Refined discrete and empirical horizontal gradients in VLBI analysis
 484 Downloads
 1 Citations
Abstract
Missing or incorrect consideration of azimuthal asymmetry of troposphere delays is a considerable error source in space geodetic techniques such as Global Navigation Satellite Systems (GNSS) or Very Long Baseline Interferometry (VLBI). Socalled horizontal troposphere gradients are generally utilized for modeling such azimuthal variations and are particularly required for observations at low elevation angles. Apart from estimating the gradients within the data analysis, which has become common practice in space geodetic techniques, there is also the possibility to determine the gradients beforehand from different data sources than the actual observations. Using raytracing through Numerical Weather Models (NWMs), we determined discrete gradient values referred to as GRAD for VLBI observations, based on the standard gradient model by Chen and Herring (J Geophys Res 102(B9):20489–20502, 1997. https://doi.org/10.1029/97JB01739) and also for new, higherorder gradient models. These gradients are produced on the same data basis as the Vienna Mapping Functions 3 (VMF3) (Landskron and Böhm in J Geod, 2017. https://doi.org/10.1007/s0019001710662), so they can also be regarded as the VMF3 gradients as they are fully consistent with each other. From VLBI analyses of the Vienna VLBI and Satellite Software (VieVS), it becomes evident that baseline length repeatabilities (BLRs) are improved on average by 5% when using a priori gradients GRAD instead of estimating the gradients. The reason for this improvement is that the gradient estimation yields poor results for VLBI sessions with a small number of observations, while the GRAD a priori gradients are unaffected from this. We also developed a new empirical gradient model applicable for any time and location on Earth, which is included in the Global Pressure and Temperature 3 (GPT3) model. Although being able to describe only the systematic component of azimuthal asymmetry and no shortterm variations at all, even these empirical a priori gradients slightly reduce (improve) the BLRs with respect to the estimation of gradients. In general, this paper addresses that a priori horizontal gradients are actually more important for VLBI analysis than previously assumed, as particularly the discrete model GRAD as well as the empirical model GPT3 are indeed able to refine and improve the results.
Keywords
VLBI GNSS Troposphere Horizontal gradients1 Introduction
During their passage through the neutral atmosphere, radio waves are delayed and bent as a result of interaction with dry gases and water particles. As there is no chance to directly measure the amount of delay with sufficient accuracy, the delays are usually modeled instead. While the elevation angledependent part of the delay is taken into account by the use of mapping functions, the delay also depends significantly on the azimuth of the observation. The ellipsoidal shape of the troposphere as well as the temporally and spatially varying refractivity of the air cause the delays to vary significantly for different observed azimuth angles. In most cases, this effect is considered through horizontal troposphere gradients multiplied with sine and cosine functions, intended to model symmetric variations over the azimuth range. Consideration of these gradients is particularly important for the realization of celestial reference frames (CRFs) (MacMillan and Ma 1997) and terrestrial reference frames (TRFs) (Böhm and Schuh 2007; Mayer et al. 2017). In the analysis of space geodetic techniques such as Global Navigation Satellite Systems (GNSS) and Very Long Baseline Interferometry (VLBI), it has become common practice to estimate gradients on the basis of a very high number of observations. In GNSS, these gradient values are determined and published for instance by the International GNSS Service (IGS), while in VLBI they are important output quantities of analysis software. However, horizontal gradients can also be determined from sources other than the actual observations. Raytracing through numerical weather models (NWMs) has proven to be well suited for deriving troposphere delays and hence has become the basis for the most accurate mapping functions currently available. In these NWMs, the lower atmosphere is discretized to a temporally varying threedimensional grid, where the raytracing beams then propagate through. Following the Eikonal equation, the raytracing beams are delayed and bent, simulating the real travel path as well as possible. As the NWMs are available globally, raytraced delays can be produced for any point on Earth. The raytracing software developed by Hofmeister and Böhm (2017) as part of the Vienna VLBI and Satellite software (VieVS) (Böhm et al. 2017) can not only be used for the derivation of highly accurate mapping functions [see Landskron and Böhm (2017)], but provides the basis for the determination of horizontal troposphere gradients through 2D raytracing at several azimuth angles, too. Depending on the underlying gradient model, the gradients can be realized at all NWM epochs for any site on Earth.
Yet, only minor importance was attached to a priori gradients in VLBI, underlined by the negligible number of existing models and realizations. The Linear Horizontal Gradients (LHG) by Böhm and Schuh (2007) represent the only existing discrete a priori gradient model for VLBI. Calculated directly from NWMs (without raytracing), these gradients are provided for all VLBI stations at each NWM epoch, intended for a priori use in VLBI analysis. The term "discrete" in this context means that the gradients are determined discretely for certain locations and times, generally from uptodate information from raytracing through NWMs. In contrast, empirical models rely on experience values from climatology instead. Hereof, two models need to be mentioned: the DAO model from the Data Assimilation Office (MacMillan and Ma 1997) which has been determined by vertical integration over horizontal refractivity gradients, as well as the APG model (Böhm et al. 2013), which first applied the technique of raytracing through monthly mean pressure level reanalysis data of the ECMWF. The gradients from these models can then be applied in VLBI analyses as a priori values.
Section 2 first gives a basic understanding of azimuthal asymmetry in troposphere delay modeling. In Sect. 3 the generation of new gradient models is described, whose performance is then assessed in Sect. 4, leading to the conclusions in Sect. 5.
2 Fundamentals of horizontal gradients

