The present availability of 18
The EUREF Permanent Network (Bruyninx et al., 2012; Ihde et al., 2013) is
the key geodetic infrastructure over Europe, currently made up by over 280 continuously
operating GNSS (global navigation satellite systems such as the USA's
NAVSTAR Global Positioning System, GPS, and Russia's Globalnaya
Navigatsionnaya Sputnikovaya Sistema, GLONASS) reference stations, and
maintained on a voluntary basis by EUREF (International Association of
Geodesy Reference Frame Sub-Commission for Europe,
However, such time series are affected by inconsistencies due to updates of the reference frame and the applied models, implementation of different mapping functions, use of different elevation cut-off angles and any other updates in the processing strategies that causes inhomogeneities over time. To reduce processing-related inconsistencies, a homogenous reprocessing of the whole GNSS data set is mandatory and, to do it properly, a well-documented, long-term metadata set is required.
This paper focuses on the tropospheric products obtained in the framework of
the second EPN reprocessing campaign (hereafter EPN-Repro2), for which, using
the latest available models and analysis strategy, GNSS data of the entire
EPN network have been homogeneously reprocessed for the period 1996–2014.
The EPN homogeneous long-term GNSS time series can be used as a reference
data set for a variety of scientific applications in meteorological and
climate research. Ground-based GNSS meteorology (Bevis et al., 1992) is very
well established in Europe and dates back to the 1990s, starting with the EC
4th Framework Programme (FP) projects WAVEFRONT (GPS Water Vapour Experiment
For Regional Operational Network Trials) and MAGIC (Meteorological
Applications of GPS Integrated Column Water Vapour Measurements in the
western Mediterranean, Haase et al., 2001a). Early this century, the ability
to estimate ZTDs in near real time has been demonstrated (COST-716, 2005),
and the EC 5th FP scientific project TOUGH (Targeting Optimal Use of GPS
Humidity Measurements in Meteorology, 2003–2006) was funded. Since 2005, the
operational production of tropospheric delays has been coordinated and
monitored by the EUMETNET GNSS Water Vapour Programme (E-GVAP, 2005–2017,
Phase I, II and III,
Promoting the use of reprocessed long-term GNSS-based tropospheric delay data
sets for climate research is one of the objectives of the “Working Group 3:
GNSS for climate monitoring” of the EU COST Action ES 1206 “Advanced Global
Navigation Satellite Systems tropospheric products for monitoring severe
weather events and climate (GNSS4SWEC)”, launched for the period of
2013–2017. The Working Group 3 enforces the cooperation between geodesists
and climatologists in order to generate recommendations on optimal GNSS
reprocessing algorithms for climate applications, and to standardise the
conversion method between propagation delay and atmospheric water vapour for
these applications (Saastamoinen, 1973; Bevis et al., 1992; Bock et al.,
2016). For climate applications, maintaining long-term stability is a key
issue. Steigenberger et al. (2007) found that the lack of consistencies over
time due to changes in GNSS processing could cause inconsistencies of several
millimetres in the GNSS-derived integrated water vapour (IWV), making climate
trend analysis very challenging. Jin et al. (2007) studied the seasonal
variability of tropospheric GPS ZTD (1994–2006) over 150 international GPS
stations and showed its relative trend in the Northern and Southern
hemispheres as well as in coastal and inland areas. Wang and Zhang (2009)
derived GPS precipitable water vapour (PWV or PW) using the International
GNSS Service (IGS, Dow et al., 2009) tropospheric products at about
400 global sites for the period 1997–2006 and analysed the PWV diurnal
variations. Nilsson and Elgered (2008) reported on PWV changes from
Against this background, EPN-Repro2 is a unique data set for the development of a climate data record of GNSS tropospheric products over Europe, suitable for analysing climate trends and variability, and calibrating/validating independent data sets at European and regional scales. However, although homogenously reprocessed, this time series still suffers from site-related inhomogeneities due, for example, to instrumental changes (receivers, cables, antennas, and radomes), changes in the station environment, etc. which might affect the analysis of the long-term variability (Vey et al., 2009). Therefore, to get realistic and reliable water vapour trend estimates, such change points in the time series need to be detected and corrected for (Ning et al., 2016a).
