Opcion 3 Ingles

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Available online at www.sciencedirect.

com

ScienceDirect
Advances in Space Research 73 (2024) 585–596
www.elsevier.com/locate/asr

A novel method for inverting coseismic 3D surface deformation


using InSAR considering the weight influence of the spatial
distribution of GNSS points
Mingkai Chen a, Guangyu Xu a,b,⇑, Tengxu Zhang c, Xiaowei Xie a, Zhiping Chen a
a
School of Surveying and Geoinformation Engineering, East China University of Technology, Nanchang 330013, China
b
Key Laboratory of Mine Environmental Monitoring and Improving around Poyang Lake of Ministry of Natural Resources, East China University
of Technology, Nanchang 330013, China
c
College of Resources and Environmental Science and Engineering, Hubei University of Science and Technology, Xianning 437100, China

Received 17 April 2023; received in revised form 5 October 2023; accepted 9 November 2023
Available online 14 November 2023

Abstract

The combination of InSAR and GNSS data can provide more accurate measurements of coseismic 3D surface deformations, which
are crucial in determining earthquake source parameters. Currently, the methods for inverting coseismic 3D surface deformations by
integrating GNSS and InSAR data using the strain model (SM) (e.g., Simultaneous and integrated strain tensor estimation from geodetic
measurements to obtain 3D displacement maps (SISTEM), SM-VCE) simply select GNSS points based on a certain number or distance,
without considering the weight influence of the spatial distribution of G NSS points on the InSAR and GNSS data fusion results. To
address this issue, this paper proposes a new method that uses Voronoi diagrams analyze the contribution of GNSS points to the defor-
mation at target points. In this paper, GNSS points with high contribution can be used to construct observation value equations and
optimize the internal weight of GNSS data. The proposed approach characterizes the contribution of GNSS data to target points by
analyzing the changes in the Delaunay triangulation area formed before and after the target points are added to the GNSS data. Sim-
ulation experiments reveal that the proposed method outperforms conventional methods in accuracy and completeness. The proposed
method is then applied to map the coseismic 3D surface deformations of the 2020 Mw6.5 Monte Cristo Range earthquake and find that
the maximum deformation in the east–west, north–south, and vertical directions is about 13 cm, 11 cm, and 24 cm, respectively.
Ó 2023 COSPAR. Published by Elsevier B.V. All rights reserved.

Keywords: 3D surface deformation; GNSS; InSAR; Strain model

1. Introduction and GNSS are two commonly techniques in the field of


deformation monitoring. InSAR technology has advan-
Coseismic surface deformation is the most intuitive tages such as all-weather capability and high spatial resolu-
reflection of ground deformation caused by earthquakes, tion. GNSS is characterized by high accuracy and high
and it is of great significance for studying the tectonic temporal resolution, enabling the acquisition of precise
and geometric structures of earthquake zones, as well as three-dimensional deformation measurements. However,
evaluating and preventing earthquake disasters. InSAR both techniques have their limitations in practical applica-
tions. For example, DInSAR can only obtain line-of-sight
(LOS) deformation and is susceptible to atmospheric
⇑ Corresponding author.
effects. Additionally, because SAR satellites flies along a
E-mail addresses: gyxu@ecut.edu.cn (G. Xu), txzhangsgg@whu.edu.cn
(T. Zhang), xwxie@ecut.edu.cn (X. Xie), zhpchen@ecut.edu.cn (Z. Chen).
polar orbit, InSAR measurements are less sensitive to

https://doi.org/10.1016/j.asr.2023.11.013
0273-1177/Ó 2023 COSPAR. Published by Elsevier B.V. All rights reserved.
M. Chen et al. Advances in Space Research 73 (2024) 585–596