The rotation of the Earth and its resulting centrifugal force not only turn the Earth into an ellipsoid, but act on the atmosphere as well. Consequently, the troposphere is thicker at the equator than at the poles by some kilometers. This effect, which is also referred to as the atmospheric bulge, systematically acts on electromagnetic signals traveling through the troposphere, more precisely on the hydrostatic part; the longer a signal’s path, the larger its delay. At the equator, the systematic effect is fairly equal for signals from the north and from the south. At the poles, it is equal for all cardinal directions. Given the site of the VLBI station WETTZELL in southern Germany at a latitude of 49\(^{\circ }\), for instance, signals arriving from the north are less delayed than signals from the south.

Space geodetic techniques as well as raytracing through NWMs usually refer to the reference ellipsoid. However, the real shape of the Earth is much more complex, being referred to as the geoid. Deflections of the vertical (DOV) are the angles between the plumb line and lines perpendicular to the reference ellipsoid at certain locations. In reverse, these DOV can also be visualized as horizontal gradients. They are particularly distinct at plate boundaries or near major mountain ranges.

Higher temperatures lead to higher convection which lifts the tropopause upwards, which is why the thickness of the troposphere is generally lower in cold conditions and higher in warm conditions (Geerts and Linacre 1997). As a consequence, the tropopause over the poles is up to 2 km higher in summer than in winter.

The refractivity along the signal path, which mainly depends on temperature, pressure, humidity, \(CO_2\) composition and density (Jones 1981), is highly variable both temporally and spatially. As a result, signals reaching a station from different cardinal directions experience different delays, which is considered as a random effect.
3 Development of new horizontal gradients
This section presents the determination of new north gradients \(G_\mathrm{n}\) and east gradients \(G_\mathrm{e}\) for the gradient formula by Chen and Herring (1997) as well as for new, higherorder gradient formulae. The main goals for the new gradients are to outperform existing models in VLBI analysis, as well as to improve the baseline length repeatability (BLR) of VLBI analysis w.r.t. estimating the gradients. The basis for the determination are raytraced delays from the VieVS raytracer applying the 2D piecewise linear approach (Hobiger et al. 2008). Unlike 1D raytracing, in the 2D approach lateral changes in refractivity are also considered.
3.1 Determination of discrete horizontal gradients for VLBI
Properties of the raytraced delays that were generated using the VieVS raytracer from 1999 to 2014
Parameter  Specification 