EPN Analysis Centres providing EPN-Repro2 solutions.
This paper describes the EPN-Repro2 reprocessing campaign in Sect. 2. Section 3 is devoted to the combined solutions, i.e. the official EPN-Repro2 products, while in Sect. 4 the combined solution is evaluated with respect to radiosonde, ERA-Interim data and in terms of ZTD trends. The summary and recommendations for future reprocessing campaigns are drawn in Sect. 5.
EPN-Repro2 processing options for each contributing solutions. AS0 solution is provided by ASI/CGS (Matera, Italy), GO0, GO1 and GO4 solutions are provided by GOP (Pecny, Czech Republic), IG0 solution by IGE (Madrid, Spain), LP0 and LP1 solutions by LPT (Waben, Switzerland), and MU2 and MU4 solutions by MUT (Warsaw, Poland). PPP is precise point positioning, EGM is Earth Gravitational Model, GMF is Global Mapping Function, VMF is Vienna Mapping Function, HOI is higher-order ionosphere, IONEX is IONospheric maps Exchange format, IGRF is International Geomagnetic Reference Field, FES is finite element solution.
EPN-Repro2 is the second EPN reprocessing campaign organised in the framework of the special EUREF project “EPN reprocessing”. The first reprocessing campaign, which covered the period 1996–2006 (Voelksen, 2011), involved the participation of all 16 EPN Analysis Centres (ACs), reprocessing their own EPN sub-network. This strategy guaranteed that each site was processed by at least three ACs, which is an indispensable condition for providing a combined product. The second reprocessing campaign covered all the EPN stations, which were operated from January 1996 to December 2013. Then, the participating ACs decided to extend this period until the end of 2014 for tropospheric products. Data from about 280 stations in the EPN historical database have been considered. As of December 2014, 23 % of EPN stations are between 15 and 18 years old, 26 % are between 10 and 14 years old, 30 % between 5 and 10 years old and 21 % less than 5 years old. Only 5 out of 16 EPN ACs (see Table 1) took part in EPN-Repro2, each providing at least one reprocessed solution. One of the goals of the second reprocessing campaign was to test the diversity of the processing methods in order to ensure the verification of the solutions. For this reason, the three main GNSS software packages Bernese (Dach et al., 2015), GAMIT (King et al., 2010) and GIPSY-OASIS II (Webb and Zumberge, 1997) have been used to reprocess the whole EPN network and, in addition, several variants have been provided. In total, eight individual contributing solutions, obtained using different software and settings, and covering different EPN networks, are available. Among them, three are obtained with different software and cover the full EPN network, while three are obtained using the same software (namely Bernese), but covering different EPN networks. In Table 2 the processing characteristics of each contributing solution are reported. Despite the software used and the analysed networks, there are few diversities among the provided solutions, the impact of which needs to be evaluated before performing the combination. In the reprocessing campaign all the ACs used for the GNSS orbit the CODE (Center for Orbit Determination in Europe) Repro2 product (Lutz et al., 2014), with one exception (see Table 2) where JPL (Jet Propulsion Laboratory) Repro2 products (Desai et al., 2014) are used. For tropospheric modelling two mapping functions are used: Global Mapping Function (GMF; Boehm et al., 2006a) and Vienna Mapping Function (VMF1; Boehm et al., 2006b). Their impact has been evaluated in Tesmer et al. (2007).
During the reprocessing period, the Russian satellite system GLONASS became
operational, and GLONASS observations have been available since 2003. However,
only from 2008 onwards the amount of GLONASS data (see
Fig. 1) is significant. The impact of GLONASS
observations has been evaluated in terms of raw differences between ZTD
estimates as well as on the estimated linear trend derived from the ZTD time
series. As a matter of fact, GPS data (from the American navigation
satellite system) are used by all ACs in this reprocessing campaign, while
two of them (namely IGE and LPT) reprocessed GPS and GLONASS observations.
