You are on page 1of 18

RESEARCH ARTICLE Global Three-Dimensional Ionospheric Tomography by

10.1029/2022SW003368
Combination of Ground-Based and Space-Borne GNSS Data
Key Points:
Dengkui Mei1, Xiaodong Ren1,2 , Hang Liu1, Xuan Le1, Si Xiong1, and Xiaohong Zhang1,2,3
• S pace-borne Global Navigation
Satellite System (GNSS) data from 1
School of Geodesy and Geomatics, Wuhan University, Wuhan, China, 2Hubei Luojia Laboratory, Wuhan, China, 3Chinese
11 Low-Earth-Orbit satellites were
Antarctic Center of Surveying and Mapping, Wuhan University, Wuhan, China
combined with the ground-based
GNSS data for global 3D ionospheric
tomography
• The incorporation of space-borne Abstract The inhomogeneous distribution of Global Navigation Satellite System (GNSS) data is
data significantly increased the total a significant problem for global GNSS computerized ionospheric tomography (CIT). By combining
number of voxels with observations
• Combining ground-based and
ground-based GNSS data from 460 global stations with space-borne GNSS data onboard eleven Low-Earth-
space-borne GNSS data improved the Orbit (LEO) satellites, we employed the EOF (empirical orthogonal function)-based multiplicative algebraic
performances of global ionospheric reconstruction technique (EMART) to reconstruct the global ionospheric electron densities. In comparison
tomography
with GNSS + LEO (EMART_GL) solutions, the proportion of crossed voxels for GNSS-only (EMART_G)
solutions can rise by up to 12.54%, from 56.83% to 69.37%. Validation experiments were conducted under quiet
Correspondence to: and disturbed conditions using GNSS slant total electron content (STEC) references that were not previously
X. Ren, used in ionospheric tomography. The EMART_GL solutions outperformed the EMART_G solutions when
xdren@whu.edu.cn
compared to the STEC references under quiet and disturbed conditions. The root mean square values of the
CIT-derived STEC for the EMART_G solutions can be decreased by 31.3% and 30.3% on average under quiet
Citation:
and disturbed conditions, respectively. The vertical TEC references from the Jason-2 satellite were also used
Mei, D., Ren, X., Liu, H., Le, X.,
Xiong, S., & Zhang, X. (2023). for validation over the ocean areas. Due to insufficient data over the ocean areas, the EMART_GL solutions
Global three-dimensional ionospheric showed no significant improvements. The reconstructed 3D IEDs were further validated with the COSMIC
tomography by combination of electron density profiles. Generally, the tomographic results of the EMART_GL solutions were more consistent
ground-based and space-borne
GNSS data. Space Weather, 21, with the COSMIC profiles when compared to those of the EMART_G solutions and the NeQuick-2 model.
e2022SW003368. https://doi. It demonstrates that it is beneficial for global ionospheric tomography by combining ground-based and
org/10.1029/2022SW003368 space-borne data.
Received 5 DEC 2022
Accepted 24 MAR 2023
Plain Language Summary The computerized ionospheric tomography technique is extensively
being applied to imaging ionospheric electron density (IED) distributions based on Global Navigation Satellite
System (GNSS) observations. The voxel-based model is usually utilized in ionospheric tomography, with
the ionosphere being divided into many cubic voxels. However, predominant issues like unevenly distributed
stations, a lack of data, and high-elevation GNSS signals could affect the precise information about the
ionosphere. The electron densities of voxels that are not crossed by GNSS rays cannot be accurately estimated.
To address this problem, incorporating multisource data, such as data from Low-Earth-Orbit (LEO) satellites,
make great sense. A growing number of LEO satellites offer new opportunities for global IED reconstruction.
In this study, the EOF (empirical orthogonal function)-based multiplicative algebraic reconstruction technique
(EMART) was used to reconstruct the 3D IED by combining ground-based GNSS data from global receivers
and space-borne GNSS data onboard eleven LEO satellites.

1. Introduction
Global monitoring and exploration of ionospheric space weather are highly demanded for accurate and precise
representations of electron density (F. S. Prol, Kodikara, et al., 2021). There have been well-established tech-
niques for measuring ionospheric electron density (IED), exemplified by ionosondes, incoherent scatter radars,
and in situ probes (Peng et al., 2021; Smirnov et al., 2021). However, such measurements are typically constrained
© 2023. The Authors. to a single place (Norberg et al., 2018; Pryse, 2003). Computerized ionospheric tomography (CIT), first proposed
This is an open access article under by Austen et al. (1988), has developed into a new approach for imaging the IED and comprehending ionospheric
the terms of the Creative Commons
Attribution-NonCommercial-NoDerivs plasma dynamics. Although the Navy Navigational Satellite System’s polar-orbiting satellites at an altitude of
License, which permits use and approximately 1,100 km can produce two-dimensional ionospheric images via radio tomography, these satellites
distribution in any medium, provided the can only be monitored via a series of ground receivers aligned in longitude but separated in latitude (Fremouw
original work is properly cited, the use is
non-commercial and no modifications or et al., 1992; Leitinger et al., 1997; Pryse & Kersley, 1992). Following that, it was demonstrated how to reconstruct
adaptations are made. the three-dimensional ionospheric structures using high-orbit satellites, specifically the global navigation satellite

MEI ET AL. 1 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

system (GNSS) (Hajj et al., 1994; Kunitsyn et al., 1997). Using GNSS receivers from international GNSS services
(IGS) or regional networks, extensive research has been conducted to extract global or regional three-dimensional
snapshots of the Earth’s ionosphere using tomographic techniques (Bust & Mitchell, 2008; Chen et al., 2019;
Kunitsyn et al., 2008; Mitchell & Spencer, 2003; Nesterov & Kunitsyn, 2011; F. S. Prol, Kodikara, et al., 2021;
Wen, Yuan, et al., 2010; Yao, Chen, et al., 2013; Yao et al., 2020; Zheng et al., 2021).

Due to the fact that GNSS receivers are usually positioned in an imbalanced manner, with high data coverage over
land but not over ocean areas, there are still regions in the ionospheric inversion that are not crossed by GNSS
rays. These regions devoid of rays lead to data gaps (Jayawardena et al., 2016; Nesterov & Kunitsyn, 2011; Ren
et al., 2021; Schmidt et al., 2008). The incompleteness of input data results in the typical ill-posed problem of
the tomographic system (Nesterov & Kunitsyn, 2011; Wen et al., 2008; Wen, Liu, & Tang, 2010; Yao, Tang,
et al., 2013).

In comparison with GNSS satellites, Low-Earth-Orbit (LEO) satellites are closer to the Earth and move signif-
icantly quicker in the sky, passing overhead in minutes rather than hours as they do in medium Earth orbit. The
addition of LEO constellations adds a wealth of geometric diversity (Jayawardena et al., 2016). The dual-frequency
GNSS receiver for precise orbit determination (POD) onboard the LEO satellite can track up to 8–12 different
GNSS satellites in a large spatial volume above orbital altitudes and can observe the ionosphere ahead, behind,
and aside of the LEO position (Zakharenkova et al., 2016). Therefore, GNSS data from LEO satellites can provide
additional upward looking slant total electron content (STEC) measurements that enable global imaging of the
topside ionosphere and plasmasphere with ionospheric tomography (Bust & Mitchell, 2008; Gerzen et al., 2020;
Jayawardena et al., 2016; H. Li et al., 2012; F. S. Prol, Hoque, & Ferreira, 2021). To maintain the high accuracy
of tomographic reconstructions, the region's ray path coverage must be enhanced using many hours of data.
Nevertheless, the hours-long temporal resolution is insufficient to capture the topside ionosphere-plasmasphere's
dynamic features during storms (Jayawardena et al., 2016).

