Jurnal VLBI Helene Oktafiola

Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023].

See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth
RESEARCH ARTICLE Tropospheric delay ray tracing applied in VLBI analysis
10.1002/2014JB011552 David Eriksson1 , D. S. MacMillan2 , and John M. Gipson2

Key Points: 1 Center for Applied Mathematics, Cornell University, Ithaca, New York, USA, 2 NVI Inc. at the Planetary Geodynamics
• Troposphere delays are computed Laboratory, Goddard Space Flight Center, Greenbelt, Maryland, USA
by ray tracing numerical weather
model data
• Ray traced line-of-sight delays
are applied for each VLBI Abstract Tropospheric delay modeling error continues to be one of the largest sources of error in
quasar observation VLBI (very long baseline interferometry) analysis. For standard operational solutions, we use the VMF1
• VLBI analysis precision is improved elevation-dependent mapping functions derived from European Centre for Medium-Range Weather
using troposphere ray trace delays
Forecasts data. These mapping functions assume that tropospheric delay at a site is azimuthally symmetric.
As this assumption is not true, we have instead determined the ray trace delay along the signal path through
Correspondence to: the troposphere for each VLBI quasar observation. We determined the troposphere refractivity fields from
D. S. MacMillan,
[email protected] the pressure, temperature, specific humidity, and geopotential height fields of the NASA Goddard Space
Flight Center Goddard Earth Observing System version 5 numerical weather model. When applied in VLBI
analysis, baseline length repeatabilities were improved compared with using the VMF1 mapping function
Citation:
Eriksson, D., D. S. MacMillan, and model for 72% of the baselines and site vertical repeatabilities were better for 11 of 13 sites during the 2
J. M. Gipson (2014), Tropospheric delay week CONT11 observing period in September 2011. When applied to a larger data set (2011–2013), we see
ray tracing applied in VLBI analysis, a similar improvement in baseline length and also in site position repeatabilities for about two thirds of the
J. Geophys. Res. Solid Earth, 119,
9156–9170, doi:10.1002/2014JB011552. stations in each of the site topocentric components.

Received 25 AUG 2014


Accepted 9 NOV 2014
1. Introduction
Accepted article online 13 NOV 2014 Although many improvements have been made in the analysis of space geodetic data, tropospheric delay
Published online 14 DEC 2014 modeling is still a source of significant random and systematic error in the analysis of very long baseline
interferometry (VLBI) data. Since current geophysical problems such as global sea level rise and post-
glacial rebound require station position determinations at better than the 1 mm level, we still need to
make improvements in tropospheric delay modeling. To simplify the situation, earlier researchers made
the assumption that the troposphere at a site only varies in the vertical direction. Making this assumption,
the tropospheric delay through uniform refractivity layers is determined by one-dimensional ray tracing.
The general approach was to ray trace tropospheric profile data at a range of elevation angles and fit the
delay variation with an elevation-dependent mapping function. For example, Niell [1996] developed the
Niell Mapping Function (NMF) using the U.S. Standard Atmosphere profiles [Cole et al., 1965] at five North-
ern Hemisphere latitudes in January and July. The resulting mapping functions are parametrized by the
time of year (annual variation), latitude, and height of the site above the reference ellipsoid. The VMF1
mapping functions [Boehm et al., 2006] improved upon NMF both spatially and temporally by using numer-
ical weather model (NWM) data to derive mapping functions. Specifically, ECMWF (European Centre for
Medium-Range Weather Forecasts) data are spatially interpolated to a geodetic site location to obtain a
vertical site tropospheric profile (pressure, temperature, relative humidity, and geopotential height) at 6
h intervals. In an alternative use of NWM data, Niell [2000] included the 200 mb geopotential height from
the NWM of the Goddard Space Flight Center (GSFC) Data Assimilation Office in the parametrization of the
hydrostatic mapping function. Since the troposphere varies horizontally, several researchers have investi-
gated the computation of the ray trace delay along the signal path through three-dimensional NWM data.
For example, Nafisi et al. [2012a], Hobiger et al. [2008], and Gegout et al. [2011] discuss different ray trac-
ing strategies and the sensitivities of these algorithms. Nafisi et al. [2012b] applied ray trace delays derived
from ECMWF data in the analysis of the CONT08 VLBI series of 15 consecutive daily sessions in August 2008.
However, they were unable to show a clear improvement in geodetic precision compared to using VMF1
mapping functions.
If there were no atmosphere, the path of the ray from the site toward the observed radio source would be a
straight line. However, due to the presence of the atmosphere, the raypath will not be straight and its path
can be retrieved by solving the eikonal equation (see Wheelon [2001] or Born and Wolf [1999]):
|∇L|2 = n2 (⃗r ), (1)

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9156
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

where ∇L are the vector components of the raypath direction, L is the optical path length, n is the index of
refraction, and ⃗r is the three dimension vector position on the raypath. This can be rewritten into a set of dif-
ferential equations that must be solved simultaneously using some standard methods such as Runge-Kutta.
Since VLBI has a large number of observations, it is important to use a computationally efficient algorithm
to perform the ray tracing. Since solving the eikonal equation is very time consuming [see, for example,
Hobiger et al., 2008], we made the assumption in the work reported here that the ray does not leave the path
of constant azimuth, which makes it possible to use a simpler 2-D ray tracing algorithm. By assuming no
out-of-plane movement, we assume that the horizontal components of the gradient is equal to zero so that
the refractivity gradient always points toward the center of the Earth. The tropospheric delay, 𝛿𝜏atm , is the
radio path delay difference between the delay along the path through the atmosphere and the geometrical
straight line vacuum path delay if no atmosphere was present.

