Potential Landslide Detection in Kelok Sembilan Flyover Using SAR Interferometry

Kelok Sembilan flyover is located in West Sumatra-Riau province connection, Indonesia. Landslide occurs along with the connection every year. However, detecting and obtaining scientific information about the potential landslide zone in the area is difficult because it is mountainous, and there is no disaster equipment monitoring installed on it. Here, we identify the area that potential landslide using Satellite Radar SAR Synthetic Aperture Radar) technology with high accuracy until millimeter unit. Also, we showed how to observe the potential landslide with a low penetrate radar signal (C-band) and shifted polarization due to the topography in the vegetation area. The observation data is extracting from satellite Sentinel1A/B provided by the European Space Agency (ESA). The satellite is C-band frequency with 5.5 cm wavelength and 5 x 20-meter range-azimuth resolution. The acquisition time of SAR data is starting from October 2014 to August 2018, which is supported by 119 images. Processing the SAR data is using the Quasi-Persistent Scatterer technique to improve the number of scatterers in a non-urban area. Both ascending and descending orbit of the satellite with both VV and VH polarization were selected to confirm potential landslide detection consistency. The result of processing was validated with GPS geodetic by DGPS network configuration. Based on the QPS processing, five areas with significant land movement have been detected; one of them is collapse after ten days since the last data observation. The land movement range in the area is approximately more than 500 millimeters in four years. Keywords— Synthetic aperture radar; landslide; radar interferometry; sentinel-1. Manuscript received 22 Nov. 2020; revised 12 Dec. 2020; accepted 18 Feb. 2021. Date of publication 30 Apr. 2021. IJASEIT is licensed under a Creative Commons Attribution-Share Alike 4.0 International License.

The Kelok Sembilan flyover was built around a V-shaped area with a high slope based on the topographic on the area. The area is also close to the Solok-Bio-bio strike fault line that existed about 1 km on the western side. Furthermore, based on the geology structures, the area is complex and consists of tension joint, anticline, and syncline on both sides. This condition leads to improve vulnerability landslides in the area. Monitoring continuously in Kelok Sembilan flyover is required to determine which area has a high potential landslide, then can be scientific information in maintenance the construction which has spent much cost to build. No equipment is installed along the area to monitor the potential landslide in the Kelok Sembilan flyover.
In this research, identifying the potential landslide area, Radar SAR (Synthetic aperture radar) satellite technology provided by European Space Agency (ESA) is a sophisticated solution to detect and overcome the unavailability of landslide monitoring equipment on site. Besides, the radar data is free to access also has large data availability with a short temporal and normal baseline of 6 days and 200 meters, respectively [4], [5]. However, the limited penetration of the radar SAR signal due to the radar data is a C-band frequency that is applied in the non-urban (canopy) area, which will affect to lower the number of persistent scatterer (PS) detection than L-band [6], [7]. Furthermore, due to the flyover located in the 'V' shape area, the radar satellite of Sentinel-1 flight parallel with the 'V' slope, the polarization radar signal will shift that induced by terrain slope [8].
Under the condition mentioned above, the quasi-persistent scatterer (QPS) technique was selected and applied to detect the potential landslide area in the Kelok Sembilan flyover. The technique is advanced of the original Persistent Scatterer Interferometry (PSI) that extracts the information from the partial coherence target [9] and then increases the spatial distribution of PS point detection [8]. The technique also allows the estimate of the signal shifting due to the slope area [10]. Furthermore, the technique overcomes the limitation of the Interferometry SAR (InSAR/D-InSAR) [11] technique that is influenced by the temporal and geometrical decorrelation and water vapor content effect [1], and extremely PS cut off in PSI technique [8]. Over the year, the QPS technique spread in many remote sensing applications included but not limited to mining applications [12], land subsidence [13], Monitoring of Tunneling [14], land deformation [15].
In processing, the QPS technique was supported by 119 number scenes of Sentinel-1 SAR data. The observation was starting from October 2014 to August 2018. The scenes were processed in ascending and descending orbits with VV (vertical transmit and vertical receive) and VH (vertical transmit and horizontal receive) polarization. The Geodetic GPS measurement (Geographical Positioned System) was installed in the area to measure the position (latitude, longitude, and high) of PS points during the observation to compare the result of QPS processing. The geodetic GPS equipment was configured with DGPS (Differential Geographical Positioned System) technique in one GCP (ground control point) (P6) and five rover points (P1-P5) (Fig.  7) to determine the GPC point carried out by selecting the stable point (no movement) of the QPS processing result.
The research aims to implement the QPS technique to observe the potential landslide area using C-band SAR applied in a non-urban area and V shape contour shape. The QPS technique compares with the Geodetic GPS equipment that is configured using the DGPS technique. Besides, in this research, we present, 1). The PS distribution in positive and negative slope due to terrain slope affects both ascending and descending orbits with VV and VH polarization. 2). The trend of land movement in potential landslide area detection. 3). Modeling the potential landslide area in the Kelok Sembilan flyover.

