IEEE Geoscience and Remote Sensing Magazine - March 2020 - 15

deformation patterns of buildings become an important
additional phase contribution; they can be separated from
the linear deformation pattern and modeled separately by
including temperature information [7].
PSI works well in areas where a high PS density can be
achieved, such as urban settings. In rural locations, the
PS density is often too low to remove the atmosphere successfully or obtain measurements in the area of interest.
SBAS [8] already addressed this issue. With the concept
of quasipermanent scatterers (QPSs) [20], the idea of scatterers that temporarily act as PSs was introduced, helping
increase the density of measurement points outside of urban areas.
To include more points, it is necessary to work with distributed scatterers (DSs). A DS is formed from a large group
of adjacent pixels sharing the same scattering mechanisms.
Through statistical analysis, the scattering parameters can
be retrieved. As a PS is dominated by one single scatterer,
there is no speckling effect, and we have a deterministic signal. When evaluating DSs, the speckling effect is apparent,
and statistical analyses are needed.
The first work in DS investigation showed that, using
the full covariance matrix, the parameters of a PS model
(such as linear deformation and DEM error) of DSs relative
to a PS can be determined [21]. In practice, this allowed the
inclusion of DSs in a postprocessing step, densifying the already established PS analysis. However, in rural areas with
very limited PSs, this is not enough.
Allowing DSs and PSs to be processed together was the
next critical step. Based on work on scattering statistics [29],
this was finally realized with the SqueeSAR approach [26].
SqueeSAR is based on an adaptive neighborhood formed
via DeSpecKS. Afterward, the estimated phase history of
a DS is derived from the phase triangulation algorithm
(PTA), which then allows use of DSs, similar to PSs, in all
standard PSI processing steps. Related approaches are also
used to extend SBAS [30]. Advanced multibaseline D-InSAR approaches are another alternative for processing SAR
stacks with sparse PS points [31].
In the new DS interferometry (DSI) approach [27], the
statistical homogeneity test of the neighborhoods is based
on the Anderson-Darling test instead of the Kolmogorov-
Smirnov test used in SqueeSAR. To reduce the computational burden, DSI uses a small baseline approach instead
of the full graph.
Another important research direction is SAR tomography [32]-[34], which makes full use of the information provided in an InSAR data stack. We can divide the
SAR tomography approaches into two main directions.
The first involves reconstructing the 3D volume of a penetrable medium, such as a forest. This is typically done
by working with long-wavelength SAR data, as the penetration capability is needed to get a SAR profile along the
elevation direction.
The second approach, which is more suitable to shortwavelength data in urban areas, entails solving the layover
MARCH 2020

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

problem and separating a limited number of scatterers
in a resolution cell, also allowing for differential tomographic SAR and deformation analysis [35]-[38]. Today,
forest tomography can be carried out only with airborne
data, as the temporal decorrelation would be prohibitive
with the revisit times and wavelengths of current satellites [39]. A P-band satellite mission devoted to this (the
ESA's Biomass) is to be launched in 2022 [40]. Airborne
glacier tomography is also possible, bettering groundpenetrating radar [41].
AMPLITUDE-BASED SURFACE-MOTION ESTIMATION
Phase-based techniques are referenced when surface-motion monitoring from SAR is discussed. However, differential interferometry is not the only possibility for measuring surface motion from SAR images. Amplitude-based
methods offer another approach for motion estimation
from SAR.
Amplitude-based methods have several advantages over
phase-based approaches. Amplitude is far less sensitive to
changes and, therefore, does
not suffer so strongly from
temporal decorrelation. It is
AT THE END OF PHASE TWO
robust with respect to atmoAND THE BEGINNING OF
sphere changes. AmplitudePHASE THREE, SERVICE
based methods not only provide
PROVIDERS AND OTHER
results in the range direction;
RELATED PARTS OF THE
they can also measure moINDUSTRY WERE
tion in the azimuth direction,
CONSOLIDATING.
which phase-based methods
cannot. However, the precision
of amplitude-based methods
depends directly on the spatial resolution of the SAR system
and is generally lower than that of differential interferometry techniques.
The first applications of pixel-offset tracking with SAR
data for sea-ice monitoring, presented in 1991, were based
on Seasat SAR images [42]. In 1992, Scambos et al. [43] demonstrated the use of pixel-offset tracking for glacier-motion
estimation. Amplitude-based methods are often used for ice
and glacier monitoring due to the lack of coherence, which
hinders InSAR measurements on ice, and to the relatively
high velocities of glaciers, which are typically too fast to be
estimated correctly from D-InSAR; pixel-offset tracking and
similar methods work well with higher velocities. However,
the technique is not limited to ice; in 1999, Michel et al. [44]
measured the deformation of the Landers earthquake using
ERS data. By using ASAR, the displacement from the 2005
earthquake in Kashmir was shown [45].
As the precision of amplitude-based methods depends
on the spatial resolution of the SAR system, it should
come as no surprise that development increased drastically with the availability of high-resolution SAR data
from the TerraSAR-X and COSMO SkyMed missions. With
high-resolution data, amplitude-based surface-motion estimation became more suitable for various applications
15

IEEE Geoscience and Remote Sensing Magazine - March 2020

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