SoftwareX 22 (2023) 101372

Contents lists available at ScienceDirect

SoftwareX
journal homepage: www.elsevier.com/locate/softx

Original software publication

RINEXAV: GNSS global network selection open-source software based

on qualitative analysis of RINEX files
∗
Filip Gałdyn , Radosław Zajdel, Krzysztof Sośnica
Wrocław University of Environmental and Life Sciences, Institute of Geodesy and Geoinformatics, Wrocław, Poland

article info a b s t r a c t

Article history: The determination of global geodetic parameters using Global Navigation Satellite System (GNSS)
Received 7 December 2022 observations requires using a global, evenly distributed network of stations. The homogeneous
Received in revised form 14 March 2023 distribution of stations is crucial for deriving Earth rotation parameters, coordinates of Earth’s center
Accepted 14 March 2023
of mass, orbit determination, and the realization of the International Terrestrial Reference Systems.
It is necessary to select the most representative group of stations, providing GNSS observations of
Dataset link: https://cddis.nasa.gov/archive
/gnss/data/daily/, https://github.com/RINEX the best quality, i.e. taking into account the factors, such as the number of observations, measured
AV/Rinexav-py-modules/ signal-to-noise ratio, or signal multipath indicator. This paper provides a description of the open-
source software RINEXAV to select and optimize the GNSS station network based on the qualitative
Keywords: analysis of RINEX files. The provided software allows users to maximize the number of observations
IGS based on specified criteria, maintain the uniform distribution of stations, and automatically select the
GNSS most suitable stations from among those available. To accomplish this, the software uses multi-criteria
MGEX
decision-making and clustering methods. The paper provides the description, use and impact of the
Data mining
Geodesy
RINEXAV software in geosciences.
© 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).

Code metadata

Current code version v.1.0

Permanent link to code/repository used for this code version https://github.com/ElsevierSoftwareX/SOFTX-D-22-00412
Code Ocean compute capsule
Legal Code License MIT License, Copyright (c) 2022 RINEXAV
Code versioning system used git
Software code languages, tools, and services used Python 3
Compilation requirements, operating environments & dependencies Python Packages: Numpy, Pandas, Streamlit, sklearn
If available Link to developer documentation/manual
Support email for questions [email protected]

1. Motivation and significance global networks, e.g., those operated by the IGS, are located in
Europe and America, as opposed to the Africa, Asia, or Oceania
Currently, the International Global Navigation Satellite Sys- regions. GNSS observations play a crucial role in geosciences. De-
tems (GNSS) Service (IGS) [1] has a well-developed global station termination of global geodetic parameters, such as Earth rotation
network that includes over 500 stations in total, from which parameters, coordinates of the Earth’s center of mass, orbit de-
more than 360 allow tracking multi-GNSS signals; Multi-GNSS termination, and realization of a terrestrial reference frame using
Experiment (MGEX) [2]. There is also a large number of perma- GNSS observations require the use of a global, evenly distributed
nent GNSS stations around the world operating within regional network of permanent GNSS stations [4]. The selection of a proper
and national networks [3]. The largest number of stations of network is a prerequisite for the representation of the Earth in the
form of a discrete global grid and has a profound impact on the
∗ Corresponding author. geophysical interpretation of information behind the temporal
changes of the estimated parameters.
E-mail addresses: [email protected] (Filip Gałdyn),
[email protected] (Radosław Zajdel), [email protected] The complexity of the calculation processes prevents the use
(Krzysztof Sośnica). of all available GNSS stations for the calculation of global geodetic

https://doi.org/10.1016/j.softx.2023.101372
2352-7110/© 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Filip Gałdyn, Radosław Zajdel and Krzysztof Sośnica SoftwareX 22 (2023) 101372

Fig. 1. A simplified diagram that shows the capabilities of the RINEXAV software, 1 (blue) - first version, 2 (red) - second.