north–south deformations. GNSS, on the other hand, Therefore, this paper aims to explore the contribution of
incurs high infrastructure costs, resulting in relatively GNSS points to the 3D deformation at the target point.
sparse point distributions and lower spatial resolution, Voronoi diagrams are introduced to analyze the spatial dis-
making it incapable of capturing continuous surface defor- tribution of GNSS points, and the GNSS data that have
mation information. Therefore, a key research focus is on the optimal contribution to the target point are obtained
how to leverage the strengths of both techniques and inte- to construct the SM observation equation. The contribu-
grate their solutions to achieve higher three-dimensional tion of different GNSS sites is evaluated to obtain the inter-
deformation precision. nal weight of the GNSS data. The VCE method is then
Incorporating GNSS and InSAR has yielded numerous used to invert the surface 3D deformation.
achievements in obtaining three-dimensional deformation
data. A combination method uses the Gibbs energy equa- 2. Methods
tion as the objective function and obtains the optimal value
of its three-dimensional deformation variable when the The method proposed in this paper aims to utilize Vor-
energy of the function reaches a minimum onoi diagrams to analyze the spatial distribution of GNSS
(Gudmundsson et al., 2002; Samsonov and Tiampo, stations, providing an alternative scheme for GNSS data
2006; Samsonov et al., 2008; Hu et al., 2021). Then Luo selection and weighting in the fusion of InSAR and GNSS
et al., (2008) proposed direct decomposition method that data. This approach is then applied to Strain model-based
uses the high-precision horizontal deformation displace- methods such as SISTEM and SM-VCE. Therefore, we
ment provided by GNSS observations as a constraint to have included introductions to the SISTEM and SM-
decompose the line-of-sight (LOS) deformation displace- VCE methods in the supplementary material Test S1.
ment of DInSAR into vertical deformation displacement.
In the other method, interpolating GNSS data into the 2.1. InSAR LOS geometric decomposition
InSAR grid and then using weighted least squares (WLS)
to solve surface deformations on a per-pixel basis (Wang InSAR technology is widely used in earthquake defor-
et al., 2016; Zhou et al., 2021). However, in surface defor- mation monitoring studies, but it can only obtain LOS
mation studies, the deformation of adjacent points is inter- deformation observations. To overcome this limitation of
related, yet the methods mentioned above primarily focus InSAR technology, it is necessary to decompose the LOS
on computing deformation for individual points. deformation and recover its east–west, north–south, and
Guglielmino et al., (2011) proposed the SISTEM method vertical components. Fig. 1 illustrates the relationship
(Simultaneous and integrated strain tensor estimation from between InSAR LOS deformation and its east–west,
geodetic measurements to obtain 3D displacement maps), north–south, and vertical components, which can be repre-
which was the first to use the strain model (SM) to solve sented by satellite geometry (Fialko et al., 2001):
3D deformations by combining GNSS and InSAR data.
Subsequently, Liu et al. (2021) considered the weighting LLOS ¼  sin h cos ad e þ sin h sin ad n þ cos hd v ð1Þ
issue between GNSS and InSAR data and proposed the where LLOS represents the LOS displacement,
SM-VCE method, which uses variance component estima- T
d ¼ ½d e ; d n ; d v  represents the unknown components in
tion (VCE) to accurately weight the observation equation
constructed by the strain model. This improves the accu-
racy of 3D deformation inversion. These methods based
on SM (e.g., SISTEM, SM-VCE method) use the InSAR
and GNSS observation values of surrounding points to
invert the 3D deformations at the target point, avoiding
the interpolation of GNSS points. However, these methods
do not discuss the contribution of surrounding points to
the 3D deformation at the target point in depth. The SIS-
TEM and SM-VCE methods simply select a certain dis-
tance or number of GNSS points around the target point
to construct the SM observation model. When GNSS
points are distributed uniformly, these methods can achieve
good results in inverting the 3D deformation field. How-
ever, when the distribution of GNSS points is uneven,
and considering that the InSAR observation values are
insensitive to the north–south deformation, the study of
the correction contribution of the spatial distribution of
GNSS points to the inversion of the 3D deformation field
at the target point becomes more important. Fig. 1. Satellite Imaging Geometry (Zhou et al., 2021).