Raytracing software  VieVS raytracer (Hofmeister and Böhm 2017) 
Raytracing method  2D piecewise linear (Hobiger et al. 2008) 
NWM  ECMWF ERAInterim pressure level data + ECMWF operational data 
Horizontal resolution of the NWM  \(1^{\circ }\times 1^{\circ }\) 
Vertical coverage  25 pressure levels 
Horizontal coverage  33 VLBI stations 
Temporal resolution  6hourly at 00:00, 06:00, 12:00 and 18:00 UTC each day from 1999 through 2014 (\(=\) 23,376 epochs) 
Outgoing elevation angles per point  7 (\(3^{\circ }\), \(5^{\circ }\), \(7^{\circ }\), \(10^{\circ }\), \(15^{\circ }\), \(30^{\circ }\) and \(70^{\circ }\)) 
Azimuth angles per point  16 (\(0^{\circ }\):22.5\(^{\circ }\):337.5\(^{\circ }\)) 
3.2 Determination of an empirical gradient grid
Apart from the discrete horizontal gradients GRAD, there is also a new empirical gradient grid determined as part of the Global Pressure and Temperature 3 (GPT3) model, providing empirical values for \(G_{\mathrm{n_h}}\), \(G_{\mathrm{e_h}}\), \(G_{\mathrm{n_w}}\) and \(G_{\mathrm{e_w}}\). Empirical gradient models are needed particularly for observations in the early years of VLBI up to about 1990 (Spicakova et al. 2011), when only few stations were observing a small number of sources, resulting in a nonuniform sky coverage that limits the ability of estimating the gradients in a leastsquares adjustment (Heinkelmann and Tesmer 2013). However, empirical gradients may also be important for recent data, for instance for the purpose of deriving terrestrial reference frames (TRFs) from VLBI or for high latitude sites in general where the effect of the atmospheric bulge is most distinct (Böhm et al. 2011).
Currently, only the empirical gradient models APG and DAO are of importance. APG is globally applicable based on a spherical harmonics expansion up to degree and order nine, whereas DAO is only available for a selected list of 174 VLBI stations (as of 2016/05), with new ones being added regularly. Both models provide only total gradients and no separated hydrostatic and wet parts. For VLBI analysis, Böhm et al. (2011) recommend using DAO rather than APG.
Properties of the gridwise raytraced delays that were generated for the derivation of the empirical gradient grids GPT3 (\(1^{\circ }\times 1^{\circ }\)) and GPT3 (\(5^{\circ }\times 5^{\circ }\))
Parameter  Specification 

Raytracing software  VieVS raytracer (Hofmeister and Böhm 2017) 
Raytracing method  2D piecewise linear (Hobiger et al. 2008) 
NWM  ECMWF ERAInterim pressure level data 
Horizontal resolution of the NWM  \(1^{\circ }\times 1^{\circ }\) 
Horizontal coverage  (1) global grid with resolution \(5^{\circ }\times 5^{\circ }\) (lat: [\(87.5^{\circ }\), \(87.5^{\circ }\)], lon: [\(2.5^{\circ }\), \(357.5^{\circ }\)]), resulting in 2592 grid points and (2) global grid with resolution \(1^{\circ }\times 1^{\circ }\) (lat: [\(89.5^{\circ }\), \(89.5^{\circ }\)], lon: [\(0.5^{\circ }\), \(359.5^{\circ }\)]) resulting in 64,800 grid points 
Vertical coverage  25 pressure levels 
Temporal resolution  Mean values for every month from 2001 through 2010 (\(=\) 120 epochs) 
Outgoing elevation angles per point  4 (\(3.3^{\circ }\), \(5^{\circ }\), \(15^{\circ }\) and \(30^{\circ }\)) for \(5^{\circ }\times 5^{\circ }\) grid and 1 elevation (\(3^{\circ }\)) for \(1^{\circ }\times 1^{\circ }\) grid 
Azimuth angles per point  8 (\(0^{\circ }:45^{\circ }:315^{\circ }\)) 
Empirical gradients, however, only have the ability to describe a small, apparently insignificant part of the actual, discrete gradients, which is outlined in Fig. 5. Unlike DAO, the GPT3 gradients possess a small timedependent component, although there is no chance to sufficiently describe the significant random, shortterm variations due to weather events dominating the behavior of the discrete gradients.
4 Comparisons and results
Setting for the VLBI analysis using VieVS
Option  Decision 

Mapping function  Vienna Mapping Functions 1 (VMF1) (Böhm et al. 2006) 
Terrestrial reference frame  VieVS TRF (Böhm et al. 2017) 
Celestial reference frame  International Celestial Reference Frame 2 (ICRF2) 
Tidal ocean loading  FES2004 (Lyard et al. 2006) 
Tidal and nontidal atmosphere loading  VIENNA (Wijaya et al. 2013) 
Estimate \(\varDelta L_w^z\) within the analysis  Yes; as piecewise linear offsets hourly using relative constraints of 1.5 cm 
Estimate gradients within the analysis  if desired; as piecewise linear offsets 6hourly using relative constraints of 0.5 mm, but no absolute constraints 
Mean BLRs (cm) from VLBI analyses for all 1338 sessions from 2006 to 2014. In column (1), only a priori gradients are used, while in column (2) the gradients are additionally estimated in the VLBI analysis using the standard gradient formula
Gradient model  (1)  (2) 

