of 0 m. The test ZWD quantity is formed by integrating


vertically through the resulting wet refractivity field
above each test site. The corresponding truth quantity is
formed by integrating through the original simulated
atmospheric wet refractivity profile.

Figure 2: Plan view of voxel footprints (dotted black


line), simulated primary (green triangle), rover (red
triangles) and test sites (blue circles).

Estimated wet refractivity profiles resulting from the


tomographic inversion are similar for both the regular and
inversion truth profiles. The estimated wet refractivity
values above the test sites are plotted in Figure 4 as black
circles. The truth profile is plotted as the solid black line.
The model differs significantly from the truth profile and
is due to the lack of height difference amongst the
reference stations. Despite the poor vertical resolution, the
ZWD test quantities agree well and RMS ZWD
differences from truth are summarized in Table 1. Results
for a regular profile have an RMS ZWD difference of 3.0
mm and 1.6 mm for the inversion. The wet refractivity
value of each voxel is plotted in Figure 5. Slight
differences in the values in several of the horizontal layers
deviate from the simulated truth values since they should
be uniform over the layers; however, they indicate that the
model allows spatial variability in the estimation.

Observations are simulated as real-time input observables


every 30 s and unknown wet refractivity parameters are
estimated at five-minute epochs. Figure 3 depicts an
example of the number of rays traversing each voxel in
the model space. This is a function of the satellite
constellation and the geometry of the reference station
network. The voxels are all well observed with over 700
rays passing each of them.

Figure 4: Truth (solid line) and estimated (circles) wet


refractivity for an inversion atmospheric profile above
a flat network.
Table 1: RMS of ZWD differences (model-truth) for
DD tomography simulations.

Figure 3: Example of the number of rays traversing


each voxel over five minutes with a 30 s sample
interval.

Atmospheric
Profile
Regular
Inversion
Regular
inversion

Height
Variation (m)
0m
0m
400 m
400 m

RMS ZWD
Difference (mm)
3.0
1.6
1.0
0.8

A flat station height distribution is used for the initial


simulations where the stations are at 0 m altitude. Test
sites, plotted as blue circles in Figure 2, are selected at the
center of each voxel in the bottom layer with an altitude

Presented at the ION GNSS-05 Conference, Long Beach, CA, September 13 16, 2005.

Figure 7: Truth (solid line) and estimated (circles) wet


refractivity for a regular atmospheric profile above a
network with station heights varying by 400 m.
Figure 5: Estimated wet refractivity for voxels in
model space based on a simulated inversion
atmospheric profile.
Figure 6 is a plot of the simulated network topography
where there is approximately a 400 m station height
differential. Simulations using this network yield wet
refractivity fields with vertical profiles matching truth
profiles more closely. Regular and inversion atmospheric
truth profiles are plotted respectively in Figures 7 and 8 as
solid black lines and estimated wet refractivity values are
plotted as black dots. The voxel values for the two
simulations are plotted in Figures 9 and 10. RMS
differences between ZWD determined from the estimated
and truth wet refractivity values are 1.0 mm and 0.8 mm
for the regular and inversion simulations respectively.

Figure 8: Truth (solid line) and estimated (circles) wet


refractivity for an inversion atmospheric profile above
a network with station heights varying by 400 m.
DATA SET AND MULTIREF PROCESSING
A data collection campaign called, Alberta GPS
Atmospheric Moisture Evaluation (A-GAME) 2004 took
place during July 12 17 2004, and was carried out by
the Department of Geomatics Engineering at the
University of Calgary and the Meteorological Service of
Canada. The purpose was to acquire a data set of colocated GPS and meteorological observations for various
applications including GPS meteorology and precise
positioning techniques. The network is the same as the
one used in the A-GAME 2003 campaign and a detailed
description can be found in Nicholson et al. [2003].

Figure 6: 3-D representation of a simulated network


with station heights varying by 400 m.

Figure 11 is a map of the Southern Alberta Network


stations used for the real data processing. Stations are
equipped with NovAtel Modulated Precision Clock
(MPC) receivers that use an OEM4 technology and have a
16 Gb hard drive [NovAtel, 2002; 2003]. NovAtel 600
series antennae are used at the sites. The majority of the
stations have a Paroscientific MET3A instrument for
precise pressure, temperature and relative humidity
measurements [Paroscientific, 2001].

Presented at the ION GNSS-05 Conference, Long Beach, CA, September 13 16, 2005.

approach, a vertical separation of at least 400 m is


required to attain a realistic vertical wet refractivity
profile. In addition, the validation process will be a
comparison of DD SWD produced by the tomography
model and the MultiRefTM DD SWD.

