Contributors: Kalev Rannat (Tallinn University of Technology, TUT), Hannes Keernik (TUT), Fabio Madonna (CONSIGLIO NAZIONALE DELLE RICERCHE – ISTITUTO DI METODOLOGIE PER L'ANALISI AMBIENTALE, CNR-IMAA), Emanuele Tramutola (CNR-IMAA), Fabrizio Marra (CNR-IMAA)

Issued by: CNR-IMAA / Fabio Madonna

Date: 03/06/2021

Last updated: 25/11/2024

...

Introduction

This document provides a short description of GNSS Integrated Precipitable Water (IPW) datasets provided in the Copernicus Data Store (CDS) catalogue and retrieved using the GNSS data of the International GNSS Service (IGS network ) and the EPN-repro2 datasetEUREF Permanent Network (EPN). Additional information about GNSS IPW measuring principles, hardware and raw data processing can be found in the related ATBD document.

The GNSS IPW datasets provided in the CDS are retrieved from ZTD that is a final product from GNSS raw data processingGNSS tropospheric products. In the remainder of this document the expression “GNSS troposphere product” refers to Zenith Total Delay (ZTD) with its formal 1σ error as ZTD uncertainty. Although the data for IGS and EPN-repro2 come from different sources and in a different format, the The methodology for converting ZTD to IPW is the same for bothall GNSS tropospheric datasets, except for some minor technical details.

CDS offers two categories of GNSS tropospheric products: Near Real-Time (NRT) data, which includes IGS daily, and reprocessed datasets, which encompass data from the reprocessing campaigns of IGS and EPN (IGS-repro3 and EPN-repro2, respectively).

The IGS collects, archives, and freely distributes GNSS observation data sets from a cooperatively operated The IGS collects, archives, and freely distributes GNSS observation data sets from a cooperatively operated global network of more than 500 ground tracking stations. The IGS network is classified as a reference network based on the basis of the Measurement System Maturity Matrix (MSMM) approach      [1], ensuring open access , to high-quality GNSS data products since 1994. The product used for retrieval of IGS IPW is produced by 12 licensed Analysis Centres (AC).

The EPN-repro2 dataset [2] IGS-repro3 is a reprocessed product from EUREF Permanent Network (EPN, http://www.epncb.oma.be/) with more than 300 continuously operating GNSS reference stations with precisely known coordinates. It consists of 18+ years of GNSS data, making it a valuable database for the development of a climate data record of GNSS tropospheric products over Europe. The product name, EPN-repro2, denotes the second reprocessing campaign of the EPN, where five ACs homogenously reprocessed the EPN network for the period 1996–2014. Finally, the individual contributions of each AC are combined to provide the official EPN reprocessed products. The combined product (based on the methods presented in Pacione, et al., 2011 [3] is described along with its evaluation against radiosonde data and European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA-Interim) data.

Meteorological information used as ancillary data required for the IPW estimation is obtained from the ECMWF ERA5 atmospheric reanalysis. The methods for requesting data co-located with the GNSS sites can be found in the ATBD document.

In the following sections, a short retrospective overview is given about GNSS IPW quality assurances, data and metadata sources (Section 2), data processing for both IGS and EPN-repro2 (Section 3), data validation issues (Section 4), examples of IPW comparison (Section 5), data and metadata format in CDS (Section 6) and data licensing (Section 7).

The maturity matrix of IGS network can be found from Appendix A. Description of data units and conversion is given in Appendix B.

Data and metadata sources

The IGS datasets are available since 22^th October 2000 through the CDDIS data portal (https://cddis.nasa.gov/archive/gnss/products/troposphere/zpd/ , Fig. 1):

...

Image Removed

Figure 1:  Map of the global distribution of IGS sites.

Metadata consists of additional necessary data used in a process of estimating GNSS IPW and its uncertainty, described in the ATBD document. The metadata can be found from IGS sites’ specifications and from the headers of SINEX TRO files (described in M311a_Lot3.3.2.3_2020, Section 4.1 and at https://igs.org/formats-and-standards/ ). However, technical implementation of IPW software package uses SEMISYS (referred in product availability and licenses section).

The GNSS troposphere product available at CDDIS is delivered by the IGS data processing agencies in a unified SINEX TRO format. However, these ZTD time series are still not corrected for any effects derived from instrumental changes (receivers, antennas, radomes) or changes in station environment. This kind of analysis needs extensive data reprocessing. A comprehensive description of the IGS GNSS troposphere data products can be found in the ATBD document. While not best suited for long-term climate studies, the IGS daily dataset is relevant in atmospheric analyses due to its worldwide network and relatively short data latency (ca 2 weeks).