In addition to imaging the ionosphere independently utilizing ground-based or space-borne STEC data, Hajj
et al. (1994) pioneered the use of ground-based and space-based dual-frequency GNSS receivers for ionospheric
imaging. Since then, radio occultation (RO) data from LEO satellites have been used to increase the spatial reso-
lution, and particularly the vertical resolution, of tomographic models (Cheng et al., 2021; Hernández-Pajares
et al., 1998; H. Li et al., 2012; Rius et al., 1997; She et al., 2017). Generally, there are a finite number of satellites
available for a single LEO mission. Ren et al. (2021) explored the performances and benefits of incorporating the
LEO-based ionospheric data for global 3D ionospheric tomography through simulation experiments. However,
it is unknown whether combining ground-based ionospheric data with real space-borne ionospheric data from a
dozen LEO satellites will enhance tomographic performance. Therefore, the purpose of this study is to focus on
the methodology, performance, and validation of LEO-augmented GNSS ionospheric tomography by combining
space-borne GNSS observations onboard 11 LEO satellites with ground-based IGS GNSS observations.

This paper is structured as follows: Section 2 describes the estimation method of STEC measurements from
GNSS and LEO satellite sounding data and the ionospheric tomographic algorithm. Section 3 describes the data
sets that were used in this study. Section 4 compares and validates the tomographic results using the TEC refer-
ences and COSMIC (FORMOSAT-3/Constellation Observing System for Meteorology Ionosphere and Climate)
profiles. Section 5 discusses the tomographic performances of different methods. Finally, Section 6 summarizes
the conclusion.

2. Methodology
2.1. Ionospheric STEC Measurement Estimation

The ionospheric STEC measurements along the signal path can be obtained by using the carrier to code leveling
(CCL) method as follows (Z. Li et al., 2015, 2021; Ren, Chen, et al., 2020):
( )
̃ 1 1
(1) 𝐿𝑟,4 = 40.3 ×
𝑠,𝑘
− 2 × STEC𝑘 + 𝑐 ⋅ DCB𝑟,𝑖1 𝑖2 + 𝑐 ⋅ DCB𝑠𝑖1 𝑖2
𝑓𝑖21 𝑓𝑖2

where 𝐿𝐿̃ 𝑠𝑠𝑠𝑠𝑠


𝐴𝐴 𝑟𝑟𝑟4
denotes the ionospheric observations that are extracted from the CCL𝐴𝐴at the 𝐴𝐴th epoch of a continuous
arc in meters;
𝐴𝐴 STEC𝑘𝑘 is the ionospheric delay along the line-of-sight𝐴𝐴at the 𝐴𝐴th epoch of a continuous arc in TECU

MEI ET AL. 2 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

(1 TECU = 10 16el/m
𝐴𝐴 2); DCB𝑟𝑟𝑟𝑟𝑟1𝐴𝐴𝑖𝑖2 and DCB𝑠𝑠𝑖𝑖1 𝑖𝑖2 represent differential code biases (DCBs) of receiver and satellite,
respectively. Equation 1 can be rewritten as:

𝑓𝑖21 𝑓𝑖22 ( )
STEC𝑘 = − ( )× 𝐿̃ 𝑠,𝑘 − 𝑐 ⋅ DCB𝑟,𝑖1 𝑖2 − 𝑐 ⋅ DCB𝑠𝑖 𝑖
(2) 𝑟,4 1 2
40.3 × 𝑓𝑖21 − 𝑓𝑖22

here, the STEC value is biased by the receiver and satellite DCBs. Therefore, receiver and satellite DCBs should
be estimated and deducted in order to get clean STEC measurements. Generally, we can use all stations over the
world to model the global ionospheric map (GIM), which will generate the receiver and satellite DCB estimates
with high accuracy. In this study, we established GIMs using an expansion of spherical harmonics of 15 × 15 with
a 2-hr interval. The following is the global ionospheric model (Ren, Chen, et al., 2020):

VTEC(𝜑, 𝜆) = MF × STEC(𝜑, 𝜆)
(3) ∑ ∑ ̃
𝑛max 𝑛 ( )
= 𝑃𝑚𝑛 (sin 𝜑) 𝐶̃𝑚𝑛 cos(𝑚𝜆) + 𝑆̃𝑚𝑛 sin(𝑚𝜆)
𝑛=0 𝑚=0

where
𝐴𝐴 𝐴𝐴 and 𝐴𝐴 denote the max degree and order of spherical harmonic function, respectively;
𝐴𝐴 𝐴𝐴 𝑃𝑃̃𝑚𝑚𝑚𝑚 represents a
regularization Legendre series of degree n and order
𝐴𝐴 m; 𝐴𝐴 and 𝐴𝐴 are the geomagnetic latitude and solar-fixed
longitude of ionospheric pierce point (IPP), respectively;
𝐴𝐴 𝐴𝐴 and 𝑆𝑆̃𝑚𝑚𝑚𝑚 are spherical harmonic coefficients to be
C̃ 𝑚𝑚𝑚𝑚
estimated. MF is a modified single-layer mapping function used to convert STEC to vertical TEC (VTEC):

1
MF(𝑧𝑧) =
(4)
( ( ))
𝑅𝑅
cos arcsin 𝑅𝑅+𝐻𝐻 sin(0.9782 × 𝑧𝑧)

where R is the earth radius, H is the effective height of the thin-shell model equaling to 450 km, and z is the
satellite zenith angle at the receiver. The receiver and satellite DCBs were estimated as constant values for each
day, and they were separated using a zero-mean condition for each used satellite system.
Similarly, the STEC measurements derived from dual-frequency GNSS receivers onboard LEO satellites using
the CCL method are also biased by receiver and satellite DCBs. The spherical harmonic function in Equation 3
is also a popular method for modeling the global topside ionosphere or plasmasphere based on the single-layer
assumption. Due to a limited number of topside ionospheric observations, the DCBs are estimated simultane-
ously with the eight-order spherical harmonic expansion every 2 hr on a global scale (Ren, Chen, et al., 2020).
Then the clean STEC measurements will be obtained by deducting receiver and satellite DCBs. To generate the
global topside ionospheric model, Foelsche and Kirchengast (2002) proposed the following F&K (Foelsche and
Kirchengast) geometric mapping function. It can be expressed as:

1 + 𝑅𝑅shell ∕𝑅𝑅orbit
MF(𝑧𝑧) =
(5) √
cos(𝑧𝑧) + (𝑅𝑅shell ∕𝑅𝑅orbit )2 − (sin(𝑧𝑧))2

where
𝐴𝐴 𝐴𝐴shell = 𝑅𝑅E +𝐴𝐴ℎshell, 𝐴𝐴𝐸𝐸 is the Earth's radius
𝐴𝐴 and 𝐴shell is the single-layer height;
𝐴𝐴 𝐴𝐴orbit denotes the alti-
tude of LEO satellite orbit; z is the zenith angle of space-borne receiver. The single-layer shell height was esti-
mated by taking into account the ionosphere's dependence on solar activity (Ren, Chen, et al., 2020; Zhong, Lei,
et al., 2016). Specifically, the thin-shell height was fixed at 1,600 km in this study.

