Numerical simulation of troposphere-induced errors in GPS-derived geodetic time series over Japan

Troposphere-induced errors in GPS-derived geodetic time series, namely, height and zenith total delays (ZTDs), over Japan are quantitatively evaluated through the analyses of simulated GPS data using realistic cumulative tropospheric delays and observed GPS data. The numerical simulations show that the use of a priori zenith hydrostatic delays (ZHDs) derived from the European Centre for Medium-Range Weather Forecasts (ECMWF) numerical weather model data and gridded Vienna mapping function 1 (gridded VMF1) results in smaller spurious annual height errors and height repeatabilities (0.45 and 2.55 mm on average, respectively) as compared to those derived from the global pressure and temperature (GPT) model and global mapping function (GMF) (1.08 and 3.22 mm on average, respectively). On the other hand, the use of a priori ZHDs derived from the GPT and GMF would be sufficient for applications involving ZTDs, given the current discrepancies between GPS-derived ZTDs and those derived from numerical weather models. The numerical simulations reveal that the use of mapping functions constructed with fine-scale numerical weather models will potentially improve height repeatabilities as compared to the gridded VMF1 (2.09 mm against 2.55 mm on average). However, they do not presently outperform the gridded VMF1 with the observed GPS data (6.52 mm against 6.50 mm on average). Finally, the commonly observed colored components in GPS-derived height time series are not primarily the result of troposphere-induced errors, since they become white in numerical simulations with the proper choice of a priori ZHDs and mapping functions.     

Assessment of ECMWF-derived tropospheric delay models within the EUREF Permanent Network

The Global Positioning System (GPS) observations from the EUREF Permanent Network (EPN) are routinely analyzed by the EPN analysis centers using a tropospheric delay modeling based on standard pressure values, the Niell Mapping Functions (NMF), a cutoff angle of 3° and down-weighting of low elevation observations. We investigate the impact on EPN station heights and Zenith Total Delay (ZTD) estimates when changing to improved models recommended in the updated 2003 International Earth Rotation and Reference Systems Service (IERS) Conventions, which are the Vienna Mapping Functions 1 (VMF1) and zenith hydrostatic delays derived from numerical weather models, or the empirical Global Mapping Functions (GMF) and the empirical Global Pressure and Temperature (GPT) model. A 1-year Global Positioning System (GPS) data set of 50 regionally distributed EPN/IGS (International GNSS Service) stations is processed. The GPS analysis with cutoff elevation angles of 3, 5, and 10° revealed that changing to the new recommended models introduces biases in station heights in the northern part of Europe by 2–3 mm if the cutoff is lower than 5°. However, since large weather changes at synoptic time scales are not accounted for in the empirical models, repeatability of height and ZTD time series are improved with the use of a priori Zenith Hydrostatic Delays (ZHDs) derived from numerical weather models and VMF1. With a cutoff angle of 3°, the repeatability of station heights in the northern part of Europe is improved by 3–4 mm.     

Assessment of troposphere mapping functions using three-dimensional ray-tracing

The troposphere delay is an important source of error for precise GNSS positioning due to its high correlation with the station height parameter. It has been demonstrated that errors in mapping functions can cause sub-annual biases as well as affect the repeatability of GNSS solutions, which is a particular concern for geophysical studies. Three-dimensional ray-tracing through numerical weather models (NWM) is an excellent approach for capturing the directional and daily variation of the tropospheric delay. Due to computational complexity, its use for positioning purposes is limited, but it is an excellent tool for evaluating current state-of-the-art mapping functions used for geodetic positioning. Many mapping functions have been recommended in the past such as the Niell Mapping Function (NMF), Vienna Mapping Function 1 (VMF1), and the Global Mapping Function (GMF), which have been adopted by most IGS analysis centers. A new Global Pressure Temperature model (GPT2) has also been developed, which has been shown to improve upon the original atmospheric model used for the GMF. Although the mapping functions mentioned above use the same functional formulation, they vary in terms of their atmospheric source and calibration approach. A homogeneous data set of three-dimensional ray-traced delays is used to evaluate all components of the mapping functions, including their underlying functional formulation, calibration, and compression method. Additionally, an alternative representation of the VMF1 is generated using the same atmospheric source as the truth data set to evaluate the differences in ray-tracing methods and their effect on the end mapping function. The results of this investigation continue to support the use of the VMF1 as the mapping function of choice when geodetic parameters are of interest. Further support for the GPT2 and GMF as reliable back-ups when the VMF1 is not available was found due to their high consistency with the NWM-derived mapping function. Additionally, a small latitude-dependent bias in station height was found in the current mapping functions. This bias was identified to be due to the assumption of a constant radius of the earth and was largest at the poles and at the equator. Finally, an alternative version of the VMF1 is introduced, namely the UNB-VMF1 which provides users with an independent NWM-derived mapping function to support geodetic positioning.     

