low-density observation problem was relieved slightly by the
algorithmic modifications such as SequuezSAR (Ferretti
et
al
., 2011) utilizing partially coherent distributed scatterers or
NSBAS
(Doin
et al
., 2011; López-Quiroz
et al
., 2009) employing
partially disconnected scatterers in the time domain, obser-
vation densities of time series techniques are not sufficient
to trace landslide risk areas especially considering that the
area of interest is frequently located at the layover side of
SAR
imaging. The observation-density issue of
InSAR
time-series
analyses was reported by Gong
et al
. (2016) in a comparative
study. Therefore, such problems of time-series analyses con-
stitute the basis to employ an error regulation two-pass
DInSAR
method over potential landslide areas. The other technical
issue of
InSAR
time-series analysis is its filtering efficiency
of the atmospheric phase error field, so called Atmospheric
Phase Screen (
APS
), and
DEM
errors in a simultaneous way.
Since sometimes error fields are not easily modeled with a
simple ramp shape, it causes failure of time series analysis to
address error component. Thus, the integrated approaches to
combine
InSAR
time series analysis with a weather forecast-
ing model as shown by Jung
et al
. (2014) and Li
et al
. (2009)
were introduced. However, it should be noted that combined
approaches will largely increase the complexities of process-
ing but have no obvious merit to address the lack of reliable
scatterers of time-series analysis for landslide applications.
Therefore, the error regulated two/multi pass
InSAR
is preferred in some application cases such as localized
landslide observations. In order to address errors caused by
atmospheric effects of single/multi-pass
InSAR
pairs, a water
vapor correction methods have been actively developed.
The most straightforward approach is to employ the
correlation between topography and the tropospheric error
components which was assigned by liner (Peltzer
et al
., 1999)
or exponential model (Taylor and Peltzer, 2006). Although
it can be more effective than complicated approximation of
APS
as suggested (Kinoshita
et al
., 2013) in some cases, the
local turbulence by sudden atmospheric instability and/or
onopraphic effect cannot be manipulated by the topography
correlated
APS
correction. Thus, those observations justify is
the employment of high resolution water vapor information.
There are two sources to extract high resolution water vapor
information for the application of
InSAR
error corrections. One
is the weather forecasting model (Nico
et al
., 2011; Yun
et al
.,
2015) and the other is the space-borne radiometry as shown
in Li
et al
. (2003, 2005, and 2006) employing contemporary
Earth observation assets such as a MEdium Resolution
Imaging Spectrometer (
MERIS
). Compared to the time series
analyses, the use of high resolution
APS
correction will allow
to estimate errors more reliably and to decrease the temporal
gap between
SAR
and error estimation. However, the tasks
required to effectively suppress errors in
DInSAR
measurements
are highly challenging. Moreover, topographic error need to be
minimized by the use of maximum accuracy base
DEM
.
Thus, our approaches in this study to build
APS
consisting
of wet component error correction from
MERIS
and dry
component from ground temperature and pressures and
combine highly reliable base topography from high resolution
lidar. Then, the constructed
InSAR
observation will achieve
sufficient observation density and reliable deformation
measurements for the landslide applications. In particular, it
is the first case to introduce direct error compensation from
both high resolution
APS
and base topography. Thus, success
of this approach for the landslide application will establish a
model case to build up deformation measurements which will
be prelude of landslide in highly harsh environment for the
forecasting of landslide risk by the conventional
InSAR
time
series analyses. We tried to demonstrate the superiority of two
pass error correction method compared to the
StaMPS
which is
one of In
SAR
time series analysis.
Test Site and Data Sets
The region of interest is located near the coastal region of the
Uljin County in the Northeastern Gyeongsang Province of
the South Korean peninsula (36.9° N, 129.4° E). Because of
the relatively high population density in this area, landslides
originating in steep and high-relief terrain often as conse-
quence of localized torrential downpours are critical issues
given high seasonal precipitation rates of up to over 200 mm
during summer monsoon seasons. Consequently, most
landslides occur between spring and fall and are caused by a
combination of physical weathering, such as frost shattering,
during winter-time, and subsequent slope wash during spring
thaws, or by extensive monsoon rainfalls in summer.
In the coastal areas of the Uljin County, industrial construc-
tion and buildings have often been placed against steep arti-
ficial wall cuts. Previous studies have revealed several major
landslide characteristics in this area: with over 80 percent of
all observations, the major landslide type originates as debris
flow (Lee and Yu, 2009), the volumetric size of 85 percent of
all observed landslides is less than 1,000 m
2
, and thus highly
localized (Choi, 2001); and a major landslide EFs is the com-
bination of geological characteristics and rainfall (Choi, 2001;
Hong
et al
., 1990). Accumulated rainfall in excess of 200 mm,
maximum rainfall of 30 mm/h, and daily rainfall of 150 mm are
particularly considered to be thresholds for landslide formation
(Choi, 2001). Furthermore, 90 percent of the landsides in Korea
are triggered in igneous and metamorphic rock regions (Lee
and Park, 2005), in particular in shale, granite gneiss, and meta-
sedimentary units which increase the feasibility of landslide
formation in this area. However, since landslide record in the
target areas have not been kept by local authorities, it remains
challenging to quantitatively investigate landslide frequency.
Although high precision
DInSAR
measurements are an
essential tool under such conditions, it is difficult to attain
a sufficiently high accuracy because of the external factors
that cause errors in wave propagation. The dense forest in
the target area is paramount to regulate the risk of landslides,
but it also presents an obstacle for accurate
DInSAR
measure-
ments. On the other hand, local climate characteristics such
as orographic effects (Smith and Evans, 2007) and the proxim-
ity to the seashore can produce significant anomalies in the
water vapor distribution, and can consequently result in the
error components of
InSAR
phase angle measurements. Figure
1 shows aspects of the unequal water-vapor distribution in
the target area. In such settings, the water vapor and the wet
phase delay will be concentrated near the eastern coastal line.
Moreover, high altitude parts of the target area cause a dry
delay error in the
DInSAR
measurements. After all, an im-
proved
DInSAR
approach is essential to address the challenges
described above. For this purpose, we analyzed eight
ENVISAT
ASAR images along with auxiliary data as shown in Table 1.
Besides the
SRTM
and
ASTER GDEM
datasets as base
DEMs
for
InSAR
processing, we employed a lidar
DEM
which was
acquired in early-2010 with Optech
ALTM
30/70 with a mini-
mum point density of 1 pt/m
2
. Since the expected geodetic ac-
curacy was less than 30 cm horizontally and 15 cm vertically,
we employed the lidar
DEM
as ground truth for the topography
as well as the base
DEM
for
InSAR
processing.
Processing Strategy
Due to the lack of available
InSAR
pairs over the target area
shown in Table 1, it is difficult to apply conventional time
series analysis techniques such as
SBAS
and
PS
to compensate
for error terms in the
DInSAR
analysis and to forecast landslide
locations. To overcome this challenge, we employed
StaMPS
,
which was developed for extracting reliable deformation val-
ues even in the presence of error terms and fewer
InSAR
pairs.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
April 2018
191