A. Material
The sequences of satellite Sentinel-1 SAR data with an acquisition time from October 2014 to Augustus 2018 were used to observe the potential landslide zone in the Kelok Sembilan flyover. The data is C-band frequency with a 5.44 cm wavelength provided by ESA (European Space Agency) [16]. Both orbits of the satellites were selected in observation to obtain the information with high precision and consistency in detecting the potential landslide in the area. Also, its polarization in co-polarization VV (vertical transmit and vertical receive) and cross-polarization VH (vertical transmit and horizontal receive) were extracted. The dataset SAR satellite of Sentinel-1 is listed in Table 1. Both orbits of Sentinel-1 A/B datasets were processed in a single look complex (SLC) format. The SLC data format consists of phase and amplitude information that has focused and georeferenced using altitude and orbit data from the satellite [17], [5]. Also, the data in zero Doppler slant-range geometry [18]. From three-part of beam swath mode of Interferometric Wide (IW) of Sentinel-1, IW1 and IW3 were selected for the study area in both ascending and descending orbits, respectively. IW1 in ascending and descending orbit has a low off-nadir angle than IW3 because its beam swath close to a perpendicular off-nadir angle, then the number of persistent scatterers (PS) detection large than IW3. The spatial resolution for both beam swath modes is 2.7 x 22.5 meters (IW1) and 3.5 x 22.6 meters (IW3) in ascending and descending orbit [19], respectively. Furthermore, the SAR satellite of Sentinel-1 A/B has a large critical baseline Bn, with an inclination angle of 98.18 0 and a repeat cycle of 6 days [20], then more data can be co-registration.

B. Methods
In processing the SAR data Q-PS (Quasi Persistent Scatterer), the Interferometry technique was selected. The Quasi Persistent scatterer technique is advanced from the original PSI (Persistent Scatterer Interferometry) technique [10], [21] that can be developed to extracting the information from the partial coherence target. The primary purpose of the technique is to improve the number of persistent scatterers (PS) in spatial distribution, particularly in a non-urban area [22], [23], [24]. Also, the technique provided tools to estimate the phase-shifted. The main reason selected the technique was the area located in V shape, high vegetation, and topography.
The MST (Minimum spanning Tree) network configuration was selected to maximize the coherence radar target  minimize the atmospheric effect. Also, the connectivity of the images, the number of the link, and the weight of each link were considered. In identification PS, QPS technique extracts the coherence target with high temporal phase-coherent that defined as [11], [17], [25], [26], [27].
Where the component ∆ , is interferometric phase acquiring that do not depend on the elevation and the liner deformation trend, ∆ , , is interferometric phase depends on the elevation, ∆ , , is the interferometric phase of deformation trend for point target T, and N is the number of images. The phase component that depends on the height and linear deformation is presented, respectively, by [11], [28].
and ∆ , , Where λ is wavelength radar signal, R is the slant range radar to target distance, $ is look-angle of radar satellite, Δℎ , is the elevation of point target T relative to control point O, '( , is normal baseline images i and j, Δ) , ) is deformation trend point target T relative to control point O and '* , is the temporal baseline between images i and j. Fig. 1 The schematic radar geometry. It is related to the orientation angle to the ground slope in the Kelok Sembilan flyover area. The x-axis represents the azimuth direction, and the y-axis represents the ground range direction. The surface normal to ground land is indicated by ( +. In-ground truth observation, the QPS technique was validated using geodetic GPS data measured in the study area. In GPS measurement, Differential GPS techniques were applied with one base station and five rover points. Also, its slope was measured with a clinometer instrument and UAV observation.