Testing of global pressure/temperature (GPT) model and global mapping function (GMF) in GPS analyses

Several sources of a priori meteorological data have been compared for their effects on geodetic results from GPS precise point positioning (PPP). The new global pressure and temperature model (GPT), available at the IERS Conventions web site, provides pressure values that have been used to compute a priori hydrostatic (dry) zenith path delay z _h estimates. Both the GPT-derived and a simple height-dependent a priori constant z _h performed well for low- and mid-latitude stations. However, due to the actual variations not accounted for by the seasonal GPT model pressure values or the a priori constant z _h, GPS height solution errors can sometimes exceed 10 mm, particularly in Polar Regions or with elevation cutoff angles less than 10 degrees. Such height errors are nearly perfectly correlated with local pressure variations so that for most stations they partly (and for solutions with 5-degree elevation angle cutoff almost fully) compensate for the atmospheric loading displacements. Consequently, unlike PPP solutions utilizing a numerical weather model (NWM) or locally measured pressure data for a priori z _h, the GPT-based PPP height repeatabilities are better for most stations before rather than after correcting for atmospheric loading. At 5 of the 11 studied stations, for which measured local meteorological data were available, the PPP height errors caused by a priori z _h interpolated from gridded Vienna Mapping Function-1 (VMF1) data (from a NWM) were less than 0.5 mm. Height errors due to the global mapping function (GMF) are even larger than those caused by the GPT a priori pressure errors. The GMF height errors are mainly due to the hydrostatic mapping and for the solutions with 10-degree elevation cutoff they are about 50% larger than the GPT a priori errors.     

Forecast Vienna Mapping Functions 1 for real-time analysis of space geodetic observations

The Vienna Mapping Functions 1 (VMF1) as provided by the Institute of Geodesy and Geophysics (IGG) at the Vienna University of Technology are the most accurate mapping functions for the troposphere delays that are available globally and for the entire history of space geodetic observations. So far, the VMF1 coefficients have been released with a time delay of almost two days; however, many scientific applications require their availability in near real-time, e.g. the Ultra Rapid solutions of the International GNSS Service (IGS) or the analysis of the Intensive sessions of the International VLBI Service (IVS). Here we present coefficients of the VMF1 as well as the hydrostatic and wet zenith delays that have been determined from forecasting data of the European Centre for Medium-Range Weather Forecasts (ECMWF) and provided on global grids. The comparison with parameters derived from ECMWF analysis data shows that the agreement is at the 1 mm level in terms of station height, and that the differences are larger for the wet mapping functions than for the hydrostatic mapping functions and the hydrostatic zenith delays. These new products (VMF1-FC and hydrostatic zenith delays from forecast data) can be used in real-time analysis of geodetic data without significant loss of accuracy.     

Effect of different tropospheric mapping functions on the TRF, CRF and position time-series estimated from VLBI

This paper compares estimated terrestrial reference frames (TRF) and celestial reference frames (CRF) as well as position time-series in terms of systematic differences, scale, annual signals and station position repeatabilities using four different tropospheric mapping functions (MF): The NMF (Niell Mapping Function) and the recently developed GMF (Global Mapping Function) consist of easy-to-handle stand-alone formulae, whereas the IMF (Isobaric Mapping Function) and the VMF1 (Vienna Mapping Function 1) are determined from numerical weather models. All computations were performed at the Deutsches Geodätisches Forschungsinstitut (DGFI) using the OCCAM 6.1 and DOGS-CS software packages for Very Long Baseline Interferometry (VLBI) data from 1984 until 2005. While it turned out that CRF estimates only slightly depend on the MF used, showing small systematic effects up to 0.025 mas, some station heights of the computed TRF change by up to 13 mm. The best agreement was achieved for the VMF1 and GMF results concerning the TRFs, and for the VMF1 and IMF results concerning scale variations and position time-series. The amplitudes of the annual periodical signals in the time-series of estimated heights differ by up to 5 mm. The best precision in terms of station height repeatability is found for the VMF1, which is 5–7% better than for the other MFs.     