Good examples of GNSS reprocessed time series are EPN-repro1 and EPN-repro2 (available for EUREF, http://www.epncb.oma.be/_productsservices/analysiscentres/repro2.php). The dataset is freely available from BKG (Bundesamt für Kartographie und Geodäsie): ftp://igs.bkg.bund.de/EPNrepro2/products/.

In the framework of the EPN-repro2, the second reprocessing campaign of the EPN, five ACs homogeneously reprocessed the EPN network (Fig. 2) for the period 1996–2014 [2]. ZTD screening, which is the process of inspecting data for errors and correcting them prior to performing data analysis, has been applied to all initial ZTD time series [2]. By doing this, all estimated shifts (biases) from GNSS antenna replacements are removed [2]. Therefore, this data record can be used as a reference for a variety of scientific applications (e.g. validation of regional numerical weather prediction reanalyses and climate model simulations) and has a high potential for monitoring trends and the variability in atmospheric water vapour.

...

The EPN-repro2 data is recorded by GNSS weeks (834–1825) with combined solutions recorded in SINEX TRO mixed format. Unlike CDDIS for IGS, the EPN ftp service supports traditional anonymous data access. One of the major differences compared to IGS datasets is that EPN-repro2 data is not recorded at full hours, but with a 30 minutes shift (00:30, 01:30, …, 23:30).

The daily ZTDs provided by the ACs are combined on a weekly basis. Only sites with a corresponding coordinate solution and with three individual AC contributions are presented. Estimates with a given STDDEV > 15 mm are excluded. Rough outlier detection is performed to find strong outliers. Details on the combination process can be found from [3].

For estimating GNSS IPW based on ZTD, ancillary meteorological data are needed. This are obtained from the ERA5 reanalysis dataset available via the CDS. Geopotential, specific humidity and temperature values are downloaded for 37 pressure levels in order to calculate water-vapor-weighted mean temperature of the atmosphere, T_m, which is crucial for converting ZTD to IPW.

GNSS data processing and retrieval of IPW

The GNSS IPW provided through the CDS is estimated based on using tropospheric products delivered by IGS data ACs and EPN-repro2. The data processing schema for retrieval of GNSS IPW for both IGS and EPN tropospheric product is described in more detail in ATDB document. Here only a simplified description of the data flow and data processing steps for retrieval of GNSS IPW from GNSS tropospheric products is depicted (Fig. 3).

...

Image Removed

Figure 3: Retrieval of IPW for IGS and EPN-repro2

• Pre-processing of GNSS troposphere product (steps 1–4). First, ZTD with its uncertainty is downloaded from IGS and EPN data repositories (steps 1–2, Fig. 3). The site metadata (coordinates, altitude, city, agency etc.) for both networks are obtained from SEMISYS [4] (step 3). Due to the fact that the site latitude and longitude are given with height from WGS84 ellipsoid, the AMSL (height above the mean sea level) used in IPW retrieval, must be calculated first (step 4).
• Pre-processing of ancillary data (steps 5–7). The supporting meteorological data at all ERA5 pressure levels and at ground surface level are downloaded, the water-vapour-weighted mean temperature of the atmosphere (T_m) is calculated and bilinearly interpolated to the site coordinates and altitude (step 5). Since the time resolution of both, ERA5 and IGS ZTDs, is one hour there is no need to interpolate T_m and ground surface pressure (p₀) in time. On the other hand, this is done in case of EPN-repro2 since its data are provided at half-hours. Depending on dry gases between receiver and satellite, the ZHD is calculated using p₀, latitude and height of the site above the mean sea level (step 6). The ZWD, on the other hand, depends on water vapour amount and is found by subtracting ZHD from ZTD. The conversion factor that is needed to relate IPW to the ZWD is calculated using a function of T_m (step 7).
• Calculating IPW and its uncertainty (steps 8–10). IPW estimation (in units kg m^-2, equivalent to mm) is given using ZWD and the conversion factor (step 8). For calculating GNSS IPW uncertainty (step 9), the approach chosen for C3S_311a_Lot3 is based on the GRUAN GNSS data processing [5], and accounts for all of the uncertainty sources due to the measurement systems. The IPW uncertainty calculation relies on previously published values for input variables used as constants and their uncertainties [6]. In case of uncertainties for p₀ and T_m, these values, statistically determined using IGRA and GRUAN radiosonde measurements, have been found to be in a good agreement with the results presented in several papers specifically focusing on ERA5 [7, 8]. After the values for IPW and its uncertainty are calculated, the results are ingested to the database (step 10).