𝛿𝜏atm = n( ⃗r ) ds − ds (2)
∫ ∫
atm vac

⎛ ⎞
𝛿𝜏atm = (n( ⃗r ) − 1) ds + ⎜ ds − ds⎟ (3)
∫ ⎜∫ ∫ ⎟
atm ⎝atm vac ⎠

The first term is the electromagnetic delay due to the troposphere. This is the dominant part of the delay,
and it can be as large as 25 m for an outgoing elevation of 5◦ . The second term is the geometric excess delay,
which is the path length difference due to bending. It is the difference between the length of the true path
and the unrefracted straight line path from the station with the same angle as the outgoing angle of the
true path. The geometric excess delay is below 1 mm for elevations larger than 60◦ , 20 mm for an elevation
of 5◦ , and as large as 2.5 m at 1◦ . It is therefore important to accurately model the geometric excess delay for
lower elevation observations.
In this paper, we discuss our investigation of the application in VLBI analysis of ray traced delays computed
for each radio source observation using troposphere profile data from the GSFC (Goddard Space Flight
Center) GMAO (Global Modeling Assimilation Office) Goddard Earth Observing System version 5 (GEOS-5)
numerical weather model. Section 2 describes the computation of the troposphere refractivity field from
numerical weather model parameter fields. We discuss the algorithms used to ray trace the refractivity
fields in section 3. After the ray traced troposphere delays were calculated for all VLBI observations, they
were applied in the VLBI analysis. In section 4, we discuss the results of this analysis and the improvement
in geodetic parameter estimates relative to using the best current mapping function approaches, VMF1
or NMF.

2. Transformation From Numerical Weather Model Parameters to Refractivity Fields


2.1. The GMAO GEOS-5 FP-IT Model
The first step in our ray trace procedure is to compute a gridded refractivity field that can be used for the
ray tracing. For the work reported in this paper, we used the GSFC GMAO GEOS 5.9.1 model. This model is a
terrain-following model from which we have extracted surface pressure, pressure level thickness, geopoten-
tial heights, temperature, and specific humidity. Here the temperature, specific humidity, and geopotential
heights are given at the midpoint between pressure levels so that one can compute the pressure values at
the midpoint by using the fact that the uppermost level is constrained to 0.01 hPA. Surfaces of constant eta
follow the surface height variations. The pressure differences between eta levels depend on location and
are provided by the model. This yields a set of levels with pressure, geopotential heights, temperature, and
specific humidity.
The GMAO model parameters are calculated in near real time and are given as 3 h time-averaged outputs
every third hour at midpoint times 01:30, 04:30,..., 22:30. There are a total of 72 vertical levels for each profile
and the horizontal resolution is 5/8◦ in longitude and 1/2◦ in latitude. The model is a part of the GEOS 5.9.1
FP-IT project and is available from 1-1-2000 up to the present with a latency of about 24 h.

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9157
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

2.2. Computation of Refractivity Values


Given the pressure, temperature, and specific humidity, the atmospheric refractivity can be computed
according to Davis et al. [1985] as
pd p p
N = Nd + Nv = k1 Zd−1 + k2 Zv−1 v + k3 Zv−1 v2 (4)
T T T
where N = (n − 1)106 and Nd and Nw are the dry and the wet components of the atmospheric
refractivity. Zd and Zv are the inverse compressibilities. To evaluate this expression, we used the values
k1 = 77.6890 (10−2 K/Pa), k2 = 71.295 (10−2 K/Pa), and k3 = 375.463 (101 K2 ∕Pa) from Rueger [2002a] and
Rueger [2002b]. According to Wallace and Hobbs [2006], the water vapor pressure pv is computed using
pq
pv = Mv Mv
(5)
Md
+ (1 − Md
)q

where Mv = 18.02 g/mol and Md = 28.96 g/mol are the molar masses of wet and dry air, respectively, p is
the total pressure, and q is the specific humidity. For ray tracing, it is important to compute both the wet and
the dry components of the refractivity fields. Rewriting the above equation using the equation of state for
nonideal gases,

pi = Zi 𝜌i Ri T (6)

where i = d, v and the Ri are the corresponding gas constants, we get

p R p p
N = k1 + [(k2 − k1 d ) v + k3 v2 ] = Nh + Nnh (7)
T Rv T T

where Nh is the hydrostatic refractivity that is proportional to the total pressure and Nnh is the nonhydro-
static refractivity. We set the water vapor compressibility Zv equal to one since it is insignificantly different
from one. The refractivity can then be evaluated using the parameters (p, q, T ) from the GMAO model.
2.3. Conversion From Geopotential Heights Into Heights Over the Ellipsoid
For VLBI, the station heights are given with respect to the surface of the reference ellipsoid. Since the numer-
ical weather models provide geopotential heights, these heights must be converted into heights over the
ellipsoid. The relation between the heights over the geoid and the geopotential height is
𝜁 ⋅ g0
H= (8)
𝛾̄ (𝜑, 𝜆, H)

where H is the height over the geoid, g0 is the conventional value of the acceleration due to gravity, and
𝜁 is the geopotential height. The mean gravitational acceleration between the surface of the geoid and a
point over the ellipsoid, 𝛾̄ , can be computed using any suitable gravitational model for the WGS-84 ellipsoid.
After converting from geopotential heights into heights over the geoid, the heights over the ellipsoid can
be obtained using

h = H + Hu (𝜑, 𝜆), (9)

where Hu is the geoid undulation, which is the difference between the surface height of the geoid and
the surface of the ellipsoid. This undulation should be computed using spherical harmonics of a degree
larger than 2000; otherwise, it can result in errors on the meter level which will affect the ray tracing. We
used a global 1’ × 1’ grid provided by EGM08 to interpolate to the model points [Pavlis et al., 2012]. EGM08
expresses these undulations using spherical harmonics up to degree 2190. As a test, when we kept spheri-
cal harmonics only up to degree 360, there were differences relative to EGM08 exceeding 10 m at multiple
points on the globe, which is too large to give accurate ray tracing results.
2.4. Geodetic Versus Geocentric Latitudes
The station positions of the VLBI stations are expressed in geodetic coordinates while numerical weather
models are based on a spherical Earth. Since much of the input data to the NWM, such as topography,
is based on an ellipsoidal Earth, the numerical weather models should be interpreted as being given in
geodetic coordinates.