586
M. Chen et al. Advances in Space Research 73 (2024) 585–596

the east–west, north–south, and vertical directions, Bgeo ¼ lated. Voronoi diagrams are a mathematical concept pro-
½be ; bn ; bv  represents the coefficient vector for the projection posed by Russian mathematician Georgy Voronoi and
of east–west, north–south, and vertical displacements onto are a geometric shape used to divide a set of points on a
the LOS direction, with be ¼ sin h cos a, bn ¼ sin h sin a, plane into non-overlapping regions (Voronoi, 1908). Voro-
bv ¼ cos h. h and a represent the incidence angle and azi- noi diagrams can be obtained by constructing a TIN trian-
muth angle of the satellite, respectively. gulation with discrete points and taking the perpendicular
bisectors of the TIN triangulation edges. Fig. 2(a)(c)
2.2. Strain model transformation illustrates the process of generating Voronoi diagrams for
GNSS stations, where the orange diamond represents the
Assuming that there are SAR images in ascending and target point to be estimated.
descending orbits and GNSS stations in the study area, In a study area with n GNSS points represented by set S,
the DInSAR and GNSS techniques can be used to obtain for any point p in a region of Voronoi diagram, and any
the InSAR LOS observation value LLOS in the ascending GNSS point s and any other GNSS point s0 in other
and descending orbits and GNSS 3D deformation observa- regions, the following properties hold:
T
tion value LGNSS ¼ ½d GNSS;e ; d GNSS;n ; d GNSS;v  . VorðsÞ ¼ fp : distanceðs; pÞ6 dis tan ceðs0 ; pÞ; 8s0 2 Sg ð5Þ
Let there be K adjacent points near the target point P 0 in
the study area (the selection of adjacent points is described Equation (5) can be interpreted as stating that for any
point p within a diagram region, the distance between point
in section 2.4 of this paper). Let P k ðk ¼ 1; 2; :::; K Þ be one of
 T p and GNSS point s is no greater than the distance between
the points, and let d 0 ¼ d 0e ; d 0n ; d 0v and point p and any other GNSS point in other regions. There-
 0 0 0 T
x ¼ xe ; xn ; xv
0
be the 3D deformation and 3D coordi- fore, this property can be used to describe the influence
 T range of GNSS points in the set S on the InSAR LOS
nates of P , respectively. Let d k ¼ d ke ; d kn ; d kv
0
and
  T observations.
xk ¼ xke ; xkn ; xkv be the 3D deformation and 3D coordi- Subsequently, GNSS points that contribute to the calcu-
nates of P k , respectively. It is known from the Strain Model lation of the 3D deformation of the target point are deter-
that the relationship between the 3D deformation and 3D mined and selected. The target point is added to the set of
coordinates between P k and P 0 can be expressed using the GNSS points and a Voronoi diagram is constructed again.
following equation (Guglielmino et al., 2011): The Voronoi diagrams that undergo changes from the orig-
d k ¼ BkSM l ð2Þ inal Voronoi diagrams are detected. The GNSS points in
the area where the original Voronoi diagram’s area
 T
l ¼ d 0e ; d 0n ; d 0v ; eee ; een ; eev ; enn ; env ; euu ; xen ; xev ; xnv decreases can be considered as contributing to the correc-
2 3 tion of the InSAR line-of-sight (LOS) observations, and
1 0 0 Dxke Dxkn Dxkv 0 0 0 Dxkn Dxkv 0
6
B kSM ¼ 4 0 1 0 0 Dxke 0 Dxkn Dxkv 0 Dxke 0
7
Dxkv 5
the corresponding decrease in the area proportion can be
0 0 1 0 0 Dxke 0 Dxkn Dxkv 0 Dxke Dxkn
used as a weighting factor. As shown in Fig. 2(d), the addi-
tion of the target point to the construction of the Voronoi
The matrix B kSM represents the coefficient matrix of the diagrams causes changes in the areas of GNSS points
Strain Model, where Dxki ¼ xki  x0i ði ¼ e; n; uÞ; l is the s1 ; s2 ;    ; s6 . The corresponding area changes are denoted
unknown vector of the Strain Model. where e and x repre- as a1 ; a2 ;    ; a6 respectively. Thus, the contribution weight
of any GNSS point to the target point can be expressed as:
sent the strain and rotation parameters of P 0 point, respec-
tively. According to (1) and (2), LkLOS can be expressed as: ai
wi ¼ k Pn ði ¼ 1; 2;    ; nÞ ð6Þ
i¼1 ai
LkLOS ¼ Bkgeo B kSM l ð3Þ
h iT where k is the precision scaling factor. Since GNSS hori-
Similarly,LkGNSS ¼ d kGNSS;e ; d kGNSS;n ; d kGNSS;v can be zontal accuracy is higher than vertical accuracy, based on
Wang et al., (2016); Wang et al., (2021), we set k to 1 for
expressed as: the GNSS horizontal direction and 0.5 for the vertical
LkGNSS ¼ B kSM l ð4Þ direction.
Considering the spatial distribution and sparsity of
T
where LGNSS ¼ ½d GNSS;e ; d GNSS;n ; d GNSS;v  is the deformation GNSS, 1 to 10 GNSS points will be selected. However,
observation. in the inversion of surface 3D deformation using InSAR
and GNSS data, a small number of GNSS stations can
2.3. Spatial analysis of GNSS using Voronoi diagrams cause instability in the solution of the 3D deformation
observation. Through experiments, SISTEM
In this section, Voronoi diagrams were used for spatial (Guglielmino et al., 2011) selected 6 GNSS points, and
analysis of GNSS points to select those that make a signif- SM-VCE (Wang et al., 2021) used 5–7 GNSS points.
icant contribution to the 3D deformation inversion of the Therefore, in this paper, when the number of selected
target point, and their corresponding weights were calcu- GNSS points is less than 6, nearby GNSS points are
587
M. Chen et al. Advances in Space Research 73 (2024) 585–596

Fig. 2. Illustrates the Voronoi diagrams of the GNSS stations. The orange point represents the target point to be estimated, and the red points represents
the GNSS station around the target point.

selected for supplementation to ensure the stability of the of InSAR observations varies among different orbits. So,
inversion of the surface 3D deformation. The weight of we consider InSAR ascending LOS observations, InSAR
the supplemented points is represented as: descending LOS observations, GNSS horizontal observa-
d tions and GNSS vertical observations as four types of
w0m ¼ k min ðw1 ; w2 ;    ; wn Þðm ¼ 1;    ; M Þ ð7Þ observations Lj ðj ¼ 1; 2; 3Þ, each with observations around
Md min
the K j target point. Based on (3) and (4), we can establish
In (7), d and d min represent the Euclidean distances the relationship between deformation observations Lj of
between the supplementary point and the point smin with point P k and 3D point deformations of point P 0 , which
the minimum weight to the target point P 0 , and M repre- can be expressed as:
sents the number of supplementary points. This approach 2 3 2 3
not only restricts the weight of the supplementary points, L1 B1
6 7 6 7
but also avoids the instability of the inversion results 4 L2 5 ¼ 4 B 2 5  l ð8Þ
caused by the small number of GNSS points selected by L3 B3
the Voronoi diagrams.
where:
2.4. VCE for GNSS and InSAR fusion h iT
K
Lj ¼ L1j ; L2j ;    ; Lj j ðj ¼ 1; 2; 3Þ
In section 2.3, GNSS (basic and supplementary) points h i
K
around the target point were selected. Due to the VCE Bj ¼ B 1j ; B 2j ;    ; Bj j
method’s requirement for roughly equal numbers of obser- ( k
vations in different categories, Therefore, we adopted three B j;geo B kj;SM ðj ¼ 1; 2Þ
Bj ¼
k
times the number of the GNSS observations centered at the B kj;SM ðj ¼ 3Þ
target point to select InSAR observations.
There is an inconsistency in the accuracy of GPS hori- In section 2.3, the selection and weight of GNSS points
zontal and vertical observations (Zhou et al., 2021). There- were analyzed. Regarding the weights for InSAR, we ini-
fore, we will categorize GNSS observations into two tially set the weight matrix to an identity matrix, and then
groups. Additionally, it’s worth noting that the accuracy utilize a variance component estimation algorithm to deter-
588
M. Chen et al. Advances in Space Research 73 (2024) 585–596