2.2. Reconstruction Method of Ionospheric Electron Density (IED)

In ionospheric tomography, the selected region is usually divided into small cubic voxels. In longitudinal, latitudi-
nal, and altitudinal directions, the ionosphere is discretized into i layers, j layers, and k layers, respectively (Zheng
et al., 2015). The three-dimensional spatial domain chosen for analysis in this study includes latitudes ranging
from −180° to +180° with a resolution of 10°; longitudes ranging from −90° to 90° with a resolution of 5°; and
altitudes ranging from 100 to 2,000 km with a resolution of 50 km and then up to 5,000 km with a resolution
of 100 km. The voxel-based tomographic model’s latitudinal and longitudinal dimensions are both 36, whereas
the altitudinal dimension is 68. In this case, the total number of voxels, or unknown variables to be estimated, is
36 × 36 × 68 = 88,268. Given the scarcity of space-borne receivers in comparison to ground-based receivers,
the time window of the input STEC data for ionospheric tomography was 5 min for GNSS and 20 min for LEO.

MEI ET AL. 3 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

In the voxel-based ionospheric tomographic model, the ground-based and space-borne STEC values can be
approximated by Equations 6 and 7:

𝑛𝑛

(6)
STEC𝑖𝑖ground−based = 𝐴𝐴𝑖𝑖𝑖𝑖 𝑥𝑥𝑗𝑗
𝑗𝑗

𝑛𝑛

STEC𝑝𝑝space−borne =
(7) 𝐴𝐴𝑝𝑝𝑝𝑝 𝑥𝑥𝑞𝑞
𝑝𝑝=1

In Equations 6 𝐴𝐴
and 7, STECground−based
𝐴𝐴 and STECspace−borne refer to ground-based and space-borne STEC measure-
ments, respectively.

The above two Equations can be rewritten in matrix form:

(8)
𝒚𝒚 𝑚𝑚×1 = 𝑨𝑨𝑚𝑚×𝑛𝑛 𝒙𝒙𝑛𝑛×1

where 𝒚𝒚 is a m × 1 column vector containing the m known STEC measurements,


𝐴𝐴 𝐴𝐴 𝑨𝑨 is a m × n matrix containing
all the intersects of the m rays traversing the corresponding n voxels,
𝐴𝐴 and 𝒙𝒙 is a n × 1 column vector containing
the n unknown electron densities of voxels (Zheng et al., 2021). Due to the data insufficiency, many voxels are not
crossed by any GNSS rays, and the number of measured STECs is much smaller than that of the unknowns; thus,
in a voxel-based tomographic model as shown in Equation 8, it is often an ill-conditioned problem. To reduce the
number of unknowns, the vertical profiles of the ionosphere can be approximated as a linear combination of a
few dominant orthonormal functions (Hong et al., 2017; Howe et al., 1998; Mitchell & Spencer, 2003; Ssessanga
et al., 2015).

The voxel-based tomographic solutions obtained using iterative algebraic reconstruction techniques, such as the
multiplicative algebraic reconstruction technique (MART), are heavily influenced by the ionosphere background.
This mainly occurs because the algebraic technique only modifies the iterative values of voxels traversed by
GNSS rays. As for the voxels that may have no intersections with signal rays, their tomographic results are
the same as the background. As a result, the background ionosphere must be represented accurately (F. D. S.
Prol & Camargo, 2015). Considering that the voxel-based tomographic model used in this study extends to
approximately 5,000 km in altitude and the NeQuick-2 model provides a more accurate representation of the
topside ionosphere (Gerzen et al., 2020; Nava et al., 2008; Wang et al., 2017), the vertical profiles at each hour
of the 30 days before the tomographic day that were derived from the NeQuick-2 model are used to construct an
original data set. Specifically, the total number of all vertical profiles from the 36 × 36 = 1,296 vertical bins is
30 × 24 × 1,296 = 933,120. The empirical orthogonal functions (EOFs) are obtained by carrying out singular
value decomposition on the data set of these vertical profiles. Generally, the three largest EOFs at each hour can
represent more than 98% of the original density profiles. Therefore, Equation 9 is applied to map the inverse
problem from voxel space into EOF space defined by three EOFs derived from the NeQuick-2 model, which is
expressed as

(9)
𝑿𝑿 = 𝑴𝑴 𝑻𝑻 𝒙𝒙

where
𝐴𝐴 𝒙𝒙 is the unknown electron density in the inversion problem (Equation𝐴𝐴 8), 𝑿𝑿 is obtained by the mapping
process of the inversion problem from voxel space to EOF 𝐴𝐴space, 𝑴𝑴 is an orthonormal matrix constructed from
the EOFs to enable the mapping. Set

(10)
𝑯𝑯 = 𝑨𝑨𝑨𝑨

Then Equation 8 can be rewritten as

(11)
𝒚𝒚 = 𝑯𝑯𝑯𝑯

MEI ET AL. 4 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

Figure 1. Distributions of IGS GNSS stations.

The unknown solution


𝐴𝐴 𝒙𝒙 in Equation 8 is usually estimated using the MART (Bender et al., 2011; F. D. S. Prol &
Camargo, 2016). Similarly,
𝐴𝐴 the 𝑿𝑿 can be obtained by applying Equations 9–11 to the MART algorithm, and we
can derive the EOF-based MART (EMART) algorithm:
| 𝜆𝜆𝑯𝑯 𝑖𝑖 |
( )||| 𝑖𝑖 𝑗𝑗 𝑖𝑖 |||
| ⟨𝑯𝑯 ,𝑯𝑯 ⟩ |
(12)
𝑿𝑿 𝑘𝑘+1
𝒚𝒚
= 𝑿𝑿 𝑘𝑘𝑗𝑗 ⋅ ⟨ 𝑖𝑖 𝑖𝑖 𝑘𝑘 ⟩
| |
𝑗𝑗
𝑯𝑯 , 𝑿𝑿

where
𝐴𝐴 𝐴𝐴 is the relaxation parameter and is set to be 0.01. It should be noted that 𝐴𝐴 the 𝑿𝑿 is not the final elec-
tron density values that we need. Once the EMART algorithm reaches the termination condition, that is,
‖ ≤ 0.01, then the normal electron density solution to Equation 8 in the voxel basis can be obtained back
‖ 𝑿𝑿 𝑘𝑘+1 −𝑿𝑿 𝑘𝑘 ‖
𝐴𝐴 ‖ 𝑿𝑿
‖ 𝑘𝑘 ‖
𝐴𝐴 as 𝒙𝒙 = 𝑴𝑴𝑴𝑴 (Hong et al., 2017).

3. Data Sets and Processing Procedure


3.1. Ground-Based GNSS Data

Ground-based STEC measurements were derived in this study from about 460 ground-based IGS stations, most of
which can provide GNSS dual-frequency observations. The station locations are shown in Figure 1. The sample
rate of ground-based GNSS observations is 30 s. An elevation mask of 15° is used to reduce multipath effects.

3.2. Space-Borne GNSS Data

The majority of LEO satellites orbiting at various altitudes are equipped with a dual-frequency GNSS receiver
and a zenith-looking antenna that can be used not only for POD, but also as a new source of information about
the topside ionosphere, namely, up-looking STEC measurements from onboard receivers to GNSS satellites.
The space-borne STEC measurements used in this study were mainly from dual-frequency data acquired by the
following LEO satellite missions: GRACE, Swarm, TerraSAR-X, Kompsat, COSMIC, and MetOp. The corre-
sponding sub-satellite points of these LEO satellites for 1 hr are illustrated in Figure 2. The sample rate of
space-borne GNSS observations is 10 s.

It should be noted that by establishing a global topside ionospheric map (GTIM), the DCBs of satellites and
space-borne receivers have been estimated and deducted from the STEC measurements (Ren, Chen, et al., 2020).
The orbit altitudes of LEO satellites are summarized in Table 1.