Two solutions were prepared and compared, using the same software and the
same processing characteristics, but different observation data: one with
GPS and GLONASS, and one with GPS data only. The difference in ZTD trends
(Fig. 2) between a GPS-only and a GPS
Time series of the number of GNSS observations for the period 1996–2014. GPS observations are shown in red, GPS/GLONASS in blue and their differences in green. The difference becomes significant starting from 2008.
ZTD trend differences between GPS only and GPS/GLONASS, computed
over 111 sites. The rate is in violet (primary
According to the processing options listed in the EPN guidelines for the
Analysis Centre
(
EPN station KLOP (Kloppenheim, Frankfurt, Germany) ZTD differences time series between solutions processed with individual and type mean antenna calibration models. Two instrumentation changes occurred at the station (marked by vertical dashed red lines): the first in 27 June 2007, when the previous antenna was replaced with a TRM55971.00 and a TZGD radome, and the second in 28 June 2013 with the installation of a TRM57971.00 and a TZGD radome.
As reported in the International Earth Rotation and Reference Systems Service (IERS) Convention (IERS, 2010), the diurnal heating of the atmosphere causes surface pressure oscillations with diurnal and semidiurnal variability and even higher harmonics. These atmospheric tides induce periodic motions of the Earth's surface (Petrov and Boy, 2004). The conventional recommendation is to calculate the station displacement using the Ray and Ponte (2003) tidal model. However, crustal motion related to non-tidal atmospheric loading has been detected in station position time series from space geodetic techniques (van Dam et al., 1994; Mangiarotti et al., 2001; Tregoning and Van Dam, 2005). Several models of station displacements related to this effect are currently available. Non-tidal atmospheric loading models are not yet considered as Class 1 models by the IERS (IERS, 2010), indicating that there are currently no standard recommendations for data reduction. To evaluate their impact, two solutions, one with and one without a non-tidal atmospheric loading model, have been compared for the year 2013. In the solution with the model, the National Centers for Environmental Prediction (NCEP) model is used at the observation level during data reduction (Tregoning and Watson, 2009).
Dach et al. (2010) have already found that the repeatability of the station coordinates improves by 20 % when applying the non-tidal atmospheric loading correction directly on the data analysis and by 10 % when applying a post-processing correction to the resulting weekly coordinates. However, the effect on the ZTDs seems to be negligible. Generally, it causes a difference below 0.5 mm with a standard deviation not larger than 0.3 mm. The difference is thus below the level of confidence. Figure 4 shows time series of the differences of the ZTDs and the up components between two solutions obtained with and without non-tidal atmospheric loading for two EPN stations: KIR0 (Kiruna, Sweden) and RIGA (Riga, Latvia). Furthermore, there is no correlation between the values of estimated differences and vertical displacements caused by non-tidal atmospheric loading, as correlation coefficients for the analysed EPN stations were below 0.2.
The EPN ZTD combined product is obtained by applying a generalised least square approach following the scheme described in Pacione et al. (2011). The first step in the combination process is the reading and checking of the SINEX TRO files delivered by the ACs. At this stage, gross errors (i.e. ZTD estimates with formal standard deviations larger than 15 mm) are detected and removed. The combination starts if at least three different solutions are available for a single site. Then, a first combination is performed to compute proper weights for each contributing solution, to be used in the final combination step. In this last step the combined ZTD estimates, their standard deviations and site/AC-specific biases are determined. The combination fails if, after the first or second combination level, the number of ACs becomes less than three. Finally, ZTD site/AC-specific biases exceeding 10 mm are investigated as potential outliers.
The EPN-Repro2 combination activities were carried out in two steps. First,
a preliminary combined solution for the period 1996–2014 was performed,
taking all the available eight homogeneously reprocessed solutions (see
Table 2) as input. The aim of this preliminary
combined solution is to assess each contributing solution and to investigate
site/AC-specific biases prior to the final combination, flag the outliers
and send feedback to the ACs. The agreement of each contributing solution
with respect to the preliminary combination is given in terms of bias and standard
deviation (not shown). The standard deviation is generally below 2.5 mm,
with a clear seasonal behaviour (larger for larger ZTD values), while the
bias is generally in the range of
VENE (Venice, Italy) time series of ZTD biases and standard deviations for the three contributing solutions AS0, GO4 and MU4 with respect to the combined solution for the period 21 July 1996–28 July2007 (GPS weeks 0863–1437). GO0 and GO1 are not shown here, since they are very close to GO4.