mine the posterior variances for different InSAR observa- field of the forward calculation is projected according to
tions. According to (6) and (7), the weights of DInSAR (1), and Gaussian errors with standard deviation of
and GNSS can be defined as: 6 mm and 7 mm are added to obtain its corresponding
 ascending/descending LOS deformation observations.
I K j j ¼ 1; 2
Pj ¼   ð9Þ GNSS data randomly selects 120-pixel points from the
diag w1 ; w2 ;    ; wK 3 j ¼ 3 3D deformation field of forward calculation and adds
Using Weighted Least Square (WLS) to solve the obser- Gaussian errors with standard deviation of 1 mm in east–
vation equation constructed by (8), its initial solution and west direction, 1 mm in north–south direction and 2 mm
corresponding residual can be obtained: in vertical direction. The selection of Gaussian noise stan-
8 !1 ! dard deviation for GNSS and InSAR was guided by the
>
< bl ¼ P B T P B
> P
3 3 parameters outlined in Wang et al., (2021). Fig. 3 (d) and
T
j j j B j Pj Lj
j¼1 j¼1 ð10Þ (e) show the ascending/descending InSAR LOS observa-
>
>
: tions used in this simulation experiment.
V j ¼ Bjbl  Lj The 3D surface deformation of simulated data is
According to the variance component estimation algo- inverted using SISTEM, SM-VCE, and the method pro-
 2 2 2 posed in this paper, as shown in Fig. 4 and Fig. 5 Specifi-
^¼ r
rithm, the unit weight variance r ^1 ; r
^2 ; r
^3 of ascend-
cally, Fig. 4(a)(c) shows the 3D deformation results
ing/descending InSAR observations and GNSS
obtained by the SISTEM method, Fig. 4(d)(f) shows
observations can be calculated:
the results obtained by the SM-VCE method, and Fig. 4
^ ¼ C1 W
r ð11Þ (g)(i) shows the results obtained by the method proposed
where: in this paper. Fig. 5 shows the corresponding residuals of
the three methods. From the inversion results and residual
X
3 3 
X  maps, it can be seen that the three methods have consistent
N¼ Nj ¼ B Tj Pj B j
3D deformation results overall. However, the proposed
j j
  method in this paper improves the inversion results in the
W ¼ V T1 P1 V 1 ; V T2 P2 V 2 ; V T3 P3 V 3 east–west, north–south, and vertical directions. Compared
2    2     3
K 1  2tr N 1 N 1 þ tr N 1 N 1 tr N 1 N 1 N 1 N 2 tr N 1 N 1 N 1 N 3 with the SISTEM method and SM-VCE method, the pro-
6      2   7
C¼6
4 tr N 1 N 2 N 1 N 1 K 2  2tr N 1 N 2 þ tr N 1 N 2 tr N 1 N 2 N 1 N 3 7
5 posed method can obtain better 3D deformation inversion
   1   1   1 2
tr N 1 N 3 N 1 N 1 1
K 2  2tr N N 3 þ tr N N 3
tr N N 3 N N 2
results in the earthquake epicentral region with uneven
GNSS distribution and sparse GNSS points. Furthermore,
Determine whether the condition r b2  rb 2min < c2 is
2  2 2 2  2 max  2 2 2 the strain model-based approach not only provides three-
met. Where b r max ¼ max r ^1 ; r
^2 ; r
^3 , b
r min ¼ min r ^1 ; r
^ 2 ;r
^3 Þ, dimensional deformation but also offers stress–strain ten-
and the threshold for setting InSAR and GNSS observation sors and rotation tensors. For simplicity, we have pre-
values is c ¼ 1mm.If it is met, the estimated value of the sented the results of three strain invariants (i.e.,
parameter to be solved at this time is regarded as the best dilatation, differential rotation, and maximum shear) in
estimate. If it is not met, update the weight matrix and iter- Fig. S2. Detailed derivations of these three strain invariants
ate (10), (11) until the condition is met. The updated weight can be found in Supplementary Material Test S2.
matrix can be expressed as: Then, we further compared the three methods in terms
^21
r of their 3D deformation calculation performance. Fig. 6
^j ¼
P ; j ¼ 1; 2; 3 ð12Þ resents a comparison between the simulated GNSS station
^2j P1
r j observations and the calculated 3D deformation results by
the three methods. It can be seen that the three methods
3. Simulation experiment generally agree well with the simulated GNSS station
observations in the east–west, north–south, and vertical
The equation for forward calculating the 3D surface directions. However, SISTEM and SM-VCE methods exhi-
deformation field of the ground based on the Okada elastic bit significant discrepancies with the GNSS observations
half-space matrix dislocation model (Okada, 1985) is: and the calculated 3D deformation results in the north–
south direction compared to the proposed method. Table 1
d obs ¼ GðmÞ þ e ð13Þ
provides the root mean square error (RMSE) of the inver-
where d obs represents the 3D deformation components sion of the 3D deformation compared with the true values,
ðd e ; d n ; d v Þ in the east–west, north–south and vertical direc- which is consistent with the situation shown in Fig. 6.
tions, G is the Green’s function, m is the vector of slip val- When using the SM model to construct the observation
ues. Fig. 3(a)(c) shows the coseismic 3D surface equation, it is necessary to select nearby InSAR and GNSS
deformation field obtained by forward calculation using data. Once the selection logic of InSAR and GNSS data
(13) and a known fault slip model in Fig. S1. points is determined, the spatial distribution and sparsity
Then according to the synthetic aperture radar parame- of GNSS points will affect the accuracy of the calculated
ters of the region, the coseismic 3D surface deformation 3D deformation results. In this study, the relationship
589
M. Chen et al. Advances in Space Research 73 (2024) 585–596

