GPS Solutions (2019) 23:41

https://doi.org/10.1007/s10291-019-0833-3

ORIGINAL ARTICLE

Fast ionospheric correction using Galileo Az coefficients and the NTCM

model
Mohammed Mainul Hoque1   · Norbert Jakowski1 · Raul Orús‑Pérez2

Received: 1 August 2018 / Accepted: 11 February 2019

© Springer-Verlag GmbH Germany, part of Springer Nature 2019

Abstract
Europe’s Global Navigation Satellite System (GNSS) Galileo broadcasts three parameters for ionospheric correction as
part of satellite navigation messages. They are called effective ionization coefficients and are used to drive the NeQuickG
model in single frequency Galileo operations. The NeQuickG is a three-dimensional electron density model based on several
Epstein layers whose anchor points, such as ionospheric peak densities and heights, are derived using the spatial and tem-
poral interpolation of numerous global maps. This makes the NeQuickG computationally more expensive when compared
with the GPS equivalent, the Klobuchar model. We propose here an alternative ionospheric correction approach for single
frequency Galileo users. In the proposed approach, the broadcast coefficients are used to drive another ionospheric model
called the Neustrelitz Total Electron Content Model (NTCM) instead of the NeQuickG. The proposed NTCM is driven by
Galileo broadcast parameters and the investigation shows that it performs better than the NeQuickG when compared with the
reference vertical total electron content (VTEC) data. It is found that the root mean squares (RMS) and standard deviations
(STD) of residuals are approx. 1.6 and 1.2 TECU (1 TECU = 1016 electrons/m2) less for the NTCM than the NeQuickG.
A comparison with the slant TEC reference data shows that the STD, mean and RMS residuals are approx. 9.5, 0.6, 10.0
TECU for the NeQuickG whereas for the NTCM, they are 9.3, 2.5, 10.1 TECU respectively. A comparison with Jason-2
altimeter datasets reveals that the NTCM performs better than the NeQuickG with RMS/STD deviations of approx. 7.5/7.4
and 8.2/7.9 TECU respectively. The investigation shows that the Galileo broadcast messages can be effectively used for
driving the NTCM.

Keywords  GNSS · Galileo ionosphere correction · NTCM · NeQuickG

Introduction System (GPS) utilizes the ionospheric correction algorithm

(ICA) commonly known as the Klobuchar model (Klobu-
Single frequency users of a Global Navigation Satellite Sys- char 1987, IS-GPS-200J 2018), the European GNSS Galileo
tems (GNSS) improve positioning and navigation results by uses NeQuickG (OS SIS ICD 2010) and the Chinese GNSS
correcting ionospheric signal delays which can cause link Beidou uses the COMPASS ionospheric model (CIM, Wu
related range errors of up to about 100 m. Current GNSSs et al. 2013). These model-based corrections allow mitigating
commonly offer the use of a specific ionospheric model ionospheric range errors of approx. 50–70% (IS-GPS-200J
enabling a first order correction of the ionospheric delay 2018, OS SIS ICD 2010). Whereas the GPS and Beidou
which is proportional to the total electron content (TEC) systems rely on simple models using a mapping function
along the ray path. For example, the Global Positioning to transform vertical TEC (VTEC) to slant TEC (STEC)
along the GNSS links used, Galileo uses the comprehen-
sive three-dimensional electron density model NeQuickG
* Mohammed Mainul Hoque
[email protected] to integrate electron density numerically along the signal
links. The comprehensive NeQuick model developed at the
1
Institute of Communications and Navigation, International Centre for Theoretical Physics (ICTP) and the
German Aerospace Center (DLR), Kalkhorstweg 53, University of Graz (Hochegger et al. 2000; Nava et al. 2008)
17235 Neustrelitz, Germany
has been adopted as the single frequency correction model
2
ESA/ESTEC Wave Interaction and Propagation Section for Galileo (OS SIS ICD 2010) commonly known as the
(TEC-EFW), Noordwijk, The Netherlands

