Earthquake-induced landslides monitoring and survey by means of 1 InSAR 2

. This study uses interferometric SAR techniques to identify and track earthquake-induced landslides as well as lands 11 prone to landslides, by detecting deformations in areas struck by earthquakes. The pilot study area investigates the Mila region 12 in Algeria, which suffered significant landslides and structural damages (earthquake: Mw5, 2020-08-07). DInSAR analysis 13 shows normal interferograms with small fringes. The Coherence Change Detection (CCD) and DInSAR analysis were able to 14 identify many landslides and ground deformations confirmed also by Sentinel-2 optical images and field inspection. The most 15 important displacement (2.5 m), located in Kherba neighborhood, caused severe damages to dwellings. It is worth notice that 16 CCD and DInSAR are very useful since they were also able to identify ground cracks surrounding a large zone (3.94 Km 2 17 area) in Grarem City whereas the Sentinel-2 optical images could not detect them. Although, displacement time-series analysis 18 of 224 interferograms (04-2015 to 09-2020) performed using LiCSBAS did not detect any pre-event geotechnical precursors, the post-event analysis shows a 110 mm/y subsidence velocity in the back-hillside of Kherba.


21
Although it is still challenging to predict exactly where and when natural hazards (earthquakes, landslides, floods, etc.) might 22 occur, the capacity to monitor and survey the zones prone to important landslides as well as the capacity to identify and locate 23 those impacted by earthquakes are key issues in risks mitigation, reduction, preparedness and adaptation. Actually, since 24 earthquakes and landslides might occur in many places worldwide, they might cause a huge number of victims, important 25 socio-economic, assets damages and losses. Their impact can be significantly reduced thanks to satellite imaging which allows

27
It is then worth detecting or predicting critical ground changes at specific places, either after a geotechnical hazard occurs 28 due to landslides and earthquakes mainly, or before it is suddenly triggered (Bakon et al., 2014;Galve et al., 2015). Such 29 challenges can be tackled by regular image processing oriented landslides areas monitoring, in the aftermath of earthquakes, 30 using SAR interferometric methods and optical images, for instance. Actually, since SAR (Synthetic Aperture Radar) is an 31 active sensor system that uses microwave signals to collect data backscattered from the earth's surface, the use of satellite 32 imaging systems like Interferometric SAR methods appears as a cost-effective way for measuring millimeter-level 33 displacements of the earth surface (Herrera et al., 2009), at a regional scale and can be used as an early warning system for the The expected outcomes are based upon the processing of SAR data as it uses Differential InSAR (DInSAR), Coherence 36 Change Detection (CCD), and time series analysis (LiCSBAS software). LiCSBAS exploits the LiCSAR data that process 37 InSAR datasets automatically (Sentinel-1), taking advantage of high-resolution SAR sensing, in order to track ground changes 38 and landslides.

39
The SAR analyses aim to detect ground deformations through DInSAR and CCD investigations as they consider, for

55
-Use time-series analysis to investigate the displacements and their velocities before and after the occurrence of the 56 main shock. For the city of Mila, the time series is performed out for a period extending from April 2015 up to October 57 2020, i.e. a long period before (5 entire years) the main shock in order to avoid a disturbance or bias that might be 58 related to seasonal effects such as rains and vegetation effects (Lazeckỳ et al., 2020a), and a short period (4 months) 59 ahead of the event date in order to investigate the historical development of the landslide.

60
-Compare and correlate the InSAR images processing results with the satellite optical images observations.

76
In the present, two areas are studied, i.e. Kherba and Grarem Cities. The altitude at the top point 1 (Fig. 2.a) in Kherba hill 77 is 654 m and 411 m for the upper point (2), located at 2.14 km distance with 11.34% slope. The maximum ground horizontal 78 offset reached 2.5 m and the vertical deformations exceed 1.8 m ( Fig. 3.b) at the top of Kherba hill (point A Fig. 2.a). The 79 slope failure boundary of Kherba City is mapped as shown in Fig. 2   with the open-source software SNAP (Sentinel Applications Platform). It is worth using data from many orbits to monitor the 96 AoIs due to different oriented directions, incidence angles of satellites, and the ground topography. The optical images of 97 Sentinel-2 satellites are obtained from ESA, whereas downloading and processing data is done via QGIS, Semi-Automatic

99
For Mila region, the AoI is covered by 3 orbits, two are ascending (66, 59) and one is descending (161) (Fig. 1). Since the 100 present study intends to detect the areas influenced by landslides, many pre-event and post-event data were used. Eighteen

101
Sentinel-1 A and 17 Sentinel-1 B images (a total of 35) were downloaded to monitor Mila's area for the period from 1 July 102 2020 to 26 October 2020. Table 1  The temporal baselines for all InSAR pairs are 6 days, except the last three pairs of the ascending orbit 66 that have 12 days.

108
Furthermore, since a bad coherence map of the IFG-24 (Orbit 161), may lead to misinterpretation of results, prior acquisition 109 data (before the 3 rd of August) is selected to generate the co-event interferogram (IFG-22). Therefore the temporal baseline is 110 12 days. The gray rows in Table 1 represent the co-event interferograms of the three orbits. The perpendicular baselines 111 guarantee also a good quality of InSAR studies (Braun, 2019

117
Four aspects are investigated and compared in the present case study:

118
-The SAR Interferometric (InSAR) methodology, which is subdivided into three sub-groups:

120
-CDD for the coherence change detection,

121
-Time series analysis and LiCSAR data.