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9158
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

2.5. Vertical Interpolation of Pressure,


Temperature
90 270
Temperature, and Specific Humidity
In order to construct a grid of refractiv-
260
ity values suitable for the ray tracing,
80
we compute the refractivity for a set of
250
320 exponentially distributed heights
as proposed by Hobiger et al. [2008]
by splining the NWM pressure, tem-
70 240
perature, and specific humidity. This
distribution of heights is consistent with
height (km)

230
the exponential decrease in pressure
60
with height. The temperature is extrap-
220
olated upward using the Committee on
Space Research (COSPAR) International
50 210
Reference Atmosphere model [Rees et
al., 1990]. This model provides yearly
200
means of the temperature as a function
40
of height for latitudes {−90,−85,...,85,90}
190
that we use to extrapolate the temper-
atures above the largest height of the
30 180
−100 −50 0 50 100 NWM. Since temperature has a clear
latitude latitude dependence that can be seen
in Figure 1, the COSPAR model is more
Figure 1. Temperature values from the COSPAR International Ref-
erence Atmosphere model. There is a clear variation in latitude suitable than the US76 Standard Atmo-
which makes this reference atmosphere preferable to the U.S. sphere model, which only provides a
Standard Atmosphere. global average temperature for each
height. After extrapolating the tempera-
ture, the pressure can be extrapolated upward as in Nafisi et al. [2012a] where it is important to note that the
height used there is actually the geopotential height.
Since the model is terrain following, there are no troposphere values below the surface terrain, which means
that we need to extrapolate these variables downward if a station is located below the height of the corre-
sponding NWM grid cell. The temperature can be extrapolated downward by using a lapse rate calculated
from heights up to 10 km over which the temperature, in general, has a linear structure. It is more prob-
lematic to extrapolate the specific humidity downward, and we therefore set the specific humidity equal
to its value at the lowest model height for all heights below this height. The pressure can be extrapolated
downward in the same way as for the upward extrapolation.

3. Ray Tracing Algorithm


Because of the large number of VLBI observations, it is necessary to use a fast ray tracing algorithm, where
the delay for each observation can be computed in a time of the order of 1 ms. To be able to make a fast
ray tracing algorithm, we assume that the horizontal components of the gradient of the refractivity field
are zero, which will force the ray to stay in the plane of constant azimuth [Hobiger et al., 2008]. To be able to
compute the delay along the raypath from both the wet and the dry part of the atmosphere, a piecewise
linear scheme is used to first find the path of the ray and then to evaluate the hydrostatic and wet delay
components along the path.
Our implementation is similar to the piecewise linear approach discussed by Hobiger et al. [2008] where
more details can be found. We used the 2-D approach, because the difference between 2-D and 3-D ray
tracing rarely exceeds the few millimeter level (3 mm) for elevation angles as low as 5◦ [Hobiger et al., 2008]
and because the solution of the system of ordinary differential equations in each iteration of 3-D ray trac-
ing requires at least a factor of 10 times as much computational time. In contrast, the ray traced delays using
different weather models can differ by decimeters [see Nafisi et al., 2012a], which indicates that the contribu-
tion to the error budget from inaccuracies in the numerical weather model is an order of magnitude larger
than the error due to constraining the ray tracing to two dimensions. If numerical weather models improve
and computational performance increases, then it might then be optimal to perform 3-D ray tracing.

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9159
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

Figure 2. Global hydrostatic and wet zenith delays for 24-9-2011 12:00 CT in picoseconds. The wet zenith delay field
shows a larger spatial variation than the hydrostatic zenith delay field.

3.1. Vertical Interpolation


To perform the ray tracing, we need to determine the refractivity at an arbitrary height in a refractivity pro-
file. This is necessary, for example, at the initial step of ray tracing where we need the refractivity at the
height of the station. We assume an exponential decrease in refractive index between each layer k and the
next layer k + 1,

n(r, 𝜑, 𝜆) = 1 + (nk (𝜑, 𝜆) − 1) exp(C(r − Rk )), (10)

where the Rk are the layer heights, the nk are the layer refractivities, and
( )
n (𝜑,𝜆)−1
log nk+1(𝜑,𝜆)−1
.
k
C= (11)
Rk+1 − Rk

This choice is made because the main contributor to the refractivity field is the pressure which varies
exponentially with height.
3.2. Horizontal Interpolation
The hydrostatic refractivity field varies slowly in space as can be seen in Figure 2. This is different from the
wet refractivity field that varies much more spatially and temporally, suggesting that it might be optimal
to use different interpolation schemes for the wet and the hydrostatic refractivity fields. The GMAO GEOS-5

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9160
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

model provides parameter values that


are representative of each grid box and
not of a single point in space, that is,
for the central grid point (M. Bosilovich,
personal communication, 2013). Since
the hydrostatic refractivity field varies
slowly spatially, we assign the hydrostatic
refractivity to the center point of the grid
box and interpolate the hydrostatic field
from the grid box centers. However, since
the wet component varies faster spa-
tially and is less certain, it makes sense
to assign the wet part of the refractivity
to the entire grid box. To avoid making
the wet refractivity field discontinuous,
smoothing is applied near the borders
of each grid box to account for spatial
uncertainty in the wet part due to effects
such as local fronts and storms.
This horizontal interpolation scheme is
illustrated in Figure 3 where the gray
Figure 3. Illustration of the horizontal interpolation of the wet refrac-
tivity field. The gray subcells have the value of the GMAO grid box, subcells have the value of the grid cell
and interpolation is carried out between the subcells to find the values and interpolation is carried out between
outside the subcells. these grid cells in order to obtain the
value for the white space in between the
subcells. In our routines, we let the gray
subcell have about 60% of the area of the
entire cell.
3.3. Earth Radius
The ray tracing is carried out at a site in a spherical coordinate system centered about the site vertical coor-
dinate using geodetic coordinates of the NWM data as local spherical coordinates. We assume that the
raypath does not move out of the vertical azimuth plane. This means that at the local level for each step of
the piecewise linear integration, we assume that the horizontal components of the refractivity gradient are
zero so that the refractivity gradient points vertically downward. Making these assumptions allows the use
the spherical Snell’s law. The most suitable choice of radius for the ray tracing is choosing a local radius of
the Earth to be the Euler radius of curvature given by
MN
R= , (12)
M sin2 𝛼 + N cos2 𝛼
where 𝛼 is the azimuthal angle between the ellipsoidal meridian plane of the site and the vertical plane
containing the normal vector at the site point and the source point and
a(1 − e2 ) a
M= ( )1.5 N= √ . (13)
1 − e2 sin2 𝜑 1 − e2 sin2 𝜑