Weekly mean ZTD biases
All the site/AC-specific biases are divided into three groups: the red group contains site/AC-specific biases with values larger than 25 mm, the orange group contains site/AC-specific biases in the range of [15, 25 mm] and the yellow group contains site/AC-specific biases in the range of [10, 15 mm]. In Table 3 the percentages of red, orange and yellow biases for each contributing solution are summarised. The majority of biases belong to the yellow group; the percentage of biases in the orange group ranges from 12 % for LP0 and LP1 solutions to 27 % for the AS0 solution, while the percentage of biases in the red group ranges from 3 % for the MU4 solution to 22 % for the IG0 solution.
The final EPN-Repro2 tropospheric combination is based on the following input solutions: AS0, GO4, IG0, LP1 and MU2. MUT AC provided the MU2 solution after the preliminary combination, its only difference with respect to MU4 is the use of type mean antenna and individual calibration models, the effect of which already been described in Sect. 2.2. For those ACs providing more than one solution, we have chosen the one carried out with the Vienna Mapping Function. The agreement in terms of bias and standard deviation of each contributing solution with respect to the final combination is shown in Fig. 6. The standard deviation had improved significantly with respect to the preliminary combination (not shown here) due to the removal of outliers detected during this early combination. The standard deviation is below 3 mm before GPS week 1055 (26 March 2000) and 2 mm thereafter. This is related to the worse quality of data and products during the first years of the EPN/IGS activities.
Percentage of red, orange and yellow biases (see text) for each contributing solution.
The final EPN-Repro2 tropospheric combination is consistent with the final
coordinate combination performed by the EPN Analysis Centre Coordinator.
During the coordinate combination all stations were analysed by comparing
their coordinates for specific ACs and the preliminary combined values. In
the cases where the differences were larger than 16 mm in the up component
(vertical displacement), the station was eliminated and the whole combination
process was repeated, up to three times if necessary. This ensures the
consistency of the individual contributing solution with respect to the final
coordinates at the level of 16 mm in the up component. As internal quality
metric, we have considered the site coordinate repeatability of the final
coordinate combination (Fig. 7). As a rule of thumb, 9 mm repeatability in
the up component (i.e. 3 mm in ZTD as explained in Santerre, 1991) are
needed to fulfil the requirement of retrieving IWV at an accuracy level of
0.5 kg m
Long-term up component repeatability of the final coordinates for all stations. The site coordinate repeatability is used as an internal quality metric. Stations are sorted by name.
The evaluation with respect to other sources or products, such as radiosonde data from the E-GVAP and numerical weather reanalysis from the European Centre for Medium-Range Weather Forecasts, ECMWF (ERA-Interim), provides a measure of the accuracy of the ZTD combined products.
VENE (Venice Italy) time series of daily repeatability (for definition; see Fig. 7) in the up component for the period 21 July 1996–28 July 2007 (GPS weeks 0863–1437).
For the GNSS and radiosonde (RS) comparisons at the EPN collocated sites, we used profiles from the World Meteorological Organization (WMO) provided by EUMETNET in the framework of the memorandum of understanding between EUREF and EUMETNET. Radiosonde profiles are processed using a software by Haase et al. (2003) that checks the quality of the profiles, converts the dew point temperature to specific humidity, shifts the radiosonde profile to correct for the altitude offset between the GPS and the radiosonde sites, and determines the ZTD and IWV compensating for the change of the gravitational acceleration g with height.