(a) No a priori gradients  1.68  1.65 
(b) LHG  1.66  1.67 
(c) GRAD1  1.58  1.66 
(d) GRAD2  1.57  1.65 
(e) GRAD3  1.58  1.65 
(f) APG  1.65  1.66 
(g) DAO  1.64  1.66 
(h) GPT3 (\(5^{\circ }\times 5^{\circ }\))  1.63  1.66 
(i) GPT3 (\(1^{\circ }\times 1^{\circ }\))  1.63  1.66 
Raytraced delays  1.57  1.64 
4.1 Comparison of BLRs
Baseline length repeatabilities are an appropriate measure to assess the quality of geodetic VLBI products (Böhm and Schuh 2004; Titov 2009). The lower the BLR, the better the performance of a certain model. Table 4 shows the resulting BLRs from VLBI analyses of several models, as averaged over 1338 VLBI sessions from 2006 to 2014. The raytraced delays, which serve as the basis for the determination of the GRAD a priori gradients, were computed following the specifications in Table 1. They were then interpolated to the VLBI observation epochs through spline interpolation. The settings for the VLBI analyses are listed in Table 3. The results of Table 4 are surprising because the estimation^{1} of gradients in the VLBI analysis degrades the resulting BLRs. Best results are achieved when using a priori gradients without estimation of the gradients. GRAD2 yields the best performance, improving the BLRs of 43% of the stations by more than 1 mm while degrading only 5% of the stations by more than 1 mm (the complementary 52% are between − 1 and + 1 mm, too small to be referred to as an improvement or degradation). Figure 6 outlines this more closely, assuming no gradients were estimated in the VLBI analysis. Figure 7 shows that the improvement from the a priori gradients is most distinct for shorter baselines. This is most likely because horizontal gradients affect horizontal positions in particular. Since baselines run straight through the Earth, their repeatability is less affected by horizontal position changes with increasing baseline length.
Mean BLRs (cm) from VLBI analyses for all those sessions from 2006 to 2014 that contain fewer than 3000 observations (1129 out of 1338 sessions, columns 2 and 3) and more than 3000 observations (209 out of 1338 sessions, columns 4 and 5). In (1), only a priori gradients are used, while in (2) the gradients are additionally estimated in the VLBI analysis
Gradient model  < 3000 observations  > 3000 observations  

(1)  (2)  (1)  (2)  
(a) No a priori gradients  2.32  2.38  1.08  0.97 
(b) LHG  2.32  2.36  1.06  1.04 
(c) GRAD1  2.21  2.39  1.00  0.99 
(d) GRAD2  2.19  2.37  1.01  0.98 
(e) GRAD3  2.19  2.38  1.01  0.99 
(f) APG  2.25  2.39  1.09  0.97 
(g) DAO  2.23  2.39  1.09  0.97 
(h) GPT3 (\(5^{\circ }\times 5^{\circ }\))  2.23  2.39  1.08  0.97 
(i) GPT3 (\(1^{\circ }\times 1^{\circ }\))  2.23  2.39  1.07  0.97 
Raytraced delays  2.18  2.33  1.01  0.99 

The raytraced delays, which represent the absolute reference values in this comparison, can be approximated perfectly well by using VMF1 plus the gradients GRAD1 as well as GRAD2. In other words, this means that better BLRs can only be attained as soon as the raytraced delays themselves become more accurate.

Unlike the commonly accepted opinion, gradients shall not always be estimated within VLBI analysis. The design matrix in the leastsquares adjustment must be sufficiently overdetermined in order to produce reliable results. A certain criterion has to be fulfilled to ensure this, where the minimum value of 3000 observations per session turned out to be an approximate, but reliable boundary. Below this number, no gradients shall be estimated in VLBI analysis.

Best results are achieved with the a priori gradients GRAD2. However, GRAD1 is only marginally worse but does not require a new gradient formula.

Empirical a priori gradients generally have a considerably smaller effect on the resulting BLRs than discrete a priori gradients. In case no discrete a priori gradients are available, empirical gradients are most useful for VLBI sessions with few observations, where its usage yields much better BLRs than estimating the gradients in the analysis. GPT3 is marginally better that APG and DAO, whereas the difference between GPT3 (\(5^{\circ }\times 5^{\circ }\)) and GPT3 (\(1^{\circ }\times 1^{\circ }\)) is even more marginal.