Data validation process

 IGS

The integrity of the data in daily SINEX TRO files is guaranteed by the IGS ACs. However, for assuring the data quality for retrieval of GNSS IPW and its uncertainty (methodology described in Ning et al., 2016 [5]) the data processing chain for CDS comprises additional steps as follows:

• Checking ZTD values for range. The valid range is [1.00, 3.00] m;
• Checking formal error of ZTD (σ_ZTD). A ZTD value is accepted if σ_ZTD <= 10 mm or σ_ZTD < 2.5 · median (the latter is applied only if the 24 hours median < 3.0 mm)

If any of these checks fails, the ZTD or σ_ZTD will be assigned a “null” value (indicating a missing or unacceptable value).

These checks are similar to those used by screening of ZTD time series for IGS repro-products [2, 9]. However, it must be noted, that a range check on formal errors should be station-specific [2] and the time series have not passed all the procedures applied on reanalysis (e.g., EPN-repro2).

IGS sites’ metadata (site coordinates) are checked on-line by SEMISYS and corrected by proofed values from SEMISYS if the horizontal displacement exceeds 90 meters (equivalent of 3 arc seconds). Coordinate check avoids using erroneous AMSL values from a geoid model.

EPN

EPN-repro2 time series do not need any additional quality check. All necessary quality checks are performed already during reprocessing. EPN-repro2 is known as a reference quality data product [2].

Unlike many other measurements in environmental physics (atmospheric temperature, humidity, etc.) the GNSS IPW is retrieved from GNSS ZTDs characterised with uncertainties expressed in terms of formal (not instrumental) errors.

The characteristic values depend on GNSS data processing software and the IGS specifies σ_ZTD limits to 4 mm [10] for observations in mostly ideal conditions with instrumental installations following strict IGS technical requirements. The sources of uncertainties contributing to the final σ_IPW are extensively analysed in Ning et al., 2016 [5] and in GAIA-CLIM D2.8, Annex IX, Product Traceability and Uncertainty for the GNSS IPW Product [11].

Comparing σ_ZTD values between IGS daily product and EPN-repro2, the user may notice remarkable differences. In case of IGS daily data these values stay around 2 mm. At the same time, EPN-repro2 σ_ZTD values can reach 5–6 mm. As a direct consequence to this (σ_ZTD contributes over 75% to the total IPW uncertainty), the total σ_IPW values may reach slightly over 1 mm. It should be noted that EPN-repro2 is a combined product [3] of the results provided by different ACs which use different software — Bernese, GIPSY and GAMIT. While the ZTD values estimated by different software agree within around 3 mm, the σ_ZTD values may differ up to three times. This is due to the fact that the initial constraints for the models used by the software packages are different (a dedicated reader may refer to the software manuals). The Bernese and GIPSY show σ_ZTD values usually around 1.5–2 mm while σ_ZTD provided by the GAMIT is around 4–5 mm. Eventually, the higher σ_ZTD values from GAMIT lead to relatively higher values of σ_ZTD in EPN-repro2. An extensive analysis by comparing the differences from diverse GNSS data processing software and data processing strategies on GNSS IPW values itself can be found from [12, 13], where mean values of IPW estimated from GNSS observations agree within 0.5 mm.

The differences between σ_ZTD values derived from different software is a reason why, in addition to using σ_ZTD as given by data providers, the uncertainty of GNSS IPW product could be calculated with fixed σ_ZTD = 4.0 mm (as suggested by IGS [10]). While also σ_ZTD values are available to the CDS users, these values should not be used mechanically. Depending on the application, these uncertainty estimates may need to be scaled according to intercomparison experiments (for example, GNSS versus VLBI, MWR or radiosonde).

Examples

A short overview about the comparison results is given in this section. The following comparisons are made by using the IGS daily tropospheric product only. The IPW estimation from 18 globally distributed IGS stations were compared for the period 2014–2019 with the IPW values calculated from four independent co-located data sets (Fig. 4):

• co-located MWR, retrieved from Onsala MWR ZWDs (January–June 2019 data only);
• ERA5 (2014–2019), using bilinear spatially interpolated values to the locations of the GNSS site (noting that to the extent ERA5 provides a priori constraints to the GNSS post-processing the two series are not entirely independent);
• simultaneous co-located radiosoundings, available in IGRA (for 15 stations, 2014–2019);
• simultaneous co-located radiosoundings from GRUAN (for Tateno and Ny-Ålesund only, 2014–2019).

The co-located stations were determined using two criteria: distance and height difference between stations were <15 km and <50 m, respectively.

...

Image Removed

Figure 4: Map of IGS stations used in comparability study. Station ONSA is co-located with MWR, others are co-located with IGRA or GRUAN radiosonde stations. The IGS stations are divided into two categories based on the distance from the co-located datasets: (1) within 2 km radius (in red) and (2) within 2–15 km radius (in blue).