A comparison of the GNSS and radiosonde ZTD time series for the EPN site CAGL (Cagliari, Sardinia Island, Italy) is shown in Fig. 9, with the mean biases and standard deviations reported in the figure. Similarly, we computed an overall bias (RS minus GNSS) and standard deviation for all the 183 EPN collocated sites, using all the data available in the considered period (Fig. 10). In this figure, the sites are sorted with increasing distance from the nearest radiosonde launch site. For instance, MALL (Palma de Mallorca, Spain) is the closest (0.5 km to the radiosonde site with WMO code 8301) while GRAZ (Graz, Austria) is the most distant (133 km to RS WMO code 14015). The amount of data available for the comparisons varies between sites, depending on the availability of the GNSS and radiosonde ZTD estimates in the considered epoch, and ranges from 121 pairs for VIS6 (Visby, Sweden, integrated in the EPN since 22 June 2014) to up to 21 226 pairs for GOPE (Ondrejov, Czech Republic, integrated in the EPN since 31 December 1995).
EPN station CAGL (Cagliari, Sardinia Island, Italy).
RS minus GNSS ZTD biases for all GNSS-RS station pairs. The error bar is the standard deviation. Sites are sorted with increasing distances from the nearest radiosonde launch site.
The mean relative [(RS-GNSS)/GNSS] bias ranges from
In agreement with Ning et al. (2012), the ZTD standard deviation generally
increases with distance from the radiosonde launch site. It is in the range
of [0.16; 0.76] %, which corresponds to [3; 18] mm in ZTD up to 15 km
(first band in Fig. 10); in [0.29; 0.78] %, corresponding to [7; 19] mm
up to 70 km (second band in Fig. 10), and in [0.43; 1.35] %,
corresponding to [10; 33] mm up to 133 km (third band in Fig. 10). The
numbers of the standard deviation are comparable with previous studies. Haase
et al. (2001b) showed a very good agreement with biases less than 5 mm in
ZTD and a standard deviation of 12 mm for most of the analysed sites in the
Mediterranean. Similar results (6.0
If we compare both the EPN-Repro1 ZTD product (completed with the EUREF operational product after 30 December 2006) and the EPN-Repro2 with the radiosonde ZTDs for the same period 1996–2014, we found an improvement of approximately 3–4 % in the overall standard deviation for the second processing.
We also compared the EPN-Repro2 ZTDs with the ZTDs calculated from
ERA-Interim (Dee et al., 2011) from the European Centre for Medium-Range
Weather Forecasts (ECMWF). The ERA-Interim is a reanalysis product of a
numerical weather prediction (NWP) model and is available every 6 h (00:00,
06:00, 12:00, 18:00 UTC) with a horizontal resolution of
Mean statistics and uncertainties, calculated from results of individual stations, provided for AC individuals and EUREF combined (EPN-Repro1 and EPN-Repro2) tropospheric parameters compared to the ERA-Interim reanalysis (ERA-Interim minus GNSS). EGRD represents east gradient and NGRD north gradient.
For the period 1996–2014 and for each EPN station, the ZTD and tropospheric linear horizontal gradients were computed using the GFZ (German Research Centre for Geosciences) ray-tracing software (Zus et al., 2014). Combined EPN-Repro1 and EPN-Repro2 products as well as individual ACs tropospheric parameters were assessed with the corresponding parameters estimated from the ERA-Interim reanalysis. The evaluation of GNSS and ERA-Interim was performed using the GOP-TropDB (Gyori and Dousa, 2016) by calculating parameter (ZTD, horizontal gradients; see below) differences for each station, using the values at every 6 h (00:00, 06:00, 12:00 and 18:00 UTC), as available from the ERA-Interim model output. A linear temporal interpolation to those four timestamps was thus necessarily applied to all GNSS products, which are available in HH:30 timestamps as required for the combination process. As all compared GNSS products have the same time resolution (1 h), the interpolation is assumed to affect all products in the same way. Therefore, we assume that all intercomparisons to a common reference (ERA-Interim) principally reflect the quality of the products. No vertical corrections were applied since ERA-Interim variables were estimated for the long-term antenna reference position of each station.