Fig. 3. Simulation experiment data. The red triangles represent the distribution of GNSS stations, and the black rectangles denote the surface projection
of the fault.

Fig. 4. Simulation experiment’s 3d deformation results. (a)(c) sistem, (d)(f) sm-vce, (g)(h) proposed method. the black boxes indicate the surface
projection of the fault location, while the red boxes highlight areas with calculated deformation anomalies.

590
M. Chen et al. Advances in Space Research 73 (2024) 585–596

Fig. 5. Residuals of Inverted 3D Surface Deformation. (a)(c) SISTEM, (d)(f) SM-VCE, (g)(h) Proposed method. The black boxes indicate the
surface projection of the fault location.

Fig. 6. Shows the comparison of inversion results and GNSS observations in the simulated experiment. The horizontal axis represents the displacement of
GNSS stations, and the vertical axis represents the average displacement of inversion results within the 500m500m range of the corresponding GNSS
stations.

between the RMSE of the deformation in the three direc- choose six nearby GNSS stations, while the proposed
tions (east–west, north–south, and vertical) and the num- method in this paper employs the Voronoi diagrams opti-
ber of GNSS stations within the study area is examined mization scheme.
(as shown in Fig. 7). This analysis is conducted under the It can be observed from Fig. 7 that as the number of
conditions where the SM-VCE and SISTEM methods GNSS points in the study area increases, the RMSE of
591
M. Chen et al. Advances in Space Research 73 (2024) 585–596

Table 1 selection, phase filtering, unwrapping, geocoding, and the


Comparison of root-mean-square errors of 3D deformation for three other post-processing steps. The GPS coseismic deforma-
methods in simulated experiment.
tion were obtained from Nevada Geodetic Laboratory.
methods West-East (mm) North-South (mm) Vertical (mm) The GPS data was processed with Gipsy software
SISTEM 1.4 6.6 2.0 (https://gipsyx.jpl.nasa.gov/), which includes non-fiducial
SM-VCE 1.1 6.4 1.5 solution, precise point positioning (PPP) as well as the spe-
Our method 1.0 4.4 1.4
cial settings for some geophysical models and site-specific
parameters.
the calculated results of the three methods gradually Fig. 9 shows the processed InSAR LOS observations
decreases. The RMSE of the north–south direction of SIS- and the distribution of GNSS stations. The black line rep-
TEM method and SM-VCE method are similar because resents the fault line (Hammond et al., 2020). The black tri-
GNSS station data contributes more to the north–south angles represent 31 GNSS candidate stations used in the
direction, and SISTEM and SM-VCE methods in this calculations for all three methods, and the black circles rep-
study adopt the same selection logic. The proposed method resent three GNSS validation stations that are not involved
can obtain good 3D deformation results even when GNSS in the three-dimensional deformation recovery which are
points are sparse. Compared to the SISTEM method, the MINA, MOHO and PILO stations respectively. Text labels
proposed method exhibits better performance in the east– have been applied to the GNSS stations within the SAR
west, north–south, and vertical directions. The difference image coverage area. The five black triangle-labeled sta-
between the proposed method and SM-VCE method is tions are utilized for goodness-of-fit analysis, while the
not significant in the east–west direction, while it is supe- three GNSS validation points are employed to verify the
rior in the north–south and vertical directions. accuracy of the three-dimensional deformation estimation.
Table S1 shows the three-dimensional deformation of 8
GNSS stations within study region. We have provided
4. Real experiment detailed data for the remaining GNSS stations in the sup-
plementary material Appendix 1.
On May 15, 2020, at 11:03 (UTC), Monte Cristo Range Fig. 10 shows the 3D deformation and residual results
Mw 6.5 earthquake occurred 56 km west of Tonopah, obtained by the SISTEM, SM-VCE, and proposed meth-
Nevada, USA (Fig. 8). The epicenter was located at ods. The results obtained by the three methods are consis-
38.169°N and 117.850°W, with a focal depth of 2.7 km. tent, indicating that the proposed method is reliable.
The earthquake was caused by the strike-slip fault of the Among them, the maximum deformation of east–west,
shallow crust of the North American plate (https://earth- north–south and vertical of the earthquake is about
quake.usgs.gov/earthquakes/eventpage/nn00725272/ 13 cm, 11 cm and 24 cm, respectively. There is little differ-
executive). ence in the 3D deformation results obtained by the three
To reproduce the 3D deformation caused by this earth- methods in the east–west and vertical directions. The
quake, the research area of this study utilized publicly north–south deformation results obtained by the SISTEM
available InSAR and GNSS data. The InSAR coseismic and SM-VCE methods are closer to each other, but the
interferograms were obtained from automated SAR- results of both methods are not continuous near the fault
VIEWS program (Meyer et al., 2016). SARVIEWS InSAR in the north–south direction due to the point selection
interferograms are processed using GAMMA Remote logic. In contrast, the north–south deformation obtained
Sensing software tools (Wegnüller et al., 2016) through by the proposed method is continuous and smooth. From
the ASF HyP3 engine, which includes interferometric pair Fig. 10(k) and (n), it can be seen that there are differences