Figure 9: Estimated wet refractivity for voxels in


model space based on simulated observations of a
regular atmospheric profile above a network with
station heights varying by 400 m.

Figure 11: Map of a portion of the Southern Alberta


Network. Stations with a raindrop have a co-located
MET3A instrument.
Airdrie is selected as the reference station and baselines
ranging from 24 73 km in length are formed to the
remaining rover stations. Table 3 lists the percentage of
fixed ambiguities on L1 and L2 along with the baseline
lengths. Ambiguities are fixed for greater than 88% of the
time over a 24-hour period for all baselines. Figure 12
plots the DD fixed ambiguities for all satellites pairs for
the Airdrie-Calgary baseline with the base satellite
indicated by green dots. If the DD L1 and L2 ambiguities
are fixed, the non-base satellite is plotted in blue
otherwise it is plotted in red.
Table 2: Station coordinates of SAN sub-network from
which DD SWD observables are derived (*denotes the
primary reference station).
Station
Name

Figure 10: Estimated wet refractivity for voxels in


model space based on simulated observations of an
inversion atmospheric profile above a network with
station heights varying by 400 m.
Table 2 lists the station coordinates of the sub-network
required for processing the data with MultiRefTM Network
RTK Software. Twenty-four hours of data from seven
stations collected on July 12, 2004 are used in the realdata testing of the DD tomography method. Shrestha
[2003] demonstrated that for a functional tomography

Airdrie*
Calgary
Cremona
Olds
Strathmore
Sundre

Latitude
(deg min sec)
51 16
51 04
51 32
51 47
51 03
51 48

57.83
45.84
40.56
31.61
39.24
27.55

Longitude
(deg min sec)

Height
(m)

-114
-114
-114
-114
-113
-114

1081.44
1116.80
1168.80
1031.80
975.40
1083.31

00
07
29
05
23
38

15.12
57.72
18.60
35.76
14.28
7.80

After attempting to resolve ambiguities, MultiRefTM is


able to generate DD IF and geometry-free (GF)
observables since the reference station coordinates are
known. The magnitudes of IF and GF misclosures
provide an indication of the level of differential
tropospheric error and ionospheric error respectively.
Table 4 lists the DD IF and GF misclosures for the
baselines processed. The RMS of the DD GF misclosures

Presented at the ION GNSS-05 Conference, Long Beach, CA, September 13 16, 2005.

ranges from 2.03 to 4.95 cm with an average ionospheric


error of 0.6 ppm. The RMS of the DD IF misclosures
ranges from 1.76 to 3.75 cm with an average tropospheric
error of 0.6 ppm. These values may be distorted as they
may include amplified noise and multipath as well as float
or incorrect network ambiguities.
Table 3: Baseline lengths for station pairs processed
with MultiRef and percentage of fixed L1 and L2
ambiguities for all baselines on July 12, 2004.

Calgary

Baseline Length
from Airdrie (km)
24.3

Cremona

44.6

97.8

Irricana

27.9

88.1

Rover Station

A time series of the magnitude of the DD IF misclosures


if plotted in Figure 13. There are several periods where
the misclosures are unreasonably high to be attributed
solely to tropospheric error. Filtering is done to remove
DD IF misclosures that exceed a 3 noise level of 0.15
cm. The DD SWD observables are formed under the
assumption that the DD IF misclosures represent unmodeled differential wet delay.

% Fixed
97.7

Olds

57.0

87.7

Strathmore

49.7

96.8

Sundre

73.0

94.1

Figure 13: DD IF misclosures for the Airdrie-Calgary


baseline on July 12, 2004.
TOMOGRAPHY RESULTS FOR SOUTHERN
ALBERTA NETWORK

Figure 12: L1 and L2 ambiguities for the AirdrieCalgary baseline. Green indicates the base satellite.
The non-base satellite is plotted in blue (fixed) or red
(float).
Table 4: MultiRef DD IF and GF misclosures for July
12, 2004.