13
Vol.:(0123456789)
 41   Page 2 of 12 GPS Solutions (2019) 23:41

NeQuickG model. The vertical electron density profile is height. The NeQuick model is adapted for real-time Gali-
based on a composition of several Epstein layer functions fed leo single frequency ionospheric corrections and is referred
by numerous coefficients adopted by the Comité Consultatif to as the NeQuickG (European Commission 2016). How-
International des Radiocommunications (CCIR) in the 1960s ever, in real-time applications, it is driven by a single input
(Jones and Gallet 1962). To achieve the best accuracy for parameter called the effective ionization level (Az) and is
Galileo single frequency users, the NeQuickG model is fed determined using three broadcast coefficients ai0, ai1, ai2, as
by an effective ionization level (Az) as a proxy measure of
the solar activity level instead of using the original monthly
Az = ai0 + ai1 𝜇 + ai2 𝜇2 (1)
−22 −2 −1
average solar radio flux index F10.7. The Az is deduced from the unit of Az is sfu (1  sfu = 10   Wm   Hz ). The
dual-frequency GNSS measurements at selected ground modified dip√ latitude µ, known as Modip, defined by
stations by optimizing the performance of the NeQuickG tan 𝜇 = I∕ cos 𝜑 in which I is the geomagnetic inclination
model. Thus, it is clear that Az contains basic information at 300 km height and 𝜑 is the geographic latitude of the loca-
regarding the actual solar activity level. Although Az opti- tion. This Az parameter is equivalent to the solar radio flux
mization is modified by the NeQuickG properties, the aim is index F10.7 in the original climatological NeQuick model.
to check its applicability with regards to the NTCM global The NeQuickG vertical electron density profile is given
TEC model (Jakowski et al. 2011a) in a similar way as to the by a sum of several Epstein layers whose shape parameters
discussions of the Klobuchar model (Hoque et al. 2017). By such as peak ionization, peak height and semi-thickness are
feeding the NTCM with Klobuchar coefficients, an improved deduced from CCIR models (recommended by the Interna-
accuracy to the order of approx. 10–40% could be achieved. tional Radio Consultative Committee under the International
However, here the focus is on the developing a simplified Telecommunication Union). The peak ionization model con-
and fast correction algorithm without a loss in performance sists of 24 maps of 988 coefficients each, one for each month
compared to the NeQuickG model. of the year and for a low and high solar activity level defined
We begin with brief descriptions of the NeQuickG and by a 12-month smoothed sunspot number of 10 and 100.
NTCM models. The subsequent section discusses the data- Similarly, the peak height model consists of 24 maps of 441
base and sources used for Galileo broadcast coefficients, and coefficients each, one for each month of the year and for two
the reference ground and space-based VTEC and STEC data levels of solar activity. Thus, the whole CCIR model consists
sets for model development and validation. The final section of (988 + 441) × 12 × 2 = 34,296 coefficients. Linear interpo-
gives a performance evaluation of the NTCM compared to lation is applied for other levels of solar activity. An iterative
the NeQuickG model. numerical integration is used for slant delay computation.
As discussed, the use of thousands of model coefficients,
frequent interpolation between maps and iterative integra-
Ionosphere broadcast models tion for slant delay computation make the NeQuickG com-
putationally expensive. So, we propose here an alternative
The Galileo system uses a version of the NeQuick model ionospheric correction approach for single frequency Galileo
called NeQuickG to aid single frequency operations. The users.
NeQuickG model is operated by ionization coefficients
broadcast in Galileo navigation messages. This section Neustrelitz TEC model (NTCM)
briefly describes the driving parameter behind the NeQuickG
and discusses how electron density distributions are obtained Similar to the NeQuick, the original NTCM is driven by a
within the NeQuickG. The formulation of the NTCM as an proxy of the solar extreme ultraviolet (EUV) radiation level
alternative to the NeQuickG is described briefly and the represented by the solar radio flux index F10.7 (Jakowski
driving parameter defined. It also discusses the complexity et al. 2011a). The NTCM was later modified for possible use
of the NeQuickG compared to the simplicity of the NTCM. in future satellite navigation systems as the broadcast model
NTCM-BC. The coefficients for the NTCM-BC require daily
NeQuick Galileo updates using worldwide GNSS TEC data similar to the GPS
Klobuchar model (Hoque and Jakowski 2015). Compared to
NeQuick is a three-dimensional and time-dependent iono- the original NTCM, the NTCM-BC, which is updated daily,
spheric electron density model based on the climatological has been simplified by ignoring the long-term TEC depend-
representation of the ionosphere (Hochegger et al. 2000; ence, such as annual, semi-annual and solar cycle variations.
Nava et al. 2008). It is driven by the solar activity input data To aid single frequency GPS operations, another version of
such as an average sunspot number or solar radio flux index the NTCM has recently been proposed. This version is driven
F10.7 and predicts the mean electron density depending on by a parameter called Klobpar derived from GPS broadcast
the month, time of day, geographical latitude, longitude, and coefficients (Hoque et al. 2017, 2018). The Klobpar replaces

13
GPS Solutions (2019) 23:41 Page 3 of 12  41

the F10.7 in the original climatological NTCM. The NTCM- where k1–k12 are model coefficients, 𝜑 and 𝜑m are geographic
Klobpar TEC prediction performance is improved during high and geomagnetic latitude respectively, λ is the geographic
and low solar activity conditions by approx. 40% and 10%, longitude, δ is the solar declination angle, LT is the local
respectively, compared to the Klobuchar model (Hoque et al. time, doy is the day of the year, 𝜑GNP and λGNP are the lati-
2017). Encouraged by the good performance achieved using tude and longitude of the geomagnetic North pole used in
the NTCM-Klobpar model, we investigated the possibility of the dipole model for geographic to geomagnetic latitude
driving the NTCM using Galileo broadcast parameters. This conversion. The quantity MF is the modified single layer
was subsequently named the NTCM-GlAzpar. model (MSLM) mapping function (Schaer 1999) in which
The NTCM-GlAzpar formulations proposed are exactly the earth’s mean radius Re = 6371 km, ionospheric pierce
the same as the NTCM-Klobpar formulations (Hoque et al. point (IPP) height hI = 450 km and β is the elevation angle at
2017). However, the equations are given below for complete- the receiver in radians. Equations (2–7) give the NTCM for-
ness. As mentioned above, a new driving parameter computed mulations and show that the model computes VTEC by mul-
from Galileo broadcast parameters, called GlAzpar, was used tiplying functions from F1 to F5. The function F1 describes
instead of Klobpar, which is computed from GPS broadcast the diurnal, semi-diurnal and tar-diurnal TEC variation, F2
parameters. As the driving parameter was changed, a different describes TEC dependence on annual and semi-annual vari-
set of model coefficients was needed. These were computed ation, F3 and F4 describe TEC dependence on geomagnetic
by fitting ionospheric TEC observations to the model. In the latitudes and equatorial crest regions and finally F5 describes
following, the proposed two-dimensional VTEC model is out- the solar cycle variation of TEC. The quantity GlAzpar is
lined together with a single layer mapping function MF for the computed from the Az and the procedure is discussed in
conversion of VTEC to STEC and vice versa: the next section. For details on different terms, we refer to
Jakowski et al. (2011a).
VTECNTCM-GlAzpar = F1 × F2 × F3 × F4 × F5 (2)

F1 = cos 𝜒 ∗∗∗ + cos 𝜒 ∗∗ (k1 cos VD + k2 cos VSD

(3) Data sources and estimation of coefficients
+ k3 sin VSD + k4 cos VTD + k5 sin VTD )