Implementation and testing of the gridded Vienna Mapping Function 1 (VMF1)

The new gridded Vienna Mapping Function (VMF1) was implemented and compared to the well-established site-dependent VMF1, directly and by using precise point positioning (PPP) with International GNSS Service (IGS) Final orbits/clocks for a 1.5-year GPS data set of 11 globally distributed IGS stations. The gridded VMF1 data can be interpolated for any location and for any time after 1994, whereas the site-dependent VMF1 data are only available at selected IGS stations and only after 2004. Both gridded and site-dependent VMF1 PPP solutions agree within 1 and 2 mm for the horizontal and vertical position components, respectively, provided that respective VMF1 hydrostatic zenith path delays (ZPD) are used for hydrostatic ZPD mapping to slant delays. The total ZPD of the gridded and site-dependent VMF1 data agree with PPP ZPD solutions with RMS of 1.5 and 1.8 cm, respectively. Such precise total ZPDs could provide useful initial a priori ZPD estimates for kinematic PPP and regional static GPS solutions. The hydrostatic ZPDs of the gridded VMF1 compare with the site-dependent VMF1 ZPDs with RMS of 0.3 cm, subject to some biases and discontinuities of up to 4 cm, which are likely due to different strategies used in the generation of the site-dependent VMF1 data. The precision of gridded hydrostatic ZPD should be sufficient for accurate a priori hydrostatic ZPD mapping in all precise GPS and very long baseline interferometry (VLBI) solutions. Conversely, precise and globally distributed geodetic solutions of total ZPDs, which need to be linked to VLBI to control biases and stability, should also provide a consistent and stable reference frame for long-term and state-of-the-art numerical weather modeling.     

Short Note: A global model of pressure and temperature for geodetic applications

The empirical model GPT (Global Pressure and Temperature), which is based on spherical harmonics up to degree and order nine, provides pressure and temperature at any site in the vicinity of the Earth’s surface. It can be used for geodetic applications such as the determination of a priori hydrostatic zenith delays, reference pressure values for atmospheric loading, or thermal deformation of Very Long Baseline Interferometry (VLBI) radio telescopes. Input parameters of GPT are the station coordinates and the day of the year, thus also allowing one to model the annual variations of the parameters. As an improvement compared with previous models, it reproduces the large pressure anomaly over Antarctica, which can cause station height errors in the analysis of space-geodetic data of up to 1 cm if not considered properly in troposphere modelling. First tests at selected geodetic observing stations show that the pressure biases considerably decrease when using GPT instead of the very simple approaches applied to various Global Navigation Satellite Systems (GNSS) software packages so far. GPT also provides an appropriate model for the annual variability of global temperature. Electronic supplementary material The online version of this article (doi: contains supplementary material, which is available to authorized users.     

VMF3/GPT3: refined discrete and empirical troposphere mapping functions

Incorrect modeling of troposphere delays is one of the major error sources for space geodetic techniques such as Global Navigation Satellite Systems (GNSS) or Very Long Baseline Interferometry (VLBI). Over the years, many approaches have been devised which aim at mapping the delay of radio waves from zenith direction down to the observed elevation angle, so-called mapping functions. This paper contains a new approach intended to refine the currently most important discrete mapping function, the Vienna Mapping Functions 1 (VMF1), which is successively referred to as Vienna Mapping Functions 3 (VMF3). It is designed in such a way as to eliminate shortcomings in the empirical coefficients b and c and in the tuning for the specific elevation angle of \(3^{\circ }\). Ray-traced delays of the ray-tracer RADIATE serve as the basis for the calculation of new mapping function coefficients. Comparisons of modeled slant delays demonstrate the ability of VMF3 to approximate the underlying ray-traced delays more accurately than VMF1 does, in particular at low elevation angles. In other words, when requiring highest precision, VMF3 is to be preferable to VMF1. Aside from revising the discrete form of mapping functions, we also present a new empirical model named Global Pressure and Temperature 3 (GPT3) on a \(5^{\circ }\times 5^{\circ }\) as well as a \(1^{\circ }\times 1^{\circ }\) global grid, which is generally based on the same data. Its main components are hydrostatic and wet empirical mapping function coefficients derived from special averaging techniques of the respective (discrete) VMF3 data. In addition, GPT3 also contains a set of meteorological quantities which are adopted as they stand from their predecessor, Global Pressure and Temperature 2 wet. Thus, GPT3 represents a very comprehensive troposphere model which can be used for a series of geodetic as well as meteorological and climatological purposes and is fully consistent with VMF3.     