3.3. Processing Procedure and Strategy

The flow chart of the processing procedure and strategy is illustrated in Figure 3. First, the GNSS data is collected
from 460 IGS GNSS stations and 11 LEO satellites. Second, the GIM and global topside ionospheric map are

MEI ET AL. 5 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

Figure 2. Sub-satellite points of LEO satellites within 1 hr.

generated, respectively, using the spherical harmonics function. Then, we can obtain clean STEC measurements
by deducting both receiver and satellite DCBs. Next, we establish a global 3D ionospheric tomographic model
and estimate the ionospheric density using the EMART algorithm. Finally, the tomographic results are validated
by the GNSS STEC references, the Jason-2 VTEC references, and the COSMIC ionospheric electron density
profiles (EDPs) on 4 March 2015, a quiet time, and on 17 March 2015, a disturbed time.
To demonstrate the superiority and reliability of GNSS + LEO (EMART_GL) solutions over GNSS-only
(EMART_G) solutions, we randomly selected 5% of the space-borne GNSS STEC measurements as references
that were not used in tomography. Generally, the number of measurements at each moment is different. Specif-
ically, the number of measurements used in each epoch is 35,342 and 4,880 on average for ground-based and
space-based measurements, respectively. And 5% STEC measurements at each moment for validation are, on
average, 2,232 and 308 for ground-based and space-based measurements, respectively. These references were
mainly located in the region where the space-borne observations were denser than the ground-based observations.
STEC values derived from tomographic results can be calculated by summing all the products of electron density
in each voxel and its corresponding intersect along the single ray, which can be expressed as follows:
N
(13)

STECCIT = 𝐴𝐴𝑛𝑛 𝑥𝑥𝑛𝑛
𝑛𝑛

where STECCIT denotes the CIT-derived STEC value, N is the total number of
𝐴𝐴
crossed voxels along the ray𝐴𝐴path,𝐴𝐴𝐴𝐴 and 𝐴𝐴 are the electron density and corre-
Table 1 sponding intersect in
𝐴𝐴 the 𝐴𝐴th voxel, respectively. The mean absolute error
Summary of LEO Satellites Used in the Study (MAE) and root mean square (RMS) errors are calculated by Equations 14
and 15, respectively.
Satellites Height (km) Orbit/Inclination Measurements/Range (km)
GRACE-1 410–430 Circular/89° STEC, 420-20200 (14)
MAE = ⟨|STECCIT − STECref |⟩
GRACE-2 410–430 Circular/89° STEC, 420-20200 √⟨
(15) (STECCIT − STECref )2

RMS =
Swarm-A 460 Circular/87.5° STEC, 460-20200
Swarm-C 460 Circular/87.5° STEC, 460-20200
Where
𝐴𝐴 STECref denotes the referenced STEC values and the angle brackets
Swarm-B 510–520 Circular/87.5° STEC, 510-20200 denote averaging.
TerraSAR-X 515 Circular/97.4° STEC, 515-20200
Kompsat-5 559–561 Circular/97.0° STEC, 550-20200 4. Results and Analysis
COSMIC-1 767–838 Circular/72° STEC, 800-20200 4.1. Improvements of Geometrical Distribution
COSMIC-6 774–829 Circular/72° STEC, 800-20200
Figure 4 shows the number of crossed voxels before and after the addition
MetOp-A 826–828 Circular/98.7° STEC, 827-20200
of LEO data at various layers. And Figure 5 illustrates the number of rays
MetOp-B 826–828 Circular/98.7° STEC, 827-20200 in crossed voxels for GNSS, LEO, and GNSS + LEO at UT12 on 4 March

MEI ET AL. 6 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

Figure 3. Flow chart of processing procedure and strategy.

2015. As illustrated in Figure 4, the number of crossed voxels increased significantly when GNSS and LEO satel-
lites were combined, whereas the number of crossed voxels below the sixth layer remained constant for GNSS
and GNSS + LEO solutions because the orbit altitudes of LEO satellites are higher than this layer at a height of
400 km.

Figure 5b also shows that there were no crossed voxels at a height of 150 km. It can be seen from Figure 4 that
as the layer count increased, the number of crossed voxels increased at first, then reduced, which was caused
by the height resolution of 50 km below and 100 km above the 36th layer. As illustrated in Figure 5, additional
measurements from LEO satellites not only increased the number of crossed voxels but also reduced data gaps
over the ocean and polar regions.

Figure 6 shows the percentage of crossed voxels in total voxels on 4 March 2015. It revealed that the addition
of LEO satellites increased the percentage of crossed voxels by 12.54% at most, from 56.83% to 69.37%, when
compared to the GNSS-only solutions.

4.2. Validation by Space-Borne GNSS STEC References


Figures 7 and 8 show the RMS values and MAE values of the CIT-derived STEC values during quiet conditions
on 4 March and during storm conditions on 17 March, respectively. The corresponding average MAE and RMS
errors are listed in Table 2. In Figure 7, the tomographic results of EMART_GL solutions in terms of RMS and

MEI ET AL. 7 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

MAE values were much better in general during quiet conditions. The MAE
values were decreased by 35.4% on average, from 2.45 TECU to 1.55 TECU,
and the RMS values were decreased by up to 30.6%, from 3.06 TECU to 2.09
TECU, indicating that tomographic results are more likely to be improved
by including LEO-based data when comparing with the LEO-based STEC
references.

In Figure 8, the average MAE value was reduced by 35.9%, from 2.74 TECU
to 1.74 TECU, and the average RMS value was reduced by 30.3%, from
3.39 TECU to 2.35 TECU, after the combination of LEO data during storm
conditions. In comparison with the space-borne GNSS STEC references, the
EMART_GL solutions outperformed the EMART_G solutions under both
quiet and disturbed conditions. This indicated the advantage of combining
ground-based and space-borne GNSS data for global ionospheric tomography.

4.3. Validation by Jason-2 VTEC References


Jason-2, an ocean altimetry satellite orbiting at about 1,336 km, can obtain
direct VTEC measurements utilizing the dual-frequency signals (Liu
et al., 2022). The Jason-2 VTEC measurements can be used to validate the
ionospheric tomographic performances over the ocean areas as an independ-
ent TEC data source. The Jason-2 VTEC products used in this study were
provided by the Madrigal Database (http://cedar.openmadrigal.org/index.
html). The CIT-derived VTEC values of 24 tomographic moments in a day
were integrated from 100 km to Jason-2's orbit altitude. All the CIT-derived
VTEC values were compared with the Jason-2 VTEC references. Figures 9
and 10 present the scatter plots between the integrated VTEC of the
Figure 4. Number of crossed voxels of each layer at UT12 on 4 March 2015.
NeQuick-2, EMART_G, and EMART_GL solutions, and the Jason-2 VTEC
on 4 March and 17 March, respectively. According to Figure 9, the tomo-
graphic solutions (as shown in Figures 9b and 9c) performed better than the NeQuick-2 model (as shown in
Figure 9a). When compared to the EMART_G solutions, the EMART_GL solutions exhibited the lowest RMS
value (7.059 TECU), enhancing the tomographic results slightly.
It should be mentioned that Jason-2 data was only available before UT 07, while data after UT 07 was unavaila-
ble due to data missing on 17 March. As a result, there weren't many VTEC values for comparison (as shown in
Figure 10). The comparison results on 17 March were similar to those on 4 March. The tomographic solutions
outperformed the empirical NeQuick-2 model, and the EMART_GL solutions performed the best.
In general, the EMART_GL solutions were slightly better than those of the NeQuick-2 model and the EMART_G
solutions for both quiet and storm conditions when compared to the Jason-2 VTEC references. It is known that
the Jason-2 VTEC references mainly cover the ocean areas, where the ground-based and space-borne data are
sparse. The comparison results of the tomographic solutions with the Jason-2 VTEC references were much worse
than those with the space-borne GNSS STEC references. Due to the lack of data in the ocean areas, it showed that
adding LEO-based data did not significantly improve the tomographic results across those ocean areas.