Fig. 7. Illustrates the relationship between the number of GNSS stations in the study area and the precision of the 3D deformation results.
592
M. Chen et al. Advances in Space Research 73 (2024) 585–596

Fig. 8. GNSS coseismic deformation of the 2020 Monte Cristo Range earthquake from Nevada Geodetic Laboratory (http://geodesy.unr.edu/news_
items/20200619/nn00725272_24hr_19-Jun-2020.txt) (a) The red arrows represent GNSS horizontal coseismic offsets. (b) The red arrows represent GNSS
vertical coseismic offsets. The blue triangle denotes GNSS stations. The red beach ball show focal mechanism of 2020 Monte Cristo Range earthquake
from GCMT. The red rectangle in inter map is the area of the figure. The blue and orange rectangles outline the footprint of InSAR ascending and
descending track. The black lines denotes Quaternary faults the USGS Quaternary fault and fold database (https://www.usgs.gov/natural-hazards/
earthquake-hazards/faults, last accessed September 2023). The red lines represent surface rupture of the 2020 Monte Cristo Range earthquake (Koehler
et al., 2021).

Fig. 9. InSAR LOS observations, (a) ascending track; (b) descending track, and distribution of GNSS stations. The black line represents the fault line.

between the proposed method and the SISTEM and SM- In addition to the calculated three-dimensional defor-
VCE methods in the area of maximum north–south defor- mation, Fig S3 also presents three dimensionless strain field
mation near the epicenter. In the simulation experiment, components (dilatation strain, differential rotation, maxi-
the proposed method’s recovery effect was better than that mum shear strain). As shown in the figure, when the distri-
of the SISTEM and SM-VCE methods in the area of max- bution of GNSS stations is relatively sparse, the SM-VCE
imum north–south deformation near the epicenter. There- method (Fig. S3(d)(f)) and the method proposed in this
fore, it can be preliminarily concluded that the proposed paper (Fig. S3(g)(i)) provide more detailed strain compo-
method provides more reliable north–south deformation nents compared to the SISTEM method (Fig. S3 (a)(c)).
results for the Monte Cristo Range earthquake in the area These three strain field components clearly demonstrate the
of maximum deformation. orientation of the fault, which is consistent with the fault

593
M. Chen et al. Advances in Space Research 73 (2024) 585–596

Fig. 10. 3D deformation and residual results obtained by the three methods. (a)(c) SISTEM, (d)(f) SM-VCE, (g)(h) Proposed method, (i)(l)
Comparison between the proposed method and SISTEM, (m)(o) Comparison between the proposed method and SM-VCE, Black line represents the
fault line.

