IEEE Geoscience and Remote Sensing Magazine - March 2020 - 12

As the third phase is only beginning, it is difficult to determine its effect, but, in terms of research work, some approaches are already changing. In the second phase of data
availability, it was common to carefully search test sites and
research areas, analyzing the situation in situ before ordering SAR data. Now, when working in a research area, we
often begin with the data, investigating Sentinel-1 stacks to
look for motion elements beTHE DEVELOPMENT OF
fore going to the field. This apPSInSAR OPENED THE DOOR
proach is driven by the ease of
data availability and the ability
FOR SUCCESSFUL SURFACEto test and play with the data,
MOTION MONITORING IN
obtaining preliminary results
URBAN AREAS.
before, for example, deciding
to order additional high-resolution data.
The basic support of this phase comes from the differential InSAR (D-InSAR) nature of the Sentinel-1 data. These
are available from 2014, their orbital tubes have minimal
width, and the acquisitions are ubiquitous, with fixed 6or 12-day revisit intervals. This period is short enough that
motion can be evaluated not only for human artifacts but
also for the majority of sites. Based on our assessment, we
expect a further increase in InSAR-related papers. With the
growing number of InSAR-related projects, we believe this
will soon be seen publications, as the bibliometric analysis
shows research trends, but always with a delay.
METHODOLOGICAL DEVELOPMENT
PHASE-BASED SURFACE-MOTION ESTIMATION
The methodologies for InSAR were presented at the end
of the 1980s with InSAR [4] and D-InSAR results [5]. The
interferometric phase { int is the phase difference between
two complex SAR images, usually acquired from similar
positions with the same sensor or sensors with identical
system properties:

{ int = { s - { m =

4r
Drlos, (1)
m

where { m and { s are the phases of the master and slave
images, respectively; m is the wavelength of the system; and
Drlos is the range difference between the master and slave
images in the line-of-sight (LOS) direction. Differences in
Drlos are caused by motions of the surveyed object between
acquisitions and also from differences in the topographic
height of the surveyed objects in relation to the difference
of the sensor positions of the slave and master images.
Therefore, it is useful to separate the different components contributing to { int, which are

{ int = { flat + { topo + { motion + { atmo + { orbit + { noise, (2)

where { flat is the phase contribution caused by the range
differences at zero height (the so-called flat Earth); { topo is
12

the phase contribution caused by the relative height of the
object; { motion is the phase contribution caused by motion
of the object between the acquisitions; { atmo is the phase
contribution caused by differences in the path delay of the
electromagnetic signal passing through the atmosphere;
and { orbit is the phase component caused by differences in
the orbit estimation and true orbit position of the sensor.
Finally, { noise contains noise terms caused by thermal noise
of the system and other noise sources not included in the
other terms.
This separation of the phase components allows us to
address each separately. { topo is calculated similarly to the
phase contribution from motion, based on the fact that a
difference in the topographic height Dh causes a difference
in Drlos, depending on the perpendicular baseline between
sensor positions B = and, thus, on the width of the orbital
tube. Therefore, { topo can be calculated as

{ topo =

B = Dh
4r
, (3)
m rm tan (i inc)

where rm is the distance (range) between the sensor and the
object in the master image, and i inc is the local incidence
angle at the object's position in the master image.
Another important point is that { int and the other phases are wrapped phases; the value is limited between - r
to r. Therefore, even for large values of Drlos, the resulting
phase is limited to values between - r and r; that is, { int is
ambiguous and allows for an unambiguous measurement
of motion for small values of Drlos 1 _ m 2 i .
In D-InSAR, the goal is to derive the LOS motion and,
therefore, separate the { motion component from { int. This
can be done by simulating { topo based on a preexisting
digital elevation model (DEM) and removing the simulated phase from { int. This has been successfully used,
for example, in seismic-event analysis [6]; however, huge
errors remain due to the unresolved phase components
{ atmo, { orbit, and { noise. Furthermore, as no DEM is perfect,
there remains a { topo contribution from the difference between the DEM and real topographic height position of
the scatterer.
The largest problem in InSAR and D-InSAR is temporal
decorrelation, which is caused by changes on the ground
between the acquisitions, mainly from growing vegetation,
displacement from wind, or displacements of the scattering centers depending on the spatiotemporal water content.
This effect is strong, causing phase differences > 2r, especially for shorter wavelengths, rendering InSAR unusable in
many heavily vegetated areas. In InSAR, this effect can be
mitigated by reducing the time between acquisitions, ideally to close to zero, but this is not possible when measuring motion, as this requires a time difference between
the acquisitions.
PSInSAR [2], introduced in 2001, plays a pivotal role
in solving this problem (Table 2). PSI circumvents the
problem of temporal decorrelation by considering only
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