IEEE Geoscience and Remote Sensing Magazine - March 2020 - 96

The presence of APD is strictly related to the temporal
separation of the acquisitions: in twin satellites, such as
TanDEM-X, APD affects the two interferometric images in
the same way [3], leading to a cancellation in the interferometric beating.
Phase measurements are restricted to the principal argument,
or, in InSAR jargon, the measured phase is said to be wrapped.
The retrieval of the absolute phase, that is, phase unwrapping
(PU), may be a complex task, especially in the presence of
strong coherence losses (decorrelation) [81]. Some aspects related to PU in the presence of multiple acquisitions are addressed
in the following sections. However, a comprehensive description of this critical nonlinear processing step is beyond scope of
this article and can be found in the existing literature [84]: here,
it is simply noted that errors at such processing stages may
drastically affect the accuracy and reliability of the results. In
the case of landslides, especially at shorter wavelengths,
PU is a more difficult task due
IN URBAN AREAS,
to the presence of topography
SIGNIFICANT COHERENCE
and APD variations as well
CAN BE OBSERVED OVER
as decorrelation.
YEARS, BUT SPARSE
Among the several factors
contributing
to reduction of
VEGETATION CAN EASILY
the
coherence
degree of interMAKE IT IMPOSSIBLE TO
ferometric
acquisitions,
one
EXPLOIT THE PHASE SIGNAL
of the most critical for repeatAFTER JUST A FEW WEEKS.
pass acquisitions is associated with the temporal variations of the scene reflectivity
(temporal decorrelation). In urban areas, significant coherence can be observed over years, but sparse vegetation can
easily make it impossible to exploit the phase signal after just
a few weeks: this is due to variations (growth/change) in the
greenery. In general, temporal decorrelation has an increasing impact as the transmitted wavelength is reduced. On one
hand, interferograms at the X band are much more affected
by temporal decorrelation with respect to the C and L bands
due to the smaller wavelength; on the other hand, smaller
wavelengths provide a higher sensitivity to displacements [1].
Decorrelation induces phase errors that affect or even, in
some cases, prejudice the process of retrieval of the absolute
phase. Typically, decorrelation can be partially mitigated by
using local spatial averaging (multilooking); however, this
leads to unavoidable spatial resolution losses.
TOWARD MULTIPASS COHERENT JOINT
DATA PROCESSING
As mentioned, a single interferogram is affected by several
error sources; this limits its application to only large-magnitude and wide-deformation events, such as earthquakes
[86] and volcanic activity [87]. Landslide monitoring typically requires handling problems related to
◗◗ limited area extension
◗◗ reduced deformation rates on the order of centimeters
per year
96

◗◗ the presence of decorrelation
◗◗ the requirements of providing deformation measure-

ments up to the full available spatial resolution.
Beginning with the ERS-1 and ERS-2 missions, large
archives of SAR images have been made available. This
stimulated the development of improved (interferometric) data analysis algorithms able to exploit the coherent
nature of the data, which frequently span time intervals of
years, to move beyond the limitations of classical singlepair interferometry.
Acquired data, formed by N multipass acquisitions performed at different epochs from different orbits, may be
stacked into a vector that, according to the model in (2), can
be expressed (after being referred to a single master image)
as a sum of contributions:
	

{ = { z + { D + { A + { n, (3)

where { z is the RT phase contribution after the external
DEM subtraction.
The ability to perform measurements in a vector space with
large dimensions (N is typically larger-even far larger-than
10), together with an appropriate modeling for the different
contributions ( { z aligned to the baseline vector and { A spatially correlated in azimuth and range but temporally uncorrelated), allows, in most cases, a reliable extraction of the deformation time series { D .
In addition to the improvement of earthquake applications
by allowing monitoring of aftershocks in seismic sequences
[88], new, formerly impossible uses were achieved. These concern the monitoring of restrained deformations, which are almost unnoticeable in single-pair interferograms but make a
systematic contribution over time; for instance, this is the case
for ground subsidence induced by underground activities (e.g.,
mining and subsurface extraction, consolidation of soil layers,
and so on). Landslide risk associated with slope instabilities
represents a particularly favorable scenario for multitemporal
DInSAR (MT-DInSAR) analysis.
In this context, two classes of algorithms, sometimes
referred in the literature as A-DInSAR or MT-DInSAR techniques, show complementary characteristics in terms of
the assumed electromagnetic scattering mechanism and,
more important for application to landslides, the scale of
analysis. These classes are the stacking of coherent interferograms (SCI) [5], for the analysis of partially coherent
scatterers at small scales, and PSI [4], for the monitoring of
scatterers retrieving coherence over years at the large (fullresolution) scale. In the following sections, we summarize
the main differences of the two approaches.
MONITORING OF DEFORMATION AT SMALL SCALES
The SCI techniques, first implemented in [89] (published
later in journals in [90]) and subsequently improved in the
small baseline subset (SBAS) method [5], are the basis of
many other implementations [91], [92], including commercial versions [93]. SCI represents the natural extension of
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