One advantage of using the Euler radius of curvature is that it gives the radius of curvature of the ellipse
cross sectioned by a vertical plane in the direction of the azimuth. For the WGS-84 spheroid, a = 6378137 m,
2 2
b = 6356752.3142 m, and e2 = a a−b 2
.
The geocentric latitude for the osculating sphere coordinate system equals the geodetic latitude of the site
position on the Earth’s surface since the radial direction of the osculating sphere is normal to the reference
ellipsoid. Using a local spherical coordinate system based on the osculating sphere with Euler radius was
compared to using a geocentric spherical ray tracing system. For the latter system, one must convert to ellip-
soidal coordinates when retrieving the refractivity values. In our tests, the difference in ray traced delays was
at most 0.5 mm at the minimum elevation angle of 5◦ . We decided to use the osculating sphere coordinate
system since it requires fewer calculations and the delay differences were on the submillimeter level.

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9161
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

Figure 4. Comparison of GSFC ray trace delays with calculations from other investigators who provided their results to
the Vienna benchmarking campaign

3.4. Time Interpolation


The GMAO model values are given every 3 h, while VLBI observations occur at arbitrary times. Hence, the
grid of refractivity values must be splined in time before each observation can be processed. The refractivity
field computed from the GMAO data is splined temporally to provide a 4-D field, n = n(t, h, 𝜙, 𝜆), that can be
used to calculate the delays at any given time.
3.5. Global Zenith Delay Maps
To get a picture of how the refractivity varies over the Earth’s surface, we computed the zenith delay from
every grid point and created global maps of zenith delays. This was done for the 12:00 CT on 16 September
2011, and the result can be seen in Figure 2. It is clear from Figure 2 that the hydrostatic zenith delay does
not vary as much spatially as the wet zenith delay. It is well known that the hydrostatic zenith delay, which
can be computed accurately using Saastamoinen’s formula [Saastamoinen, 1972], is a linear function of sur-
face pressure which has a much larger spatial scale of variation and also varies as a function of topographic
height. The spatial scale of pressure variation ranges from a few hundred to a few thousand kilometers.
Synoptic scale weather patterns (which are characteristic of pressure variations) have a spatial scale of
1000–2000 km and move with periods of a few days [see, for example, Holton, 2006]. Wet delay variation is
due to turbulence, which varies on much smaller spatial scales than the hydrostatic delay.
3.6. Benchmarking Campaign
Nafisi et al. [2012a] summarize the results of a ray trace delay campaign undertaken to compare ray trace
software from different investigators. Each investigator in the campaign used his software to compute the
total slant path delay given the same tropospheric profile data. We obtained this troposphere data and per-
formed the ray tracing calculations as a test of our ray tracing procedure. Figure 4 provides a comparison of

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9162
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

slant path delays computed at an elevation of 5◦ as a function of azimuth. Our (GSFC) slant path delays are
close to the average of the other calculations and have an azimuthal dependence similar to that of the other
ray trace calculations. However, it should be noted that there are 1–5 cm level biases between the different
calculations. Nafisi et al. [2012a] also assessed ray trace delay differences arising from different NWMs. The
computed slant delays at 5◦ using ECMWF, Japan Meteorological Agency, or Canadian Meteorological Cen-
ter NWM differed by up to 20 cm. They concluded that the differences in ray trace delays are caused much
more by differences in NWMs than to errors in ray tracing programs.

4. Validation of Ray Tracing and Improvement in VLBI Analysis


In order to validate our calculation of ray trace delays, we used our ray tracing software to generate total
(hydrostatic + wet) a priori delays and the wet partial derivatives (wet mapping function) for all VLBI obser-
vations in each 24 h daily observing session and apply these in VLBI analysis solutions of different sets of
VLBI data. We considered two time periods: (1) CONT11 (continuous VLBI campaign 2011) observed during
the period 15–29 September 2011 with 134,986 group delay observations and (2) all daily 24 h sessions (R1,
R4, CONT11, and RDV (research and development very long baseline array (VLBA)) network sessions) from
January 2011 through June 2013 consisting of 912,733 group delay observations. The VLBI analysis reported
in this paper used the Calc/Solve analysis system. Ma et al. [1990] provide a general description of the theo-
retical models that are used and the Solve least squares estimation program. The models that are employed
generally follow the recommendations of IERS (International Earth Rotation and Reference Systems Service)
Conventions 2010 [Petit and Luzum, 2010]. Our analysis also included application of models for antenna
thermal deformation [Nothnagel, 2009] and atmospheric pressure loading [Petrov and Boy, 2004].
4.1. VLBI Solution Descriptions
We evaluated the improvement resulting from using ray trace delays by comparing site position time series
solutions that used ray traced delays versus VMF1 delays since the VMF1 mapping functions are the best
current mapping functions available. Specifically, we evaluated the improvement in baseline length and
station coordinate estimate repeatabilities. The advantage of considering VLBI baseline length measure-
ments is that the baseline length is independent of orientation in contrast to the local station coordinates
that depend upon reference frame orientation. In these VLBI solutions, each 24 h session is analyzed inde-
pendently. For each session we estimate site positions (applying a no net translation constraint), station
clock offsets (piecewise linear with 60 min intervals), wet zenith residual delays (piecewise linear with 20 min
intervals), and gradients (piecewise linear with 6 h intervals).
The a priori VMF1 delay is the sum of the hydrostatic and wet delay

