IEEE Geoscience and Remote Sensing Magazine - March 2020 - 138

successful application of a timely early warning for landslides by in situ measurements [32].
Based on the advantages, limitations, and complementarity of different EO methods, a landslide early warning
framework is proposed to increase the resilience of local
communities to landslide hazards by informing populations of when to leave for short-term evacuations.
This article makes the case that obtaining a landslide
early warning from EO is now within our grasp. We believe
that this message is both important and timely. It is significant because landslides kill thousands of people every
year, predominantly in parts of the world that are poorest
and thus least able to protect themselves. It is well-timed
because, although early warning has long been touted as
a "golden bullet" in landslide
risk mitigation, it requires accurate predictions that have
TO MONITOR A SINGLE
generally been out of reach
SLOPE IN ITS TERTIARY
until now.
PHASE, InSAR AND IN SITU

METHODOLOGY
The InSAR data set for the
COMPLEMENTARY
time series displacement exCOVERAGE IN SPACE
traction of Xinmo landslides
AND TIME.
includes 29 descending SAR
images acquired by Sentinel1A/1B satellites from 9 November 2015 to 19 June 2017. The European Space Agency's (ESA's) Sentinel-1A/1B satellites operate day and night,
performing C-band microwave SAR imaging and providing radar imagery with wide coverage (e.g., 250 × 250 km)
and a short repeat cycle (6-24 days). The SAR data in this
study were interferometrically processed with GAMMA
software. A shuttle radar topography mission (SRTM)
with 30-m horizontal resolution was used to simulate and
eliminate the topographic phase. Interferograms were filtered by the adaptive filtering method to reduce noise. Coherent pixels were detected using the full-rank matrix approach demonstrated in [33] and their time series analysis
was performed following the InSAR time series integrated
atmospheric estimation model (InSAR TS+AEM) described
in [34]. Both the coherent pixel detection approach and
the InSAR TS+AEM method have been successfully used in
previous InSAR studies. The mean velocity map and time
series displacements results were finally geocoded into the
WGS84 coordinate system.
Researchers from the State Key Laboratory of Geohazard Prevention and Geoenvironment Protection (SKLGP)
at Chengdu University of Technology have been monitoring the Heifangtai area with a range of in situ sensors including seven GNSS receivers, 34 crackmeters, two range
gauges, and 13 piezometers since 2017. The sensors collected data that were transmitted to SKLGP in real time
with the General Packet Radio Service. Note that the real-time adaptive crackmeter developed by SKLGP [35] acquired one sampling per hour in normal conditions but
SENSORS CAN PROVIDE

138

automatically increased its samples when a displacement
acceleration was detected.
RESULTS
PREFAILURE MOVEMENT SIGNALS REVEALED
WITH SPACEBORNE INTERFEROMETRIC SYNTHETIC
APERTURE RADAR
On 24 June 2017, a 13 million-m3 landslide suddenly buried Xinmo, Sichuan, China, causing 10 deaths; 73 people
are still missing. Xinmo is a village on the left bank of the
Songping River, a first-order tributary of the upper reaches
of the Minjiang River [36]. The surrounding steep slopes are
prone to rock falls, landslides, and debris flows [37]. The region is tectonically active: several active faults nearby have
generated three ≥ 6.7-magnitude (Mw) earthquakes since
the 1930s [Figure 1(a)]. Xinmo itself was built on the deposits of an old landslide triggered by the 1933 7.3-Mw Diexi
earthquake [36], [38] [Figure 1(a)].
To explore the prefailure displacement history of
the Xinmo landslide, InSAR analysis was performed on
Sentinel-1 data to determine a mean velocity map and
a time series of landslide motion for an approximately
1.5-year period prior to failure (Figure 2). The accumulative displacement map during the period from November
2015 to June 2017 [Figure 2(a)] shows that the area near
the head scarp of the landslide exhibited clearly detectable
displacements, with a maximum of 3 cm preceding failure.
Figure 2(c)-(e) shows the displacement times series results
for three selected points (P1, P2, and P3), whose locations
are shown in Figure 2(b). The last three acquisition dates
are 26 May 2017, 7 June 2017, and 19 June 2017 (five days
before the failure), respectively. A dramatic acceleration
can be observed during the period from 7 June 2017 to
19 June 2017 (from 17 days before the failure). It should
also be noted that all interferograms were carefully checked
to avoid phase unwrapping errors, and the InSAR time series was performed pixel by pixel. We did not apply strong
spatial filtering; hence, our InSAR mean velocity map is not
as smooth as those in previous studies. However, the overall pattern of our InSAR mean velocity map is consistent
with those in previous results (e.g., [28] and [29]).
These findings clearly demonstrate that quantitative
time-series analysis from satellite radar observations can
detect accelerated movements prior to catastrophic failure,
occurring 5-17 days before the landslide. It should be noted
that the source area of the Xinmo landslide is located on a
steep slope at an altitude of ~3,400 m above sea level, where
in situ sensors would be difficult to install, highlighting one
notable advantage of InSAR over in situ monitoring sensors.
EARLY WARNING FOR THE DANGCHUAN 4#
LANDSLIDE USING IN SITU SENSORS
The Heifangtai loess terrace, located in Yongjing County,
Gansu, China [Figure 3(b)], with an area of 13.7 km2, is
formed from a terrace of quaternary aeolian loess deposits
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