With only two exceptions out of 18 sites, ERA5−IGS, GRUAN RS−IGS and IGRA−IGS IPW differences follow the same pattern (i.e. have the same sign for specific stations; an example for selected sites is given in Fig. 5). As seen from Table 1 the average IPW bias over all comparison techniques included ranges from 0.15 mm (ERA5–IGS) to 0.64 (MWR–IGS), while its average RMSD ranges from 0.91 mm (GRUAN RS–IGS) to 1.64 mm (IGRA–IGS).

Some researchers have reported a seasonal behaviour of the GNSS–RS IPW bias [14, 15], others have pointed out only a dependency between its RMSD and latitude [16, 17] without seeing a clear effect on biases. Based on the presented examples, there is no dependency between the IPW difference and latitude. On the contrary, a strong correlation between IPW difference, RMSE and latitude is found similarly to previous studies, with highest values at the lowest latitudes.

In addition, a seasonal cycle in the standard deviation of the differences is present for most of the stations, especially at continental stations like ALIC, ANKR, BAKE and PICL. The highest values for standard deviation of differences occur in summertime when IPW reaches its maximum. Based on GNSS and GRUAN RS data collected at NYA1 and TSKB, this kind of seasonal pattern in standard deviation of IPW differences is mostly due the seasonal cycle in the IPW estimated from radiosounding data. Compared to wintertime, the uncertainty of GRUAN RS-estimated IPW in summer is three and five times higher at NYA1 and TSKB, respectively.

The uncertainty of GNSS-estimated IPW was calculated using two different approaches: (1) using σ_ZTD = 4 mm as claimed by IGS and (2) using σ_ZTD values provided by CDDIS. The annual average uncertainties of GNSS-estimated IPW when using σ_ZTD = 4 mm in its calculations are 0.66 and 0.72 mm at NYA1 and TSKB. If using σ_ZTD values provided by CDDIS, the uncertainty decreases to 0.43 and 0.49 mm, respectively. At the same time, GRUAN RS-estimated average IPW uncertainty at these stations was 0.31 and 1.07 mm. It can be concluded that the magnitude of the GRUAN RS and IGS GNSS data uncertainties is sufficiently large to explain the differences between the IPW measurements.

...

Comparison

...

No. of co-located sites

...

Avg. points per site

...

Bias

...

RMSD

...

Avg. [mm]

...

Range [mm]

...

Avg. [mm]

...

Range [mm]

...

ERA5–IGS

...

18

...

46 894

...

0.15

...

-0.9 to 0.93

...

1.46

...

0.93 to 2.6

...

GRUAN RS–IGS

...

2

...

704

...

0.23

...

0.11 to 0.36

...

0.91

...

0.6 to 1.23

...

IGRA–IGS

...

15

...

2 054

...

0.43

...

-0.9 to 1.84

...

1.64

...

0.73 to 2.27

...

MWR–IGS

...

1

...

3 120

...

0.64

...

–

...

1.21

...

–

...

Image Removed
Figure 5: IPW differences in IGRA−IGS, ERA5−IGS, MWR−IGS and GRUAN RS-IGS co-location datasets.

An example of IPW comparison between EPN-repro2, ERA5 and GRUAN RS during 2014 for NYA1 is shown in Fig. 6. While the two datasets have different scope of application, the results obtained using EPN-repro2 are comparable to the outcome derived from IGS tropospheric product. Considering only the year 2014 and IGS data, the average ERA5–GNSS and GRUAN RS–GNSS IPW differences at NYA1 are –0.07 and 0.31 mm (RMSD values 0.67 and 0.6 mm), respectively. When EPN-repro2 is used instead of IGS daily data, these values become 0.02 and 0.44 mm (RMSD values 0.53 and 0.58 mm, respectively). Therefore, it can be concluded that in terms of IPW RMSD, EPN-repro2 shows slightly better agreement with GRUAN RS compared to IGS. This is in line with expectations due to the fact that IGS provides ZTD on a weekly basis with a delay up to 20 days without any reprocessing and homogenization.

...

Image Removed

Figure 6.: IPW differences in ERA5−EPN-repro2 and GRUAN RS−EPN-repro2 co-location datasets at NYA1.

GNSS data and metadata format in the CDS

The CDS web interface provides data in two CSV formats:

• Level-wise rows;
• Observation-wise rows.

The option to download the data NetCDF4 format will become available soon.

All CDS in-situ observations share a common data model (CDM). This format is described in the CDS documentation.

GNSS product analyses and ancillary information for C3S users

...