line orientation mentioned in Hammond et al., (2020), posed method. However, in the calculated results, a
which is from south-west to north-east. noticeable jump in the three-dimensional deformation near
To analyze the goodness of fit near the integrated GNSS the fault was observed for the SISTEM and SM-VCE
points within the study area. Table S2 presents the discrep- methods (highlighted by the red box in the Fig. 4). We
ancy between integrated three-dimensional deformation focused our analysis on a point at the center of the box
results and the GNSS three-dimensional deformation with the most significant deformation change, located at
observations at 5 participating GNSS station locations. coordinates (86.815, 33.183). Fig S4 illustrates the selection
The fusion result of this method near faults is closer to logic for this point in the simulated true three-dimensional
GNSS stations (such as COLU, MONT). Table S3 repre- deformation field for all three methods, with black trian-
sents the Root Mean Square Error (RMSE) for these five gles representing GNSS stations and red dots representing
integrated GNSS stations. Analyzing the results from the target point at coordinates (86.815, 33.183). Fig. S4 (a)
Table S2 and Table S3, it can be observed that, compared  (c) depict the selection logic for the SISTEM and SM-
to the SISTEM method and SM-VCE method, the values VCE methods, denoted by white circles around black
integrated in this study are closer to the GNSS station GNSS stations. Similarly, Fig. S4(d)  (f) represent the
results, indicating a better fit for the proposed method in selection logic for our proposed method.
this paper. Upon comparison, it was evident that the selection logic
Subsequently, to further verify the accuracy of this in SISTEM and SM-VCE methods relied solely on dis-
method. Table S4 presents the differences between the tance, leading to the selection of nearby points with oppo-
fusion results and three GNSS validation points (MOHO, site deformation characteristics, resulting in the observed
MINA, and RHIL). Table S5 shows the RMSE values of jump in the calculated three-dimensional deformation. This
the 3D deformation solved by the GNSS stations in the phenomenon is more likely to occur near the fault. In con-
study area. The RMSE of the three methods are consistent trast, our proposed method utilizes Voronoi diagrams to
in the east–west and vertical directions, while the proposed represent the influence range of each GNSS station (as
method in this paper has improved to a greater extent for shown in Fig. 2(c) and expressed in Eq. (5). When a target
the north–south deformation that is insensitive to InSAR. point is incorporated into the GNSS station set, resulting
in the reconstruction of the Voronoi diagram, the change
5. Discussion in the proportion of influence range for each GNSS station
serves as the basis for weighting. This approach mitigates
In this paper, we conducted an analysis of the simula- the impact of the sparsity or density of GNSS stations on
tion experiment using the SISTEM, SM-VCE, and the pro- the calculated three-dimensional deformation results.
594
M. Chen et al. Advances in Space Research 73 (2024) 585–596