𝜏tot = mhyd 𝜏hyd


zen
+ mwet 𝜏wet
zen
(14)

where mhyd and mwet are the VMF1 hydrostatic and wet mapping functions, 𝜏hyd zen
is the Saastamoinen zenith
delay [Saastamoinen, 1972], and 𝜏wet is the ECMWF wet zenith delay. We performed solutions with three
zen

different data weighting strategies: (1) baseline reweighting, (2) elevation-dependent weighting, and (3)
correlated noise. For each baseline in each daily session, baseline reweighting adds a constant noise con-
tribution in quadrature to the observation uncertainties such that the 𝜒 2 per degree of freedom over the
observations in a daily VLBI session is unity. The noise value is computed over all the observations for
each baseline observed in the daily 24 h session. For elevation-dependent weighting, noise of the form
𝛼∕ sin(elev) is added in quadrature to the observation uncertainties, introducing more noise for low eleva-
tion angles. In this study we set 𝛼 = 6 ps. The correlated noise strategy [Gipson, 2007] adds an off-diagonal
correlation between simultaneous observations on baselines that share a common station, using the same
elevation-dependent noise term. Two such observations are correlated because they each contain the tro-
pospheric delay noise associated with the common station. These three different weighting strategies are
described in MacMillan and Gipson [2009].
If wet residual zenith and gradient delay parameters are not estimated, the resulting ray trace delay solu-
tions will be much worse. Even though the ray trace delays contain troposphere wet zenith delay and
gradient delay information, it is still necessary to estimate the residual wet zenith delay and gradient
parameters to account for biases between the GMAO model and the true site weather parameters. If either
gradient or residual wet zenith delay parameters are not estimated, applying ray trace delays significantly

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9163
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

improve the solution results [Nafisi et al.,


2012a]. However, this will significantly
decrease the quality of the solution,
increasing the variance of site position
and baseline length estimates com-
pared with solutions in which these
troposphere parameters are estimated.
We therefore only considered the case
where these troposphere parameters
were estimated.

4.2. CONT11 VLBI Sessions


In order to test the ray tracing pro-
cedures, we used the data set from
CONT11, which was a 2 week cam-
paign of continuous 24 h VLBI sessions
in September 2011 observed with the
same 13-antenna network for each
Figure 5. Comparison of ray traced wet zenith delay (dashed red line)
at Westford and wet residual zenith delay (solid blue line) estimated day. The advantage of studying this
from observed VLBI data. data set is that it provides a temporally
dense set of high-quality VLBI observa-
tions in contrast to the operational R1
and R4 sessions that each observe with 8–10 antennas on 1 day each week, but the networks vary from
week to week. Initially, we investigated how much of the observed wet zenith delay is modeled by the
ray traced wet zenith delay. Figure 5 shows how closely the wet zenith delay at Westford (Massachusetts)
calculated by ray tracing follows the wet zenith delay estimated from the VLBI data. The VLBI estimate
was made by applying only an a priori hydrostatic delay using the VMF1 hydrostatic mapping function
so that the estimated residual zenith delay is predominantly the wet zenith delay. Ray tracing provides a
good a priori model for the wet zenith delay and accounts for about 90% of the RMS (root-mean-square)
of the wet residual zenith delay estimated from the VLBI data. Figure 6 compares the estimated resid-
ual wet zenith delays at Westford after applying a priori ray trace or VMF1 delays in VLBI analysis. Table 1
summarizes the wet residual delay statistics from the four different a priori tropospheric delay options
(NMF, VMF1 hydrostatic, VMF1 total, and ray trace delay) for four representative VLBI sites. These statis-
tics show that VMF1 and ray tracing delays are both acceptable a priori models for the wet zenith delay,
but that overall ray tracing is the best option. The estimated wet residual zenith delay is reduced more
at some sites than others, indicating how well the numerical weather model refractivity agrees with the
actual tropospheric refractivity close to the station. We can, for example, see that of the two sites that were
improved the most on average in baseline length repeatability (see Figure 7), only Hobart12 (Tasmania,
Australia) had a significant decrease in RMS wet residual zenith delay. A possible reason why Tsukub32
(Japan) is not improved as much might be because of the narrow and complicated shape of Japan which
means that many of the GMAO grid cells will contain both ocean and land.
Total (hydrostatic + wet) delays from ray tracing were applied as a priori tropospheric delays in Solve
analysis, and a comparison was made with analysis using VMF1 total delays given by equation (14) or NMF
hydrostatic delays. Table 2 shows that on
Table 1. RMS of Residual Wet Zenith Delay (ps) a average, ray tracing improves the base-
NMF VMF1 hyd VMF1 total RT
line length precision (repeatability) for
all three weighting strategies compared
Fortleza 690.8 699.1 80.9 60.0
Hobart12 327.8 327.1 42.6 25.2 to using either VMF1 or NMF. The corre-
Kokee 328.6 327.3 62.1 50.7 lated noise strategy performs better than
Tsukub32 739.8 746.0 61.3 62.8 the other weighting options. Figure 7
Westford 644.3 646.2 55.7 48.4 shows how the improvement in base-
Weighted average 391.9 393.1 46.8 41.0
line length repeatability using ray traced
a The weighted average was computed using all stations with the
delays versus VMF1 is distributed among
number of observations as weights.
the stations, where the improvement

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9164
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

bars are ordered by baseline length.