Image Removed

 Figure 7: Ancillary data for retrieval of GNSS IPW

Being the ancillary data recorded in the same database of the GNSS IPW values, the users may access also the meteorological conditions at the same time of the GNSS IPW retrieval. Moreover, the availability of the IPW values estimated from ERA5 gives the possibility to compare the results (GNSS IPW) with ERA5 reanalysis, which is a common practice in climate studies.

For both IGS and EPN, the site metadata are obtained from SEMISYS and the sites' amsl-values calculated by EGM2008 geoid model.

 IGS and EPN dataset tabular description

Table 2. Metadata table

...

standard name

...

description

...

report_id

...

This parameter starts from 1 for the first data report provided in the data file, and is incremented for each new report.

...

Table 3. Data table 

...

standard name

...

description

...

product of the 3rd IGS reprocessing campaign, performed by IGS ACs. The repro3 product [2] chosen for CDS, covers the period 2000-2019 and is processed by the Center for Orbit Determination in Europe (CODE, https://www.aiub.unibe.ch/research/code___analysis_center/index_eng.html).

Both IGS and EPN follow the same high technical and quality assessment requirements. The EPN-repro2, the product of the 2nd EPN reprocessing campaign, similar to the IGS dataset, is also of reference quality [3]. It consists of 18+ years of GNSS data belonging to the EUREF Permanent Network (EPN,http://www.epncb.oma.be/), making it a valuable database for the development of a climate data record of GNSS tropospheric products over Europe. For EPN-repro2, five ACs homogenously reprocessed the EPN network for the period 1996–2014. Finally, the individual contributions of each AC are combined to provide the official EPN reprocessed products. The combined product is described along with its evaluation against radiosonde data and European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA-Interim) data by Pacione, et al. [4].

Meteorological information used as auxiliary data required for the IPW estimation is obtained from the ECMWF ERA5 atmospheric reanalysis. The methods for requesting data co-located with the GNSS sites can be found in the ATBD document. The IPW derived from the reprocessed tropospheric products can be recommended for climate research.

In the following sections, a short retrospective overview is given about GNSS IPW quality assurances, data and metadata sources (Section 2), data processing (Section 3), data validation process (Section 4), examples of IPW comparison (Section 5), data and metadata format in CDS (Section 6), GNSS product analysis and ancillary products (Section 7) as well as data licensing (Section 8). The maturity matrix of the IGS network can be found in Appendix A. Description of data units and conversion is given in Appendix B.

Data and metadata sources

 IGS-daily

The daily (NRT) tropospheric products for IGS sites (Fig. 1) are publicly available through the CDDIS data portalhttps://cddis.nasa.gov/archive/gnss/products/troposphere/zpd/. The troposphere products utilize the IGS final satellite, orbit, and Earth Orientation Parameters (EOP) products and are therefore available approximately three weeks following the observation day. Since the inception of the troposphere product, three analysis groups (GFZ, JPL, and currently USNO) have generated the IGS combination solution each using different methodologies. Therefore, the resulting file types have changed through the years.The current product consists of files containing daily ZTD estimates for selected sites in the IGS network. The original product consisted of one file per site per GPS week [https://igs.org/products/#troposphere].

Image Added

Figure 1:  A map showcasing the worldwide locations of IGS sites with both NRT and reprocessed datasets accessible through CDS, along with sites offering access to EPN repro2 data.

The GNSS tropospheric products available at CDDIS are delivered by the IGS ACs in a unified SINEX TRO format (description available at https://files.igs.org/pub/data/format/sinex_tropo.txt). For IGS-daily the ZTD time series are still not corrected for any effects derived from instrumental changes (receivers, antennas, radomes) or changes in the station environment. This kind of analysis needs extensive data reprocessing. A comprehensive description of the IGS GNSS troposphere data products can be found in the ATBD document. While not best suited for long-term climate studies, the IGS daily dataset is relevant in atmospheric analyses due to its worldwide network and relatively short data latency (ca 2-3 weeks).

IGS-repro3

The reprocessed GNSS IWV time series in CDS, which relies on tropospheric products from IGS, is denoted as IGS-repro3. The ZTD data utilized for these series can be accessed at https://cddis.nasa.gov/archive/gnss/products/repro3 , and are also accessible through a few other IGS global data centers. IGS-repro3 consists of the results of 10 ACs (https://igs.org/news/repro3-solutions-now-available/ ), each processing its own set of GNSS sites according to their best practices and not all of them providing the corresponding tropospheric products. It must be noted that the IGS-repro3 does not resemble a combined product but, instead, the combination of the single dataset provided by each AC.

Details about the available products, main updates in the modelling since the repro2 version, and the combination strategy can be found at https://lists.igs.org/pipermail/igsmail/2021/008022.html. 