The topography has a significant influence on the resulting gradients, e.g., the presence of mountain ranges causes variant gradient values. For this reason, the provision on a grid with a coarse mesh size of \(5^{\circ }\) seems to be insufficient, as the grid points are up to 550 km away from each other that makes it impossible to consider any topography in between. The mesh size of \(1^{\circ }\) comes closer to reality in theory; however, the results are only slightly better. Probably, the provision of new empirical gradients for individual sites would yield better results for VLBI purposes than on a global grid. The provision on a global grid, however, allows GPT3 to be used for many more purposes than VLBI.

In general, GRAD provides better BLRs than the LHG from Böhm and Schuh (2007).

When deciding to estimate gradients, the use of a priori gradients only slightly affects resulting BLRs.
4.2 Comparison of modeled delays with raytraced delays
Unlike the comparison in section 4.1 where gradients were determined for VLBI stations located at discrete spots on Earth, in this section a comparison is done for a \(5^{\circ }\times 5^{\circ }\) global grid containing 2592 grid points. Raytraced delays were generated for each grid point according to the specifications listed in Table 2. The raytraced delays, regarded as the "true" values, are then compared to delays modeled with the three gradient formulae Eqs. (2), (5) and (6). The gradients LHG as well as DAO cannot be considered here, as they are only available for VLBI station locations and not for arbitrary points such as grid intersections.
Mean absolute residuals (mm) in slant total delay between raytracing and applying no gradient formula, the three GRAD gradients and empirical gradients, each for 5\(^{\circ }\) elevation and different azimuths \(\alpha \), averaged over all 2592 grid points and 120 epochs from January 2001 to December 2010
Gradient model  Mean abs. diff. in \(\varDelta L\) (cm)  

\(\alpha = 0^{\circ }\)  \(\alpha = 45^{\circ }\)  \(\alpha = 90^{\circ }\)  \(\alpha = 135^{\circ }\)  \(\alpha = 180^{\circ }\)  Mean \(\alpha \)  
No a priori gradients  25.6  19.6  9.7  19.0  26.0  20.0 
GRAD1  4.1  1.1  4.1  1.1  4.2  2.9 
GRAD2  1.4  0.8  1.1  0.8  1.3  1.1 
GRAD3  1.4  0.8  1.1  0.8  1.3  1.1 
APG  16.4  14.4  10.8  13.0  16.8  14.3 
GPT3  9.4  7.5  7.4  7.5  9.5  8.3 

Due to the presence of an atmospheric bulge, azimuthal asymmetry is most pronounced in north and south direction and is least pronounced in east and west direction.

The consideration of azimuthal asymmetry is of particular importance especially for small elevation angles like \(5^{\circ }\).

With the standard gradient formula of Chen and Herring (1997) (\(=\) GRAD1), an improvement in the slant total delays of up to 20 mm can be reached at 5\(^{\circ }\) elevation. On average, it improves the residuals by 86%.

Using the second gradient formula (\(=\) GRAD2) further improves the slant total delays, although to a smaller degree. On average, the residuals are lower by notable 95% compared to not considering azimuthal asymmetry.

The third gradient formula (\(=\) GRAD3) is not meaningful as it is not capable of further reducing the residuals compared to GRAD2. This is most likely owing to insufficient overdetermination in the leastsquares adjustment, where six gradient variables shall be estimated from eight azimuths.