Table 4 summaries the mean total statistics of individual (ACs) and combined (EUREF) tropospheric parameters, ZTDs and horizontal gradients, over all available stations. The EUREF combined solution does not provide tropospheric gradients and these could therefore be evaluated for individual solutions only. In Table 4, a common ZTD bias (ERA-Interim minus GNSS) of about 1.8 mm is found for all GNSS solutions compared to ERA-Interim, but a large station to station variability could be noted, as is obvious from the estimated uncertainties. ZTD standard deviations are generally at the level of 8 mm between GNSS and ERA-Interim ZTDs, but with the IG0 solution performing about 25 % worse than the others, as already detected during the combination. Two solutions, AS0 and LP1, are slightly better than GO4 and MU2: with a standard deviation of 7.7 mm, their accuracy is at the level of the EUREF combined solution. The better performance of the AS0 solution can be explained by applying a stochastic troposphere modelling using original (not double-difference) observations that are sensitive to the absolute tropospheric delays, so that the true dynamics in the troposphere are better taken into account. LP1 included roughly one-third of the EPN stations, properly selected according to the station quality, hereby making it difficult to interpret this difference with respect to those solutions processing the full EPN.
Distributions of station mean ERA-Interim minus GNSS ZTD
biases
The comparison of tropospheric linear horizontal gradients (east and north) from GNSS and ERA-Interim revealed a problem with the MU2 solution (see Table 4). This solution shows a high inconsistency over different stations, which is not visible in the total statistics, but mainly in the uncertainties, which are an order of magnitude higher compared to all other solutions. A geographical plot (not shown here) confirmed this site-specific systematic effect, both in positive and negative sense. The impact was, however, not observed in the MU2 ZTD results. Additionally, the GO4 solution performed slightly worse than the others. This was identified as a consequence of estimating 6 h gradients using a piecewise linear function without any absolute or relative constraints. In such a case, higher correlations with other parameters occurred and increased the uncertainties of the estimates. For this purpose, the GO6 solution (not shown) was derived, fully compliant with the GO4, but stacking tropospheric gradients into 24 h piecewise linear modelling. In comparison with the former GO4 solution (Dousa and Vaclavovic, 2017), the GO6 standard deviations dropped from 0.38 to 0.28 mm and from 0.40 to 0.29 mm for eastern and northern gradients, which corresponds to the LP1 solution that applied the same settings. Additionally, Dousa and Vaclavovic (2017) found a strong impact of a low-elevation receiver tracking problem on the estimation of the horizontal gradients, which was particularly visible when comparing with ERA-Interim horizontal gradients. Looking for systematic behaviour in monthly mean differences in the gradients therefore seems to be a useful indicator for instrumentation-related issues and should be applied as one of the tools for cleaning the EPN historical archive.
For completeness, we also evaluated the EPN-Repro1 ZTD product with respect to ERA-Interim using the same period, i.e. 1996–2014 (after completing again with the EUREF operational product; see above). Comparing EPN-Repro1 and EPN-Repro2 with the numerical weather model reanalysis showed an 8–9 % improvement of EPN-Repro2 in both overall standard deviation and bias. Figure 11 shows the distributions of station mean biases and standard deviations of EPN-Repro1 and EPN-Repro2 ZTDs compared to ERA-Interim ZTDs using the whole period 1996–2014. Common reductions of both statistical characteristics are clearly visible for the majority of all stations. From the data of Fig. 11, we also illustrate the site-by-site improvements in terms of ZTD bias, standard deviation and rms in Fig. 12. The calculated median improvements for these statistics reached 21.1, 6.8 and 8.0 %, which correspond to the above-mentioned improvement of 8–9 %. A degradation of the standard deviation was found at three stations: SKE8 (Skellefteaa, Sweden, integrated in the EPN since 28 September 2014), GARI (Porto Garibaldi, Italy, integrated in the EPN since 8 November 2009) and SNEC (Snezka, Czech Republic, former EPN station since 14 June 2009). These three stations provide much less data compared to other stations: only 1, 30 and 3 % of data pairs available at other stations. All other stations (290) showed improvements. We found 72 stations with increased absolute bias in EPN-Repro2 compared to EPN-Repro1 while the other 221 stations (75 %) had a reduced bias with ERA-Interim ZTD.