4.4. Validation by COSMIC Electron Density Profiles (EDP)

The COSMIC-1 satellites provided thousands of high vertical resolution profiles of electron density every day in
the ionosphere in 2015. Figures 11 and 12 give the comparisons of EDPs between the NeQuick-2, EMART_G,
and EMART_GL solutions and the COSMIC profiles on 4 March and 17 March, respectively. Due to incomplete
data, several tomographic moments of the 2 days were not supplied.
As illustrated in Figure 11, the tomographic EDPs obtained by the EMART_G and EMART_GL solutions
showed better agreement with the COSMIC profiles than those of the NeQuick-2 model in most cases, as shown
in Figures 11a, 11b, 11e, 11f, 11i, 11n, 11r, 11u, and 11v, and so on. In several cases, as shown in Figures 11c,
11j, 11l, 11o, 11q, and 11r, the EMART_GL solutions performed the best, showing the most consistent variations
with the COSMIC profiles. The maximum electron density values of the EMART_GL solutions were also very
close to those of the COSMIC profiles. In some cases, as shown in Figures 11a, 11b, 11d, 11e, 11i, and 11n, the

MEI ET AL. 8 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

Figure 5. Number of rays in crossed voxels at different layers for (a) GNSS, (b) LEO and (c) GNSS + LEO at UT12 on 4 March 2015.

EMART_GL solutions' tomographic profiles were nearly identical to those of the EMART_G solutions, with no
significant improvements. The iterative tomographic results of the EMART_GL algorithm were only improved
when additional LEO-based rays passed through the voxels. If not, the tomographic results of the EMART_G
algorithm will most likely be equivalent to those of the EMART_G algorithm.
As illustrated in Figure 12, the tomographic results of both the EMART_G and EMART_GL solutions were
superior to those of the NeQuick-2 model in most cases, as shown in Figures 12d, 12e, 12f, 12g, and 12j. In some
cases, as shown in Figures 12a, 12b, 12g, and 12i, the EMART_G solutions were almost the same as the empirical
model. However, the EMART_GL solutions were much closer to the COSMIC profiles. There were also some
cases where the tomographic results almost showed no improvement when compared to the empirical model, as
shown in Figures 12c, 12h, and 12i. Additionally, the tomographic results during storm conditions were slightly
worse than those during quiet conditions. On 17 March, for instance, there were some discrepancies between the
reconstructed profiles and the COSMIC ones, as shown in Figures 12i, 12k, and 12l.

Figure 6. Percentage of crossed voxels in total voxels on 4 March 2015.

MEI ET AL. 9 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

Figure 7. (a) MAE values and (b) RMS values of the CIT-derived TEC values on 4 March 2015.

In order to further quantitatively assess the improvements when compared to the COSMIC profiles, we calculated
the MAEs and RMS values of the maximum electron densities of all tomographic EDPs in Figures 11 and 12,
respectively. Table 3 shows the MAE and RMS values of the maximum electron density values obtained by the
NeQuick model, the EMART_G and EMART_GL solutions on 4 March and 17 March 2015. From Table 3, we
can see that the EMART_GL algorithm performed the best with the smallest RMS of 1.761 × 10 11el/m 3 and
2.440 × 10 11el/m 3 in quiet and disturbed conditions, respectively, followed by the EMART_G algorithm, while
the NeQuick-2 model performed the worst. In addition, the tomographic performances of the EMART algorithm
in quiet time were better than those in disturbed time.

Figure 8. (a) MAE values and (b) RMS values of the CIT-derived TEC values on 17 March 2015.

MEI ET AL. 10 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

Table 2
Average MAE and RMS Values of CIT-Derived TEC for the EMART_G and EMART_GL Solutions on 4 March and 17
March 2015
Mean absolute error (×10 11el/m 3) Root mean square error (×10 11el/m 3)

Time EMART_G EMART_GL Improvement EMART_G EMART_GL Improvement


03.04 2.45 1.55 35.4% 3.06 2.09 30.6%
03.17 2.74 1.74 35.9% 3.39 2.35 30.3%

5. Discussion
The advantage of adding LEO satellite observation value is that it can make up for the ionospheric observation
gap around the world (F. S. Prol, Kodikara, et al., 2021), especially over the ocean and polar areas. Based on
simulated data, it has been demonstrated that LEO-based ionospheric observations have better coverage and a
higher density of IPPs in polar and ocean regions (Ren, Zhang, et al., 2020). Besides, the rays of LEO satellites
flying rapidly had better geometric distributions than ground-based GNSS rays, which can be used to reduce the
number of voxels with no observations (Ren et al., 2021).
In this study, the number of LEO satellites is limited (a total of 11). Figures 13a–13c illustrate an example of
the three-dimensional distributions of rays at UT12 on 4 March 2015, for GNSS, LEO, and GNSS + LEO,
respectively. Figure 13d depicts the vertical view of rays for GNSS + LEO. During each global CIT, the LEO
space-borne observations cannot completely cover the ocean, and there are still observation gaps in most of the
ocean areas (as shown in Figure 13d). To investigate the improvements of the EMART_GL algorithm over differ-
ent areas with dense or sparse LEO observations, Figures 14 and 15 present the comparisons of the CIT-derived
STEC values of the EMART_G and EMART_GL solutions with the GNSS-only and LEO-only STEC references
on 4 March and 17 March, respectively. It can be seen that the accuracy of the EMART_GL algorithm is signif-
icantly improved compared to EMART_G when using LEO-only STEC as references; however, the accuracy
of the EMART_GL algorithm is slightly improved at some moments when compared to GNSS-only STEC as
references. The model accuracy is improved because the LEO-only references are primarily distributed over the
areas (see the red areas in Figure 13d) where LEO satellites flew. Nevertheless, the GNSS-only references are
mainly distributed over land areas (most of the blue areas in Figure 13d), where there are few LEO-based obser-
vations and the improvement of the model accuracy is limited. It indicated that the model accuracy is usually
improved over the areas with denser LEO observations, while the areas with few LEO-based observations will
not see significant improvements. In comparison with the COSMIC profiles, there were also some cases where
the tomographic results almost showed no improvement when compared to the empirical model, as shown in
Figures 12c, 12h, and 12i. This was most likely because the regions where RO events occurred were not covered
by the additional LEO-based measurements. The EMART_GL solutions were nearly identical to the EMART_G
solutions along the COSMIC profiles.

Figure 9. Comparison of the CIT-derived VTEC values of the (a) NeQuick-2, (b) EMART_G and (c) EMART_GL solutions with the Jason-2 VTEC values on 4
March 2015.

MEI ET AL. 11 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

Figure 10. Comparison of the CIT-derived VTEC values of the (a) NeQuick-2, (b) EMART_G and (c) EMART_GL solutions with the Jason-2 VTEC values on 17
March 2015.