F2 = 1 + k6 cos VA + +k7 cos VSA (4) Since March 2013, Galileo satellites have broadcast three
effective ionization coefficients ­ai0, ­ai1 and ­ai2 as part of
F3 = 1 + k8 cos 𝜑m (5) satellite navigation messages (Orús-Pérez et  al. 2018).
( ) ( ) The coefficients are designed and computed as the driving
F4 = 1 + k9 exp EC1 + k10 exp EC2 (6)
parameters of the NeQuickG for aiding single frequency
F5 = k11 + k12 GlAzpar (7) Galileo operations. The coefficients are updated at least once
per day. However, in many days more than one set and up
where cos 𝜒 ∗∗∗ = cos (𝜑 − 𝛿) + 0.4  , cos 𝜒 ∗∗ = cos (𝜑 − 𝛿)
2𝜋 (LT−LTD ) to 4 sets of coefficients are broadcast by the Galileo space
− 𝜋2 × 𝜙 × sin 𝛿 , VD = , VSD = 2𝜋LT , VTD = 2𝜋LT ,
24 12 8 segment. This paper uses 5 years of effective ionization
2

VA =
2𝜋 (doy−doyA )
, VSA =
4𝜋 (doy−doySA )
,  EC1 = −
(𝜑m −𝜑c1 )
, broadcast data from 2013 to 2017. The effective ionization
2
365.25
2
365.25 2𝜎c1
data were provided by the European Space Agency (ESA).
EC2 = −
(𝜑m −𝜑c2 )
2  , LTD = 14,doyA = 18, doySA = 6, However, limited data of ionization coefficients can be
found as header information in the International GNSS Ser-
2𝜎c2

𝜑c1 = 16◦ N, 𝜑c2 = 10◦ S, 𝜎c1= 12◦ , 𝜎c2 = 13◦  , 𝜑GNP = vice (IGS) processed broadcast ephemeris brdm product
79.74◦ N, 𝜆GNP = −71.78◦ E   , 𝛿 = 23.44 × sin (0.9856× (http://mgex.igs.org/IGS_MGEX_Produ​cts.php). The brdm
) × 𝜋∕180) × 𝜋∕180
(doy( − 80.7) (   ,) 𝜑(m = sin (sin
))(𝜑)× product is available in the annual Multi-GNSS Experiment
sin 𝜑GNP + cos (𝜑) × cos 𝜑GNP × cos 𝜆 − 𝜆GNP , (MGEX) archives of the Crustal Dynamics Data Information
System (CDDIS) (ftp://ftp.cddis​.eosdi​s.nasa.gov/gnss/data/
1 campai​ gn/mgex/daily/​ rinex3​ /) and the Institut Géographique
MF = √ (8)
1 − (sin z)2 National (IGN). Although ionization coefficients have been
broadcast since 2013, the values are missing from the brdm
Re ( (
𝜋
)) product for numerous days in the years 2013, 2014, 2016 and
sin z = sin 0.9782 × −𝛽 2017 which significantly reduces the data that is publicly
Re + hI 2
available. However, this type of data can also be found in
the station-specific receiver independent exchange format
STEC = MF × VTECNTCM-GlAzpar , (9) (RINEX) navigation data files. The brdm file records one set
of ionization coefficients per day even though several sets

13
 41   Page 4 of 12 GPS Solutions (2019) 23:41

are broadcasted during the course of a single day. Figure 1 Thus, the use of Az in driving another ionosphere model
shows a comparison between the ai0, ai1, and ai2 coefficients is not straightforward. Moreover, the Az computed varies
broadcast and the corresponding IGS values during 2015. as a function of Modip of the user location. For detailed
The maximum and minimum differences (ESA − IGS val- information on the Az parameter and its generation in the
ues) are shown in the corresponding panels. NeQuickG, we refer to the paper by Hoque and Jakowski
In comparing the plots in Fig 1, it is shown that the broad- (2015) and references therein. In the NTCM model formu-
cast coefficients and the values provided by the IGS are not lation, the solar activity dependence is modeled by a linear
identical. The maximum and minimum differences between function (see 7). The latitude dependency of the TEC is
these values are given in each subplot. For example, the ai0 modeled by separate functions such as (5) and (6). There-
value broadcast has a maximum and minimum difference fore, the NTCM does not need a solar activity proxy depend-
of 55 and − 62.5 sfu when compared with the brdm values. ing on the geographic location (i.e., Modip dependency).
Whereas, the min/max values suggest a major mismatch, The main purpose of this paper is to simplify the ionospheric
overall the figure indicates a fair agreement of the ai0, ai1, correction method for Galileo users. Thus, a single solar
ai2 coefficients collected by the IGS with those provided by proxy measure was derived by averaging Az values. Differ-
the ESA. The Galileo navigation message does not provide ent procedures for generating the average Az were tested.
an issue-of-data for the ionospheric correction parameters For example, the Az values were computed at different lati-
nor does it ensure that identical coefficients are transmitted tudes along a certain meridian (e.g., 0°E) and also across
by all satellites. As such the differences shown in the figure the globe using a 2.5 × 5-degree latitude and longitude grid.
are likely to be representative of the “uncertainty” faced by These were then averaged to produce the final Az value.
a user receiver in terms of the actual data set to be used at a However, no significant differences were detected in the
certain epoch. The RINEX standard supports multiple, time- model performance for different Az averaging schemes.
stamped sets of ionospheric correction parameters within the So, the averaging procedure was kept as simple as possible.
header of a broadcast ephemeris file, even though this option Therefore, as a potential scheme GlAzpar was computed by
is not yet used in the brdm ephemeris product. averaging Az values for Modip values between 70° North
The Az level was computed by (1) using the ioniza- and South as suggested by Orús-Pérez (2018). The Modip
tion coefficients provided by ESA for our TEC modeling. Eq. (1) is integrated analytically between the two limits and
Although Az represents current solar conditions, it is opti- divided by the Modip range. However, the contribution of ai1
mized for the NeQuickG by minimizing model errors with coefficient is canceled out in the integration process. Consid-
respect to GNSS measurements at Galileo sensor stations. ering this, finally, GlAzpar was computed by taking the root
So, the Az estimated not only depends on the current solar mean squared (RMS) of Az values for Modip values between
conditions, but also on the performance of the NeQuickG. 70° North and South. The square of the Modip Eq. (1) is
integrated analytically between the two limits and divided
by the Modip range 140. Thus, GlAzpar is determined by
|√( 2 )|
GlAzpar = || ai0 + 1633.33a2i1 + 4802000a2i2 + 3266.67ai0 ai2 ||
| |
(10)
the units are sfu. The G1Azpar values computed during
the selected period 2013–2017 are plotted together with
the solar activity proxy F10.7 in the top panel of Fig. 2 as
a time series plot. The F10.7 data are obtained from the
NOAA’s (National Oceanic and Atmospheric Administra-
tion) National Geophysical Data Center (NGDC) (archive
ftp://ftp.ngdc.noaa.gov/). The bottom panel of Fig. 2 shows
the corresponding scatter plot of GlAzpar versus F10.7. The
arithmetic mean and standard deviation (STD) of GlAzpar
are determined considering a bin size of 10 sfu. The mean
(continuous blue line) and mean ± STD (broken blue line)
values are plotted in the bottom panel of Fig. 2.
Both panels in Fig. 2 indicate that the GlAzpar is highly
correlated with the F10.7 as expected. Thus, it can be used
as a proxy measure of ionospheric activity instead of F10.7
Fig. 1  Galileo effective ionization coefficients obtained from the ESA for driving ionosphere models. When comparing GlAzpar
and IGS