Time series of monthly mean biases and standard deviations for ZTD differences of EPN-Repro2 and ERA-Interim are shown in Fig. 13. The small negative bias slowly decreases towards 2014, but the high uncertainty of the mean bias indicates site-specific behaviour, depending mainly on latitude and altitude of the EPN station and the quality of both ERA-Interim and GNSS products. There is almost no seasonal signal observed in the time series of ZTD mean biases or uncertainties, but there clearly is in the ZTD mean standard deviation and the uncertainties. The increase of standard deviation in summer is due to more humidity in the troposphere, which is more difficult to model accurately in both GNSS and ERA-Interim. The slightly increasing standard deviation towards 2014 can be attributed to the increase of number of stations in EPN, starting from about 30 in 1996 and with more than 250 in 2014. A higher number of stations reduces the variability in monthly mean biases; however, site-specific errors then contribute more to higher values of standard deviation.
Site-by-site ZTD improvements of EPN-Repro2 versus EPN-Repro1 compared to ERA-Interim.
Time series of monthly mean biases
Geographical distribution of ZTD biases
Figure 14 displays the geographical distribution of total ZTD biases (ERA-Interim minus GNSS) and standard deviations for all sites. Prevailing positive biases seem to become lower or even negative in the mountain areas. There is no latitudinal dependence observed for ZTD biases in Europe, but a strong one is observed for standard deviations. This corresponds mainly to the increase of water vapour content and its variability towards the equator.
To illustrate the impact of the new processing on the resulting ZTD trends and related uncertainties, we considered five EPN stations among those with the longest time span: GOPE (Ondrejov, Czech Republic, integrated in the EPN since 31 December 1995), METS (Kirkkonummi, Finland, integrated in the EPN since 31 December 1995), ONSA (Onsala, Sweden, integrated in the EPN since 31 December 1995), PENC (Penc, Hungary, integrated in the EPN since 3 March 1996) and WTZR (Bad Koetzting, Germany, integrated in the EPN since 31 December 1995). For these five stations, we have computed ZTD trends using EPN-Repro2, EPN-Repro1 (again completed with the EUREF operational products), radiosonde and ERA-Interim data. Furthermore, those five stations also belong to the IGS Network, for which IGS Repro1, completed with the IGS operational products, are available and extracted from the GOP-TropDB, so that we could also calculate ZTD trends from this data set.
First, we removed the annual signal from the original time series and marked
all outliers according to the 3
In Fig. 15, the ZTD trends and uncertainties are
presented for the five sites and for all ZTD data sets. First of all, it
should be noted that the trends between the three GNSS ZTD data sets are
very consistent (as long as the same homogenisation procedure is applied).
The overall rms among trends estimated from GNSS measurements is 0.02 mm year
ZTD trend comparisons at five EPN stations for five different ZTD data sets. The error bars are the formal errors of the estimated trend values.
In this paper, we described the activities carried out in the framework of
the second reprocessing campaign of the EPN. We focused on the tropospheric
products that were homogenously reprocessed by five EPN Analysis Centres for the
period 1996–2014 and we described the ZTD combined product. We evaluated the
impact of few diversities among the provided GNSS solutions. The inclusion
of additional GLONASS observations in the GNSS processing has a neutral
impact on the ZTD trend analysis, indicating that the ZTD trends might be
determined independently of the satellite systems used in the processing
(see Sect. 2.1). The inconsistencies in the ZTD time series due to
different antenna calibration models (see Sect. 2.2) are not large enough
to be captured during the combination process (see Sect. 3), in which a 10 mm
threshold in the ZTD bias (about 1.5 kg m
Both individual and combined tropospheric products, along with reference coordinates and other metadata, are stored in a SINEX TRO format (Gendt, 1997), and are available to the users at the EPN regional data centres (RDC), located at BKG (Federal Agency for Cartography and Geodesy, Germany). For each EPN station, plots on ZTD time series, ZTD monthly means, comparison with radiosonde data (if collocated), and comparison versus the ERA-Interim data will be available at the EPN Central Bureau (Royal Observatory of Belgium, Brussels, Belgium).