The residuals when using empirical gradients are far off those from discrete gradients. However, GPT3 considerably improves the delays with respect to APG.
5 Conclusions
On the basis of raytraced delays through numerical weather models (NWMs) using the highly sophisticated VieVS raytracer (Hofmeister and Böhm 2017), we developed new discrete horizontal gradients for a priori use in VLBI analysis referred to as GRAD, as well as a new empirical gradient model GPT3 in the two grid sizes \(1^{\circ }\times 1^{\circ }\) and \(5^{\circ }\times 5^{\circ }\). All of these models are capable of outperforming existing models in our comparisons; this is shown through baseline length repeatabilities (BLRs) from VLBI analyses as well as theoretical delays. An extended gradient formula including higherorder terms (GRAD2) is able to simulate the raytraced delays with even higher precision than the standard gradient formula by Chen and Herring (1997). We found that the common estimation of gradients in VLBI analysis shall only be carried out under certain conditions. If the respective VLBI session exhibits fewer than 3000 observations, the gradient estimation rather degrades than improves the results. The sole usage of a priori gradients GRAD without additional gradient estimation is to be preferable in 90% of the VLBI sessions. However, as in general only a comparably small improvement can be achieved with the new models, we are forced to the conclusion that a big leap in the accuracy may only be achieved when the raytraced delays and NWMs themselves become more accurate. This is supported by the fact that the raytraced delays can be approximated already very well through the modeled gradients in all comparisons.
6 Data and code availability
Text files containing GRAD gradients can be downloaded from http://ggosatm.hg.tuwien.ac.at/DELAY/ETC/GRAD/. Information on the usage of the files is found in http://ggosatm.hg.tuwien.ac.at/DELAY/readme.txt.
Footnotes
 1.
 2.
There were several strategies tested for a boundary value regarding total number of observations, mean number of observations per station, number of scans and some more. However, the most general criterion (total number of observations) turned out to be the most appropriate one at the same time. See Landskron (2017) for details.
Notes
Acknowledgements
Open access funding provided by Austrian Science Fund (FWF). The authors would like to thank the Austrian Science Fund (FWF) for financial support of this study within the projects RADIATE VLBI (P25320) and RADIATE ORD (ORD 86).
References
 Böhm J, Schuh H (2004) Vienna mapping functions in VLBI analyses. Geophys Res Lett 31:L01603. https://doi.org/10.1029/2003GL018984 CrossRefGoogle Scholar
 Böhm J, Schuh H (2007) Troposphere gradients from the ECMWF in VLBI analysis. J Geod 81:403–408. https://doi.org/10.1007/s0019000701442 CrossRefGoogle Scholar
 Böhm J, Werl B, Schuh H (2006) Troposphere mapping functions for GPS and VLBI from European Centre for MediumRange Weather Forecasts operational analysis data. J Geophys Res 111:B02406. https://doi.org/10.1029/2005JB003629 CrossRefGoogle Scholar
 Böhm J, Spicakova H, Urquhart L, Steigenberger P, Schuh H (2011) Impact of a priori gradients on VLBIderived terrestrial reference frames. In: Alef W, Bernhart S, Nothnagel A (eds) Proceedings of the 20th meeting of the European VLBI group for geodesy and astrometry. Schriftenreihe des Instituts für Geodäsie und Geoinformation der Universität Bonn, No. 22, pp 128–132, ISSN 18641113Google Scholar
 Böhm J, Urquhart L, Steigenberger P, Heinkelmann R, Nafisi V, Schuh H (2013) A priori gradients in the analysis of space geodetic observations. In: Collileux X, Altamimi Z (eds) Proceedings of the IAG commission 1 symposium 2010 (REFAG2010), Paris, France, 4–8 October 2010, International Association of Geodesy Symposia SeriesGoogle Scholar
 Böhm J, Möller G, Schindelegger M, Pain G, Weber R (2015) Development of an improved blind model for slant delays in the troposphere (GPT2w). GPS Solut 19:433. https://doi.org/10.1007/s1029101404037 CrossRefGoogle Scholar
 Böhm J, Böhm S, Boisits J, Girdiuk A, Gruber J, Hellerschmied A, Krasna H, Landskron D, Madzak M, Mayer D, McCallum J, McCallum L, Schartner M, Teke K (2017) Vienna VLBI and Satellite Software (VieVS) for Geodesy and Astrometry. Publications of the Astronomical Society of the Pacific, Accepted 2018Google Scholar
