IEEE Geoscience and Remote Sensing Magazine - March 2020 - 36

introduced the maximum a posteriori probability estimation framework into the tiling strategy to resolve local and
global inconsistencies. Zhang et al. [124] partitioned the
data set that exceeded computer hardware capabilities into a
group of blocks and utilized an approximate L1-norm solution of a network flow problem to direct the block unwrapping. Then, the full-size unwrapped result was obtained by
assembling the unwrapped blocks. Similarly, the authors in
[125] introduced a simulated annealing idea into Goldstein's
branch-cut algorithm to parallelize the set of branch cuts,
and the corresponding parallel computing nature-based tiling strategy guaranteed the globality of phase unwrapping.
Moreover, in [83] and [126], the authors proposed a residue
clustering-based phase unwrapping algorithm that not only
improved speed and saved memory but also guaranteed the
consistency of local and global unwrapping results.
ATMOSPHERE
Atmosphere has always been a key factor affecting timeseries InSAR and one of its main error sources [127]-[129].
The electromagnetic signal passes through the ionosphere
and troposphere. Delays in the ionosphere are caused by
spatial and temporal disturbances of free-electron density.
The ionospheric effect is mostly observed by long-wavelength SAR sensors caused by the dispersive property of
the atmospheric medium. Meanwhile, tropospheric delays
are nondispersive effects affected by pressure, temperature,
and humidity [11].
Considering that the ionosphere is a dispersive medium,
the delay is different when systems operate in two different
microwave frequencies, such as the Global Navigation Satellite Systems [132]. Several approaches exist to estimate the
ionospheric phase delay using InSAR data. One approach
applies azimuth offsets, which can be integrated along the
azimuth direction to estimate the ionosphere phase delay
[133]. Another approach relies on the dispersive characteristic of the ionosphere for microwave signals and divides
the spectrum of the radar signal in range direction into two
subbands, from which two lower-resolution SAR images at
different center frequencies are formed [134].
As for the troposphere, the atmospheric signal shows
vertical stratification, which is obviously influenced by
terrain and has strong randomness in time. Current solutions focus on two different aspects. One is using external
data to compensate for the APS, including the use of local
or global weather models, large GPS network data [129],
and satellite multispectral imagery, such as moderateresolution imaging spectroradiometers [133], [134] and
medium-resolution imaging spectrometers [135], [136].
Numerical models can estimate the atmospheric phase
from meteorological observation data [137], such as using
weather research and forecasting [138], [139] to simulate
the atmospheric phase. Combined methods that use GPS
data together with weather models to produce atmospheric
delay maps on SAR acquisition dates have also been developed [140], [141].
36

Another group of methods focuses on mitigating or
correcting delays based solely on the InSAR data themselves. Typical methods involve averaging N independent interferograms [142] or utilizing the space-time
characteristics of the atmospheric signal to estimate the
APS. Because the stratified component of APS in interferograms can be empirically estimated from the best-fitting linear relation between the phase delay and topography, a more adaptive method has been proposed as a
quadtree-aided joint model for correcting stratified APS
[11]. As the relationship with topographic height might
vary from place to place, the stratified delays should be
modeled locally.
MULTIDIMENSIONAL DEFORMATION
Due to the limitations of the SAR imaging mode, the deformation obtained from time-series InSAR is along the
line-of-sight direction [143] and sometimes cannot meet
the demand. Monitoring multidimensional deformations
along north-south, east-west, and vertical directions is a
difficult problem in time-series InSAR [144]. Current solutions include the fusion of ascending and descending data
[18], [145] and the fusion of external data, such as leveling
and GPS [146]-[148].
Because SAR satellites are all in near-polar orbit and
SAR imaging is underside looking, it is insensitive to the
deformation in the north-south direction. The joint ascending and descending data can obtain the deformation
along only the east-west and vertical directions. GPS can
acquire high-precision 3D deformation results and assist
InSAR with acquiring 3D deformation. After the GPS results are interpolated, the global optimization method
is used to fuse time-series InSAR observation results and
GPS data to obtain high-precision 3D deformation information. In [149], the authors used the Markov random
field model technique and Bayesian statistical theory to
fuse the deformation results of GPS and InSAR, thus obtaining the 3D deformation map for California. Considering that the optimization model generates ill-conditioned
equations when the GPS points are sparse, Hu et al. [150]
used a Broyden-Fletcher-Goldfarb-Shanno algorithm to
iteratively optimize the fusion model and so overcome the
instability of the solution. Because the fusion model may
easily fall into the local extremums, the authors in [151]
introduced the ant colony optimization method to solve
the model, which obtained considerable results. Similarly,
GPS-based methods are limited by the sparse spatial distribution of GPS points.
DISCUSSION AND CONCLUSIONS
As an effective surface deformation monitoring method,
time-series InSAR breaks through the low noise capacity and low accuracy of DInSAR. After nearly 20 years
of development, the practicality and applicability of
time-series InSAR technology have been continuously
improved and widely verified in different fields. It has
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

MARCH 2020

IEEE Geoscience and Remote Sensing Magazine - March 2020

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2020