We showed in Sect. 4.1 that EPN-Repro2 led to an improvement of approximately 3–4 % in the overall standard deviation in the ZTD differences with radiosonde data compared with EPN-Repro1.
The assessment of the EPN-Repro2 comparison with the ERA-Interim reanalysis
showed an 8–9 % improvement in both the overall ZTD bias and standard
deviation with respect to EPN-Repro1 for the majority of the stations (see
Sect. 4.2). Comparisons of the GNSS solutions with ERA-Interim, showed the
agreement in ZTD at the level of 8–9 mm; however, site performance ranged
from 5 to 15 mm for standard deviations and from
The use of ground-based GNSS long-term data for climate research is an emerging field. For example, for the assessment of Euro-CORDEX (Coordinated Regional Climate Downscaling Experiment) climate model simulation, the IGS Repro1 data set (Byun and Bar-Sever, 2009) has been used as a reference for reprocessed GPS products (Bastin et al., 2016). However, this data set is quite sparse over Europe (only 85 stations over the 280 EPN stations) and covers only the period 1996–2010. As pointed by Baldysz et al. (2015, 2016) an additional 2 years of ZTD data can change the estimated trends up to 10 %. Therefore, with data after 2010 and with a better coverage over Europe, EPN-Repro2 can be used as a reference data set with a high potential for monitoring the trends and variability in atmospheric water vapour as reported in Sect. 4.3. As a matter of fact, a comparison between GNSS IWV, computed from EPN-Repro2 ZTD data for SOFI (Sofia, Bulgaria) by the Sofia University, and ALADIN-Climate IWV simulations conducted by the Hungarian Meteorological Service, is performed for the period 2003–2008 at the moment. The preliminary results show a tendency of the model to underestimate IWV. Clearly, a larger number of model grid points needs to be investigated in different regions in Europe and the EPN-Repro2 data are well suited for this.
The reprocessing activity of the five EPN ACs was a huge effort, generating homogeneous products not only for station coordinates and velocities, but also for tropospheric products. The knowledge gained will certainly help for a next reprocessing activity. A next reprocessing will most likely include Galileo and BeiDou data and therefore it will be started in some years from now after having successfully integrated these new data in the current operational near real-time and daily products of EUREF. The consistent use of identical models in various software packages is another challenge for the future and would enable to improve the consistency of the combined solution. Prior to any further reprocessing, it was agreed that EUREF should focus on cleaning and documenting the data in the EPN historical archive as this should highly facilitate any future work. For this purpose, all existing information needs to be collected from all the levels of data processing, combination and evaluation, which includes initial GNSS data quality checking, generation of individual daily solutions, combination of individual coordinates and ZTDs, long-term combination for velocity estimates and assessments of ZTDs and gradients with independent data sources.
The EPN historical data centre is publicly available at
R. Pacione coordinated the writing of the manuscript and wrote Sects. 1, 2, 3 and 4.1. A. Araszkiewicz wrote Sect. 2.2 and 2.3, 4.3 and contributed to Sect. 4.1. E. Brockmann wrote Sect. 2.1. J. Dousa wrote Sect. 4.2. All authors contributed to Sect. 5. All authors approved the final manuscript before its submission.
The authors declare that they have no conflict of interest.
The authors would like to acknowledge the support provided by COST – (European Cooperation in Science and Technology) for providing financial assistance for the publication of the paper. The authors thank the members of the EUREF project “EPN reprocessing”. e-GEOS work is done under ASI Contract 2015-050-R.0. The assessments of the EUREF combined and individual solutions in the GOP-TropDB were supported by the Ministry of Education, Youth and Science, the Czech Republic (project LH14089). The MUT AC contribution was supported by statutory founds at the Institute of Geodesy, Faculty of Civil Engineering and Geodesy, Military University of Technology (No. PBS/23-933/2016). Finally, we thank the two anonymous referees and the Associate Editor Roeland Van Malderen for their comments which greatly helped to improve the paper. Edited by: R. Van Malderen Reviewed by: two anonymous referees