ZTD screening, which is the process of inspecting data for errors and correcting them before performing data analysis, has been applied to all initial ZTD time series [3, 5]. Data reprocessing removes all estimated shifts (biases) from GNSS antenna replacements [5]. Therefore, this data record can be used as a reference for various scientific applications (e.g. validation of regional numerical weather prediction reanalyses and climate model simulations) and has a high potential for monitoring trends and the variability in atmospheric water vapour.

EPN-repro2

GNSS IPW retrieved from the reprocessed ZTD time series from EPN-repro2 (http://www.epncb.oma.be/_productsservices/analysiscentres/repro2.php) is offered as the third product to the CDS users. In the framework of the EPN-repro2, the second reprocessing campaign of the EPN, five ACs homogeneously reprocessed the EPN network (Fig. 1) for the period 1996–2014 [3]. The reprocessing techniques conducted and the quality control applied by EPN ACs closely resemble those employed by IGS ACs. EPN-repro2 is known as a reference quality data product [3]. The ZTD dataset is freely available from BKG (Bundesamt für Kartographie und Geodäsie): ftp://igs.bkg.bund.de/EPNrepro2/products/. 

The EPN-repro2 data is recorded by GNSS weeks (834–1825) with combined solutions recorded in SINEX TRO mixed format. Unlike CDDIS for IGS, the EPN ftp service supports traditional anonymous data access. One of the major differences compared to IGS datasets is that EPN-repro2 data is not recorded at full hours but with a 30-minute shift (00:30, 01:30, …, 23:30).

Contrary to IGS-repro3, the EPN-repro2 is a combined product. The daily ZTDs provided by the ACs are combined weekly. Only sites with a corresponding coordinate solution and with three individual AC contributions are presented. Estimates with a given STDDEV > 15 mm are excluded. Rough outlier detection is performed to find strong outliers. Details on the combination process can be found in [5].

Metadata

The metadata consists of additional necessary data used in estimating GNSS IPW and its uncertainty, described in the ATBD document. The GNSS-related metadata can be retrieved from GNSS sites’ specifications (logfiles) and the headers of SINEX TRO files (described in M311a_Lot3.3.2.3_2020, Section 4.1 and athttps://kb.igs.org/hc/en-us/articles/201096516-IGS-Formats). However, the technical implementation of the IPW software package uses SEMISYS (the service of GFZ, referred to in the product availability and licenses section).

For estimating GNSS IPW based on ZTD, ancillary meteorological data are needed. These are obtained from the ERA5 reanalysis dataset available via the CDS. Geopotential, specific humidity and temperature values are downloaded for 37 pressure levels to calculate the water-vapor-weighted mean temperature of the atmosphere, Tm, which is crucial for converting ZTD to IPW.

GNSS data processing and retrieval of IPW

The GNSS IPW provided through the CDS is estimated based on using tropospheric products delivered by IGS data ACs and EPN-repro2. The data processing schema for retrieval of GNSS IPW for both IGS and EPN tropospheric products is described in more detail in the ATBD document. Here only a simplified description of the data flow and data processing steps for retrieval of GNSS IPW from GNSS tropospheric products is depicted (Fig. 2).

Image Added

Figure 2: Retrieval of IPW for IGS and EPN-repro2

• Pre-processing of GNSS troposphere product (steps 1–4). First, ZTD with its uncertainty is downloaded from IGS and EPN data repositories (steps 1–2, Fig. 3). The site metadata (coordinates, altitude, city, agency etc.) for both networks are obtained from SEMISYS [6] (step 3). Due to the fact that the site latitude and longitude are given with height from WGS84 ellipsoid, the AMSL (height above the mean sea level) used in IPW retrieval, must be calculated first (step 4).
• Pre-processing of ancillary data (steps 5–7). The supporting meteorological data at all ERA5 pressure levels and at ground surface level are downloaded, the water-vapour-weighted mean temperature of the atmosphere (T_m) is calculated and bilinearly interpolated to the site coordinates and altitude (step 5). Since the time resolution of both, ERA5 and IGS ZTDs, is one hour there is no need to interpolate T_m and ground surface pressure (p₀) in time. On the other hand, this is done in the case of EPN-repro2 since its data are provided at half-hours. Depending on dry gases between receiver and satellite, the ZHD is calculated using p₀, latitude, and height of the site above the mean sea level (step 6). The ZWD, on the other hand, depends on the water vapour amount and is found by subtracting ZHD from ZTD. The conversion factor that is needed to relate IPW to the ZWD is calculated using a function of T_m (step 7).
• Calculating IPW and its uncertainty (steps 8–10). IPW estimation (in units kg m^-2, equivalent to mm) is given using ZWD and the conversion factor (step 8). For calculating GNSS IPW uncertainty (step 9), the approach chosen for C3S_311a_Lot3 is based on the GRUAN GNSS data processing [7], and accounts for all of the uncertainty sources due to the measurement systems. The IPW uncertainty calculation relies on previously published values for input variables used as constants and their uncertainties [8]. In case of uncertainties for p₀ and T_m, these values, statistically determined using IGRA and GRUAN radiosonde measurements, are in good agreement with the results presented in several papers specifically focusing on ERA5 [9, 10]. After the values for IPW and its uncertainty are calculated, the results are ingested into the database (step 10).