Combination of long time-series of troposphere zenith delays observed by VLBI

Within the International Very Long Baseline Interferometry (VLBI) Service for Geodesy and Astrometry (IVS), long time-series of zenith wet and total troposphere delays have been combined at the level of parameter estimates. The data sets were submitted by eight IVS Analysis Centers (ACs) and cover January 1984 to December 2004. In this paper, the combination method is presented and the time-series submitted by the eight IVS ACs are compared with each other. The combined zenith delays are compared with time-series provided by the International Global Navigation Satellite System (GNSS) Service (IGS), and with zenith delays derived from the European Centre for Medium-Range Weather Forecasts (ECMWF). Before the combination, outliers are eliminated from the individual time-series using the robust BIBER (bounded influence by standardized residuals) estimator. For each station and AC, relative weight factors are obtained by variance component estimation. The mean bias of the IVS ACs’ time-series with respect to the IVS combined time-series is 0.89 mm and the mean root mean square is 7.67 mm. Small differences between stations and ACs can be found, which are due to the inhomogeneous analysis options, different parameterizations, and different treatment of missing in-situ pressure records. Compared to the IGS zenith total delays, the combined IVS series show small positive mean biases and different long-term trends. Zenith wet delays from the ECMWF are used to validate the IVS combined series. Inconsistencies, e.g., long-term inhomogeneity of the in-situ pressure data used for the determination of VLBI zenith delays, are identified.     

Troposphere gradients from the ECMWF in VLBI analysis

Modeling path delays in the neutral atmosphere for the analysis of Very Long Baseline Interferometry (VLBI) observations has been improved significantly in recent years by the use of elevation-dependent mapping functions based on data from numerical weather models. In this paper, we present a fast way of extracting both, hydrostatic and wet, linear horizontal gradients for the troposphere from data of the European Centre for Medium-range Weather Forecasts (ECMWF) model, as it is realized at the Vienna University of Technology on a routine basis for all stations of the International GNSS (Global Navigation Satellite Systems) Service (IGS) and International VLBI Service for Geodesy and Astrometry (IVS) stations. This approach only uses information about the refractivity gradients at the site vertical, but no information from the line-of-sight. VLBI analysis of the CONT02 and CONT05 campaigns, as well as all IVS-R1 and IVS-R4 sessions in the first half of 2006, shows that fixing these a priori gradients improves the repeatability for 74% (40 out of 54) of the VLBI baseline lengths compared to fixing zero or constant a priori gradients, and improves the repeatability for the majority of baselines compared to estimating 24-h offsets for the gradients. Only if 6-h offsets are estimated, the baseline length repeatabilities significantly improve, no matter which a priori gradients are used.     

对流层映射函数对GNSS反演可降水量的影响分析

对流层映射函数是将对流层天顶延迟转化为信号传播路径上总延迟的重要模型,选择合适的映射函数对反演大气可降水量(PWV)精度的提高具有十分重要的意义．本文研究了对流层映射函数对反演PWV精度的影响,选取VMF1、GMF、NMF 3种映射函数,利用GAMIT解算比较3种映射函数在不同季节、不同高度角对网基线解算以及反演PWV的精度影响．结果表明,在进行PWV反演时,选择10°高度角作为解算截止高度角的GMF函数模型反演精度最佳,为进一步提高GNSS水汽反演的实时精度提供了参考．      