parameters. These are, in particular, computation time and mem- GFZ, the station selection is based on the pre-selection and station
ory. Therefore, it is crucial to select the most representative group classification complemented with checks for formal correctness
of stations, which provide the GNSS observations of the highest and consistency of the RINEX files with the official site logs.
quality [5–8]. However, there is a lack of details on how this task is actually
The joint use of multi-criteria decision-making algorithms and performed.
clustering methods, e.g., k-means, can help make a quick and This study provides the RINEXAV software, the first open-
quality-sensitive selection of stations. The article presents the source software for automatic multi-purpose station selection
RINEXAV software, which is a GNSS permanent station selec- based on the geometric and qualitative criteria adopted by the
tion software dedicated to users who need to find an optimal user. Section 2 provides description of the architecture of RINEXAV
network of GNSS sites from among those available for various software and gives insight in the details of the station selection
scientific purposes, but especially focusing on the support of methodology. Section 2.1 focuses on the process of downloading
determining the global geodetic parameters. RINEXAV consists of data and creating status files based on the quality assessment
two main modules. The role of the first is to retrieve Receiver of the RINEX files. Section 2.2 provides the key features and
Independent Exchange System (RINEX) files collected in the user assumptions associated with the decision-making algorithm for
database [9] and create daily status files on the availability and selecting a network of stations. The description of the graphical
quality of RINEX files. This is possible using BKG NTrip Client user interface (GUI) and Python functions are demonstrated in
2.12.17 [10] and GFZRNX [11] programs. The second module Sections 3.1 and 3.2. Finally, Section 3.3 shows an example of how
allows for analyzing the available network in the selected time the algorithm works on a single cluster basis.
interval and making a decision based on the quality parameters
as acquired in the first module. The outcome of the first module 2. Software description
is a precondition for the initialization of the second module
operation. Every study related to the determination of global geode-
Some authors proposed methods for the automatic selection tic parameters, precise orbit determination, or the realization
of stations based on the geometrical parameters aiming to im- of terrestrial reference systems based on GNSS measurements,
prove some GNSS-based global parameters. Zhang [12] used the relies on the selected global network of stations that provides
orbit dilution of precision (ODOP) of ground stations searching raw observations. Of all the available, publicly available, and
for the geometrically best network of stations for the precise documented GNSS-related tools, to the knowledge of the authors,
orbit determination. Furthermore, the geometrical DOP (GDOP) none offers a solution leading to the selection of such a network
of stations with respect to the geocenter was used by Xue and automatically, based also on signal quality parameters. Due to
Yang [13] aiming to facilitate the determination of global geodetic the demand, software has been implemented in the Python pro-
parameters. On the other hand, Yang et al. [14] developed a gramming language using a number of libraries (numpy, pandas,
method of global network selection considering both geometri- scikit-learn). The software consists of two modules. The general
cal and qualitative factors of stations based on the grid control scheme of the modules is shown in Fig. 1. The first is responsible
probabilities and random optimization algorithm. The automatic for the quality assessment of RINEX files creating daily status
algorithms of station selection were also proposed for the long- files. Due to the fact that the current studies related to the
term ionospheric anomaly monitoring based on the optimization determination of global geodetic parameters and the realization
of thresholds for station quality monitoring [15,16]. of terrestrial reference systems are produced largely on the basis
Most frequently, information about the station selection algo- of data from the network of IGS stations, two versions of this
rithm used in the specific case study is simply missing and prob- module were made. The first one uses the ability to connect
ably performed manually based on the researcher’s experience. via FTP server to the Crustal Dynamics Data Information System
In the case of IGS Analysis Centers (ACs), which deliver official (CDDIS) database. This allows the user to download all the data
GNSS-based products, the station selection is often based on the from the IGS network at a time interval selected by the user and
priority list agreed upon between the ACs, such as IGS Repro3 if the station is receiving multi-GNSS signals. The second version
priority list [17] or EPN station classification document1 , consid- of the first module allows the user to analyze 30-second RINEX
ering aspects, such as co-location sites with other space-geodetic files of their choice from any station network by placing them in a
techniques, long history of the time series of observations. For dedicated folder \src\status_file_from_folder\data. If the user wants
the Center for Orbit Determination in Europe (CODE) AC [18], to use one of the versions, a different tool must be launched. They
the global distribution, availability, and content of observation are marked in red and blue and indicated by numbers 1 and 2
files are also evaluated prior to the definitive station selection. respectively in Figs. 1 and 2. The second module is responsible
Maennel et al. [19] explain that for the IGS products provided by for deciding on the best possible network of stations in a given
time interval based on the status files from the first module.
1 http://www.epncb.oma.be/_productsservices/ReferenceFrame/Station_ This is done by maintaining the uniformity of distribution of the
Classification.php network.
2
Filip Gałdyn, Radosław Zajdel and Krzysztof Sośnica SoftwareX 22 (2023) 101372

Fig. 2. The data flow scheme during status files creation, 1 (blue) - first version, 2 (red) - second version.