Through simulation experiments and real-world analysis, Acknowledgements


our method demonstrates superior performance near fault
zones. This research was supported by the National Natural
Certainly, our method in this paper cannot entirely Science Foundation of China (42104008, 42204006), and
avoid selecting GNSS stations, near the fault, that may the Jiangxi Provincial Natural Science Foundation
exhibit different deformation characteristics on opposite (20232BAB213075). The authors would like to thank the
sides of the fault. Similar situations can also arise in the Alaska SARVIEWS project for providing the coseismic
process of selecting neighboring points for InSAR, and InSAR data, and Nevada Geodetic Laboratory for provid-
how to eliminate heterogeneous points using fault informa- ing coseismic GNSS offsets.
tion is a challenge that Strain model-based methods need to
address. As research progresses, various approaches and Appendix A. Supplementary material
methodologies, such as those discussed by researchers like
Liu et al., (2021), Hu et al., (2021), Chen et al., (2023), have Supplementary data to this article can be found online
explored how to eliminate heterogeneous points in InSAR at https://doi.org/10.1016/j.asr.2023.11.013.
and have made relevant advancements. The fusion of
GNSS and InSAR data near faults requires further investi- References
gation, and future research could explore optimized fusion
methods based on fault information for GNSS and InSAR. Chen, M.K., Xu, G.Y., Wang, L.Y., 2023. InSAR 3D coseismic surface
deformation inversion: a combination method considering deforma-
tion gradient. Geomatics Information Sci. Wuhan Univ. 48 (8), 349–
1358. https://doi.org/10.13203/j.whugis20220284 (in Chinese with
English abstract).
Fialko, Y., Simons, M., Agnew, D., 2001. The complete (3-D) surface
6. Conclusion
displacement field in the epicentral area of the 1999 M W 7.1 Hector
Mine Earthquake, California, from space geodetic observations.
In this paper, we proposed a method for spatial analysis Geophys. Res. Lett. 28, 3063–3066. https://doi.org/10.1029/
of GNSS station distribution using Voronoi diagrams to 2001GL013174.
obtain optimal site selection and optimized weighting fac- Gudmundsson, S., Sigmundsson, F., Carstensen, J.M., 2002. Three-
dimensional surface motion maps estimated from combined interfer-
tors. We applied this method to construct a strain model
ometric synthetic aperture radar and GPS data. J. Geophys. Res.:
and to internally weight InSAR and GNSS data in the Solid Earth. 107, ETG 13-1-ETG 13-14. https://doi.org/10.1029/
VCE method. Our method uses the change in area of Vor- 2001JB000283.
onoi diagrams generated before and after the addition of a Guglielmino, F., Nunnari, G., Puglisi, G., et al., 2011. Simultaneous and
target point to characterize the contribution of GNSS sta- integrated strain tensor estimation from geodetic and satellite defor-
mation measurements to obtain three-dimensional displacement maps.
tions to the target point and its weight in the solution. The
IEEE Trans. Geosci. Remote Sens. 49, 1815–1826.
logic of GNSS site selection and weight allocation, com- Hammond, W.C., Blewitt, G., Kreemer, C., et al., 2020. Geodetic
bined with the strain model that considers the spatial cor- observation of seismic cycles before, during, and after the 2020 Monte
relation of adjacent points and the VCE method that can Cristo Range, Nevada earthquake. Seismol. Res. Lett. 92, 647–662.
precisely weight multiple data sources, accurately inverts https://doi.org/10.1785/0220200338.
Hu, J., Liu, J., Li, Z., et al., 2021. Estimating three-dimensional coseismic
the 3D deformation of the study area. We compared the
deformations with the SM-VCE method based on heterogeneous SAR
inversion results of 3D deformation using SISTEM, SM- observations: selection of homogeneous points and analysis of
VCE, and our proposed method in both simulation and observation combinations. Remote Sens. Environ. 255. https://doi.
real experiments. The experiments show that our proposed org/10.1016/j.rse.2021.112298 112298.
method obtains more accurate coseismic deformation Koehler, R.D., Dee, S., Elliott, A., et al., 2021. Field response and surface-
rupture characteristics of the 2020 M 6.5 Monte Cristo Range
results than SISTEM and SM-VCE methods, especially
earthquake, Central Walker Lane, Nevada. Seismol. Res. Lett. 92,
for north–south deformation, demonstrating the reliability 823–839. https://doi.org/10.1785/0220200371.
and superiority of our proposed method. However, the Liu, J.H., Hu, J., Li, Z.W., et al., 2021. Estimation of 3D seismic
methods in the paper cannot entirely avoid the selection deformation with InSAR: An improved SM-VCE method by window
of heterogeneous points on the other side of the fault. optimization. Acta Geodetica Et Cartographica Sinica 50 (9), 1222–
1239. https://doi.org/10.11947/j.AGCS.2021.20200610 (in Chinese
How to utilize fault information to enhance the fusion
with English abstract).
accuracy of GNSS and InSAR near the fault is a direction Luo, H.B., He, X.F., Liu, Y.X., 2008. Estimation of three-dimensional
worth investigating in the future. surface motion velocities using integration of DInSAR and GPS. Acta
Geodaetica Et Cartographica Sinica 2, 168–171 (in Chinese with
English abstract).
Meyer, F.J., Webley, P.W., Dehn, J., et al., 2016. The SARVIEWS
Declaration of competing interest project: automated SAR processing in support of operational near
real-time volcano monitoring 2016, NH43D-02.
Okada, Y., 1985. Surface deformation due to shear and tensile faults in a
The authors declare that they have no known competing half-space. Bull. Seismol. Soc. Am. 75, 1135–1154.
financial interests or personal relationships that could have Samsonov, S., Tiampo, K., 2006. Analytical optimization of a DInSAR
appeared to influence the work reported in this paper. and GPS dataset for derivation of three-dimensional surface motion.
595
M. Chen et al. Advances in Space Research 73 (2024) 585–596

IEEE Geosci. Remote Sens. Lett. 3, 107–111. https://doi.org/10.1109/ Wang, X.Y., Zhang, J.Q., Zhang, Q., 2016. Inferring Multi-dimensional
LGRS.2005.858483. Deformation Filed in Xi’an by Combining InSAR of Ascending and
Samsonov, S.V., Tiampo, K.F., Rundle, J.B., 2008. Application of Descending Orbits with GPS Data. Acta Geodaetica Et Cartographica
DInSAR-GPS optimization for derivation of three-dimensional sur- Sinica 45, 810–817 (in Chinese with English abstract).
face motion of the southern California region along the San Andreas Wegnüller, U., Werner, C., Strozzi, T., et al., 2016. Sentinel-1 support in
fault. Comput. Geosci. 34, 503–514. https://doi.org/10.1016/j. the GAMMA software. In: Procedia Computer Science, International
cageo.2007.05.013. Conference on ENTERprise Information Systems/International Con-
Voronoi, G., 1908. Nouvelles applications des paramètres continus à la ference on Project MANagement/International Conference on Health
théorie des formes quadratiques. Premier mémoire. Sur quelques and Social Care Information Systems and Technologies, CENTERIS/
propriétés des formes quadratiques positives parfaites. Journal Für Die ProjMAN / HCist 2016 100, 1305–1312. https://doi.org/10.1016/j.
Reine Und Angewandte Mathematik (crelles Journal) 1908, 97–102. procs.2016.09.246.
https://doi.org/10.1515/crll.1908.133.97. Zhou, W., Zhang, W., Yang, X., et al., 2021. An improved GNSS and
Wang, Y.J., Hu, J., Liu, J.H., et al., 2021. Measurements of three- InSAR fusion method for monitoring the 3D deformation of a mining
dimensional deformations by integrating InSAR and GNSS: An area. IEEE Access 9, 155839–155850. https://doi.org/10.1109/
improved SISTEM method based on variance component estimation. ACCESS.2021.3129521.
Geomatics Information Sci. Wuhan Univ. 46, 598–1608 (in Chinese
with English abstract).

596

You might also like