高精度GNSS服务中对流层映射函数的影响性分析

针对GNSS服务中如何获取高精度的大地高以及GNSS气象中对流层湿延迟项,文中采用3种试验方案,利用对流层变化复杂的香港区域CORS,分析不同映射函数存不同高度角时对精密服务的影响。结果表明：存高度角为15°时VMFl、GMF、NMF映射函数获取基线重复性相当,天顶延迟变化规律一致;在低高度角为10°时,基于VMFl映射函数的基线重复率更好,对大气变化的敏感性更强,GMF映射函数获取的天顶延迟精度较好,NMF映射函数获取的精度较弱。     

Multi-technique comparison of tropospheric zenith delays derived during the CONT02 campaign

In October 2002, 15 continuous days of Very Long Baseline Interferometry (VLBI) data were observed in the Continuous VLBI 2002 (CONT02) campaign. All eight radio telescopes involved in CONT02 were co-located with at least one other space-geodetic technique, and three of them also with a Water Vapor Radiometer (WVR). The goal of this paper is to compare the tropospheric zenith delays observed during CONT02 by VLBI, Global Positioning System (GPS), Doppler Orbitography Radiopositioning Integrated by Satellite (DORIS) and WVR and to compare them also with operational pressure level data from the European Centre for Medium-Range Weather Forecasts (ECMWF). We show that the tropospheric zenith delays from VLBI and GPS are in good agreement at the 3–7 mm level. However, while only small biases can be found for most of the stations, at Kokee Park (Hawaii, USA) and Westford (Massachusetts, USA) the zenith delays derived by GPS are larger by more than 5 mm than those from VLBI. At three of the four DORIS stations, there is also a fairly good agreement with GPS and VLBI (about 10 mm), but at Kokee Park the agreement is only at about 30 mm standard deviation, probably due to the much older installation and type of DORIS equipment. This comparison also allows testing of different DORIS analysis strategies with respect to their real impact on the precision of the derived tropospheric parameters. Ground truth information about the zenith delays can also be obtained from the ECMWF numerical weather model and at three sites using WVR measurements, allowing for comparisons with results from the space-geodetic techniques. While there is a good agreement (with some problems mentioned above about DORIS) among the space-geodetic techniques, the comparison with WVR and ECMWF is at a lower accuracy level. The complete CONT02 data set is sufficient to derive a good estimate of the actual precision and accuracy of each geodetic technique for applications in meteorology.     

Ray-traced slant factors for mitigating the tropospheric delay at the observation level

Three-dimensional ray tracing through a numerical weather model has been applied to a global precise point positioning (PPP) campaign for modeling both the elevation angle- and azimuth-dependence of the tropospheric delay. Rather than applying the ray-traced slant delays directly, the delay has been parameterized in terms of slant factors, which are applied in a similar manner to traditional mapping functions, but which can account for the azimuthal asymmetry of the delay. Five strategies are considered: (1) Vienna Mapping Functions 1 (VMF1) and estimation of a residual zenith delay parameter; (2) VMF1, estimation of a residual zenith delay and estimation of two tropospheric gradient parameters; (3) three-dimensional ray-traced slant factors and estimation of a residual zenith delay; (4) using only ray-traced slant factors and no estimation of any tropospheric parameters and; (5) using both ray-traced slant factors and estimating a residual zenith delay and two tropospheric gradient parameters. The use of the ray-traced slant factors (solution 3) showed a 3.8% improvement in the repeatability of the up component when compared to the assumption of a symmetric atmosphere (solution 1), while the estimation of two tropospheric gradient parameters gave the best results showing an 7.6% improvement over solution 1 in the up component. Solution 4 performed well in the horizontal domain, allowing for sub-centimeter repeatability but the up component was degraded due to deficiencies in the modeling of the zenith delay, particularly for stations located at equatorial latitudes. The magnitude of the differences in the mean coordinates between solution 2 and solution 3, and the strong correlation with the differences between the north component and the ray-traced gradients (coefficient of correlation of 0.83), as well as the impact of observation geometry on the gradient solution indicate that the use of the ray-traced slant factors could have an implication on the realization of reference frames. The estimated tropospheric products from the PPP solutions were compared to those derived from ray tracing. For the zenith delay, a root mean square (RMS) of 5.4 mm was found, while for the gradient terms, a correlation coefficient of 0.46 for the N–S and 0.42 for the E–W was found for the north–south and east–west components, suggesting that there are still important differences in the gradient parameters which could be due to either errors in the NWM or to non-tropospheric error sources leaking into the PPP-estimated gradients.     