2.1. Quality assessment of RINEX files motion, requires using a network of stations that adequately
represent the Earth’s shape. The inaccurate representation of the
Fig. 2 shows the data flow in the module that creates daily Earth’s figure with the geodetic network may lead to so-called
status files. The entire module provides the ability to create daily ‘network effects’ in the estimated parameters [5,24]. When con-
status files related to the availability, and quality of RINEX files. sidering the scale realization, phase center antenna calibrations
The first version of the module allows direct downloading of would be a key factor [25,26]. Other regional or local research
RINEX files from IGS network stations. For the user, it is possi- purposes may impose further requirements, e.g., multipath se-
ble to download them from the CDDIS. Then, due to Hatanaka lection criteria for sensing snow height and surface temperature
compression of the RINEX file [20], unpacking and conversion to variations [27].
RINEX format are performed with the RNXCMP program in ver- Multi-criteria decision-making is a complex process, and with
sion 4.1.0.2 The file in RINEX format in the next step is processed a large number of observations and conflicting criteria, it re-
in two programs, GFZRNX and BKG NTrip Client, from which re- quires advanced algorithms to support the process [28]. The
ports summarizing statistics are generated. The outputs are then application of multi-criteria decision-making algorithms is broad
processed, aiming to extract information about the availability and can be used to solve any problem where a decision needs
of observations for each satellite and the quality parameters of to be made based on multiple factors [29]. Each of the multi-
the observations at each available frequency. This information is criteria decision-making methods has its procedural approach
saved in comma-separated values (CSV) format, which is called according to which rankings are carried out and selection alterna-
status file. tives are determined. These methods have been applied to many
actual problems in recent years, e.g., business and marketing
2.2. Decision-making algorithm for selecting an optimum network of management or supply chain management and logistics [30].
stations Spatial data clustering is a process of grouping a set of objects
into classes or clusters in such a way that objects within one
cluster are very similar to each other, but they are not similar
The selection of an optimal network of stations depends on
to objects in any other cluster [31]. Using spatial data clustering
the application, and one should consider different criteria. When
algorithms and the coordinates of individual stations, it is pos-
realizing the particular datum, the regional overabundance of
sible to group stations into clusters that are close to each other
observations, as well as the asymmetry between hemispheres,
in terms of distance. The k-means algorithm [32] is a simple and
can disturb its realization [21–23]. Determination of global geode-
efficient algorithm for finding clusters in data. It is one of the most
tic parameters, such as Earth rotation parameters or geocenter
popular methods for iterative descending clustering. It is designed
for quantitative variables, and the dissimilarity is calculated using
2 https://terras.gsi.go.jp/ja/crx2rnx.html the square of the Euclidean distance. The k-means algorithm is
3
Filip Gałdyn, Radosław Zajdel and Krzysztof Sośnica SoftwareX 22 (2023) 101372

Fig. 3. The data flow scheme for data analysis and station selection.

characterized by calculations performed relative to the center of a value is stored for a given channel (e.g. for C2I). The aggregation
the cluster. In the case of the hierarchical clustering algorithm, of quality parameters for each frequency narrows the scope of the
it is based on the composition of a hierarchy in which clusters analysis, but also allows the algorithm to be faster and simplify
at each level of the hierarchy are created by merging or splitting the subsequent calculations. The frequency that has the highest
clusters at the next lower level [33]. number of observations at the selected time is chosen for further
The created status files provide the dataset, which is analyzed steps. The full process workflow is illustrated in Fig. 3.
for the optimum selection of a global GNSS network. The example The RINEXAV software is designed with the user’s choice of
files are located in the \src\make_decision\rv3_stats folder. The other parameters as well. It is possible to choose a clustering and
RINEXAV software is designed to analyze files from any time multi-criteria decision method. Due to the fact that in each time
period. The user can also select frequencies at which the signals interval selected by the user the network of available stations
are received. The combined observations with special properties may change, but also due to the possibility of the user to select
can be formed by the linear combination of multi-frequency any number of stations, the k-means [32] and the hierarchical
carrier-phase and pseudorange observations. The characteristics clustering method [33] were implemented for this purpose after
of the combined observations are directly affected by the type studying the operation of various methods. We excluded other
and number of combined frequencies [34]. Zajdel et al. [25], con- algorithms, such as DBSCAN [36] search for core objects, i.e. ob-
sidered using different signals for forming ionosphere-free linear jects that have dense neighborhoods and operate automatically
combinations namely B1I/B3I and B1C/B2a for the realization without being able to specify the number of clusters. Currently,
of the scale of terrestrial reference frame. Strasser et al. [35] the chosen algorithms have many scientific applications also in
demonstrated that considering a network of stations that tracks the field of geoinformatics [37,38]. Once the time interval and the
as many signals as possible is beneficial for the determination number of stations have been selected, the clustering algorithm
of global geodetic parameters in the so-called raw processing creates, from all the available stations, the clusters (groups) that
approach. Using a multi-frequency approach preserves the orig- are closest to each other in terms of spatial distance. The user
inal measurement accuracy and allows full exploitation of the can choose between two available algorithms, k-means and hier-
information contained in each individual observation type. archical clustering. The algorithm splits the whole set of stations
The selection of particular frequencies is of key interest to into groups, which are then used in the next step to perform
potential software users. Once the time period and the frequen- the decision-making process of the best station in each group.
cies are selected, the files are processed. The average quality A multi-criteria decision algorithm is then used; the choice of
parameters and availability of RINEX files at a given station at which depends on the user (TOPSIS or COPRAS). For each group,
the selected frequencies are calculated. The RINEX file availability the algorithm ranks the stations from the best to the worst for
is calculated from the availability files as the ratio between the later selection. This process is carried out on the basis of the
number of available RINEX files to the number of specified days in data contained in the daily status files of the specified period.
the analysis for each station. The user may have 30 values for each The decision is made taking into account the signal quality pa-
quality parameter such as signal-to-noise ratio (SNR), multipath, rameters, which are: SNR, multipath, number of observations, and
etc. if specifying 30 days of data on a given channel e.g. C2I. The number of gaps. All these parameters are calculated by the BNC
arithmetic mean of each quality parameter is then calculated and software, thus, we refer to the documentation to get more insight
4
Filip Gałdyn, Radosław Zajdel and Krzysztof Sośnica SoftwareX 22 (2023) 101372