A. Persistent Scatterer (PS) Distribution and its Density for
Both Ascending and Descending Orbit PS distribution and its density is an essential parameter to determine the accuracy of the land deformation. Besides the object's structure and material in every resolution cell, we found some factors that induced PS distribution and density: SAR frequency, resolution of SAR images, bandwidth, incidence angle θi, and polarization. Fig. 3 presents the distribution and density of PS points on the Kelok Sembilan flyover. The zones toward satellite perpendicular (near range) have high distribution and density of PS for both satellite directions and polarization. However, ascending orbit with VV polarization (Fig. 3A-1) has the highest PS point distribution and density than other satellite data experiments. This result is because the dataset of the satellite is single polarization (SV) VV, with bandwidth twofold in range resolution than double polarization (DV) VV + VH (Fig. 3A-2, 3A-3 and 3A-3). High resolution in the SAR data range improves the number of PS in the range direction.
Besides, the satellite observed in low local incidence angle θi of 34.1 degrees. Meanwhile, the other experimental data have an angle of 44.1 degrees (Fig. 3A-2, 3A-3, and 3A-3). Then the observation with a low incidence angle contributes to improving the density and number of PS detection.
In equal bandwidth and incidence angle, the distribution of PS and its density observed by co-polarization VV is higher than cross-polarization VH but less sensitive to detect the land movement. Moreover, regarding the area with 'V' contour shape, VV, and VH polarization less distribution and density in positive slope for ascending and descending tracks (positive slope for ascending = negative slope in descending and vice versa). It is because the radar signal is affected by the terrain azimuth slope. When the range slope is toward the radar (positive), the orientation of angle measurements is overestimated and vice versa for the negative slope.   Table 2 compares the distribution and density PS point of SAR data with VV and VH polarization for ascending and descending orbits in the positive and negative slopes area. The data was classified based on the number and density of PS inside the square (black) line in both areas.