13
GPS Solutions (2019) 23:41 Page 5 of 12  41

Table 1  NTCM-GlAzpar model NTCM driven by GlAzpar

coefficients
Coefficients Values
(dimension-
less)

k1 0.92519
k2 0.16951
k3 0.00443
k4 0.06626
k5 0.00899
k6 0.21289
k7 − 0.15414
k8 − 0.38439
k9 1.14023
k10 1.20556
k11 1.41808
k12 0.13985

Like the NTCM and NTCM-Klobpar models (Jakowski

et al. 2011a; Hoque et al. 2017), the 12 model coefficients
for the NTCM-GlAzpar were computed from the empiri-
cal model approach given in (2–7) fitted to the worldwide
ionospheric behavior derived from CODE TEC data. Thus,
CODE TEC data, as well as the GlAzpar values computed
during the selected period of 2013–2017, were used in the
model coefficients estimation. The model coefficients are
constant and the only driving parameter refers to the current
solar activity via the GlAzpar value. The computed model
coefficients are given in Table 1.
Fig. 2  A comparison of solar condition given by F10.7 and GlAz-
par for the 2013–2017 period. The top panel shows time series plots In addition, the VTEC data from the final combined IGS
whereas the bottom panel shows scatter plots of GlAzpar versus product (labeled as igsg) and the Jason-2 altimetry mission
F10.7. The corresponding mean and mean ± STD plots are shown as (https​://www.ospo.noaa.gov/Opera​tions​/Jason​2/index​.html)
continuous and broken blue lines were used as independent reference data sets to compare the
quality of the models. The igsg maps are computed by com-
with F10.7, there are strong deviations from the expected bining independent global VTEC maps from different IGS
values for flux values > 160 sfu. However, this will not analysis centers. Like the CODE data, the daily IONEX files
degrade the performance of the model during periods of igsg data were downloaded from the CDDIS archive. The
of strong solar activity conditions since the model coef- altimeter data were downloaded from the NOAA’s National
ficients are generated based on the direct TEC depend- Center for Environmental Information archive (ftp://ftp.
ency on GlAzpar instead of F10.7. Reference data sets are nodc.noaa.gov/pub/data.nodc/jason​2/gdr/gdr/). The altim-
needed to model ionospheric delay or TEC. We used post eter TEC data were originally measured almost every second
processed VTEC data from the Center for Orbit Deter- along the satellite orbit. However, for the present analysis,
mination System (CODE) at the University of Bern (see every 10th observation was used for validation purposes.
http://aiuws​.unibe​.ch/ionos​phere​/ and Schaer et al. 1998). The distance between 10th observation points is found about
Daily ionosphere exchange (IONEX) files of hourly global 70 km at Jason-2 height (1340 km). As the ionosphere does
TEC maps were downloaded from the CDDIS (Crustal not change significantly between such distances we exclude
Dynamics Data Information System) archive (ftp://cddis​ the intermediate measurement points.
.gsfc.nasa.gov/pub/gps/produ​cts/ionex​). CODE data was Additionally, the GPS L1 and L2 carrier-phase and P1
used from the 2013–2017 for which effective ionization and P2 code pseudorange from about 200 globally distrib-
data was available. The time period includes high-to-low uted IGS (Dow et al. 2009) stations were downloaded from
solar activity conditions covering the decreasing phase of the CDDIS archive and STECs are determined for numer-
the current solar cycle 24. ous receiver-satellite link geometries during the year 2015.