In Section 4.2, the average MAE and RMS values of the CIT-derived TEC for the EMART_G and EMART_GL
solutions on 17 March were larger than those on 4 March. Similarly, the RMS values were also higher during the
storm when compared to the Jason TEC data. Besides, we noted that the RMS values for the maximum electron
density values obtained by the EMART algorithm during the storm day were slightly larger than those during the
quiet day. It was shown that the tomographic performances of both the EMART_G and EMART_GL algorithms
were worse during the storm time than those during the quiet time. As shown in Figure 16, the geomagnetic
condition on 4 March (Day of year: DOY = 63) was quiet, while on 17 March (DOY = 76), a strong geomag-
netic storm occurred with a minimum Dst value of −234 nT. It has been verified that the analyzed St. Patrick’s
Day storm on 17 March 2015 was the strongest storm during the 24th solar cycle, causing strong ionospheric
disturbances during the storm's main phase (Astafyeva et al., 2015; Cherniak & Zakharenkova, 2016; F. S. Prol,
Kodikara, et al., 2021). Due to the fact that STEC is an integral parameter along the line of sight between a
receiver and a GNSS satellite, it was revealed that the topside TEC contribution to ground-based TEC may vary
by 10%–60% during geomagnetic storms (F. S. Prol, Hoque, & Ferreira, 2021). Previous studies have reported
the substantial differences in the storm-induced responses during different storm events, including the unusual
topside ionospheric density responses to large magnetic storms (Danilov, 2013; Yizengaw et al., 2006; Zhong,
Lei, et al., 2016; Zhong, Wang, et al., 2016; Zhu et al., 2016). According to Lei et al. (2015), the ionospheric
responses to geomagnetic storms may also differ on the bottom-side and topside ionosphere. The complex effects
of the storm on the topside ionospheric total electron contents were also observed and reported during the strong
storm on 17 March 2015 (Astafyeva et al., 2015; Zhong, Lei, et al., 2016; Zhong, Wang, et al., 2016). The poor
tomographic performances during the storm may therefore be explained by the possibility that the GNSS-based
and LEO-based ionospheric observations were affected and thus with low accuracy. Also, the violent ionospheric
changes that occurred during the storm make it more challenging to accurately describe the ionosphere using the
global-scale tomographic results.

6. Conclusion
A growing number of LEO satellites offer new opportunities for global IED reconstruction. In this study,
ground-based and space-borne GNSS sounding data were used to create a global 3D ionospheric tomographic
model. The increase in the tomographic model's observed voxels is the main benefit of merging ground-based
and space-borne GNSS data. Two aspects of tomographic performances were evaluated. On the one hand, the
integrated TEC values from the reconstructed ionospheric electron densities (IEDs) were validated with the
space-borne GNSS STEC references and Jason-2 VTEC references. The EMART_GL solutions outperformed
the EMART_G solutions when compared to GNSS STEC references under quiet and disturbed conditions.
The RMS values of the CIT-derived STEC for EMART_G solutions can be reduced by 31.3% and 30.3%
on average under quiet and disturbed conditions, respectively. Since there weren't many GNSS observations

MEI ET AL. 12 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

Figure 11. Comparisons of EDPs between the NeQuick-2, EMART_G, and EMART_GL solutions and the COSMIC
profiles on 4 March 2015. (a–v) represent the comparison results at different universal time. The latitude and longitude of the
edmax point, that is, maximum electron density, for each COSMIC profile are also presented.

MEI ET AL. 13 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

Figure 12. Comparisons of EDPs between the NeQuick-2, EMART_G, and EMART_GL solutions and the COSMIC
profiles on 17 March 2015. (a–l) represent the comparison results at different universal time. The latitude and longitude of the
edmax point, that is, maximum electron density, for each COSMIC profile are also presented.

across the ocean areas, neither the EMART_G nor EMART_GL solutions exhibited significant improvements
when compared to the NeQuick-2 model. It found that areas with more dense LEO observations typically have
better model accuracy, whereas areas with fewer LEO-based observations will not be significantly improved.
Therefore, the ionospheric TEC measurements that are derived from the Jason-2 data as well as space-borne
GNSS-R (GNSS Reflectometry) observations (Ren et al., 2022) can also be incorporated to reduce the data
gaps over the ocean areas for global ionospheric tomography. On the other hand, the reconstructed 3D IEDs
were further validated with the COSMIC profiles. When compared to those of the EMART_G solutions and

Table 3
Mean Absolute Error and RMS Values of Maximum Electron Density Values Obtained by the NeQuick Model, the
EMART_G and EMART_GL Solutions on 4 and 17 March 2015
Mean absolute error (×10 11el/m 3) Root mean square error (×10 11el/m 3)

Time NeQuick-2 EMART_G EMART_GL NeQuick-2 EMART_G EMART_GL


03.04 2.821 1.896 1.314 3.421 2.473 1.761
03.17 3.488 2.618 1.699 4.536 3.466 2.440

MEI ET AL. 14 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

Figure 13. Three-dimensional distributions of rays for (a) GNSS, (b) LEO, and (c) GNSS + LEO; (d) top side view of rays for GNSS + LEO at UT12 on 4 March 2015.

Figure 14. Comparison of the CIT-derived STEC values of the EMART_G and EMART_GL solutions with (a) the GNSS-
only STEC references and (b) the LEO-only STEC references, respectively, on 4 March 2015.

MEI ET AL. 15 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

Figure 15. Comparison of the CIT-derived STEC values of the EMART_G, and EMART_GL solutions with (a) the GNSS-
only STEC references and (b) the LEO-only STEC references, respectively, on 17 March 2015.

the NeQuick-2 model, the tomographic results of the EMART_GL solutions were, on the whole, more in line
with the COSMIC profiles. The results indicated that tomographic performances can be improved using a
combination of GNSS and multiple LEO satellites under quiet conditions, but there is still much work to be
done under geomagnetic conditions.

In conclusion, it is advantageous for global ionospheric tomography to combine ground-based and space-borne
GNSS data. However, the improvements in the GNSS + LEO tomographic performances were not noticeable
due to the limited number of LEO satellites available. With increased data coverage in the future from massive
LEO satellites orbiting at various altitudes, it will be possible to establish a high-precision and high-resolution
three-dimensional ionospheric model. Accurately estimated ionospheric electron densities can provide us with a
more reliable data source for studying the ionosphere's activities.

Figure 16. Dst variations between 1 March 2015 and 20 March 2015.

MEI ET AL. 16 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

Data Availability Statement


ESA (European Space Agency) provides the Swarm relevant data via https://swarm-diss.eo.esa.int/#swarm%-
2FLevel2daily%2FEntire_mission_data%2FPOD, CDAAC (COSMIC Data Analysis and Archive Center)
provides relevant data of the other LEO satellites via https://data.cosmic.ucar.edu/gnss-ro/, Crustal Dynamics
Data Information System (CDDIS) provides GNSS data via https://cddis.nasa.gov/archive/gnss/data/daily/, the
Madrigal Database provides Jason-2 VTEC products via http://cedar.openmadrigal.org/list, and the International
Centre for Theoretical Physics (ICTP) provides the NeQuick2 model code via https://t-ict4d.ictp.it/nequick2/
source-code.