Fig. 4. Exemplary selection of 40 stations (red) with observations in the period 01.01.2021–01.03.2021 on the GPS L1 with RINEX file availability above 90%, even
global distribution and the best quality of RINEX files. Each cluster from which the station was selected is marked using a different color. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)

into the algorithm for calculating the individual parameters [39]. with RINEX file availability above 90% with k-means clustering
In addition to all the parameters that are acquired at any given method and the TOPSIS decision algorithm.
time by the measurement made by the receiver located at the The output of module 2 results in the selection of a network
station, external studies can also provide evidence of quality that is characterized by global homogeneity and the best possible
and availability. For this purpose, we used the classification of signal quality in each group. Due to the algorithmic selection,
the global network of GNSS stations made by Dr. Paul Rebis- objectivity is also maintained when selecting the best of the
chung [17]. The classification was carried out in preparation for stations. This means that the selection was made according to
the IGS Repro3 initiative, a direct IGS contribution to International the facts, without any user bias, influence of opinion, or interest
Terrestrial Reference Frame (ITRF) 2020 [40]. This classification in the subject of station quality. The algorithm also provides an
was based on an analysis of the full history of GNSS data collected opportunity for the user to retrieve station names along with
by the IGS global network from 1994 to early 2019. The station average statistics. This is possible in the form of a table containing
were classified in subsequent priority level groups according to statistics for the selected period both before and after selection.
their significance for the terrestrial reference frame realization.
Each priority level is described and specified in a document (IGS 3. Illustrative examples
Repro3 station priority list3 ). The user can specify the priorities
of the stations as one of the criteria of the decision-making The created selection algorithm is a fully self-sufficient tool
algorithm. for performing analysis related to signal quality and observation
In addition, each parameter can be given weight by the soft- from a given network. The provided modules, in combination
ware user, which will influence the software decision-making. For with each other, can analyze the data and select the best stations
example, if the user wants to get a network of stations with as while maintaining their uniform distribution. The program makes
many observations as possible, user needs to indicate a higher the selection according to the parameters set by the user on a
weight for this parameter. RINEXAV allows choosing between two database placed in a folder, which ensures maximum simplicity
decision-making methods: TOPSIS (Technique for Order Prefer- for the end user. The first module operates in two versions and
ence by Similarity to Ideal Solution; [41]) and COPRAS (Complex when being used with own user dataset of RINEX files, the user
Proportional Assessment; [42]). They are based on calculating must remember to complete the file with information about the
the euclidean distance to the ideal solution and the least ideal stations (X, Y, Z, longitude, latitude, height, priority_level). This
solution. Then, the value evaluated as the best choice must be step must be accomplished before the second module can take
closest to the ideal solution and at the same time farthest from effect. The automatic version, which allows downloading data
the least ideal solution. The ideal solution is one that maximizes directly from CDDIS, also requires entering the date or time range
the benefit criteria and minimizes the cost criteria, while the for which the user wants to retrieve the data.
least ideal solution maximizes the cost criteria and minimizes the Module 2 can be used from the Python function console mode
benefit criteria. Both methods are based on a decision matrix and and from the Streamlit GUI. Section 3 will describe these ca-
the weights entered into it. However, the methods differ in the pabilities and show an example of station selection in a single
final calculation of the indicators on which the ranking is based. cluster.
In the case of TOPSIS, this is the overall preference score, and in
the case of COPRAS, the relative significance value. These methods 3.1. Making decision using Python script version
were chosen because of their frequent occurrence in research [43,
44], the possibility of ranking alternatives with them, and the
To run the algorithm that decides which station to select for a
possibility of weighting parameters. The result of a sequence of
given time period, use the files from \src\make_decision and open
steps (Fig. 3) is a set of stations. Fig. 4 shows the result of such
make_decision.py function. The function requires all status files
a selection. The user query concerned 40 stations selected with
from the selected period in the rv3_stat folder. Before running
observations in the period 01.01.2021–01.03.2021 on the GPS L1
the script, the user needs to customize the variables defined at
the beginning of the make_decision.py function code. To allow
3 http://acc.igs.org/repro3/repro3.html the user to configure the settings of the decision algorithm, the
5
Filip Gałdyn, Radosław Zajdel and Krzysztof Sośnica SoftwareX 22 (2023) 101372