13
 41   Page 6 of 12 GPS Solutions (2019) 23:41

As an example, for January 1, 2015, about 206 GPS sta- Comparison with VTEC data from CODE and IGS
tions are selected which recorded, simultaneously, GPS L1,
L2, P1, P2 signals. The station location (blue triangle) and The daily IONEX files of hourly global CODE and IGS
measurements points, known as IPPs (green dots), are shown combined VTEC maps (labeled as codg and igsg, respec-
on a global map in Fig. 3. The daily GPS ephemeris data tively) contain 24/25 gridded maps with geographic lati-
were downloaded from the Scripps Orbit and Permanent tude and longitude resolution of 2.5° × 5°, respectively, and
Array Center (SOPAC) archive (ftp://garne​r.ucsd.edu). The are used as VTEC references. The NTCM and Klobuchar
STEC is derived here by combining both the dual-frequency model-derived VTECs are computed at each grid location
carrier-phase and code-pseudorange measurements. The and time step and then compared with the corresponding
low-noise carrier-phase derived relative STEC is used to reference VTEC values. Selected histogram plots of residu-
smooth the code-derived relative STEC. For details about als ­(VTECmodel − VTECcodg) containing mean, STD, and root
STEC calibration and separation of inter-frequency satellite mean squares (RMS) are shown in Fig. 4 for the years 2014
and receiver biases, we refer to (Jakowski et al. 2011b). The and 2015.
STEC data were then used as independent reference data sets The top panels of Fig. 4 show a global performance anal-
to compare the quality of the models in STEC predictions. ysis of the NTCM driven by GlAzpar in comparison with
the GPS Klobuchar model and other versions of the NTCM
during 2014 (left panel) and 2015 (right panel). They show
Performance evaluation of NTCM driven that the RMS and STD of residuals remain within 8.6–9.1
by Galileo Az parameter and 6.8–6.9 TECU during 2014 and 2015 respectively for all
three versions of the NTCM. They correspond to an approx.
Here, the performance of the proposed NTCM was com- 1.1–1.5 m range error for the GPS/Galileo L1 frequency
pared with the NeQuickG for GNSS single frequency opera- signal (1575.42 MHz). The mean residuals are − 1.4 and
tion. The source code of NeQuickG used for this paper can 0.2 TECU for the NTCM driven by GlAzpar, and − 1.1 and
be found at the European space software repository (https​:// − 0.7 TECU for both the NTCM driven by Klobpar and the
essr.esa.int/proje​ct/nequi​ckg-galil​eo-ionos​pheri​c-corre​ction​ NTCM driven by F10.7. The Klobuchar model gives the
-model​). The comparative study also includes the Klobuchar highest RMS, mean and STD values for both years which
GPS broadcast model as well as other available versions of are approx. 13.5, − 5.1 and 12.5 TECU for 2014. The global
the NTCM. The comparisons have been done in terms of analysis includes data from all latitudes, longitudes and local
VTEC and STEC residuals which are defined as the differ- time.
ence between a reference and its corresponding modeled The bottom panel of Fig. 4 compares the same model
TECs. The STEC is equivalent to the range errors experi- performance for the low latitude region 30°N–30°S cover-
enced by a GNSS single frequency user. As reference obser- ing all longitude and the hours of 06:00–18:00 local time.
vations, the ground-based VTEC data from CODE, IGS and The latitude region and time period are chosen in such a
space-based VTEC data from Jason-2 altimetry mission way that the analysis includes data where the ionospheric
were considered. Additionally, GPS STEC data from over TEC are highest. It was found that the RMS and STD values
200 worldwide IGS stations were considered as reference varied within approx. 9.8–10.1 TECU, for the three NTCM
observations. versions for 2015. However, during 2014 they differed and
the NTCM-G1Azpar performs slightly better than other
NTCM versions. As was found for the global analysis, the
Klobuchar model also gives the highest RMS, mean and
STD values for the regional analysis. A comparison of the
mean residuals from three NTCM versions shows that in
2015 the model driven by F10.7 gave the lowest value of
− 0.2 TECU, whereas GlAzpar driven model gives the high-
est value of 1.8 TECU.
The following compares the performance of the pro-
posed NTCM driven by GlAzpar with the NeQuickG
model. Both the NTCM and NeQuickG models are driven
by Galileo broadcast ionosphere coefficients. Consider-
ing that the codg data were used in the NTCM coefficients
estimation, here the igsg data were taken as independent
Fig. 3  Location of selected GPS stations marked as blue triangles and reference datasets. As before the NTCM and NeQuickG
measurement points are marked as green dots for January 1, 2015 model-derived VTECs are computed at each grid location

13
GPS Solutions (2019) 23:41 Page 7 of 12  41

Fig. 4  Histograms of GPS Klobuchar and NTCM residuals with respect to reference codg data showing their performances for global day and
nighttime (top panel) and low latitude daytime analysis (bottom panel)

and time step and then compared with the corresponding ref- daytime analysis. It was found that the NTCM mean residu-
erence VTEC values. Selected histogram plots of residuals als are approx. 2–3 and 6–8 TECU less in absolute values
­(VTECmodel − VTECigsg) containing mean, STD, and RMS than those of the NeQuickG for global and regional analysis,
are shown in Fig. 5 for the years 2014 and 2015. respectively.
The top panels of Fig. 5 show the combined results for
global day and night. When compared with reference VTEC Comparison with altimeter TEC data
data, we found that the RMS and STD are approx. 9.6 and
9.0 TECU for the NeQuickG model, whereas for the NTCM This section compares the performance of the proposed
model they are 7.8 and 7.7 TECU in 2014. The correspond- NTCM model driven by the Galileo broadcast coefficients
ing values are 7.8 and 7.4 TECU for the NeQuickG and with the NeQuickG with independent TEC data coming
they are both 6.4 TECU for the NTCM in 2015. The bot- from the non-GNSS source. Other versions of the NTCM
tom panel compares the model performance for the low that are available are included in the comparative study. The
latitude region 30°N–30°S covering all longitude and the reference data are for 2015 and originated from the Jason-2
hours of 06:00–18:00 local time. It was found that the RMS altimetry satellite. The Jason-2 satellite is orbiting at a mean
and STD values varied within approx. 9.5–10.7 TECU, height of approx. 1340 km and the TEC data are obtained
for the NTCM-GlAzpar. These RMS and STD estimates from a dual transmitter–receiver in C-band (5.3 GHz) and
are 0.6–1.0 TECU less than the estimates shown in Fig. 4 Ku-band (13.6 GHz). The Jason-2 data contains VTEC data
(bottom panels). As was found for the global analysis, the from the sea surface up to the height of Jason-2 at different
NTCM-GlAz also gives the smaller RMS, mean and STD geographic locations and points in time. Smoothed Jason-2
values for the regional analysis. A comparison of the mean data were used instead of raw data. First, the outliers (if any)
residuals shows that the NeQuickG always experiences were removed from the Jason-2 raw data. The data were
negative biases whereas the NTCM shows negative biases grouped into geographic regions and individual observations
for the global analysis and positive biases for the regional rejected if they exceeded the corresponding median value