In this section results are presented from preliminary tests


of a DD tomography approach developed to incorporate
MultiRefTM-derived DD SWD observables. The
configuration of the reference stations used in the tests is
not ideal for forming regularly spaced voxels so two
model spaces are tested. The first model space has 1 x 2 x
11 voxels where each voxel is 2.0 in latitude x 1.5 in
longitude x 1000 m in height. The second has 2 x 2 x 11
voxels where each voxel is 1.5 in latitude x 1.5 in
longitude x 1000 m in height. A plan view of footprint of
the 2 x 2 voxel configuration is plotted in Figure 14. As
with the simulations, the estimation of the wet refractivity
field is conducted at five-minute intervals; however, to
increase the number of observations in each batch, the
sample interval is reduced to 10 s.
The North-East and South-West voxels have only one
station in each of them, situated near the periphery. Many
of the ray paths only cross vertical voxel boundaries,
however, with the stations in the North-East and SouthWest quadrants so closely situated to voxel boundaries
that many observations at low elevation angles may cross
into neighboring voxels. This potentially leaves voxels in
these quadrants frequently under-observed. The total
number of ray traversing each voxel will be dependent on
the changing satellite constellation and also the quality of
the DD IF misclosures used to form the DD SWD as the
filter may reject observations.

Presented at the ION GNSS-05 Conference, Long Beach, CA, September 13 16, 2005.

atmospheric model, DD IF residuals are formed from the


estimated DD SWD and validated against the original DD
IF misclosure values. In this way the tomography model
can be assessed to determine if it produces smaller
positioning residuals than using the Modified Hopfield
model currently implemented with MultiRefTM.

Figure 14: Plan view of voxel footprints (dotted black


line) and primary (green triangles) and rover (red
triangles) stations used in the real data tests.
When the geometry and data quality is good, the number
of observations of each voxel is high and relatively
consistent in all voxels. An example of this is plotted in
Figure 15 where the minimum number of rays traversing
a voxel is ~500. During this period, around 0900 hours
UTC, the number of observations remains at a
comparable level and the wet refractivity fields produce
realistic results (Figure 16). The wet refractivity ranges
between 0 mm/km in the highest layers and 35 mm/km in
the lowest layer. Variations of up to 10 mm/km occur
within horizontal layers with the most spatial variability
in the lower layers. The model constraints added to the
design matrix are flexible enough to allow for variation
within the horizontal layers and do not overly smooth the
solution.

Figure 15: Number of rays traversing each voxel over


a five-minute interval starting at 0855 hours UTC (10
s sample interval).

In contrast to these results, periods with poor


observability exist and the solution degenerates. The
majority of the voxels around 1015 hours UTC, as plotted
in Figure 17, do not have many rays traversing them. The
primary reference station is located in the South-East
quadrant and, due to the way the DD pairs are formed, the
voxels in this quadrant are well observed. Periods with
low numbers of observations may be due to the
configuration of the satellite constellation or misclosures
exceeding the 3 noise level. The wet refractivity
estimates deteriorate as in the example plotted in Figure
18.
In order to validate the estimated wet refractivity fields
produced by the model, the estimated field is used to
reconstruct model DD SWD along the same ray paths
traversed as the input DD SWD observables. These are
compared to the original DD SWD observations. The
current version of MultiRefTM employs a Modified
Hopfield model with standard atmospheric parameters to
model the tropospheric delay. To investigate the potential
of using tomography model output in place of the current

Figure 16: Estimated wet refractivity field above


reference network at 0929 hours UTC.
Table 5 lists the RMS value of the DD IF misclosures and
the misclosures from using the DD tomography model for
the 1x 2 and 2 x 2 test cases. Only epochs where model
solutions were available were compared so MultiRefTM

Presented at the ION GNSS-05 Conference, Long Beach, CA, September 13 16, 2005.

results were distorted by the addition of misclosures


during times with significant residual ionospheric error.
Results indicated that misclosures can be reduced by 18
20 % using the DD tomography approach.

for this period, using the tomography approach should


improve positioning performance as well as or better than
using a Modified Hopfield atmospheric model with
standard met data. Future research will focus on
optimizing the DD tomography model presented here and
validating against radiosonde and microwave radiometer
observations.
Table 5: MultiRef and tomography model misclosures
for July 12, 2004.
Voxel
Dimensions
1 x 2 x 11
2 x 2 x 11

RMS of Misclosures (cm)


MultiRefTM
DD Tomography
1.9
1.6
1.9
1.5

Figure 17: Number of rays traversing each voxel over


a five-minute interval starting at 1015 hours UTC (10
s sample interval).

