methods, and Fey
et al
. (2015), who utilized three-dimension-
al displacement vector extraction. In terms of response time
and spatial resolution, data acquisition from unmanned aerial
vehicles have shown to be beneficial for landslide risk assess-
ment (Shahbazi
et al
., 2014).
However, recent work covering remote sensing tools for
landslide forecasting has quickly been moving towards direct
measurement of subtle dynamic changes in topography, which
can precede and potentially indicate a landslide event (e.g.,
Tralli
et al
., 2005). For such applications small amounts of creep
at the level of a few centimeters are usually to be measured and
only few methods for detecting and monitoring topographic
change by remotely sensed data are applicable to achieve that
precision. A pixel tracing method of high resolution optical
and/or image radar was suggested by Leprince
et al
. (2007) but
it has also been shown that surface deformation can be detected
very well using
SAR
applications as they are easily obtainable
using spaceborne sensors such as C-band of
ERS
,
ENVISAT
and
RADARSAT
, L-band
ALOS
PALSAR
and recent high resolution
X-band TerraSAR-X and CoSMO skyMed (e.g., Herrera
et al
.,
2011; Notti
et al
., 2010; Bovenga
et al
., 2012).
It has been known that
InSAR
is theoretically capable of
tracing surface deformation up to a few millimetres after
correcting for the interferogram phase produced by common
topography (Gabriel
et al
., 1989). Therefore,
InSAR
techniques
accommodating a common topographic basis, called
DInSAR
,
can be used to monitor landslides, as shown by Cascini
et al
.
(2009), Catani
et al
. (2005), Colesanti and Wasowski (2006),
Dehghan-Soraki
et al
. (2015), and Ye
et al
. (2004). A more
recent development in conducting
DInSAR
analyses to monitor
landslides involves the integration of ground surveys, the
geological context, and a numerical model (Hilley
et al
., 2004;
Tomas
et al
. 2014 and 2015), or else using remote sensing data
sources such as aerial photographs (Strozzi
et al
., 2004 and
2010) to clearly identify hazard zones and recent landslide areas.
DInSAR Error Regulation for Landslide Monitoring
In order to carry out landslide monitoring using
DInSAR
,
highly accurate and continuous monitoring techniques should
be developed to suppress external error elements. Given, for
example, that the potential risk area for wall-rock failure is
located along a steeply inclined artificial slope, the
DInSAR
ac-
curacy is affected by a number of atmospheric properties that
have to be dealt with.
The key technical element for the atmospheric error
correction of
DInSAR
is mainly due to the electromagnetic
phase delay by troposphere water vapor turbulence, i.e., the
so-called “wet error term”, and the stratified pressure and
temperature, which are summarized in the “dry error term”.
Additionally, there is an ionospheric error component due to
charged particles, which affects the observation of potential
surface deformation. Thus, in terms of landslide monitoring
in sloped area, the key issue is to effectively address the wet
and dry components.
Also, there are a number of additional errors such as
the error produced by the base
DEM
, and potentially less
well-constrained or even incorrect orbital information. It is
generally accepted that the influence by the accuracy of the
base topography in
DInSAR
analysis is not significant. Hanssen
(2001) estimated that a one-meter height difference in the base
DEM
might cause a scale deformation error of one centimeter,
which is used for the topographic correction during
DInSAR
analyses. What is missing in this assessment is the absolute
accuracy of the
DEMs
that are used for
DInSAR
topographic
correction in a high-relief and steep-sloped area where the
prediction of landslides is crucial but the
DEM
accuracy is
not reliable due to potential horizontal and vertical mis-
registrations. For example, a Shuttle Radar Topography
Mission (
SRTM
)
DEM
with a resolution of 3 arcsec (~90 m)
which is usually employed for
DInSAR
analysis, can show
vertical errors of more than ten meters due to the local slope,
especially in high-relief areas (Kim
et al
., 2017). The very
same problem might be shared also by other
DEM
sources such
as
ASTER GDEM
with a resolution of 1 arcsec corresponding to
a ~30 m grid resolution. Finally, the weak phase coherence
over vegetated canopy also deteriorates the quality of surface
deformation assessments using
DInSAR
observation.
The total error of the interferometric phase angle can be
expressed through the basic relationship of interferograms:
Φ
=
ψ
m
–
ψ
s
=
4
π
λ
(
L
M
– L
S
) =
4
π
λ
[(
L
Mg
– L
Sg
)+(
L
Me
– L
Se
)] (1)
where
Ф
is the observed phase difference,
ψ
m
,
ψ
s
are master
and slave phase angles, is the wavelength of
SAR
,
L
M
and
L
S
are the path lengths of master and slave
SAR
observation,
L
Mg
and
L
Sg
are the geometric path lengths of master and slave
SAR
observations, and
L
Me
and
L
Se
are the path length delay by
error terms.
The observed phase difference is the sum of the phase
angle differences of geometric movement/position and the
error terms so that:
Φ
=
Φ
G
+
Φ
E
(2)
Φ
=
Φ
a
+
Φ
o
+
Φ
t
+
Φ
n
(3)
Φ
a
=
Φ
tropospheric
+
Φ
ionospheric
=
Φ
dry
+
Φ
wet
+
Φ
ionospheric
(4)
where
Φ
G
is the phase difference by geometric movement/
position,
Φ
E
is the phase difference by whole error factors,
Φ
a
is the phase difference by atmospheric error terms,
Φ
o
is
the phase difference by inaccurate orbital information,
Φ
t
is
the phase difference by inaccurate base topography,
Φ
n
is the
phase noise,
Φ
tropospheric
is the phase difference by tropospheric
components, and
Φ
ionospheric
is the phase difference by delay
in the ionosphere. The latter component is not in the scope
of this study considering its insignificant contribution in the
C-band
SAR
phase difference. Furthermore,
Φ
dry
is the phase
difference due to the stratified temperature and pressure
effects (dry term), and
Φ
wet
is the phase difference based on
the water vapor anomaly (wet term).The technical challenge
for employing
DInSAR
for landslide monitoring over a steep
slope is targeted on effectively addressing the phase error
component in
Φ
t
,
Φ
dry
,
and
Φ
wet
.A stacked time series of
InSAR
interferograms and their interpolation have shown to be very
effective in addressing all these error components, thus, few
approaches utilizing interferogram time series such as the
Small Baseline Subset (
SBAS
) by Berardino
et al
. (2002) and
Permanent Scatterers (
PS
) by Ferretti
et al
. (2000) have been
developed. The difficulty of such interferogram time series
techniques is that it is necessary to process a large number
of
SAR
image sequences, and it usually not straightforward to
collect enough (usually 25 to 30)
InSAR
pairs in order to build
reliable
PS
results (Colesanti
et al
., 2003). Therefore, another
recent trend in time series entails using advanced time series
analysis techniques, which impose fewer constraints upon
the number of
InSAR
pairs and accuracy. Notable studies
that have used advanced time series analysis with
InSAR
to
monitor landslides include Liu
et al
. (2013) and Tomas
et al
.
(2014). Although time series analysis is an effective method
for addressing not only atmospheric errors but also potential
DEM
errors, it exhibits significant problems when monitor-
ing localized deformation over potential landslide areas. A
technical basis for time-series analysis is to find a consistently
strong scatterer for the atmospheric and
DEM
error compensa-
tions using interpolation and filtering in both spatial and tem-
poral domains. Therefore, the density of observations points
from time series analysis is highly limited. Although such a
190
April 2018
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