13
 41   Page 8 of 12 GPS Solutions (2019) 23:41

Fig. 5  Histograms of NeQuickG and NTCM residuals with respect to the reference igsg data showing their performances for global day and
nighttime (top panel) and low latitude daytime analysis (bottom panel)

by a certain threshold (e.g., 2-sigma value). The data were bias of approx. − 2.0 TECU, whereas the NTCM gives a
then smoothed using a moving average filter. The NeQuickG positive bias of approx. 1.1 TECU. Thus, it is in agreement
and NTCM-derived VTECs were calculated for the same with the expectations of the plasmaspheric contributions
location and time and subsequently, their differences from (Jakowski and Hoque 2018). The other two NTCM versions
Jason-2 data, i.e., ­VTECmodel − VTECJason2 were computed. show similar results for the RMS and STD to the NTCM
Figure 6 shows the histogram plots of the VTEC differences driven by GlAzpar although the mean bias is found much
and the corresponding RMS, mean and STD estimates. less as about 0.1–0.2 TECU. For regional daytime analysis,
Contemporary altimetry satellites operate at a much lower we find that the RMS and STD residuals are approx. 1.7–2.3
height (e.g., Jason-2 at 1330 km) compared to GNSS sat- TECU less for the different versions of NTCM compared to
ellites and provide ionospheric electron content up to that those of NeQuickG. We find a positive bias of approx. 2.5,
height. Consequently, altimeter-derived TEC data should 1.8 and 4.8 TECU for the NTCMs driven by F10.7, Klobpar
be smaller than the GNSS-derived TEC estimates because and GlAzpar, respectively, whereas the NeQuickG gives a
the latter includes the plasmaspheric electron content up to negative bias of approx. -3.4 TECU.
approx. 20,000 km height (Jakowski and Hoque 2018). Thus,
at middle to lower latitudes, the altimeter VTEC measure- Comparison with STEC data
ments are known to have a systematic bias of 2–5 TECU
above the real ionospheric VTEC values when compared This section compares the performance of the proposed
with other independent VTEC measurements (Orús-Pérez NTCM with the NeQuickG with independent STEC meas-
et al. 2002). urements computed at about 200 worldwide GPS stations.
In the case of global analysis, we find that RMS and STD The station distribution for a sample day can be seen in
residuals are approx. 0.5–0.7 TECU less for the NTCM Fig. 2 in a global map. The NTCM and NeQuickG model-
driven by GlAzpar than the NeQuickG. When comparing derived STECs are computed at each station location for
the mean residuals, we find that NeQuickG gives a negative the same link geometries above 20° elevation angles. The

13
GPS Solutions (2019) 23:41 Page 9 of 12  41

Fig. 6  Histograms of the NeQuickG and NTCM residuals with

respect to the reference Jason-2 data showing their performances for
global day and nighttime (top panel) and low latitude daytime analy-
sis (bottom panel)
Fig. 7  A performance comparison of NeQuickG and NTCM with
global STEC observations. c–e The daily STD, mean and RMS of
STEC residuals, and corresponding monthly average values are plot-
NeQuickG is a 3D electron density model and STECs are ted in red and blue lines. The number of samples and a comparison of
computed by integrating the electron density along the ray computation time are shown in a and b 
paths. However, the NTCM is a VTEC model that needs
an elevation dependent mapping function (8) for VTEC to
STEC conversion (9). Thus, STEC results derived in this both models. The same input files containing ai0, ai1, ai2,
way depend also on the accuracy of the mapping func- month, universal time, station longitude, latitude, height and
tion (Hoque and Jakowski 2013). The comparisons have satellite longitude, latitude and height are used for running
been done in terms of STEC residual which is defined as the NeQuickG and the NTCM. Both models produced simi-
the difference between the STEC observation and its cor- lar output files containing the inputs listed above, and addi-
responding modeled STEC (i.e., S ­ TECmodel − STECobs). tionally Modip/GlAzpar and corresponding STEC values.
The residual statistics such as RMS, mean and STD are An Intel machine (CPU E5-2698 [email protected] GHz, 16 CPUs)
computed over about 4.6 million daily datasets for many with Linux operating system (Scientific Linux release 7.4)
days in 2015 and their values are plotted in Fig. 7. Some was used for the computation.
days, i.e., doy 43–64, are missing in the analysis since Az The top two panels of Fig. 7 show the number of STEC
coefficients are not available for those days in our datasets. computed and corresponding computation time for each day
The computational complexity of the NeQuickG and the during 2015. On average, about 4.6 million STEC values
NTCM are compared in terms of model run time. The STEC are computed on each day and the average computation
computation time is recorded for daily input data files for time is found as 4024 and 62 s for NeQuickG and NTCM,