The improvement shown (weighted
root-mean-square (WRMS) reduction
in quadrature) is the square root of the
reduction in variance for each baseline.
A one-sided t test yielded a probability
of 0.04% that the average reduction in
variance using ray traced delays could
have occurred if the difference between
the ray traced delay correction and VMF1
was noise.
For each station, ray traced delays reduce
the variance on average over that sta-
tion’s baselines for all stations except
for Kokee (Hawaii) where VMF1 per-
forms better. Kokee is located on the
Hawaiian Island of Kauai about 800 m
above the ocean surface. This is prob-
lematic in the sense that many of the
GMAO grid cells that contribute to the
ray trace delay contain both land and
ocean that likely do not represent the
true variation in the troposphere across
the land-ocean boundary. Hobart12 and
Tsukuba are improved the most with
ray traced delays. Both of these sites are
observed on many long baselines, where
the baseline length variation is largely
due to site vertical variation, which is
highly correlated with troposphere mod-
eling error. Hence, one expects that ray
tracing, if correctly modeled, should lead
to a larger improvement for these sites.
Figure 8 shows the length WRMS reduc-
Figure 6. Estimated residual wet zenith delay after applying as the ation over all baselines, where ray tracing
priori delay either the VMF1 total delay (solid red line) or the ray traced
improves the WRMS repeatability for
total delay (dashed blue line) for Hobart12 (Tasmania, Australia) and
Tsukub32 (Tsukuba, Japan), during the CONT11 sessions. 72% of the baselines compared to
VMF1. The improvement increases with
baseline length because troposphere modeling improvements predominantly affect the site vertical com-
ponent and the projection of the vertical direction onto the baseline length increases with baseline length.
Figure 8 also shows that site vertical position variance is reduced for 11 of the 13 CONT11 sites. Although
ray tracing should contain a horizontal gradient signal, there is no clear improvement for the horizontal
site components.
With past improvements in tropospheric delay modeling (for example, better mapping functions and the
estimation of gradient parameters), the level of reference frame-scale error associated with the troposphere
has decreased [see, for example, MacMillan and Ma, 1997]. We estimated the scale bias difference between
CONT11 solutions that used ray traced delays versus NMF or VMF1. To do this, we computed the mean
length differences for each baseline for all baselines in each solution. Fitting these differences with a linear
function in baseline length, the slope is the scale bias. The ray trace VMF1 scale bias difference was 0.0009
±0.046 ppb and, the ray trace NMF scale difference was −0.038 ± 0.046 ppb.

4.3. The 2011–2013 VLBI Sessions


In order to evaluate the performance of ray tracing over a longer time period, we analyzed all 24 h VLBI
sessions for the period 2011.0–2013.5. We again determined the improvement in baseline length and

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9165
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

Figure 7. Reduction in baseline length WRMS for all baselines in the CONT11 sessions. The baselines for each station are
sorted by baseline length which shows that the improvement generally increases as a function of baseline length.

station coordinate estimate repeatabilities when ray trace a priori delays were applied. This data set is much
more inhomogeneous than the CONT11 data set and included the R1 and R4 weekly operational network
sessions, the CONT11 sessions, the bimonthly European network sessions, and the RDV (research and devel-
opment VLBA) bimonthly sessions. The RDV sessions use the 10 VLBA (very large baseline array) network
stations along with 8–10 additional IVS antennas.
Figure 9 shows that the baseline length WRMS for most stations are improved on average when ray tracing
is applied instead of VMF1. By a one-sided t test, the probability that the average reduction in variance using
ray traced delays could have occurred if the difference between the ray traced delay correction and VMF1
was noise was only 0.78%. Fortaleza (Brazil), Svetloe (Russia), Tigo Concepcion (Chile), Tsukuba (Japan),
Westford (Massachusetts), Zelenchukskaya (Russia), and Yebes (Spain) show improvement for almost all
baselines. However, the Hobart26 results are anomalous, since there is improvement for Hobart12, which
is only a short distance of only 300 m from the Hobart26 station. Hobart26 participated in a much smaller
number of sessions and observed on fewer baselines than Hobart12.
Figure 10 shows that ray tracing improves the WRMS repeatability for 71% of the baseline lengths compared
to VMF1. The improvement shown is not as large compared with the improvement seen for CONT11, where
most of the repeatabilities on most long baselines were improved. Possible reasons to explain this are the
following: (1) the quality of the data was better in CONT11 since much more effort is made to ensure station
performance quality, whereas the longer period consists of predominantly operational 24 h sessions, and (2)
there are more sources of error over the longer 2.5 year period than the short CONT11 period.
The improvement in station position estimates is plotted in Figure 10, and we see that for the vertical com-
ponent 71% of the stations are improved relative to VMF1. The results for Hobart26 are still anomalous and
the vertical repeatability is not improved at Hobart12 even though the baseline length repeatabilities were
improved significantly. However, there was significant reduction of the WRMS repeatabilities for the hori-
zontal components at Hobart12. Considering the horizontal components, we see that similar to the vertical

Table 2. Average and Weighted Average Baseline Length WRMSa

NMF VMF1 Total RT


Weights Average Weight Average Average Weight Average Average Weight Average
(bas) 6.92 6.89 6.77 6.75 6.42 6.41
(elev) 6.53 6.50 6.32 6.31 6.05 6.04
(cor) 6.37 6.35 5.97 5.96 5.73 5.73
a The weighted average used the number of observing days to weight each baseline.

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9166
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

Figure 8. Reduction in baseline length and UEN station position WRMS for the CONT11 session.

component, we have an improvement for about two thirds of the stations compared to VMF1. There is a
large improvement in the north-south component for Hobart12 while Hobart26 is still worse.
We also investigated the effect on EOP (Earth orientation parameters) estimation when ray trace delays were
applied in analysis instead of VMF1. To evaluate the results, we compared the VLBI estimates of polar motion
and length of day (LOD) with GPS estimates to determine whether ray tracing improves the consistency
between VLBI and GPS. Model improvements to either of the two independent geodetic techniques should
decrease the differences between EOP parameters estimated by the two techniques. VLBI EOP parameters
from the ray trace and VMF1 solutions were compared with the IGS (International GNNS Service) combina-
tion EOP series. From the results in Table 3, ray tracing improves agreement with IGS compared with using
VMF1 for all EOP parameters except for Y pole rate.