Data validation process

 IGS

IGS daily

The integrity of the data in SINEX TRO files is guaranteed by the IGS ACs. However, to ensure the data quality for retrieval of GNSS IPW and its uncertainty (methodology described in Ning et al., 2016 [7]) the data processing chain for CDS comprises additional steps as follows:

• Checking ZTD values for the range. The valid range is [1.00, 3.00] m;
• Checking formal error of ZTD (σ_ZTD). A ZTD value is accepted if σ_ZTD <= 10 mm or σ_ZTD < 2.5 · median (the latter is applied only if the 24-hour median < 3.0 mm)

If any of these checks fail, the ZTD or σ_ZTD will be assigned a “null” value (indicating a missing or unacceptable value).

These checks are similar to those used by a screening of ZTD time series for IGS repro-products [3, 5]. However, it must be noted, that a range check on formal errors should be station-specific [3] and the time series have not passed all the procedures applied on reanalysis (e.g., EPN-repro2).

IGS sites’ metadata (site coordinates) are checked online by SEMISYS and corrected by proofed values from SEMISYS if the horizontal displacement exceeds 90 meters (equivalent to 3 arc seconds). Coordinate check avoids using erroneous AMSL values from a geoid model.

IGS-repro3

For IGS-repro3, although a reprocessed product, the same QC algorithm is applied for ZTD and σZTD values as in case of IGS daily. The epochs with data not passed QC are unexposed. The same site metadata and ancillary meteorological data from ERA5 are used for both IGS daily and IGS-repro3.

EPN

EPN-repro2 time series do not need any additional quality checks. All necessary quality checks are performed already during reprocessing. 

Unlike many other measurements in environmental physics (atmospheric temperature, humidity, etc.) the GNSS IPW is retrieved from GNSS ZTDs characterised with uncertainties expressed in terms of formal (not instrumental) errors.

The characteristic values depend on GNSS data processing software and the IGS specifies σ_ZTD limits to 4 mm [11] for observations in mostly ideal conditions with instrumental installations following strict IGS technical requirements. The sources of uncertainties contributing to the final σ_IPW are extensively analysed in Ning et al., 2016 [7] and in GAIA-CLIM D2.8, Annex IX, Product Traceability and Uncertainty for the GNSS IPW Product [12].

Comparing σ_ZTD values between IGS and EPN-repro2, the user may notice remarkable differences. In the case of IGS, these values stay around 2 mm. At the same time, EPN-repro2 σ_ZTD values can reach 5–6 mm. As a direct consequence of this (σZTD contributes over 75% to the total IPW uncertainty), the total σIPW values may reach slightly over 1 mm. It should be noted that EPN-repro2 is a combined product [4] of the results provided by different ACs that use different software — Bernese, GIPSY, and GAMIT. While the ZTD values estimated by this software agree within around 3 mm, the σZTD values may differ up to three times. This is because the initial constraints for the models used by this software are different (a dedicated reader may refer to the software manuals). The Bernese and GIPSY show σ_ZTD values usually around 1.5–2 mm while σ_ZTD provided by the GAMIT is around 4–5 mm. Eventually, the higher σ_ZTD values from GAMIT lead to relatively higher values of σ_ZTD in EPN-repro2.

The differences between σ_ZTD values derived from different software is a reason why, in addition to using σ_ZTD as given by data providers, the uncertainty of GNSS IPW product is calculated with fixed σ_ZTD = 4.0 mm (as suggested by IGS [11]). While also σ_ZTD values are available to the CDS users, these values should not be used mechanically. Depending on the application, these uncertainty estimates may need to be scaled according to intercomparison experiments (for example, GNSS versus VLBI, MWR or radiosonde).

An extensive analysis by comparing the differences from diverse GNSS data processing software and data processing strategies on GNSS IPW values itself can be found in [13, 14], where mean values of IPW estimated from GNSS observations agree within 0.5 mm.