13
 41   Page 10 of 12 GPS Solutions (2019) 23:41

respectively. Thus, the NTCM is found approx. 65 times NeQuickG computationally expensive in terms of time and
faster than the NeQuickG. processing power when compared with the GPS Klobuchar
Comparing STD, mean and RMS residual plots in pan- model. On the other hand, the proposed NTCM-based cor-
els (c), (d), (e) it can be seen that the performance of both rection approach is based on 12 model coefficients and a
models is very similar. Considering all samples the average few empirically fixed parameters. The TEC values can be
STD, mean and RMS are found as 9.5, 0.6, 10.0 TECU for computed at any location and time without using any spa-
NeQuickG and 9.3, 2.5, 10.1 TECU for NTCM, respectively. tial or temporal interpolation of parameters and numerical
To compare model performance during different seasons the integration of electron density. This means that the NTCM
monthly average STD, mean and RMS of STEC residuals is very fast running in operational applications and performs
were determined and given in Table 2. Additionally, the well when fed with the Az parameter. From the DLR point
monthly average values (labeled as mavg-NeQG and mavg- of view, the safety of life (SoL) applications would certainly
NTCM-GlAz) are plotted in the corresponding panel with benefit from the reduced complexity of the algorithm that
red and blue horizontal lines. greatly facilitates certification for aviation users. The com-
When comparing the monthly average STD, mean and pact NTCM algorithm is also favorable for “standard” users.
RMS values in Table 2 of both models, it is found that their It is assumed that most mass market and geodetic receiver
differences lie in the range of 1.0–1.6, 1.6–6.0 and 1.7–2.7 manufacturers would favor a compact algorithm.
TECU, respectively. When comparing the monthly mean This investigation using the NTCM proves that Galileo
values it is found that the NeQuickG model shows less bias Az parameters can be used to drive another ionosphere
than the NTCM model. The relatively higher magnitudes of model independent of the NeQuickG. This type of approach
STD and RMS residuals during the March and September does not require major technology changes for Galileo
equinoxes indicate higher STEC errors for both models. This users. Instead they need only introduce the NTCM formu-
is due to the relatively higher STEC observation values dur- las while maintaining the simplicity of ionospheric range
ing these periods. error mitigation.
The investigation using STEC reference data from over
200 GPS stations for more than 320 days in 2015 shows
that there is no significant difference in the performance Conclusions
of the NeQuickG and the NTCM. It is concluded that the
NTCM can be successfully driven by Galileo broadcast This paper presents an ionospheric algorithm based on the
ionization coefficients. The global and regional perfor- TEC model NTCM which provides TEC values with lower
mance analysis with reference VTEC data and independ- error than those of NeQuickG. The investigation shows that
ent Jason-2 data shows that the NTCM performance was the NTCM can be successfully driven by Galileo broadcast
at least equal to, or better than, the NeQuickG model. One ionization coefficients. The global and regional performance
hand, the NeQuickG is a 3D electron density model based on analysis with reference VTEC data from CODE, IGS and
several Epstein layers whose ionospheric peak density and from independent Jason-2 data shows that the perfor-
height parameters are derived using the spatial and tempo- mance of the NTCM was at least equal to, or better than,
ral interpolation of numerous global maps. This makes the the NeQuickG model in the global average. A comparison

Table 2  Model comparison Month NeQuickG NTCM-GlAzpar

in terms of monthly average
RMS, mean and STD of STEC RMS (TECU) Mean (TECU) STD (TECU) RMS (TECU) Mean (TECU) STD (TECU)
residuals in 2015
Jan 11.37 0.80 10.41 11.98 2.77 11.40
Feb 15.68 − 7.47 13.74 12.95 − 4.27 12.19
Mar 13.43 1.03 13.05 14.80 7.07 12.92
Apr 12.72 0.03 12.11 12.49 2.71 11.71
May 10.28 − 0.73 10.08 10.37 2.01 9.83
Jun 9.58 1.27 8.59 9.09 0.55 8.51
Jul 7.52 0.28 7.32 7.32 −  1.27 7.08
Aug 6.60 − 0.08 6.50 7.24 2.09 6.58
Sep 8.54 1.80 8.30 10.24 6.20 8.12
Oct 10.52 3.92 9.41 8.94 2.72 8.26
Nov 9.92 1.32 9.69 9.36 2.16 8.95
Dec 8.82 − 0.24 8.74 10.36 4.59 9.23

