Monitoring and analysis of Woda landslide stability (China) combined with InSAR, GNSS and meteorological data

Detecting the slow motions of high and distant landslides in remote mountain areas has always been a problem. This paper takes the Woda landslide along the Jinsha River as an example to monitor landslide movement. Although some parts of the landslide body have been found to have moved in recent years, the timing and magnitude of motion have not been 10 systematically monitored or interpreted. Here, we apply the SBAS time series strategy using 65-scene Sentinel-1A/B satellite InSAR images and study the spatial distribution and temporal behaviour of landslide movements between July 4, 2018, and August 29, 2020. Our research results show that the cumulative deformation on the left side of the landslide body with concentrated deformation was approximately 200 mm during the 2-year observation period. By calculating the relationship between the InSAR time series and the precipitation around the landslide, it is found that the landslide deformation is closely 15 related to rainfall. GNSS technology is also deployed on the landslide mass and effectively complements InSAR technology. Simultaneously, based on the results of field surveys and hydrological data analysing the landslide's spatial deformation characteristics and deformation factors, the landslide deformation can also be inferred to be related to precipitation. The method used in this paper can be used for early recognition and early warning of high and remote landslides.

accuracy of InSAR time series monitoring. Due to equipment issues, the data quality for sites G2 and G4 is poor, and the data from these two sites are not used in this article. Finally, we collected rainfall data from the Dege weather station 50 km from the landslide body, i.e., daily precipitation data from July 2018 to August 2020, to verify whether the landslide deformation was related to precipitation.

Interferometric processing 115
Localized large movements of and heavy vegetation on the Woda landslide lead to difficulty in phase unwrapping. Therefore, we needed to carefully consider the number of looks and the window size of the filters applied to interferograms. A large number of looks reduces phase noise but at the cost of lower resolution. A larger window size of filters smooths the fringes but may cause loss of detail at fringe boundaries. In our study, we applied 4 looks in the range (i.e., ~m ground range pixel spacing) and 1 look in the azimuth (i.e., ~25 m azimuth pixel spacing). Interferograms were generated using the GAMMA software. To 120 improve the computational efficiency of Sentinel-1 SAR interferometric calculations based on the existing interferometric processing methods, for the key time-consuming steps (registration, resampling and ESD estimation), a graphics processing unit (GPU)-assisted processing algorithm framework was constructed (Yu et al., 2019) to solve the problem of slow data interference processing efficiency.
Original interferograms may experience a mixture of topography-correlated and turbulent atmospheric errors, manifesting as 125 either short-or long-wavelength signals and degrading spatial-temporal filters when extracting deformation signals by InSAR time series analysis (Yu et al., 2018). Typical spatial and temporal changes of 20% in water vapour have been reported to lead to 10-to 14-cm errors in SIR-C/X-SAR-derived displacements, which is large enough to mask actual ground motions caused by a landslide. Therefore, the atmospheric water vapour delay should not be ignored. We also applied a polynomial function to estimate the orbital ramps based on the unwrapped interferograms processed by the GAMMA minimum-cost-flow module. 130 In addition, we applied a linear polynomial function using those pixels with coherence larger than 0.35 and not located in areas with potential motion for each selected interferometric pair.

Time series InSAR analysis
Traditional D-InSAR technology is susceptible to the influences and limitations of the time baseline and the atmosphere during processing, which cause certain errors in the deformation results. The application of time series InSAR technology reduces the 135 effects of temporal decoherence, spatial decoherence and atmospheric phases. Among them, Small Baseline Subset (SBAS)-InSAR (Hooper, 2008) technology can yield better results with less data and is widely used in the field of deformation monitoring (Yang et al., 2014;Zhao et al., 2019;Bayer et al., 2017). The actual processing of SBAS-InSAR data is divided into the following three main steps ( Figure 7): 1) generation of all the interferograms; 2) removal of the residual errors; and 3) SBAS time series analysis. The specific process is as follows: + 1 SAR images arranged in the same area in time sequence 140 0 , … , are obtained; one of them is selected as the main image, and the other SAR images are registered to the main image.
+ 1 SAR images generate M interferograms. Note that each differential interferogram after unwrapping needs to be corrected absolutely with respect to a certain stable region or a reference pixel with known deformation in the figure. Spatial baseline and temporal baseline thresholds are set to construct a small baseline dataset of radar images. Differential interference processing is based on a small baseline dataset and mainly includes removal of the flat ground effect, interferogram filtering 145 and phase unwrapping. For the j-th differential interferogram generated from the SAR image obtained from the image and the main image ( > ), the interference phase of the pixel with the azimuth coordinate x and the range coordinate r can https://doi.org/10.5194/nhess-2021-101 Preprint. Discussion started: 13 April 2021 c Author(s) 2021. CC BY 4.0 License.

Results
The deformation results from Sentinel-1A/B data are obtained by using the algorithm described in Section 2. We investigate how the deformation is distributed spatially and evolves over time. Figure 9 shows that the deformation is mainly concentrated 195 on the right side of the viewing angle and centred on the landslide body. The red line represents the clearly identifiable boundary of the landslide body. The geocoded deformation rate results are superimposed on a Landsat-8 image in Figure 9. A negative value of the deformation rate indicates that the deformation point moves away from the satellite sensor, and a positive value indicates that the deformation point moves towards the satellite sensor. The results of the InSAR view-direction deformation analysis show that the deformation values in Figure 9 are all negative, indicating that the landslide body slides downwards 200 towards the Jinsha River as a whole, which conforms to the sliding law of a landslide body. The figure illustrates that strong deformation of the landslide is located in the middle part and the front edge of the landslide. We selected three representative points (P1-P3) for the locations of the largest local deformation rates in Figure 9 and analysed the deformation characteristics of these points over time in Figure 10. All the deformation characteristics are coincident, and the maximum cumulative deformation of point 1 (P1) is approximately 200 mm. Figure 10 shows that there is a clear acceleration trend in deformation 205 at each point, which may be caused mainly by precipitation. The particular causes of the deformation are discussed in Section 5 for specific analysis.

Comparison of InSAR and GNSS observations
Any kind of surface deformation can be regarded as composed of three directional components: east, north, and up (E, N, and 210 U). Assuming that the east-west, vertical, and north-south deformations measured by GNSS are is the incident angle of the radar side view observation, and is the angle between the flight direction of the radar satellite and the direction of true north. According to equation (7), the three-dimensional deformation measurements at the three GNSS