 Chen G, Herring TA (1997) Effects of atmospheric azimuthal asymmetry on the analysis of space geodetic data. J Geophys Res 102(B9):20489–20502. https://doi.org/10.1029/97JB01739 CrossRefGoogle Scholar
 Gardner CS (1977) Effects of horizontal refractivity gradients on the accuracy of laser ranging to satellites. Radio Sci 11(12):1037–1044. https://doi.org/10.1029/RS011i012p01037 CrossRefGoogle Scholar
 Geerts B, Linacre E (1997) The height of the tropopause. University of Wyoming, Atmospheric Science, LaramieGoogle Scholar
 Heinkelmann R, Tesmer V (2013) Systematic inconsistencies between VLBI CRF and TRF solutions caused by different analysis options. In: IAG Symposium, vol 138, Springer, Berlin. https://doi.org/10.1007/9783642329982_27 Google Scholar
 Herring TA (1992) Modeling atmospheric delays in the analysis of space geodetic data. In: Publications on Geodesy. In: DeMunck JC, Th. Spoelstra TA (eds) Proceedings of refraction of transatmospheric signals in Geodesy, vol 36. Netherlands Geodetic Commission Publications in Geodesy, The Hague, pp 157–164Google Scholar
 Hobiger T, Ichikawa R, Koyama Y, Kondo T (2008) Fast and accurate raytracing algorithms for realtime space geodetic applications using numerical weather models. J Geophys Res 113:D20302. https://doi.org/10.1029/2008JD010503 CrossRefGoogle Scholar
 Hofmeister A, Böhm J (2017) Application of raytraced tropospheric slant delays to geodetic VLBI analysis. J Geod 91:945. https://doi.org/10.1007/s0019001710007 CrossRefGoogle Scholar
 Jones FE (1981) The refractivity of Air. J Res Nat Bur Stand 86(1):27CrossRefGoogle Scholar
 Lagler K, Schindelegger M, Böhm J, Krasna H, Nilsson T (2013) GPT2: empirical slant delay model for radio space geodetic techniques. Geophys Res Lett 40(6):1069–1073. https://doi.org/10.1002/grl.50288 CrossRefGoogle Scholar
 Landskron D (2017) Modeling tropospheric delays for space geodetic techniques. Dissertation, Department of Geodesy and Geoinformation, TU Wien, Supervisor: J. Böhm. http://repositum.tuwien.ac.at/urn:nbn:at:atubtuw:1100249
 Landskron D, Böhm J (2017) VMF3/GPT3: refined discrete and empirical troposphere mapping functions. J Geod. https://doi.org/10.1007/s0019001710662 CrossRefGoogle Scholar
 Landskron D, Möller G, Hofmeister A, Böhm J, Weber R (2015a), Refined and siteaugmented tropospheric delay models in the analysis of VLBI observations. In: 26th IUGG general assembly, Prague, 2015/06/222007/07/02Google Scholar
 Landskron D, Hofmeister A, Böhm J (2015b), Refined tropospheric delay models for CONT11. In: International association of Geodesy symposia. Springer, Berlin, pp 1–5. https://doi.org/10.1007/1345_2015_56 CrossRefGoogle Scholar
 Landskron D, Hofmeister A, Mayer D, Böhm J (2016), Six hourly time series of horizontal troposphere gradients in VLBI analysis. In: European geosciences union general assembly 2016, Vienna, 2016/04/172016/04/22Google Scholar
 Lyard F, Lefevre F, Letellier T, Francis O (2006) Modelling the global ocean tides: modern insights from FES2004. Ocean Dyn 56:394–415CrossRefGoogle Scholar
 MacMillan DS, Ma C (1997) Atmospheric gradients and the VLBI terrestrial and celestial reference frames. Geophys Res Lett 24(4):453–456. https://doi.org/10.1029/97GL00143 CrossRefGoogle Scholar
 Mayer D, Böhm J, Krasna H, Landskron D (2017) Tropospheric delay modelling and the celestial reference frame at radio wavelengths. Astron Astrophys. https://doi.org/10.1051/00046361/201731681 CrossRefGoogle Scholar
 Nilsson T, Böhm J, Wijaya DD, Tresch A, Nafisi V, Schuh H (2013) Path delays in the neutral atmosphere. In: Böhm J, Schuh H (eds) Atmospheric Effects in Space Geodesy. Springer, Berlin ISBN: 9783642369315Google Scholar
 Spicakova H, Plank L, Nilsson T, Böhm J, Schuh H (2011) Terrestrial reference frame solution with the Vienna VLBI Software VieVS and implication of tropospheric gradient estimation. In: Proceedings of the 20th meeting of the European VLBI group for Geodesy and sstrometry (EVGA)Google Scholar
 Titov O (2009) A new estimator for VLBI baseline length repeatability. J Geod 83:1041–1049. https://doi.org/10.1007/s0019000903225 CrossRefGoogle Scholar
 Wijaya DD, Böhm J, Karbon M, Krasna H, Schuh H (2013) Atmospheric pressure loading. Chapter 4 in atmospheric effects in space geodesy. In: Böhm J, Schuh H (eds) Springer, Berlin. ISBN: 9783642369322 (online) Google Scholar
Copyright information
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.