三种对流层投影函数的比较及对定位的影响

介绍了NMF、VMF1和GMF的干湿分量投影函数的基本数学模型和计算方法,比较了其间的差异和对应的静态PPP结果,得出了相关的结论。     

Troposphere delays from space geodetic techniques, water vapor radiometers, and numerical weather models over a series of continuous VLBI campaigns

Continuous, very long baseline interferometry (VLBI) campaigns over 2 weeks have been carried out repeatedly, i.e., CONT02 in October 2002, CONT05 in September 2005, CONT08 in August 2008, and CONT11 in September 2011, to demonstrate the highest accuracy the current VLBI was capable at that time. In this study, we have compared zenith total delays (ZTD) and troposphere gradients as consistently estimated from the observations of VLBI, Global Navigation Satellite Systems (GNSS), and Doppler Orbitography and Radiopositioning Integrated by Satellite (DORIS) at VLBI sites participating in the CONT campaigns. We analyzed the CONT campaigns using the state-of-the-art software following common processing strategies as closely as possible. In parallel, ZTD and gradients were derived from numerical weather models, i.e., from the global European Centre for Medium-Range Weather Forecasts (ECMWF) analysis fields, the High Resolution Limited Area Model (European sites), the Japan Meteorological Agency-Operational Meso-Analysis Field (MANAL, over Japan), and the Cloud Resolving Storm Simulator (Tsukuba, Japan). Finally, zenith wet delays were estimated from the observations of water vapor radiometers (WVR) at sites where the WVR observables are available during the CONT sessions. The best ZTD agreement, interpreted as the smallest standard deviation, was found between GNSS and VLBI techniques to be about 5–6 mm at most of the co-located sites and CONT campaigns. We did not detect any significant improvement in the ZTD agreement between various techniques over time, except for DORIS and MANAL. On the other hand, the agreement and thus the accuracy of the troposphere parameters mainly depend on the amount of humidity in the atmosphere.     

融合大气数值模式的高精度对流层天顶延迟计算方法

针对现有对流层天顶延迟模型改正法因水汽参数难以精确获取所导致的时空分辨率与精度上的不足问题,提出了一种融合WRF(weather research and forecasting model)大气数值模式的对流层天顶延迟估计方法。通过分析WRF模式的数值模拟机理及其数据结构特征,采用直接积分与模型改正相结合的混合计算方式,实现了全球任意位置上小时级的对流层天顶延迟估计。验证结果表明,该方法计算的小时级ZTD再分析值精度为13.6mm,日均值精度更是可达9.3mm,比传统模型UNB3m的49.6mm以及目前标称精度最高模型GPT2w的34.6mm,精度分别提高了约5倍和3.5倍。在30h的预报时段内,预报值精度也可达22mm。无论是ZTD再分析值还是预报值比现有模型的估计值精度均有明显提高。     

Comparisons of homogeneously reprocessed GPS and VLBI long time-series of troposphere zenith delays and gradients

Troposphere parameters estimated from space-geodetic techniques, like the Global Positioning System (GPS) or Very Long Baseline Interferometry (VLBI), can be used to monitor the atmospheric water vapor content. Although the troposphere can only be monitored at discrete locations, the distribution of the instruments, at least the GPS antennas, can be assumed to be quasi-global. Critical in the data analysis are systematic effects within each single technique that significantly degrade the accuracy and especially the long-term stability of the zenith delay determination. In this paper, consistent time-series of troposphere zenith delays and gradients from homogeneously reprocessed GPS and VLBI solutions are compared for a time period of 11 years. The homogeneity of these completely reprocessed time-series is essential to avoid misinterpretations due to individual model changes. Co-located sites are used to investigate systematic effects and the long-term behavior of the two space-geodetic techniques. Both techniques show common signals in the troposphere parameters at a very high level of precision. The biases between the troposphere zenith delays are at the level of a few millimeters. On the other hand, long-term trends significantly differ for the two techniques, preventing climatological interpretations at present. Tests assume these differences to be due to mathematical artifacts such as different sampling rates and unmodeled semi-annual signals with varying amplitudes.     

Refined discrete and empirical horizontal gradients in VLBI analysis

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). So-called 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 ray-tracing 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, higher-order 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/s00190-017-1066-2), 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 short-term 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.