Figure 9. Reduction in baseline length WRMS for all baselines in the CONT11, R1, R4, and RDV sessions in the time period
2011–2013. The baselines for each station are sorted by baseline length which shows that the improvement generally
increases as a function of baseline length. Note that some stations do not have as many baselines as others.

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9167
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

Figure 10. Reduction in baseline length and UEN station position WRMS for the CONT11, R1, R4, and RDV sessions in the
time period 2011–2013.

4.4. LOD From Intensive One-Baseline Sessions


We next investigated whether there was any improvement in analysis of intensive one-baseline UT1 ses-
sions if we used ray trace delays in the analysis. There are two series of Intensive 1 h sessions that are
performed operationally each week. The Kokee-Wettzell sessions are observed Monday through Friday and
the Tsukuba-Wettzell sessions on the weekend (Saturday and Sunday). In order to see whether our ray trace
delays improve results, we compare VLBI LOD estimates with simultaneous GPS LOD data using all Intensive
sessions observed during the time span January 2011 and June 2013. We ran solutions with NMF, VMF1, or
ray tracing as the a priori troposphere delay model. In the analysis of Intensive sessions, we estimate UT1,
only one wet zenith troposphere delay parameter for each site and clock parameters for one site of the base-
line, because only a limited number of observations (typically 15–30 observations are made during these
1 h sessions).
To perform a comparison with GPS LOD, we computed a VLBI estimate of LOD at the midpoint epoch
between each pair of 1 h Intensive sessions using UT1 estimated for each session in a pair, which is similar to
a method used by Boehm et al. [2010] and Nafisi et al. [2012b]. The resulting LOD series was then compared
with the IGS LOD daily series, interpolating IGS to the VLBI epochs. The results of the comparisons are given
in Table 4. We see that the results using VMF1 and NMF are almost the same while in the case of ray tracing
we see an improvement in LOD agreement with GPS of about 4%.

Table 3. Comparison Between EOP Parameters and Their Agreement


With GPS

VMF1 Total RT
WRMS 𝜒 2 /dof WRMS 𝜒 2 /dof
X pole (𝜇 as) 106.3 2.8 104.4 2.7
Y pole (𝜇 as) 94.4 2.4 94.1 2.4
X pole rate (𝜇 as/d) 237.5 2.7 232.5 2.6
Y pole rate (𝜇 as/d) 252.0 3.0 255.9 3.1
LOD (𝜇 s) 14.1 5.0 14.0 5.0

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9168
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

Table 4. WRMS Difference of LOD (𝜇 s/d) From VLBI Intensive Sessions and
LOD From GPS

Baseline NMF VMF1 Ray Trace Number of Sessions


Kokee-Wettzell 25.4 25.2 24.3 80
Tsukuba-Wettzell 28.2 28.3 26.1 59

5. Conclusions and Future Work


We have seen that ray tracing has the potential to significantly improve VLBI analysis and that this depends
on carefully choosing interpolation methods and other components of the ray tracing system. The appli-
cation of ray tracing for the CONT11 period improved the baseline length repeatabilities for 72% of the
baselines compared to using the VMF1 model, which is the best troposphere model currently available.
In addition, the site vertical precision was better with ray tracing than with VMF1 for 11 out of 13 stations.
Based on data from the 2 week period of CONT11 observing, the reference frame scale bias of
0.001 ± 0.046 ppb between ray tracing and VMF1 is insignificant. Further scale tests with larger data sets
should be made in the future to confirm this result. Applying ray tracing to the larger data set of R1, R4, and
RDV observed from 2011 to 2013 showed that ray tracing improves baseline length precision for 71% of the
baselines compared with applying VMF1. In addition, site coordinate precision (up, east, north) was better
for two thirds of the 28 stations that observed during the period.
Our work shows the potential of ray tracing, and the hope is that with improvements in weather models, ray
tracing will only get better. Ray tracing did not perform as well for the station Kokee, located on the Hawai-
ian Island of Kauai, suggesting that ray tracing might not work as well where the coastline is not resolved
within the numerical weather model grid. We expect ray tracing to perform better as numerical weather
modeling evolves, and the spatial and temporal resolutions of the models improve. This could lead to better
geodetic results for VLBI sites located either along the coast or on islands.
We are currently producing ray tracing results for the entire GEOS 5 FP-IT time period (1-1-2000 to present)
and expect to extend this period back to 1990 as part of a ray tracing service for VLBI data users located at
http://lacerta.gsfc.nasa.gov/tropodelays/. The current latency is 24 h which allows the ray tracing service
to provide delays quickly and makes it possible for the VLBI community to apply tropospheric corrections
within a day after the VLBI experiment was observed. The ray tracing service provides the tropospheric cor-
rections on a 24 h session basis in the form of TRP files. These files are grouped by year and are available for
all VLBI sessions since the year 2000. If there is interest, there are plans to extend the ray tracing service to
other techniques such as GPS, DORIS, and SLR.