Fig. 5. Example configuration - (a) and (b); the resulted outcome - (c).

settings in the \src\make_decision\config.yml file must be com- Table 1

pleted. Parameters to be set include the time period, the systems Statistics of the parameters of selected stations in an exemplary cluster receiving
the BeiDou-3 B1 for the first 60 days of 2022.
from which frequencies are used, the signals used, the weights
Station Prior. No. obs. SNR Gaps Multipath Rank
of each parameter, the percentage availability of RINEX files, the
clustering method, the multi-criteria decision method, and the WROC00POL 5 50757 45.57 120 0.136 1
GRAZ00AUT 2 45054 42.46 61 0.556 2
number of stations after selection. WARN00DEU 5 53483 44.70 112 0.477 3
LEIJ00DEU 11 53282 44.80 57 0.436 4
3.2. Making decision using Streamlit GUI OBE400DEU 11 35909 42.51 51 0.387 5
GOP600CZE 11 32893 42.02 116 0.424 6
BOR100POL 2 24815 41.97 241 0.586 7
An open-source app using the Streamlit framework was cre-
BAUT00DEU 11 52512 44.63 191 0.459 8
ated in order to make using the software easier for the end users. POTS00DEU 2 50872 40.91 398 0.606 9
A user can run the GUI interface from the file \src\make_decision GOPE00CZE 11 42720 41.55 361 0.623 10
\streamlit_app.py. The interface can be triggered by running a
command streamlit run streamlit_app.py from the console. The
application is displayed in a default internet browser window
environment on the local host. The selection of parameters is predetermined parameters such as the level of importance of the
done through a user-friendly interface, and the results, in addition station, the number of observations, the signal-to-noise ratio, the
to interactive lists of stations, are displayed in the form of maps. number of gaps in the signals, and the multipath.
Fig. 5 shows the window of an example configuration and the In Fig. 6, the red circles mark the stations that have been
ranked the best among all the competitors and are the first
resulting outcome. The whole software is run using localhost and
choice for the user in terms of quality. This selection is made
a web browser. The sections labeled a and b are the example
using the TOPSIS method, assuming the identity of the weight
configuration as proposed through the config.yml file. The part
of each parameter. The green polygon highlights the cluster, for
marked c is the result in the form of tables and maps of the
which the next analysis will be discussed in detail. It contains
station network.
10 stations, which cover the area of western Poland, Eastern
Germany, the Czech Republic, and Austria. Within this group, the
3.3. Example of software use case
stations are considered individually based on the multi-criteria
analysis. Table 1 shows the average quality parameters occurring
According to the methodology, the decision is made based on
in the analyzed interval and the station priority level parameter.
the ranking of each cluster. This example focuses on a global
The last column records the ranking place that the algorithm
network of 232 stations that provided observations on the B1
assigns to a given station.
frequency in the BeiDou-3 system between 01.01-01.03.2022. The
Among the stations belonging to the marked cluster, three of
data for this particular frequency is received on channels I and X. them have a priority level of 2 (GRAZ00AUT, BOR100POL and
The average parameters were calculated for each channel. Even- POTS00DEU), and two are at a priority level of 5 (WROC00POL and
tually, the channel on which the most observations were received WARN00DEU). The next parameter is the number of observations,
was selected at each station. Furthermore, the algorithm filtered which is the highest for the mentioned stations WARN00DEU,
out those stations that had less than 90% availability of the RINEX POTS00DEU and WROC00POL. In the case of the SNR parameter,
file for the specified period. The next step was to perform cluster- the number of gaps and multipath, the best of the five stations
ing on stations that fulfilled the conditions for receiving signals with a deviant level of priority are WROC00POL, GRAZ00AUT,
on B1 of the BeiDou-3 system. We chose the k-means method and WARN00DEU. These stations are also the best ranked in the
as the clustering method for further consideration. This choice is ranking created automatically by the algorithm. A major influence
only for the purpose of presenting an example solution. It was on the final selection of the most suitable station was the low
decided to divide all stations into 100 clusters, resulting in the multipath parameter at station WROC00POL. It was more than 3
selection of the top 100 stations. The analysis concerns the global times lower than at the other stations.
network, however, we will focus on the region of Europe, just
to provide an illustrative example of the multi-criteria decision- 4. Impact
making algorithm in action. When focusing on this area, we can
see 16 clusters formed out of about 60 stations located in Europe. The proposed solution is a novel approach to solving a scien-
In each of the clusters, the algorithm created a ranking based on tific problem related to the selection of a global network, focusing
6
Filip Gałdyn, Radosław Zajdel and Krzysztof Sośnica SoftwareX 22 (2023) 101372

Fig. 6. Station selection in the region of Europe with classification and marking the best choices in terms of quality.