122
-The optical image processing.

123
Every image contains the description of its source, i.e. IFG-ID (Tables 1, 2, and 2bis) or the image's acquisitions dates.
Where is the line-of-sight (LOS) displacement, the flat earth phase, the topographic phase, is an 132 atmospheric phase, , the baseline phase and is noise phase contribution (Kim, 2013).

133
The main steps of processing data using SNAP software (DInSAR and CCD) are depicted in Fig. 4. It's worth notice that 134 for CCD processing, it is not necessary to follow the whole (DInSAR, Phase Unwrapping, and Phase to displacement). `6

147
Where: 1 , 2 , are the complex signal values of the SAR image pair, N is the window of neighboring pixels, * is the complex 148 conjugate.

149
The coherence values range between 0 and 1 so that the map is represented as a gray color which 0 is white and 1 is black.

157
In addition, the mechanism of landslides can be thoroughly studied through LiCSBAS analyses. They rely on the InSAR   `7 The optical data collected from the ESA platform (Sentinel-2) is treated and plotted using QGIS software to generate true color 166 images (bands 2, 3, and 4 corresponding to RGB). The present study skips the image of 3 rd , Aug 2020 due to bad weather 167 conditions, so that only the two images collected and mentioned in the

175
The adopted methods are applied for Mila case study to:

176
-detect and measure the co-event surface displacements and landslides, caused by the earthquake (CCD and DInSAR)

177
-monitor their dynamic evolution in the first weeks and months, at the post-event period (CCD and LiCSAR data).

178
-analyze their possible initiation ahead of the earthquake by months and years, at the pre-event period (Time-series 179 methods and LiCSAR data).

182
The quality of the SAR image is consistent with the topography slopes and area roughness. Actually, the AoI has rough 183 topography, hills, and rivers (Fig. 2). Selecting either ascending or descending passes, relying on which will avoid some 184 limitation of InSAR is an extremely essential action to infer the deformation from various angles. Therefore, considering the 185 regional topography and geology of the AoI is necessary to process InSAR and results interpreting.

190
Processing DInSAR analysis may then lead to misinterpretation due to atmospheric contribution in differential phase 191 interferograms (Fig. 5). In the study case, no regional deformation due to the earthquake is observed and there is no need to 192 continue investigating the dam and the two bridges by simple DInSAR. However, to monitor the dam and bridges, it is highly      to mask out the rest of the non-changed area using a pre-event to co-event image ratio and filter values equal to or less than 1 239 (see Fig. 13).

256
To quantify the change, an RoI represented in Fig. 13 (green rectangle) is selected for analysis. The plots in Figs. 14 and 257 15 show also the frequency distributions of coherence values within the RoI. Table 3

264
The last orbit 161 pairs make an exception due to the initial bad coherence maps (IFG-23 and 24), as shown in Fig. 15 269 The lines in Figure 14 indicate the frequency distributions of coherence time series maps. The green line in Fig.14

283
The surface area derived from the coherence images covers 2.1 km 2 , and the shape ends by two toes. The runout distance       Fig. 17

354
In this paper, active and passive space-based satellite data are used to monitor and study the impact of natural hazards 355 (earthquakes and landslides) on struck areas. The C-band Sentinel-1 SAR datasets (active sensing) and optical images of 356 Sentinel-2 data (10 m spatial resolution) were used in this study to investigate the area, the passive images were used only to 357 validate the active sensing results. For the InSAR processing, the use of DInSAR, CCD methods, and the LiCSBAS tools have 358 been able to generate a detailed time series analysis of ground changes.

359
InSAR techniques have proved their efficiency to extract useful geodetic information, such as the ground movement and track 360 surface deformation over large areas with centimetric accuracy in coherent change cases. The present research study has `15 demonstrated that the InSAR processing is adapted to study earthquake and landslides zones. As a result, three primary land 362 failures were detected over the study area using InSAR.

363
DInSAR is poorly suited to track and detect landslides. It is represented as a pixel decorrelation in phase interferograms and 364 high decay in coherence values. CCD is further suitable to map earthquake-induced landslides that may remain undetected 365 using coherent methods (DInSAR). The estimation of their horizontal/vertical displacements is a challenge to be inferred.

366
The Grarem deformation looks like a landslide that has just been initiated but might extend under an upcoming triggering 367 event. Actually, the failure plane rim is presented as a dark line in the coherence map or as the fringe circumference in phase 368 maps (estimated area 3.94 sq. km). This impending land failure needs therefore a thorough and real-time monitoring by the 369 PS-InSAR method, which can provide efficient and low-cost monitoring method able to obtain millimeter-level precision 370 displacement measurements over selected points in the area (Jia et al., 2019), and adequate geotechnical studies.

371
It is worth increasing awareness of possible future geotechnical threats in a timely manner, through on-site monitoring using 372 GPS, crack meters, and by placing inclinometers in the Grarem area, in order to develop a model of the slope stability.