Examples

Impact of reprocessing

The IGS NRT (daily) dataset provides global coverage with relatively short latency (about 2-3 weeks) and, therefore it is not the ideal choice for climate studies. Reprocessed GNSS datasets, which account for instrument and data processing changes, are preferred for such research. Figure 3 illustrates the effect of correcting instrument-related biases on IPW for station BAMF (Bamfield, Canada, 48.84°N, -125.14°E).

Image Added

Figure 3. Differences between IGS Near Real-Time (NRT) and ERA5 data, as well as IGS Repro3 and ERA5 data, both on an hourly and monthly basis, at the BAMF station. The most substantial discrepancies in the IGS NRT data become evident from March 2018 to February 2020. This noticeable change aligns with the installation of the new receiver, as indicated by the accompanying metadata on the right. Notably, such a significant shift is not observed in the reprocessed IGS time series. 

Comparison with third-party data

For a thorough examination of comparison results, which encompasses IGS repro3, EPN repro2, IGS daily, RHARM, IGRA, GRUAN, and ERA5 datasets for the period 2000–2021, we recommend reading a paper authored by Rannat et al. [15].

The following section provides a concise summary of the comparison results obtained from IGS daily, EPN repro2, radiosonde data, and ERA5. The IPW estimation from 18 globally representative IGS stations was compared for the period 2014–2019 with the IPW values calculated from four independent co-located data sets (Fig. 4):

• co-located MWR, retrieved from Onsala MWR ZWDs (January–June 2019 data only);
• ERA5 (2014–2019), using bilinear spatially interpolated values to the locations of the GNSS site (noting that to the extent ERA5 provides a priori constraints to the GNSS post-processing the two series are not entirely independent);
• simultaneous co-located radiosoundings, available in IGRA (for 15 stations, 2014–2019);
• simultaneous co-located radiosoundings from GRUAN (for Tateno and Ny-Ålesund only, 2014–2019).

The IPW datasets from EPN repro2, GRUAN radiosonde and ERA5 were compared based on data collected in 2014 at Ny-Ålesund.

The co-located stations were determined using two criteria: distance and height difference between stations were <15 km and <50 m, respectively.

Image Added

Figure 4: Map of IGS stations used in comparability study. Station ONSA is co-located with MWR, others are co-located with IGRA or GRUAN radiosonde stations. The IGS stations are divided into two categories based on the distance from the co-located datasets: (1) within 2 km radius (in red) and (2) within 2–15 km radius (in blue).

With only two exceptions out of 18 sites, ERA5−IGS daily, GRUAN RS−IGS daily and IGRA−IGS daily IPW differences follow the same pattern (i.e. have the same sign for specific stations; an example for selected sites is given in Fig. 5). As seen from Table 1 the average IPW bias overall comparison techniques included ranges from 0.15 mm (ERA5–IGS daily) to 0.64 (MWR–IGS daily), while its average RMSD ranges from 0.91 mm (GRUAN RS–IGS daily) to 1.64 mm (IGRA–IGS daily).

Some researchers have reported a seasonal behavior of the GNSS–RS IPW bias [16, 17], others have pointed out only a dependency between its RMSD and latitude [18, 19] without seeing a clear effect on biases. Based on the presented examples, there is no dependency between the IPW difference and latitude. On the contrary, a strong correlation between IPW difference, RMSE, and latitude is found to be similar to previous studies, with the highest values at the lowest latitudes.

In addition, a seasonal cycle in the standard deviation of the differences is present for most of the stations, especially at continental stations like ALIC, ANKR, BAKE, and PICL. The highest values for the standard deviation of differences occur in the summertime when IPW reaches its maximum. Based on GNSS and GRUAN RS data collected at NYA1 and TSKB, this kind of seasonal pattern in the standard deviation of IPW differences is mostly due to the seasonal cycle in the IPW estimated from radiosounding data. Compared to wintertime, the uncertainty of GRUAN RS-estimated IPW in summer is three and five times higher at NYA1 and TSKB, respectively.

The uncertainty of GNSS-estimated IPW was calculated using two different approaches: (1) using σ_ZTD = 4 mm as claimed by IGS and (2) using σ_ZTD values provided by CDDIS. The annual average uncertainties of GNSS-estimated IPW when using σ_ZTD = 4 mm in its calculations are 0.66 and 0.72 mm at NYA1 and TSKB. If using σ_ZTD values provided by CDDIS, the uncertainty decreases to 0.43 and 0.49 mm, respectively. At the same time, GRUAN RS-estimated average IPW uncertainty at these stations was 0.31 and 1.07 mm. It can be concluded that the magnitude of the GRUAN RS and IGS GNSS data uncertainties is sufficiently large to explain the differences between the IPW measurements.