on the support of determining global geodetic parameters. In different signals, the selection algorithm seems to be of the most
many cases, e.g., climate studies or reference frame definition, urgent need. The algorithm can be successfully implemented for
the long-term stability and reliability of stations would be crucial. any analysis center that collects data from GNSS stations for anal-
Determination of global geodetic parameters such as Earth’s ro- yses of multi-GNSS data to acquire a well-distributed network of
tation parameters, the displacement of the Earth’s center of mass observing stations.
(the origin of the terrestrial reference frame), geodynamic and
geophysical studies, and the realization of ITRF using GNSS obser- Declaration of competing interest
vations requires the use of a global, evenly distributed network of
ground stations. The application of the proposed algorithms may
The authors declare that they have no known competing finan-
provide a basis for research into the impact of selecting a network
cial interests or personal relationships that could have appeared
of permanent stations on the quality of global geodetic parame-
to influence the work reported in this paper.
ters. The software can be connected to data centers such as the
Crustal Dynamics Data Information System (CDDIS) or support
the regional computing centers, e.g., those of EUREF Permanent Data availability
GNSS Network (EPN), or IGS by supplementing their GNSS data
analysis services. The files used in this study can be downloaded from the
Crustal Dynamics Data Information System https://cddis.nasa.
5. Conclusions gov/archive/gnss/data/daily/. The permanent link to the reposi-
tory with software is available at the link: https://github.com/
The final product is based on programs for qualitative analysis RINEXAV/Rinexav-py-modules/.
of RINEX files, the station clustering process, and multi-criteria
decision-making. It is fully written in Python programming lan- Acknowledgments
guage and is linked from its level to other specialized programs
such as GFZRNX, BKG Ntrip Client, and CRX2RNX. In an automated
This work was realized in the frame of the grant funded by the
and objective manner, it can increase the probability of obtaining
Polish National Science Centre (NCN) Grant UMO-2021/42/E/ST10/
better results when determining global geodetic parameters or
calculations related to the realization of terrestrial reference sys- 00020. The APC is financed by Wrocław University of Environ-
tems. The software code is open to the public and flexible for fur- mental and Life Sciences (UPWr). Part of the research was carried
ther improvements, i.e, considering additional parameters, which out during the student internship funded by UPWr statutory
could support the decision-making and the selection algorithm. research funding. Eventually, we want to acknowledge IGS for
Among many possibilities, including the discontinuity file at a providing and distributing raw data from IGS stations.
given station4 and the availability of antenna calibration5 , 6 , 7 for
References
4 https://itrf.ign.fr/ftp/pub/itrf/itrf2020/ITRF2020-soln-gnss.snx
5 http://acc.igs.org/repro3/repro3_antenna_info_060819.pdf [1] Johnston G, Riddell A, Hausler G. The international GNSS service. In:
Teunissen PJ, Montenbruck O, editors. Springer handbook of global nav-
6 http://acc.igs.org/repro3/repro3_receiver_info_060819.pdf
igation satellite systems. Cham: Springer Handbooks, Springer; 2017, http:
7 http://ftp.aiub.unibe.ch/users/villiger/igsR3_2077.atx //dx.doi.org/10.1007/978-3-319-42928-1_33.