13
GPS Solutions (2019) 23:41 Page 11 of 12  41

with reference STEC data shows that there is no signifi- Hoque MM, Jakowski N, Berdermann J (2017) Ionospheric correc-
cant difference between both models performance in terms tion using NTCM driven by GPS Klobuchar coefficients for
GNSS applications. GPS Solut 21(4):1563–1572. https​://doi.
of residual statistics such as RMS, mean and STD. Here org/10.1007/s1029​1-017-0632-7
an improved mapping function could even reduce corre- Hoque MM, Jakowski N, Berdermann J (2018) Positioning perfor-
sponding errors when transforming NTCM-derived VTEC mance of the NTCM model driven by GPS Klobuchar model
to STEC values used for comparison. When compared the parameters. Space Weather Space Clim 8:A20. https​: //doi.
org/10.1051/swsc/20180​09
computation time it is found that the NTCM is in average IS-GPS-200J (2018) Global positioning system directorate systems
65 times faster than the NeQuickG. The proposed NTCM- engineering and integration interface specification. Michael J D,
based correction approach only uses 12 model coefficients Technical Director, Global Positioning Systems Directorate
and a few empirically fixed parameters. The necessary iono- Jakowski N, Hoque MM (2018) A new electron density model of the
plasmasphere for operational applications and services. J Space
spheric corrections can be computed at any location and Weather Space Clim 8:A16. https​://doi.org/10.1051/swsc/20180​
time without using any spatial or temporal interpolation of 02
internal parameters and time consuming numerical integra- Jakowski N, Hoque MM, Mayer C (2011a) A new global TEC model
tion of electron density along measured satellite-receiver for estimating transionospheric radio wave propagation errors.
J Geodesy 85(12):965–974. https ​ : //doi.org/10.1007/s0019​
links. This makes the model very fast running for operational 0-011-0455-1
applications. The current model coefficients are adapted to Jakowski N, Mayer C, Hoque MM, Wilken V (2011b) TEC models and
the available datasets of effective ionization coefficients their use in ionosphere monitoring. Radio Sci 46:RS0D18. https​
broadcast by the Galileo during the 2013–2017 period. The ://doi.org/10.1029/2010R​S0046​20
Jones WB, Gallet RM (1962) The representation of diurnal and geo-
model coefficients can easily be updated when additional graphical variations of ionospheric data by numerical methods.
data become available thus further improving the model per- ITU Telecommun J 29(5):129–149
formance. It is expected that the NTCM results will further Klobuchar JA (1987) Ionospheric time-delay algorithm for single
improve if the Az coefficients are directly adapted to the frequency GPS users. IEEE Trans Aerosp Electron Syst AES
23(3):325–332. https​://doi.org/10.1109/TAES.1987.31082​9
NTCM instead of the NeQuickG. Nava B, Coisson P, Radicella SM (2008) A new version of the
NeQuick ionosphere electron density model. J Atmos Terr Phys
Acknowledgements  We would like to thank the sponsors and operators 70(15):1856–1862. https​://doi.org/10.1016/j.jastp​.2008.01.015
of NASA’s Earth Science Data Systems and the CDDIS for archiving Orús-Pérez R (2018) Using tensor flow-based neural network to esti-
and distributing the IGS data as well as all institutes and partners who mate GNSS single frequency ionospheric delay (IONONet).
took part in data capture and made the data available to the IGS. We Adv Space Res 63(5):1607–1618. https​: //doi.org/10.1016/j.
thank the NOAA for making the Jason-2 data available to us via the asr.2018.11.011
National Center for Environmental Information. We are grateful to the Orús-Pérez R, Hernández-Pajares M, Juan JM, Sanz J, García-
ESA for providing historical data on the Galileo ionization coefficients Fernández M (2002) Performance of different TEC models to
broadcast in the 2013–2017 period. provide GPS ionospheric corrections. J Atmos Solar Terr Phys
64(18):2055–2062
Orús-Pérez R, Parro-Jimenez JM, Prieto-Cerdeira R (2018) Status of
NeQuick G after the solar maximum of cycle 24. Radio Sci. https​
://doi.org/10.1002/2017R​S0063​73
References OS SIS ICD (2010) European GNSS Galileo open service signal in
space interface control document, SISICD-2006, European Space
Dow JM, Neilan RE, Rizos C (2009) The international GNSS ser- Agency, Issue 1.1
vice in a changing landscape of global navigation satellite sys- Schaer S (1999) Mapping and predicting the earth’s ionosphere using
tems. J Geodesy 83(3–4):191–198. https​://doi.org/10.1007/s0019​ the global positioning system. Geod.-Geophys. Arb. Schweiz,
0-008-0300-3 vol. 59, Eidg. Technische Hochschule Zürich
European Commission (2016) Galileo open service, ionospheric cor- Schaer S, Beutler G, Rothacher M (1998) Mapping and predicting the
rection algorithm for Galileo single frequency receiver, issue 1, ionosphere. In: Dow JM, Kouba J, Springer T (eds) Proceedings
revision 2. https:​ //www.gsc-europa​ .eu/system
​ /files/​ galile​ o_docum​ of the 1998 IGS analysis center workshop, Darmstadt, February
ents/Galil​eo_Ionos​pheri​c_Model​.pdf. Accessed 5 Mar 2017 9–11. pp 307–318, ESA/ESOC, Darmstadt
Hochegger G, Nava B, Radicella SM, Leitinger R (2000) A family Wu X, Hu X, Wang G, Zhong H, Tang C (2013) Evaluation of COM-
of ionospheric models for different uses. Phys Chem Earth Part PASS ionospheric model in GNSS positioning. Adv Space Res
C Solar Terr Planet Sci 25(4):307–310. https​://doi.org/10.1016/ 51(6):959–968
S1464​-1917(00)00022​-2
Hoque MM, Jakowski N (2013) Mitigation of ionospheric mapping Publisher’s Note Springer Nature remains neutral with regard to
function error. In: Proceedings of ION GNSS 2013, Institute of jurisdictional claims in published maps and institutional affiliations.
Navigation, Nashville, Tennessee, USA, September 16–20, pp
1848–1855
Hoque MM, Jakowski N (2015) An alternative ionospheric correc-
tion model for global navigation satellite systems. J Geodesy
89(4):391–406. https:​ //doi.org/10.1007/s00190​ -014-0783-z (ISSN
0949-7714)

13
 41   Page 12 of 12 GPS Solutions (2019) 23:41

Mohammed Mainul Raul Orús Pérez  has a Physics

Hoque  received his Bachelor in degree from the University of
Electrical and Electronics Engi- Barcelona (UB) in 2000 and a
neering in 2000 and was awarded Ph.D. in Aerospace Science and
a Ph.D. in 2009 from the Univer- Technology from the gAGE/UPC
sity of Siegen, Germany. He has research group of the Technical
been working on ionospheric University of Catalonia (UPC) in
modeling including higher order 2005. Since 2010 he has been
propagation effects at the Ger- working in the Electromagnetics
man Aerospace Center since and Space Environment Division
2004. He is involved in several of ESA/ESTEC as a Propagation
national, ESA and EU projects Engineer. His main focus is on
related to ionospheric research. activities related to radio-wave
propagation in troposphere and
ionosphere for GNSS and other
ESA missions.

Norbert Jakowski  received his

Diploma in Physics in 1973 and
was awarded a Ph.D. in 1974
from the University of Rostock,
Germany. He has been working
at the German Aerospace Center
(DLR) in Neustrelitz, Germany
since 1991. His research activi-
ties include monitoring, mode-
ling and predicting ionospheric
processes related to space
weather and studying their
impact on radio wave propaga-
tion focusing on GNSS.

13