B. Modeling the Potential Landslide
Using the Inverse Distance Weight (IDW) interpolation technique, the potential landslide area was depicted based on the distribution and density of PS points on the study area. Fig.  5 presents the potential landslide area modeling based on the IDW interpolation technique in contour shape. Satellite data observation with ascending orbit and VV polarization is modeled in (Fig. 4B-1), and descending orbit with VV polarization is in (Fig 4B-2). With this data, five areas have significant land movement detected in both direction and VV polarization. Four areas are located on the slope and one in the flat with 500 millimeters of movement. The area out of the black square line in the positive and negative slope (Fig. 3) is not an object in this research.
The observation using ascending orbit with VH polarization depicted in Fig. 4B-3 and descending orbit with VH polarization show in Fig. 4B-4. Observation using ascending and descending orbit with VH polarization only detect the land movement in negative slope and failed for positive slope (negative slope for ascending = positive slope for descending and positive slope for ascending = negative slope for descending) orbit. The observation using copolarization VV success identifies the potential landslide on both positive and negative slope that is observed from ascending and descending orbit. However, in crosspolarization (VH) observing from both ascending and descending orbit is sensitive in negative slope (low-look angle) and loss for the positive slope. Based on the model in (Fig. 4), there are two areas (zone A and zone B) of five zones with significant land movement detected consistently in both orbit direction and polarization. The area has movement with range -217.6 mm to -561.9 mm and located on a positive slope for ascending orbit or negative slope for descending orbit. The land deformation maximum on the area reaches 700 millimeters in four years (Fig.4B-4). Detailed profile of two areas with significant land deformation in zone A and zone B is presented in Fig. 5 and Fig. 6. Fig. 5 and Fig. 6 show the profile of two of five zones that has significant land deformation observed from both satellite orbits and polarization. The PS point ID that represented the area depicted in the elevation and slope area was measured based on a Digital Elevation Model. For zone A (Fig. 5), the area observation is represented by ID 476 and ID 264 located approximately at 737.5 m and 737 m above sea level, with the local slope of 74.1 degrees and 73.8 degrees and the incidence angle θi of 34.1 degrees and 44.1 degrees for ascending and descending orbits, respectively. The different position of elevation detection of the target is due to the different duration in acquisition time, where the descending orbit observation is nine months longer than that of the ascending. Consequently, the land movement value for descending orbit is also higher than that of the ascending, -561.9 millimeters and -217.6 millimeters for ascending orbit in the LOS. For zone B (Fig.  6), it has a similar land deformation pattern with zone A, represented by ID 950 and ID 126 for both ascending and descending orbits. The point target moves far away from the satellite with a maximum LOS of -253.3 millimeters and -535.3 millimeters and local slopes of 76.7 degrees and 73.5 degrees, respectively. The ground truth campaigns validate the QPS processing taken in two periods. The first term was from 16 September to 18 September 2016, and the second term was from 3 September to 16 September 2018. The collecting data was using the geodetic GPS of LEICA instrument 1200 plus series, UAV phantom 3 professional series, Clinometer, and measuring tape. Using geodetic GPS, the collecting data by DGPS network configuration with post-processing technique in six-point measurements for both terms of the ground survey.
Point 1 (P1) has a significant change than the other points, approximately 170.  Moreover, Clinometer was used to measure the slope of the flyover, especially for the flyover in zone B. For seven points measurement, the flyover has tilted by 6-9 degrees. The flyover's wall is no longer symmetrical on both sides, and each side moves around 2-4 cm. This movement happened because in the south part of the bridge was subsidence (Fig.  8). Fig. 8 depicts the potential landslide area obtained from ascending and descending SAR orbits which were modeled based on the IDW (Inverse Distance Weighted) interpolation technique. The technique was selected because it has successfully demonstrated the land movement until small, covered areas. There are five zones of potential landslide detection on the positive and negative slopes of the Kelok Sembilan flyover. Two of them are critical, namely, zone A and zone B that have a significant area covering potential landslides with more than 500-millimetre maximum land movement in four years.
One month since the last data observation, the landslide occurred on September 26, 2018, in zone B (opposite to the flyover). On the left part of the flyover, the blue cycle shows a landslide area in May 2016 and March 2017. In the middle of the valley, some land deformation was also detected, which was caused by the construction of the park. Based on the ground truth survey that was observed using UAV, there is a clear distinction of regions that have moved with their surroundings in zone A and zone B. The comparing result of measurement with the QPS technique and geodetic GPS is presented in Table 3.

IV. CONCLUSION
A combination of ascending and descending orbit and VV+VH polarization with C-band of sentinel-1A has successfully detected the potential landslide in Kelok Sembilan flyover build in the middle 'V' contour shape of the mountain. Furthermore, for modelled the landslide, both orbit and polarization overlaid and interpolated using the IDW technique and then projected into a 12-meter resolution of digital elevation model (DEM). Two of the five zones have significant displacement with more than 500 millimeters and a 160 mm/year velocity from the analysis. One of five areas has collapsed after a ten-days after the last ground truth observation.
Other than that, this research shows the effect of shifting radar signal in V contour shape of the area to the number of persistent scatterers (PS) point distribution and density in both positive and negative slopes. The distribution of PS and its density observed by co-polarization VV is higher than crosspolarization VH but less sensitive to the land movement for ascending and descending orbits. Therefore, in dealing with the V shape effect, besides estimate and compensate using software, satellite datasets such as incidence angle, polarization, bandwidth, and orbit direction should be considered in data selection to obtain more PS points detection. The number of PS points correlated with the level of accuracy and precision of the potential landslide area detected.
Furthermore, the result of QPS processing was validated with ground truth data in situ measurement recorded using Geodetic GPS LEICA 1200 plus instruments, 3D Photogrammetry technique, Clinometer, and measuring tape. Finally, it can be summarized that 1) The combination of SAR data from both orbit and polarization can be used to monitor and model potential landslide in the V shape area with shortwavelength (C-band) that applied in a non-urban area, 2). The V slope affected PS points' distribution and density in VV and VH polarizations with different off-nadir angles and range resolution.