7
Filip Gałdyn, Radosław Zajdel and Krzysztof Sośnica SoftwareX 22 (2023) 101372

[2] Montenbruck O, Steigenberger P, Prange L, Deng Z, Zhao Q, Perosanz F, [24] Collilieux X, Altamimi Z. Impact of the network effect on the origin and
et al. The multi-GNSS experiment (MGEX) of the international GNSS scale: case study of satellite laser ranging. In: Sideris MG, editor. Observing
service (IGS) - achievements, prospects and challenges. Adv Space Res our changing earth. International association of geodesy symposia, vol.
2017;59:1671–97. http://dx.doi.org/10.1016/j.asr.2017.01.011. 133. Berlin, Heidelberg: Springer; 2009, http://dx.doi.org/10.1007/978-3-
[3] Bruyninx C, Legrand J, Fabian A, et al. GNSS metadata and data validation 540-85426-5_4.
in the EUREF Permanent Network. GPS Solut 2019;23:106. http://dx.doi.
[25] Zajdel R, Steigenberger P, Montenbruck O. On the potential contribution
org/10.1007/s10291-019-0880-9.
of BeiDou-3 to the realization of the terrestrial reference frame scale. GPS
[4] Altamimi Z, Rebischung P, Metivier L, Collilieux X. ITRF2014: A new release
Solut 2022;26(4):1–18. http://dx.doi.org/10.1007/s10291-022-01298-0.
of the International Terrestrial Reference Frame modeling nonlinear station
motions. J Geophys Res: Solid Earth 2016;121:6109–31. http://dx.doi.org/ [26] Villiger A, Dach R, Schaer S, Prange L, Zimmermann F, Kuhlmann H,
10.1002/2016JB013098. et al. GNSS scale determination using calibrated receiver and Galileo
[5] Zajdel R, Sośnica K, Dach R, Bury G, Prange L, Jäggi A. Network effects and satellite antenna patterns. J Geod 2020;94(9):1–13. http://dx.doi.org/10.
handling of the geocenter motion in multi-GNSS processing. J Geophys Res: 1007/s00190-020-01417-0.
Solid Earth 2019;124:5970–89. http://dx.doi.org/10.1029/2019JB017443. [27] Jin S, Najibi N. Sensing snow height and surface temperature variations in
[6] Hadas T, Kazmierski K, Sośnica K. Performance of galileo-only dual- Greenland from GPS reflected signals. Adv Space Res 2014;53(11):1623–33.
frequency absolute positioning using the fully serviceable galileo constella- http://dx.doi.org/10.1016/j.asr.2014.03.005.
tion. GPS Solut 2019;23:108. http://dx.doi.org/10.1007/s10291-019-0900- [28] Khosravi K, Shahabi H, Pham BT, Adamowski JF, Shirzadi A, Pradhan B, et
9. al. A comparative assessment of flood susceptibility modeling using Multi-
[7] Hadas T, Krypiak-Gregorczyk A, Hernández-Pajares M, Kaplon J, Criteria Decision-Making Analysis and Machine Learning Methods. J Hydrol
Paziewski J, Wielgosz P, et al. Impact and implementation of higher-order
2019. http://dx.doi.org/10.1016/j.jhydrol.2019.03.073.
ionospheric effects on precise GNSS applications. J Geophys Res: Solid
[29] Ishizaka A, Nemery P. Multi-criteria decision analysis: methods and
Earth 2017;122:9420–36. http://dx.doi.org/10.1002/2017JB014750.
[8] Hadas T, Hobiger T. Benefits of using galileo for real-time GNSS meteo- software. 2013, http://dx.doi.org/10.1002/9781118644898.
rology. IEEE Geosci Remote Sens Lett 2021;18(10):1756–60. http://dx.doi. [30] Behzadian M, Otaghsara S, Yazdani M, Ignatius J. A state-of the-art survey
org/10.1109/LGRS.2020.3007138. of TOPSIS applications. Expert Syst Appl 2012;39:13051–69. http://dx.doi.
[9] Romero I, editor. RINEX the receiver independent exchange format version org/10.1016/j.eswa.2012.05.056.
4.00. 2021, https://files.igs.org/pub/data/format/rinex_4.00.pdf. [31] Han J, Kamber M, Tung A. Spatial clustering methods in data mining: a
[10] Weber Georg, Mervart Leoš, Stürze Andrea, Rülke Axel, Stöcker Dirk. BKG survey. Data Mining and Knowledge Discovery - DATAMINE. 2001, https:
ntrip client (BNC) version 2.12.18. 2016, https://igs.bkg.bund.de/ntrip/bnc. //www.comp.nus.edu.sg/atung/publication/gkdbk01.pdf.
[11] Nischan T. GFZRNX - RINEX GNSS data conversion and manipula- [32] MacQueen J, et al. Some methods for classification and analysis of multi-
tion toolbox. 2016, http://dx.doi.org/10.5880/GFZ.1.1.2016.002, GFZ Data variate observations. In: Proceedings of the fifth berkeley symposium on
Services. mathematical statistics and probability. 1967, p. 281–97.
[12] Zhang L, Dang Y, Cheng Y, Xue S, Gu G, Han D. Analysis and opti-
[33] Hastie T, Tibshirani R, Friedman J. Unsupervised learning. In: The elements
mization of BDS GEO/IGSO/MEO ground monitoring stations configuration
of statistical learning. Springer series in statistics, New York, NY: Springer;
for determining GNSS orbit. Acta Geod Cartogr Sin 2016;45(S2):82. http:
2009, http://dx.doi.org/10.1007/978-0-387-84858-7_14.
//dx.doi.org/10.11947/j.AGCS.2016.F029.
[13] Shuqiang X, Yuanxi Y. Nested cones for single-point-positioning configu- [34] Guo Z, Yu X, Hu C, Jiang C, Zhu M. Research on linear combination