Acknowledgments References
We thank the NASA Global Modeling
and Assimilation Office (GMAO) for Boehm, J., B. Werl, and H. Schuh (2006), Troposphere mapping functions for GPS and very long baseline interferometry from European
providing data from the numerical Centre for Medium-Range Weather Forecasts operational analysis data, J. Geophys. Res., 111, B02406, doi:10.1029/2005JB003629.
weather model GEOS 5.9.1 and the Boehm, J., T. Hobiger, R. Ichikawa, T. Kondo, Y. Koyama, A. Pany, H. Schuh, and K. Teke (2010), Asymmetric tropospheric delays from
International VLBI Service for Geodesy numerical weather models for UT1 determination from VLBI Intensive sessions on the baseline Wettzell-Tsukuba, J. Geod., 84, 319–325,
and Astrometry (IVS) for providing VLBI doi:10.1007/s00190-10-0370-x.
data for our analysis. GMAO data are Born, M., and E. Wolf (1999), Principles of Optics: Electromagnetic Theory of Propagation, Interference and Diffraction of Light, Cambridge
available at http://gmao.gsfc.nasa.gov/ Univ. Press, New York.
products/, and IVS data products can Cole, A. E., A. Court, and A. J. Cantor (1965), Model atmospheres, in Handbook of Geophysics and Space Environments, edited by S. L.
be accessed from http://ivscc.gsfc.nasa. Valley, pp. 2-1–2-22, McGraw-Hill, New York.
gov/products-data/products.html. We Davis, J. L., T. A. Herring, I. I. Shapiro, A. E. E. Rogers, and G. Elgered (1985), Geodesy by radio interferometry: Effects of atmospheric
acknowledge our support from NASA modeling errors on estimates of baseline length, Radio Sci., 20(6), 1593–1607.
contract NNG12HP00C. We would Gipson, J. (2007), Incorporating correlated station dependent noise improves VLBI estimates, in Proc. 18th European VLBI for Geodesy and
also like to thank two anonymous Astrometry Working Meeting, edited by J. Boehm, A. Pany, and H. Schuh, pp. 129–134, Tech. Univ. Wien, Vienna, Austria.
reviewers for their helpful comments. Gegout, P., R. Biancale, and L. Soudarin (2011), Adaptive mapping functions to the azimuthal anisotropy of the neutral atmosphere,
J. Geod., 85(10), 661–677, doi:10.1007/s00190-011-0474-y.
Hobiger, T., R. Ichikawa, Y. Koyama, and T. Kondo (2008), Fast and accurate ray-tracing algorithms for real-time space geodetic
applications using numerical weather models, J. Geophys. Res., 113, D20302, doi:10.1029/2008JD010503.
Holton, J. R. (2006), An Introduction to Dynamic Meteorology, Academic, San Diego, Calif.
Ma, C., J. M. Sauber, L. J. Bell, T. A. Clark, D. Gordon, W. E. Himwich, and J. W. Ryan (1990), Measurement of horizontal motions in Alaska
using very long baseline interferometry observations, J. Geophys. Res., 95(B13), 21,991–22,011.
MacMillan, D., and J. Gipson (2009), Recent modeling improvements in SOLVE analysis, paper presented at 19th European VLBI for
Geodesy and Astrometry Working Meeting, 24-25, March 2009, Bourdeaux, pp. 54-57.
MacMillan, D., and C. Ma (1997), Atmospheric gradients and the VLBI terrrestrial and celestial reference frames, Geophys. Res. Lett., 24(4),
453–456.

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9169
21699356, 2014, 12, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2014JB011552 by Nat Prov Indonesia, Wiley Online Library on [06/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1002/2014JB011552

Nafisi, V., et al. (2012a), Comparison of ray-tracing packages for troposphere delays, IEEE Trans. Geosci. Remote Sens., 50(2), 469–481,
doi:10.1109/TGRS.2011.2160952.
Nafisi, V., M. Madzak, J. Boehm, A. A. Ardalan, and H. Schuh (2012b), Ray-traced tropospheric delays in VLBI analysis, Radio Sci., 47,
RS2020, doi:10.1029/2011RS004918.
Niell, A. E. (1996), Global mapping functions for the atmosphere delay at radio wavelengths, J. Geophys. Res., 101(B2), 3227–3246.
Niell, A. E. (2000), Improved atmospheric mapping functions for VLBI and GPS, Earth Planets Space, 52(10), 699–702.
Nothnagel, A. (2009), Conventions on thermal expansion modelling of radio telescopes for geodetic and astrometric VLBI, J. Geod., 83,
787–792, doi:10.1007/s00190-008-0284-z.
Pavlis, N. K., S. A. Holmes, S. C. Kenyon, and J. K. Factor (2012), The development and evaluation of the Earth Gravitational Model 2008
(EGM2008), J. Geophys. Res., 117, B04406, doi:10.1029/2011JB008916.
Petit, G., and B. Luzum (Eds.) (2010), IERS Conventions (2010), IERS Technical Note 36, 179 pp., Verlag des Bundesamts fur Kartographie
und Geodasie, Frankfurt am Main.
Petrov, L, and J.-P. Boy (2004), Study of the atmospheric pressure loading signal in very long baseline interferometry observations,
J. Geophys. Res., 109, B03405, doi:10.1029/2003JB002500.
Rees, D., J. J. Barnett, and K. Labitzke (1990), COSPAR International Reference Atmosphere: 1986. Pt. 2: Middle atmosphere models, Adv.
Space Res., 10(12), 525.
Rueger, J. M. (2002a), Refractive indices of light, infrared and radio waves in the atmosphere, Tech. Rep., UNISURV S-68 School of
Surveying and Spatial Information Systems, Univ. New South Wales, Australia.
Rueger, J. M. (2002b), Refractive index formulae for radio waves, paper presented at FIG XXII Int. Congr., Int. Fed. Surveyors (FIG),
Washington, D. C., 19–26 April.
Saastamoinen, J. (1972), Atmospheric correction for the troposphere and stratosphere in radio ranging satellites, in The Use of Artificial
Satellites for Geodesy, Geod. Geophys. Monogr. Ser., vol. 15, edited by S. W. Henriksen, pp. 247–251, AGU, Washington, D. C.
Wallace, J. M., and P. V. Hobbs (2006), Atmospheric Science: An Introductory Survey, Academic, San Diego, Calif.
Wheelon, A. D. (2001), Electromagnetic Scintillation, I. Geometrical Optics, Cambridge Univ. Press, Cambridge, U. K.

ERIKSSON ET AL. ©2014. American Geophysical Union. All Rights Reserved. 9170

You might also like