Acknowledgments References
This research was funded by the National
Natural Science Foundation of China Astafyeva, E., Zakharenkova, I., & Förster, M. (2015). Ionospheric response to the 2015 St. Patrick’s Day storm: A global multi-instrumental
(no. 42174031, no. 42230104), and the overview. Journal of Geophysical Research: Space Physics, 120(10), 9023–9037. https://doi.org/10.1002/2015ja021629
Special Fund of Hubei Luojia Laboratory Austen, J. R., Franke, S. J., & Liu, C. H. (1988). Ionospheric imaging using computerized tomography. Radio Science, 23(03), 299–307. https://
(no. 2201000038), and the Knowledge doi.org/10.1029/rs023i003p00299
Innovation Program of Wuhan-Shuguang. Bender, M., Dick, G., Ge, M., Deng, Z., Wickert, J., Kahle, H. G., et al. (2011). Development of a GNSS water vapor tomography system using
The numerical calculations have been algebraic reconstruction techniques. Advances in Space Research, 47(10), 1704–1720. https://doi.org/10.1016/j.asr.2010.05.034
done on the supercomputing system in Bust, G. S., & Mitchell, C. N. (2008). History, current state, and future directions of ionospheric imaging. Reviews of Geophysics, 46(1), RG1003.
the Supercomputing Center of Wuhan https://doi.org/10.1029/2006rg000212
University. We would like to thank the Chen, B., Wu, L., Dai, W., Luo, X., & Xu, Y. (2019). A new parameterized approach for ionospheric tomography. GPS Solutions, 23(4), 1–15.
anonymous reviewers for their helpful https://doi.org/10.1007/s10291-019-0893-4
suggestions. Cheng, N., Song, S., Jiao, G., Jin, X., & Li, W. (2021). Global monitoring of geomagnetic storm-induced ionosphere anomalies using 3-D iono-
spheric modeling with multi-GNSS and COSMIC measurements. Radio Science, 56(2), 1–16. https://doi.org/10.1029/2020rs007074
Cherniak, I., & Zakharenkova, I. (2016). High-latitude ionospheric irregularities: Differences between ground-and space-based GPS measure-
ments during the 2015 St. Patrick’s Day storm. Earth Planets and Space, 68(1), 1–13. https://doi.org/10.1186/s40623-016-0506-1
Danilov, A. D. (2013). Ionospheric F-region response to geomagnetic disturbances. Advances in Space Research, 52(3), 343–366. https://doi.
org/10.1016/j.asr.2013.04.019
Foelsche, U., & Kirchengast, G. (2002). A simple “geometric” mapping function for the hydrostatic delay at radio frequencies and assessment of
its performance. Geophysical Research Letters, 29(10), 111–1. https://doi.org/10.1029/2001gl013744
Fremouw, E. J., Secan, J. A., & Howe, B. M. (1992). Application of stochastic inverse theory to ionospheric tomography. Radio Science, 27(05),
721–732. https://doi.org/10.1029/92rs00515
Gerzen, T., Minkwitz, D., Schmidt, M., & Erdogan, E. (2020). Analysis of different propagation models for the estimation of the topside iono-
sphere and plasmasphere with an ensemble Kalman filter. In Annales geophysicae (Vol. 38, pp. 1171–1189). Copernicus GmbH.
Hajj, G. A., Ibanez-Meier, R., Kursinski, E. R., & Romans, L. J. (1994). Imaging the ionosphere with the Global Positioning System. Interna-
tional Journal of Imaging Systems and Technology, 5(2), 174–187. https://doi.org/10.1002/ima.1850050214
Hernández-Pajares, M., Juan, J. M., Sanz, J., & Solé, J. G. (1998). Global observation of the ionospheric electronic response to solar events using
ground and LEO GPS data. Journal of Geophysical Research, 103(A9), 20789–20796. https://doi.org/10.1029/98ja01272
Hong, J., Kim, Y. H., Chung, J. K., Ssessanga, N., & Kwak, Y. S. (2017). Tomography reconstruction of ionospheric electron density with
empirical orthonormal functions using Korea GNSS network. Journal of Astronomy and Space Sciences, 34(1), 7–17. https://doi.org/10.5140/
jass.2017.34.1.7
Howe, B. M., Runciman, K., & Secan, J. A. (1998). Tomography of the ionosphere: Four-dimensional simulations. Radio Science, 33(1),
109–128. https://doi.org/10.1029/97rs02615
Jayawardena, T. S. P., Chartier, A. T., Spencer, P., & Mitchell, C. N. (2016). Imaging the topside ionosphere and plasmasphere with iono-
spheric tomography using COSMIC GPS TEC. Journal of Geophysical Research: Space Physics, 121(1), 817–831. https://doi.
org/10.1002/2015ja021561
Kunitsyn, V. E., Andreeva, E. S., & Razinkov, O. G. (1997). Possibilities of the near-space environment radio tomography. Radio Science, 32(5),
1953–1963. https://doi.org/10.1029/97rs00837
Kunitsyn, V. E., Andreeva, E. S., Tsai, L. C., Liu, C. H., Nesterov, I. A., & Padokhin, A. M. (2008). Ionospheric imaging using various radio
tomographic systems. In Proceedings of the XXIXth URSI general assembly (pp. 7–16).
Lei, J., Zhu, Q., Wang, W., Burns, A. G., Zhao, B., Luan, X., et al. (2015). Response of the topside and bottomside ionosphere at low and
middle latitudes to the October 2003 superstorms. Journal of Geophysical Research: Space Physics, 120(8), 6974–6986. https://doi.
org/10.1002/2015ja021310
Leitinger, R., Ladreiter, H. P., & Kirchengast, G. (1997). Ionosphere tomography with data from satellite reception of Global Navigation
Satellite System signals and ground reception of Navy Navigation Satellite System signals. Radio Science, 32(4), 1657–1669. https://doi.
org/10.1029/97rs01027
Li, H., Yuan, Y., Li, Z., Huo, X., & Yan, W. (2012). Ionospheric electron concentration imaging using combination of LEO satellite data
with ground-based GPS observations over China. IEEE Transactions on Geoscience and Remote Sensing, 50(5), 1728–1735. https://doi.
org/10.1109/tgrs.2011.2168964
Li, Z., Wang, N., Liu, A., Yuan, Y., Wang, L., Hernández-Pajares, M., et al. (2021). Status of CAS global ionospheric maps after the maximum
of solar cycle 24. Satellite Navigation, 2(1). https://doi.org/10.1186/s43020-021-00050-2
Li, Z., Yuan, Y., Wang, N., Hernandez-Pajares, M., & Huo, X. (2015). SHPTS: Towards a new method for generating precise global ionospheric
TEC map based on spherical harmonic and generalized trigonometric series functions. Journal of Geodesy, 89(4), 331–345. https://doi.
org/10.1007/s00190-014-0778-9
Liu, H., Mei, D., Xu, G., Yang, P., Ren, X., & Zhang, X. (2022). Evaluation and validation of various rapid GNSS global ionospheric maps over
one solar cycle. Advances in Space Research, 70(8), 2494–2505. https://doi.org/10.1016/j.asr.2022.07.008
Mitchell, C. N., & Spencer, P. S. (2003). A three-dimensional time-dependent algorithm for ionospheric imaging using GPS. Annals of Geophysics.
Nava, B., Coisson, P., & Radicella, S. M. (2008). A new version of the NeQuick ionosphere electron density model. Journal of Atmospheric and
Solar-Terrestrial Physics, 70(15), 1856–1862. https://doi.org/10.1016/j.jastp.2008.01.015

MEI ET AL. 17 of 18
15427390, 2023, 4, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022SW003368 by Nat Prov Indonesia, Wiley Online Library on [08/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Space Weather 10.1029/2022SW003368