ration with minimal GDOP. 39, (11):2014, p. 1369–74, WHDXXBXXKXB. models of BDS multi-frequency observations and their characteristics.
http://ch.whu.edu.cn/en/article/id/3126. Sustainability 2022;14(14):8644. http://dx.doi.org/10.3390/su14148644.
[14] Yang X, Wang Q, Xue S. Random optimization algorithm on GNSS moni- [35] Strasser S, Mayer-Gürr T, Zehentner N. Processing of GNSS constellations
toring stations selection for ultra-rapid orbit determination and real-time and ground station networks using the raw observation approach. J Geod
satellite clock offset estimation. Math Probl Eng 2019;(2019). http://dx.doi. 2019;93(7):1045–57. http://dx.doi.org/10.1007/s00190-018-1223-2.
org/10.1155/2019/7579185. [36] Ester Martin, Kriegel Hans-Peter, Sander Jörg, Xu Xiaowei. A density-
[15] Lee J, Kim M. Optimized GNSS station selection to support long-term based algorithm for discovering clusters in large spatial databases with
monitoring of ionospheric anomalies for aircraft landing systems. IEEE noise. In: Proceedings of the second international conference on knowledge
Trans Aerosp Electron Syst 2017;53(1):236–46. http://dx.doi.org/10.1109/
discovery and data mining. AAAI Press; 1996, p. 226–31.
TAES.2017.2650038.
[37] Gong J, Ma R, Li D, et al. GNSS multipath mitigation method based on
[16] Wang S, Huang S, Fang H. Ground GNSS station selection to generate
K-means classification in urban environment. J Indian Soc Remote Sens
the global ionosphere maps using the information content. Space Weather
2022;20(1). http://dx.doi.org/10.1029/2020SW002675. 2022;50:805–13. http://dx.doi.org/10.1007/s12524-022-01494-y.
[17] Villiger Arturo. Upcoming switch to IGS20/igs20.atx and repro3 standards. [38] Takahashi A, Hashimoto M. Cluster analysis of dense GNSS velocity
2022, IGS Mail No. 8238, 26 Jul 2022. https://lists.igs.org/pipermail/igsmail/ field reveals characteristics associated with regional tectonics in New
2022/008234.html. Zealand. J Geophys Res: Solid Earth 2022;127. http://dx.doi.org/10.1029/
[18] Dach R, Selmke I, Villiger A, Arnold D, Prange L, Schaer S, et al. Review 2022JB024793, e2022JB024793.
of recent GNSS modelling improvements based on CODEs Repro3 contri- [39] BKG Ntrip Client (BNC) version 2.9 manual. http://acc.igs.org/misc/bnchelp.
bution. Adv Space Res 2021;68(3):1263–80. http://dx.doi.org/10.1016/j.asr. pdf.
2021.04.046. [40] Altamimi Z, Rebischung P, Collilieux X, Metivier L, Chanard K. ITRF2020:
[19] Männel B, Brandt A, Bradke M, Sakic P, Brack A, Nischan T. Status of main results and key performance indicators. 2022, http://dx.doi.org/10.
IGS reprocessing activities at GFZ. In: Freymueller JT, Sánchez L, editors.
5194/egusphere-egu22-3958.
Beyond 100: the next century in geodesy. Cham: Springer International
[41] Hwang C-L, Yoon K. Methods for multiple attribute decision making. In:
Publishing; 2020, p. 37–43. http://dx.doi.org/10.1007/1345_2020_98.
Multiple attribute decision making: methods and applications a state-of-
[20] Hatanaka Y. A compression format and tools for GNSS observation data.
Bull Geogr Surv Inst 2008;55. https://www.gsi.go.jp/common/000045517. the-art survey. Berlin, Heidelberg: Springer Berlin Heidelberg; 1981, p.
pdf. 58–191. http://dx.doi.org/10.1007/978-3-642-48318-9_3.
[21] Blewitt G. Terrestrial reference frame requirements for studies of geody- [42] Razavi Hajiagha SH, Hashemi SS, Zavadskas EK. A complex proportional
namics and climate change. In: van Dam T, editor. International association assessment method for group decision making in an interval-valued in-
of geodesy symposia, vol. 146. Cham: Springer; 2015, http://dx.doi.org/10. tuitionistic fuzzy environment. Technol Econ Dev Econ 2013;19(1):22–37.
1007/1345_2015_142. http://dx.doi.org/10.3846/20294913.2012.762953.
[22] Couhert Alexandre, Mercier Flavien, Moyard John, Biancale Richard. Sys- [43] Hezer S, Gelmez E, Özceylan E. Comparative analysis of TOPSIS, VIKOR and
tematic error mitigation in DORIS-derived geocenter motion. J Geophys COPRAS methods for the COVID-19 Regional Safety Assessment. J Infect
Res: Solid Earth 2018;123(11):10,110–42,161. http://dx.doi.org/10.1029/ Public Health 2021;14(6):775–86. http://dx.doi.org/10.1016/j.jiph.2021.03.
2018JB015453. 003.
[23] Zajdel R, Sośnica K, Drozdzewski M, et al. Impact of network constraining
[44] Dhiman HS, Deb D. Fuzzy TOPSIS and fuzzy COPRAS based multi-criteria
on the terrestrial reference frame realization based on SLR observations to
decision making for hybrid wind farms. Energy 2020;202:117755. http:
LAGEOS. J Geod 2019;93:2293–313. http://dx.doi.org/10.1007/s00190-019-
01307-0. //dx.doi.org/10.1016/j.energy.2020.117755.