Figure 19: RMS of the MultiRef (black) and DD


tomography (red) misclosures from each five-minute
batch adjustment.
SUMMARY

Figure 18: Estimated wet refractivity field above


reference network at 1015 hours UTC.
Figure 19 is a time series plot of the RMS of the
misclosures from each five-minute batch adjustment. The
MultiRefTM misclosures are shown in black and the DD
tomography model misclosures are shown in red. Results
are only plotted up to 1700 UTC as there is an extended
period of low observations following this time. Based on
the misclosures magnitude, the time series indicates that,

A double difference tomography model approach has


been developed at the University of Calgary. The novel
aspect of this work is that it has been developed for use
with an existing, real-time precise positioning software
package, MultiRefTM, and DD SWD observables can be
incorporated directly into the model. It circumvents the
common approach that requires batch estimation of ZWD
with commercial GPS software (i.e. Bernese, GAMIT)
and the zero-differencing technique that retrieves single
slant path observables from DD information. The
tomography model estimates the discrete wet refractivity
values in voxels. Voxel and model space dimensions are
flexible to accommodate different networks. This
technique has been implemented in the Southern Alberta
Network; however the flexibility of the technique makes
it applicable for regional GPS networks in general.
The DD tomography model is first tested with simulations
over a flat network as well as a network with a 400 m
vertical station separation. Simulations of both regular

Presented at the ION GNSS-05 Conference, Long Beach, CA, September 13 16, 2005.

10

and inversion atmospheric wet refractivity profiles are


used to produce DD SWD observables.
The tomography model successfully recovers regular and
inversion profiles for the network with some topography.
ZWD values can be estimated over any point within the
model space. Estimated ZWD values were produced by
integrating vertically through the model at four test sites.
Truth ZWD values are derived by integrating through the
simulated atmospheric profile above the same test sites.
ZWD truth and model comparisons differ by less than 1
mm when station heights vary by ~400 m.
GPS data from the Southern Alberta Network were
processed using MultiRefTM to produce DD SWD
observables. There are no assumptions made with respect
to the azimuthal distribution of the overlying wet
refractivity field when creating the observables for the
model so no gradient information is lost by using the DD
values directly. Ambiguities must be fixed on both L1 and
L2 before the DD IF combination is formed and the level
of residual ionospheric error must be assessed.
MultiRefTM DD SWD observables are based on the DD IF
misclosures and the wet delay component of a Modified
Hopfield model computed using standard atmospheric
values.
Wet refractivity fields estimated by the model during
times with good observability display reasonable profiles
even though the network geometry is relatively flat. The
model incorporates a horizontal nearest-neighbor
averaging constraint to strengthen the model solution
which may result in smoothed horizontal distributions.
However, the model values from the real data testing
indicate that the horizontal constraints allow variability
with horizontal layers and have the potential to recover
atmospheric refractivity gradients. When the number of
observations is low for several epochs in the majority of
the voxels, the estimates deteriorate. Therefore satellite
geometry and high ionospheric conditions may limit the
performance of the tomography model.
By integrating through the 3-D distribution along the ray
paths traversed by the original DD observations a model ,
DD SWD value is estimated. Misclosure values produced
from the model estimates are validated against the
original MultiRefTM misclosures. In both model space
dimensions tested, the RMS of the model misclosures is
lower by 18 - 20 %. This indicates that tropospheric
corrections derived from a tomography model may
provide improved positioning results for users within a
reference network. An additional benefit of this approach
is that a reference network can implement it without the
requirement of having precise pressure sensors to reduce
atmospheric error.

This method incorporates DD SWD information directly


into a tomography model to produce estimates of 3-D wet
refractivity fields. Preliminary results presented here
indicate that there is potential to integrate this technique
as a component of existing RTK software packages such
as MultiRefTM and will lead to improved positioning for
users within regional networks.
ACKNOWLEDGEMENTS
We would like to acknowledge that this research was
supported by the Canadian Foundation for Climate and
Atmospheric Sciences and the Informatics Circle of
Research Excellence.
REFERENCES
Alber, C., Ware. R., Rocken, C., Braun, J., Obtaining
single path phase delays from GPS double differences,
Geophysical Research Letters, 27, pp. 2661-2664,
2000.
Bevis, M., S. Businger, T.A. Herring, C. Rocken, R.A.
Anthes and R.H. Ware, GPS meteorology: remote
sensing of atmospheric water vapor using the Global
Positioning System, Journal of Geophysical Research,
Vol. 97, No. D14, pp. 15787-15801, 1992.
Brunner, F. K. and M. Gu, An improved model for the
dual frequency ionospheric correction of GPS
observations, Manuscripta Geodetica, 16, pp. 205-214,
1991.
Flores, A. L. P. Gradinarsky, P Elsegui, J. L. Davis, and
A. Rius, Sensing atmospheric structure: Tropospheric
tomographic results of the small-scale GPS campaign at
the Onsala Space Observatory, Earth, Planets and
Space, 52, 941-945, 2000a.
Flores, A., G. Ruffini, and A. Rius, 4D tropospheric
tomography using GPS slant wet delays, Annales
Geophysicae, 18, 223-234, 2000b.
Flores, A., A. Rius, J. Vil-Guerau de Arellano, and A.
Escudero, Spatio-temporal tomography of the lower
troposphere using GPS signals, Physics and Chemistry
of the Earth, A, 26, 6-8, 405-411, 2001.
Gradinarsky, L. and P. Jarlemark, GPS tomography using
the permanent network in Goteborg: simulations,
Proceedings of IEEE Positioning, Location and
Navigation Symposium, Palm Springs, CA, April 15-18,
pp. 128-133, 2002.
Hirahara, K. Local GPS tropospheric tomography, Earth,
Planets and Space, 52, 935-939, 2000.
Hoyle, V. A., S. Skone, N. A. Nicholson, 4-D
Tropospheric Tomography using a Ground-Based
Regional GPS Network and Vertical Profile
Constraints, Proceedings of the Institute of Navigation
GNSS 2004, Long Beach, USA, 21-24 September 2004.