Nesterov, I. A., & Kunitsyn, V. E. (2011). GNSS radio tomography of the ionosphere: The problem with essentially incomplete data. Advances in
Space Research, 47(10), 1789–1803. https://doi.org/10.1016/j.asr.2010.11.034
Norberg, J., Vierinen, J., Roininen, L., Orispää, M., Kauristie, K., Rideout, W. C., et al. (2018). Gaussian Markov random field priors in iono-
spheric 3-D multi-instrument tomography. IEEE Transactions on Geoscience and Remote Sensing, 56(12), 7009–7021. https://doi.org/10.1109/
tgrs.2018.2847026
Peng, Y., Scales, W. A., Hartinger, M. D., Xu, Z., & Coyle, S. (2021). Characterization of multi-scale ionospheric irregularities using ground-based
and space-based GNSS observations. Satellite Navigation, 2(1). https://doi.org/10.1186/s43020-021-00047-x
Prol, F. S., Kodikara, T., Hoque, M. M., & Borries, C. (2021). Global-scale ionospheric tomography during the March 17, 2015 geomagnetic
storm. Space Weather, 19(12), e2021SW002889. https://doi.org/10.1029/2021sw002889
Prol, F. D. S., & Camargo, P. D. O. (2015). Review of tomographic reconstruction methods of the ionosphere using GNSS. Brazilian Journal of
Geophysics, 33(3), 445–459. https://doi.org/10.22564/rbgf.v33i3.947
Prol, F. D. S., & Camargo, P. D. O. (2016). Ionospheric tomography using GNSS: Multiplicative algebraic reconstruction technique applied to the
area of Brazil. GPS Solutions, 20(4), 807–814. https://doi.org/10.1007/s10291-015-0490-0
Prol, F. S., Hoque, M. M., & Ferreira, A. A. (2021). Plasmasphere and topside ionosphere reconstruction using METOP satellite data during
geomagnetic storms. Journal of Space Weather and Space Climate, 11, 5. https://doi.org/10.1051/swsc/2020076
Pryse, S. E. (2003). Radio tomography: A new experimental technique. Surveys in Geophysics, 24(1), 1–38. https://doi.org/10.1023/a:1022272607747
Pryse, S. E., & Kersley, L. (1992). A preliminary experimental test of ionospheric tomography. Journal of Atmospheric and Terrestrial Physics,
54(7–8), 1007–1012. https://doi.org/10.1016/0021-9169(92)90067-u
Ren, X., Chen, J., Zhang, X., Schmidt, M., Li, X., & Zhang, J. (2020). Mapping topside ionospheric vertical electron content from multiple LEO
satellites at different orbital altitudes. Journal of Geodesy, 94(9), 1–17. https://doi.org/10.1007/s00190-020-01415-2
Ren, X., Liu, H., Zhang, J., Mei, D., & Zhang, X. (2022). An improved method for ionospheric TEC estimation using the spaceborne GNSS-R
observations. IEEE Transactions on Geoscience and Remote Sensing, 60, 1–12. https://doi.org/10.1109/tgrs.2022.3192983
Ren, X., Mei, D., Zhang, X., Freeshah, M., & Xiong, S. (2021). Electron density reconstruction by ionospheric tomography from the combi-
nation of GNSS and upcoming LEO constellations. Journal of Geophysical Research: Space Physics, 126(10), e2020JA029074. https://doi.
org/10.1029/2020ja029074
Ren, X., Zhang, X., Schmidt, M., Zhao, Z., Chen, J., Zhang, J., & Li, X. (2020). Performance of GNSS global ionospheric modeling augmented
by LEO constellation. Earth and Space Science, 7(1), e2019EA000898. https://doi.org/10.1029/2019ea000898
Rius, A., Ruffini, G., & Cucurull, L. (1997). Improving the vertical resolution of ionospheric tomography with GPS occultations. Geophysical
Research Letters, 24(18), 2291–2294. https://doi.org/10.1029/97gl52283
Schmidt, M., Bilitza, D., Shum, C. K., & Zeilhofer, C. (2008). Regional 4-D modeling of the ionospheric electron density. Advances in Space
Research, 42(4), 782–790. https://doi.org/10.1016/j.asr.2007.02.050
She, C., Wan, W., Yue, X., Xiong, B., Yu, Y., Ding, F., & Zhao, B. (2017). Global ionospheric electron density estimation based on multisource
TEC data assimilation. GPS Solutions, 21(3), 1125–1137. https://doi.org/10.1007/s10291-016-0580-7
Smirnov, A., Shprits, Y., Zhelavskaya, I., Lühr, H., Xiong, C., Goss, A., et al. (2021). Intercalibration of the plasma density measurements in
Earth’s topside ionosphere. Journal of Geophysical Research: Space Physics, 126(10), e2021JA029334. https://doi.org/10.1029/2021ja029334
Ssessanga, N., Kim, Y. H., & Kim, E. (2015). Vertical structure of medium-scale traveling ionospheric disturbances. Geophysical Research
Letters, 42(21), 9156–9165. https://doi.org/10.1002/2015gl066093
Wang, N., Yuan, Y., Li, Z., Li, Y., Huo, X., & Li, M. (2017). An examination of the Galileo NeQuick model: Comparison with GPS and JASON
TEC. GPS Solutions, 21(2), 605–615. https://doi.org/10.1007/s10291-016-0553-x
Wen, D., Liu, S., & Tang, P. (2010). Tomographic reconstruction of ionospheric electron density based on constrained algebraic reconstruction
technique. GPS Solutions, 14(4), 375–380. https://doi.org/10.1007/s10291-010-0161-0
Wen, D., Yuan, Y., Ou, J., & Zhang, K. (2010). Ionospheric response to the geomagnetic storm on August 21, 2003 over China using GNSS-based
tomographic technique. IEEE Transactions on Geoscience and Remote Sensing, 48(8), 3212–3217. https://doi.org/10.1109/tgrs.2010.2044579
Wen, D., Yuan, Y., Ou, J., Zhang, K., & Liu, K. (2008). A hybrid reconstruction algorithm for 3-D ionospheric tomography. IEEE Transactions
on Geoscience and Remote Sensing, 46(6), 1733–1739. https://doi.org/10.1109/tgrs.2008.916466
Yao, Y., Chen, P., Zhang, S., & Chen, J. (2013). A new ionospheric tomography model combining pixel-based and function-based models.
Advances in Space Research, 52(4), 614–621. https://doi.org/10.1016/j.asr.2013.05.003
Yao, Y., Tang, J., Chen, P., Zhang, S., & Chen, J. (2013). An improved iterative algorithm for 3-D ionospheric tomography reconstruction. IEEE
Transactions on Geoscience and Remote Sensing, 52(8), 4696–4706.
Yao, Y., Zhai, C., Kong, J., Zhao, C., Luo, Y., & Liu, L. (2020). An improved constrained simultaneous iterative reconstruction technique for
ionospheric tomography. GPS Solutions, 24(3), 1–19. https://doi.org/10.1007/s10291-020-00981-4
Yizengaw, E., Moldwin, M. B., Komjathy, A., & Mannucci, A. J. (2006). Unusual topside ionospheric density response to the November 2003
superstorm. Journal of Geophysical Research, 111(A2), A02308. https://doi.org/10.1029/2005ja011433
Zakharenkova, I., Astafyeva, E., & Cherniak, I. (2016). GPS and in situ Swarm observations of the equatorial plasma density irregularities in the
topside ionosphere. Earth Planets and Space, 68(1), 1–11. https://doi.org/10.1186/s40623-016-0490-5
Zheng, D., Hu, W., & Nie, W. (2015). Multiscale ionospheric tomography. GPS Solutions, 19(4), 579–588. https://doi.org/10.1007/
s10291-014-0418-0
Zheng, D., Yao, Y., Nie, W., Chu, N., Lin, D., & Ao, M. (2021). A new three-dimensional computerized ionospheric tomography model based on
a neural network. GPS Solutions, 25(1), 1–17. https://doi.org/10.1007/s10291-020-01047-1
Zhong, J., Lei, J., Dou, X., & Yue, X. (2016). Assessment of vertical TEC mapping functions for space-based GNSS observations. GPS Solutions,
20(3), 353–362. https://doi.org/10.1007/s10291-015-0444-6
Zhong, J., Wang, W., Yue, X., Burns, A. G., Dou, X., & Lei, J. (2016). Longduration depletion in the topside ionospheric total electron content
during the recovery phase of the March 2015 strong storm. Journal of Geophysical Research: Space Physics, 121(5), 4733–4747. https://doi.
org/10.1002/2016ja022469
Zhu, Q., Lei, J., Luan, X., & Dou, X. (2016). Contribution of the topside and bottomside ionosphere to the total electron content during two strong
geomagnetic storms. Journal of Geophysical Research: Space Physics, 121(3), 2475–2488. https://doi.org/10.1002/2015ja022111

MEI ET AL. 18 of 18

You might also like