Presented at the ION GNSS-05 Conference, Long Beach, CA, September 13 16, 2005.

11

Jerrett, D. and J. Nash, Potential uses of surface based


GPS water vapor measurements for meteorological
purposes, Physics and Chemistry of the Earth (A), Vol.
26, No. 6-8, pp. 457-461, 2001.

Ware, R., C. Alber, C. Rocken and F. Solheim, Sensing


integrated water vapor along GPS ray paths,
Geophysical Research Letters, 24, pp/ 417-420, 1997.

Kuo, Y.-H., Y.-R. Guo and E. R. Westwater, Assimilation


of precipitable water measurements into a mesoscale
numerical model, Monthly Weather Review, 121, pp.
1215-1237, 1993.
Kuo, Y.-H., X. Zou, Y.-R. Guo, Variational assimilation
of precipitable water using nonhydrostatic mesoscale
adjoint model, Part I: Moisture retrieval and sensitivity
experiments, Monthly Weather Review, 124, pp. 122147, 1996.
Liou, Y.-A., Y.-J. Lin, C.-C. Chiang, and C.-Y. Huang,
Reconstructing 3D wet refractivity structures of the
lower troposphere from GPS measurements,
Proceedings of the GPS-MET Conference, Tsukuba,
Japan, January 2003.
Nicholson, N., V. Hoyle, S. Skone, M. E. Cannon and G.
Lachapelle, 4-D Troposphere Modeling Using a
Regional GPS Network in Southern Alberta,
Proceedings of the Institute of Navigation GNSS 2003,
Portland, USA, 9-12 September 2003.
NovAtel, Modulated Precision Clock User Manual, 2002.
NovAtel Website, http://www.novatel.ca, Accessed:08/03
Paroscientific Inc., MET3A Meteorological Measurement
System Installation, Operation, and Maintenance
Guide, 8360-002, August 11, 2001.
Rocken, C., T. Van Hove, and R. Ware, Near real-time
GPS sensing of atmospheric water vapor, Geophysical
Research Letters, Vol. 24, No.24, pp. 3221-3224, 1997.
Rocken, C., J. Braun, T. Van Hove, R. Ware, GPS
Networks for Atmospheric Sensing. Proceedings of the
ION National Technical Meeting, Jan 26, pp. 439-446,
2000.
Shrestha, S. M., Investigations into the Estimation of
Tropospheric Delay and Wet refractivity using GPS
Measurements. M.Sc. Thesis, Dept. of Geomatics
Engineering, University of Calgary, UCGE Report
Number 20180, 2003.
Skone, S. and S. Shrestha, 4-D modeling of water vapor
using a regional GPS network, Proceedings of the ION
National Technical Meeting, Anaheim, CA, Jan., 2003.
Troller, M., Brki, B., Cocard, M., Geiger, A., Kahle, H.G., 3-D refractivity field from GPS double difference
tomography, Geophysical Research Letters, 29, 24, pp.
2149-2153, 2002.
Tsonis, A. A., An Introduction to Atmospheric
Thermodynamics, Cambridge University Press, 2002.
University Technologies International Inc., MultiRefTM
Software RTK Software Package, UTI Ref# 256.4,
http://www.uti.ca/multiref-r.pdf, University of Calgary,
2004.

Presented at the ION GNSS-05 Conference, Long Beach, CA, September 13 16, 2005.

12

You might also like